ComfyUI Workflows
Browse real AI workflows from Flux, SDXL, Pony, ControlNet, video generation and commercial AI creators.
Lonecat's Simple Workflows
I've been told numerous times that a lot of the workflows that I put out there are complex and intimidating. I wanted to put something out there that was easy to use and understand, but without sacrificing the quality and level of organization that has become my trademark.These are simple versions of my workflows broken down into easy to understand sections with plenty of notes and limited options for the person new to Comfy.A complete user guide can be found here: lonecats-simple-workflow-guide也有中文说明FeaturesControlnet (except Anima)I2I (except Anima)Florence prompt assistUltimate Upscaler for prescaling and hi res fixDetailing Suite for face, hands, & eyesFull tutorial available.也有中文说明Note: I spend a lot of time on these to make sure they are top notch. If you have ANY issues, I'm very responsive. hit me up and I'll resolve them for you.Instagram: https://www.instagram.com/synth.studio.models/Buy me a☕ https://ko-fi.com/lonecatoneThis represents many of hours of work. If you enjoy it, please 👍like, 💬 comment , and feel free to ⚡tip 😉
iGEN ONE (ZImage, Ernie, FLUX1/2, Qwen, HiDream, Nunchaku)
iGEN ONE — Workflow Guide579+ nodes · 40 groups · 13 component subgraphs · 2 pipeline rows161 unique node types — 71% Eclipse nodesBuilt with ComfyUI_Eclipse custom nodesWhat Is This?iGEN ONE is a modular, all-in-one image generation and post-processing pipeline for ComfyUI. It supports a wide range of diffusion models — Flux, Stable Diffusion, HiDream, and more — and covers everything from initial image generation through face detailing, upscaling, and watermarking in a single workflow.The key design principle is modularity: every feature lives in its own group that can be independently enabled or disabled by simply muting or bypassing it. You never need to reconnect anything — the pipeline automatically adapts to whatever groups are active.How It Works — The BasicsLayoutThe workflow is arranged in two horizontal rows that you read left to right:Row 1 (27 groups) — Everything needed to generate an image: image inputs, prompts, model loading, and renderingRow 2 (13 groups) — Everything that happens after generation: refining, face swap, detailing, upscaling, watermarks, and savingBetween the two rows sits a routing banner — a strip of utility nodes that resolves shared resources (model, VAE, CLIP, prompts, image dimensions) so that every Row 2 group can find what it needs automatically.Toggling Features On and OffEach group has a Fast Mode Switcher panel — a small control panel that lets you mute or bypass individual sub-features within the group. Think of it like a row of toggle switches for that section's optional capabilities. To disable an entire group, you mute/bypass the group itself in the ComfyUI canvas.To mute/activate an entire group: right-click the group header → "Set Group Nodes to Never" (mute all) or "Set Group Nodes to Always" (activate all).When a group is muted/bypassed, downstream groups automatically skip it and pick up from the last active group. This works because of a priority-based fallback system: each group tries a list of possible input sources in order and uses the first one that's actually active.You can enable any combination of groups and the pipeline will always find the right data path. There is no need to manually reconnect anything.Data RoutingInstead of visible noodle connections between groups, iGEN ONE uses Set/Get nodes — named value channels that work like wireless connections. A SetNode in one group publishes a value (like "ref_image" or "model_init"), and a GetNode in another group retrieves it by name. This keeps the visual layout clean and makes it easy to rearrange groups.Row 1 — Generation PipelineImage Sources (Groups 1–3 + 5–9)The workflow offers four ways to get a starting image. Only one should be active at a time — the pipeline automatically picks whichever source is enabled. A loaded image can serve two purposes: as a visual reference for img2img generation, or simply as input for the Image to Prompt group (group 8) to generate a text description — you don't have to use it for img2img.You can also load an image and skip the Initial Render entirely — disable the Initial Render switch, and the loaded image goes straight to Row 2 for detailing, upscaling, face swap, or any other post-processing. This lets you bring in images from anywhere (other workflows, other tools, photographs) and run them through the full post-processing pipeline.1. Image LoadLoad a single image from disk. This is the simplest option — pick an image and go. It also extracts any embedded generation metadata (model name, prompt, sampler, seed) from the image, and can optionally override the workflow's settings with those extracted values. This is useful for "remix" workflows where you want to re-generate with the same settings that produced the original.2. Image Load from FolderBatch processing mode. Loads images one by one from a folder, with controls for sorting (by name or date) and subfolder traversal. Like Image Load, it can extract and apply metadata from each image. Great for re-processing an entire folder of images through the pipeline.Set the index to -4 for shuffle mode (random order, no repeats). The optional seed_input slot controls when special modes advance — connect a seed and keep it the same value to freeze the image selection while you tweak other settings. Change the seed value to advance to the next image.3. Input Video FrameExtracts a single frame from a video file, with a configurable frame skip offset. Useful when you want to use a video still as your starting image.4. Text-Only GenerationWhen none of the above are active, the workflow generates purely from text prompts using an empty latent. This is the default "txt2img" mode. Even when an image source is active, it only becomes img2img if you also enable one of the i2i sub-features in the Initial Render group (i2i Denoise, Flux Preproc, DiffSynth Qwen, etc.) — otherwise the loaded image is only used for reference purposes like Image to Prompt.After selecting a source, the image can pass through several optional processing steps:5. Remove BackgroundRemoves the background using BiRefNet, isolating the subject on transparency. Useful when the background would interfere with generation or when you want to focus on the subject only.6. Image Crop — AutoAutomatic subject-aware cropping. Uses SegmentAnything (SAM) to detect the main subject, centers the crop on it, and resizes to your target dimensions. Best for single-subject images where you want tight framing.7. Image Crop — CustomManual bounding-box cropping with pixel-level controls. For when auto-crop doesn't frame things the way you want.8. Preview Cropped ImageA preview checkpoint with a Stop node. Enable this to see the crop result and halt execution before proceeding — useful for verifying your crop settings.9. Resize ImageSimple resize to specific dimensions. Used when your input image doesn't match the target generation size.The image source chain has a built-in priority system: it checks from the last processing step backward (resize → crop_preview → crop_custom → crop_auto → rembg → video → folder → load) and uses the first active result. So you can stack processing steps and the last one wins.Prompt Construction (Groups 8, 18–23)There are multiple ways to build your prompt, and they can be combined:8. Image to PromptUses an AI vision model (Qwen 9B, Q4_K_M quantization) to analyze your reference image and generate a text description. Runs via Eclipse's Smart LM Loader with a "Detailed Description" task. The result feeds into the prompt assembly group as one of the possible prompt sources.18. RaffleRandom prompt generation from a curated tag system. Raffle builds prompts by randomly selecting tags from categories (subject, pose, clothing, etc.) with seed-controlled reproducibility. Includes a negative output filter for excluding unwanted content.19. Read Prompt from FilesReads prompts from external text files (one prompt per line). Uses index-based indexing to select which prompt to use. Good for working through a prepared list of prompts in sequence.Like Image Load from Folder, set the index to -4 for shuffle mode. Connect a seed to the seed_input slot and keep it fixed to freeze the prompt selection while tweaking other settings — change the seed value to advance to the next prompt.20. PromptThe central prompt assembly hub. This is where all prompt sources come together into the final positive and negative prompts.What's inside:Wildcard Processor — Template-based prompting with __wildcard__ placeholders for varietySmart Prompt v2 (Subject) — A structured subject builder with dropdowns for gender, age, hair, clothing, etc.Smart Prompt v2 (Settings) — Environment builder with dropdowns for location, time of day, weather, etc.Join nodes — Combines all active prompt inputs (from Image-to-Prompt, Raffle, file reader, manual text)String DeDuplicate — Automatically removes duplicate tags or phrases from the combined promptPrefix / Suffix — Optional quality tags (like "masterpiece, 8K") added before or after your promptNegative Prompt — A multiline text field for your negative promptEach prompt source has its own Mode Bridge toggle, so you can enable any combination: just the manual prompt, manual + raffle, image-to-prompt + files, or any other mix.The Prompt group has three sub-feature toggles that control other groups: enabling Image to Prompt activates the Image to Prompt group (you still need to manually enable an image source group like Image Load — follow the arrow from Image to Prompt back to find it), Raffle activates the Raffle group, and Read from Files activates the Read Prompt from Files group.21. Prompt StylerWraps your positive prompt in a style template. Uses Eclipse's Prompt Styler node to apply a consistent style (like "photo-hdr") to the prompt text.22. Prompt EditAI-powered prompt rewriting. Uses the same Qwen 9B model but with a "Rewrite Style" task — it takes your prompt and creatively rewrites it while preserving the core meaning. Good for generating variations or improving prompt quality.23. Save PromptsSaves the final combined prompt to a text file. Can append to an existing file, letting you build a collection of prompts over time.Model Loading & Enhancement (Groups 9–17)9. Folder / SizeThe configuration hub for the workflow. Sets:Output folder structure (with date-based subfolders)Image dimensions — default is 896×1152 (3:4 aspect ratio, good for portrait)Batch size — how many images to generate per runLatent type — SD3/Flux/Wan/HunyuanVideoVRAM purge behavior10. Model LoaderLoads the main checkpoint using Eclipse's Smart Model Loader. The default configuration loads Flux Kreamania fp16 in UNet mode with fp8_e4m3fn weight quantization and flash-attention2. Uses external CLIP models (ViT-L-14 + t5xxl_fp8) and an external VAE (flux_vae). The Smart Model Loader handles all the complexity of model configuration in one node. The usual main model is darkBeast Blitz8.The Smart Model Loader has a built-in template system — you can save your entire loader configuration (model, CLIP, VAE, sampler settings, etc.) as a named template and restore it later with one click. The workflow ships with pre-built templates, but those reference specific checkpoints you may not have. You can either download the matching model or load a shipped template and swap in your own checkpoint. Creating your own templates for your favorite models is the recommended approach.11. LoRAsDual LoRA Stack setup with two Lora Stack nodes feeding into a Lora Stack Apply. Supports both model-only and model+clip modes with up to 9 combined LoRA slots. Each slot has its own model path and strength controls.You can mix model-only LoRAs and LoRAs that also modify CLIP in the same stack — each slot independently selects its mode.12. Model PatcherA collection of 11 optional model modifications, each independently toggleable via a Fast Muter panel:ModelSamplingFlux — Flux-specific guidance parametersModelSamplingAuraFlow — AuraFlow sampling overrideDynamicThresholdingFull — CFG thresholding for better prompt adherencePerturbedAttentionGuidance (PAG) — Self-attention manipulation for more detailSelfAttentionGuidance (SAG) — Feature map attention enhancementDifferentialDiffusion — Mask-based selective denoisingCFGZeroStar — Alternative CFG guidance techniquePatchSageAttention — Memory-efficient attention (reduces VRAM usage)TorchCompileModel — JIT compilation for faster inferenceTeaCache — Token caching for speed improvementUNetTemporalAttentionMultiply — Temporal attention modificationMost of these are bypassed by default. Enable them one at a time to see their effect on your output.13. PuLID — FluxIdentity preservation using PuLID. Load a reference face photo and PuLID will guide the generation to maintain that person's facial features in the output. Uses pulid_flux_v0.9.0 with a configurable strength weight.14. PuLID — Flux NunchakuSame concept as PuLID Flux, but optimized for Nunchaku-quantized Flux models. Uses pulid_flux_v0.9.1 + EVA02_CLIP.15. Flux ReduxStyle transfer using Flux Redux. Load one or two style reference images and the workflow applies their visual style to your generation via CLIP Vision encoding and StyleModelApply. Supports blending two references at different strengths.16. PreprocessorImage preprocessing for ControlNet. Uses DepthAnything for depth map extraction. Only needed when the ControlNet group is active.This group is also needed when using Flux ControlNet LoRAs (like depth) or the DiffSynth Qwen LoRA — their sub-feature toggles are i2i (Flux Preproc) and i2i (DiffSynth: Qwen Lora) in the Initial Render group. You must activate the Preprocessor group manually when using either of these.17. ControlNetStructural conditioning with four toggleable modes:Standard ControlNet — xinsir union-promax (strength 0.75)Union Type — Select specific control type (depth, canny, etc.)Negative Zero — Zero-out negative conditioningDiffSynth Qwen/ZIT ControlNet — Alternative ControlNet using Z-Image-Turbo model (strength 0.65)Each mode has its own Mode Bridge toggle. You can use standard ControlNet for structure while also enabling negative zero-out, for example.Rendering (Groups 24–26)24. Initial RenderThe core generation step. Contains a component subgraph (42 internal nodes) that handles the actual sampling process.Sub-features controlled by individual toggles:Initial Render — The main txt2img or img2img sampling passSeed Enhancer — Adds noise variation to the seedNoise Injection — Additional noise patterns injected into the latent (strength 0.45)Detail Daemon — Micro-detail enhancement during samplingFlux Guidance — CFG control specifically for Flux modelsi2i (Denoise) — Standard img2img with configurable denoise strengthi2i (Flux Preproc) — Flux ControlNet LoRA pathway (e.g. depth) — requires the Preprocessor group to be activated manuallyi2i (DiffSynth: Qwen/ZIT) — Qwen-based img2img pathwayi2i (DiffSynth: Qwen Lora) — Qwen LoRA variant pathway — requires the Preprocessor group to be activated manuallyNegative Prompt — Enable/disable negative conditioningStop — Halt execution after this renderDefault sampler: euler / simple / 25 steps / cfg 3.5 / denoise 1.0 — configured via Smart Sampler Settings v2.25. Latent UpscaleSecond-pass latent-space upscaling. Takes the initial render's latent output, upscales it 1.25× with bicubic interpolation, and runs a second sampling pass using a ClownShark Sampler component (7 internal nodes from the RES4LYF pack). This is a more advanced sampler with detail boost, SDE, and sigma scaling options.Default sampler: dpmpp_2m / sgm_uniform / 36 steps / denoise 0.5. ClownShark sub-sampler: multistep/dpmpp_2m / beta / 11 steps / denoise 0.23.26. Initial Render — PreviewPreview and save checkpoint. Shows the generated image and optionally saves it with full metadata embedding (workflow JSON + generation data). Includes a Stop node so you can halt here before entering the post-processing pipeline in Row 2.This is the boundary between generation and post-processing. If you just want to generate and save without any post-processing, enable the Stop node here.Row 2 — Post-Processing PipelineThe second row handles everything after initial generation. A routing banner of ~39 ungrouped nodes sits above the row, resolving shared resources: latent dimensions, reference image, MODEL (with a 6-source priority chain), VAE, CLIP, conditioning, and string prompts.Each Row 2 group automatically picks up the image from whichever previous group was last active, so you can enable any combination and the pipeline chains them correctly.Refining1. Flux2/ZIT Refiner — 3rd PassA third-pass refinement using a dedicated checkpoint (darkBeast Klein2) via a component subgraph (17 internal nodes). Uses SamplerCustomAdvanced with wavelet color matching (strength 0.75) to preserve the original color palette while refining details. Low denoise (0.3) for subtle improvement without major changes.Face Swap2. Flux2: Face SwapDiffusion-based face replacement using a two-pass BFS architecture built entirely with Eclipse and core ComfyUI nodes — no third-party face swap package needed. Contains two component subgraphs — BFS_1ST (26 internal nodes) and BFS_2ND (22 internal nodes) — for progressive face re-generation.How it works:Smart Detection finds the face in the image using the Anzhc face segmentation modelThe face region is cropped and encoded to latentBFS_1ST re-generates the face region using a dedicated checkpoint (darkBeast Klein2) via SamplerCustomAdvanced with full denoise (1.0)BFS_2ND refines the result with a second sampling pass for seamless blendingAn Image Comparer shows before/after for quality checkingBecause it uses actual diffusion sampling rather than a face-swap model, the results respect the art style and lighting of the original image. The BFS subgraphs need a Flux2 model trained for face re-generation — darkBeast Klein2 works well, but any BFS-capable Flux2 checkpoint should work. There are also LoRAs that add BFS capability to a standard Flux2 model. If your main pipeline uses a different model, the BFS checkpoint must be loaded in the BFS model loader.Upscaling (Groups 3, 9, 10)3. Upscale ImageFirst upscale stage with three independently toggleable methods:Scale to Total Pixels — Resize to 2 megapixels using lanczos interpolationSmart Sharpen+ — 4-pass adaptive sharpening (strength 0.75)Upscale with Model — Neural network upscaler (4x AnimeSharp) — muted by defaultAn Image Comparer shows before/after.9. SeedVR2 UpscaleAI-powered upscaling using the SeedVR2 7B DiT diffusion model — a video upscaler repurposed for single images. Loads its own dedicated DiT model and VAE, processes in LAB color space for better color accuracy. Includes optional pre-resize and RAM cleanup controls. Bypassed by default (resource-heavy).10. Rescale ImageFinal size adjustment with three chained operations:Reinhard Color Match (strength 0.3) — Matches colors back to the original referenceBicubic Rescale at 1.25× with supersample enabled (on by default) — supersampling renders at a higher internal resolution then downscales for cleaner results. Can be turned off for a simpler resizeSmart Sharpen (2 passes) — Final sharpening passDetailing (Groups 4–8)All five detailer groups share an identical architecture built around a component subgraph (SEGS Detailer, 41 internal nodes each). Each detailer:Detects a specific region in the image (face, eye, mouth, etc.)Creates a precise mask using SAM2.1 + VITMatte for clean edgesInpaints just that region at a low denoise to enhance detail without changing the restCompares before/after so you can check the resultEach detailer can optionally load its own dedicated model (separate from the main pipeline), has its own LoRA stack, and its own sampler settings — making them fully independent. Sub-features (model loader, LoRAs, flux guidance, negative prompt, differential diffusion, CFG zero star) are individually toggleable via Mode Bridge controls.4. Detailer: FaceEnhances facial details. Uses Florence-2 VLM with "face" detection → SAM2.1 + VITMatte masking. Dedicated model: darkBeast Blitz6. Denoise: 0.2 (subtle refinement — just enough to sharpen features without changing the face).5. Detailer: EyeEnhances eye details. Same architecture, "eye" detection prompt. Denoise: 0.35 (slightly more aggressive than face to bring out iris detail and reflections).6. Detailer: MouthEnhances mouth/teeth details. "Mouth" detection prompt. Denoise: 0.4 (the most aggressive of the face-area detailers — teeth and lips benefit from more rework).7. Detailer: X-1Body region detailer using YOLO object detection instead of Florence-2. Denoise: 0.3.8. Detailer: X-2Specialized region detailer using YOLO detection. Denoise: 0.4.The detailers run in sequence: face → eye → mouth → X-1 → X-2. Each picks up the output of the previous one automatically. Disable any you don't need — the chain adapts.Watermarks & Save (Groups 11–13)11. Create Watermark — TextOverlays a text watermark ("© Eclipse") on the image. Configurable font, size, color, and position (default: bottom-right). Includes gradient effects (cyan→blue) and optional Drop Shadow + Outer Glow from LayerStyle.12. Create Watermark — LogoOverlays a logo image as a watermark. Loads a logo file, positions it bottom-right with blue gradient effects. Optional desaturation, resize, Drop Shadow, and Outer Glow.13. Save ImageThe final output node. Collects the finished image from the entire pipeline using a priority chain that checks all possible sources in reverse order:watermark_logo → watermark_text → rescale → seedvr2 → yolo2 → yolo1 → mouth → eye → face → upscale → bfs → refiner → init → ref_imageThis means it always saves the output from the last active processing stage, regardless of which groups are enabled. The image is saved with full embedded metadata — workflow JSON, generation data (all models, VAEs, and LoRAs collected from across the entire workflow, plus prompts, dimensions), and all relevant settings.This is the only group you should always keep active. Everything else is optional.Quick Start GuideSimplest Setup — Text to ImageMake sure the image input groups are bypassed (Image Load, Image Load from Folder, Input Video Frame)In the Prompt group, type your prompt in the Wildcard Processor text field (the main prompt input — set to fixed mode by default) and your negative prompt in the Negative Prompt fieldIn the Folder / Size group, set your desired image dimensionsMake sure Model Loader is active with your preferred checkpointMake sure Initial Render and Save Image are activeBypass everything else you don't needQueue the promptImage to ImageEnable Image Load and select your source imageEnable the Resize Image group so your image is resized to match the dimensions set in Folder / Size — this avoids issues with oversized images. You can skip this if your image already matches, but large images may cause problemsIn the Initial Render group, enable the i2i (Denoise) toggle and set your denoise strength (0.3–0.7 is typical)Queue the promptPost-Process an Existing Image (Skip Render)You can load any image and send it straight to Row 2 — bypassing the entire generation step:Enable Image Load and select your imageDisable the Initial Render switch to skip rendering entirelyEnable whichever Row 2 groups you want (Refiner, Detailer: Face, Upscale Image, Face Swap, etc.)Queue — the pipeline picks up your loaded image and runs it through the active post-processing chainThis is one of the most useful features of the workflow. You can bring in any image — from a different workflow, a different tool, or even a photograph — and run it through the full detailing, upscaling, and watermarking pipeline without generating anything.Adding Post-ProcessingGenerate your base image firstEnable the Row 2 groups you want (Refiner, Upscale Image, detailers, etc.)Re-queue — the pipeline will process through all active Row 2 groups automaticallyUsing DetailersEnable any detailer groups you want (Detailer: Face, Detailer: Eye, Detailer: Mouth, Detailer: X-1, Detailer: X-2)Each detailer auto-detects its target region — no manual masking neededCheck the Image Comparer in each group to verify the resultAdjust denoise strength if the changes are too subtle or too aggressiveTroubleshootingThe workflow stops halfway throughMany groups have a Stop toggle that halts execution after that group finishes. This is useful for checking intermediate results, but they are enabled by default in some groups. If the workflow stops unexpectedly, check the Stop toggles in these groups:Image source groups (Image Load, Image Load from Folder, Preview Cropped Image)Initial Render and Initial Render — PreviewEach detailer (Face, Eye, Mouth, X-1, X-2)Disable the Stop toggle in any group where you want execution to continue through to the end.If you queue the workflow and it seems to finish too early — before reaching Save Image — a Stop toggle is almost always the reason. Check the last group that produced output and disable its Stop switch.Toggles reset when activating a groupWhen you change a group's state (mute → active or bypass → active), all toggles in that group reset to their defaults — which means all enabled. This can turn on sub-features you didn't expect, including the Stop toggle. After activating a group, always review its toggle panel and disable anything you don't need.This is the most common source of confusion. If something behaves differently after you re-activate a group, check its toggles — they've all been reset to enabled.Custom Node Packages UsedPrimary (author's own):ComfyUI_Eclipse — The backbone of this workflow. Provides loaders, pipes, Set/Get routing, Mode Bridges, Mute/Bypass Repeaters, Smart Prompt, Smart Folder, Smart Detection, Smart LM Loader, Smart Sampler Settings, Save Images, Image Comparer, and many more.RES4LYF — ClownShark Sampler (advanced sampling with detail boost) — fork of ClownsharkBatwing/RES4LYFThird-party:Raffle — Random prompt generation from tag categoriespysssss Custom-Scripts — ShowText for prompt preview displayKJNodes — Image resize, PatchSageAttentionSeedVR2 VideoUpscaler — AI-powered upscalingNunchaku — Quantized model support and PuLID integrationImpact Pack — SEGSPreview for detailer visualizationLayerStyle — Drop shadow, outer glow, SAM2Ultra, MaskGrow, ImageAutoCrop, and moreLayerStyle Advance — Extended LayerStyle nodes (SAM2 Ultra V2, VITMatte)Advanced ControlNet — ACN_AdvancedControlNetApply_v2BiRefNet — Background removalVHS (VideoHelperSuite) — Video frame loadingIf you made it this far — you're a legend. Now go generate something beautiful. 🌒
Max quality Qwen Edit 2511 outputs: minimal workflows + lots of info
IntroAlright this has been a long time coming. I'm the dude who figured out Qwen Edit 2509 a while back, and I've been on-and-off trying to figure out the same for 2511. Results in Comfy have always been worse than the examples shown by the Qwen team, and worse than the official Qwen chat implementation online. Well, I finally cracked it and it only took 5 months lol.Anyway, turns out Qwedit 2511 is fucking sick. IMO it particularly excels at making new shots of characters while maintaining their likeness.As usual, I'll start off with all the setup stuff at the top and then give an explanation + tips & info below that. Also I'm gonna be calling Qwen Edit "Qwedit" most of the time.The posted images are all raw outputs from Qwedit, without being upscaled (despite mentioning it later in this post). They're also all done with only 20 steps instead of the hypothetical 30 I'd do if I wasn't planning to upscale them. Read further for more on that too.The ref images were all made with Z-image base (workflow here), except for the anime one which came from Anima (workflow here).What is thisThese are minimalistic workflows for Qwen Image Edit 2511 that give the highest quality outputs. All other Qwen Edit workflows are now bad, use this one until those get updated with the new info. Aside from generally improving output quality (by a LOT), these changes also enable high-res edits and have better prompt adherence.As for why, basically ComfyUI has some serious issues with how it's implemented Qwen Edit and there aren't any workflows out there (that I've found) which have resolved them. These issues result in poor prompt adherence and low resolution/quality outputs. Thankfully the fix is fairly straightforward.The configuration for this is 100% portable and can be migrated to existing workflows to make them better; it works by changing how the reference inputs are handled, and uses 100% native comfy nodes. Feel free to update other workflows without crediting me.There are also a couple of (optional) related upscale workflows inside; read on for info on why those are here.WorkflowsNormal WorkflowsThese are separated into single / 2 image workflows. It's done this way because the setup for multi-image is complicated and I didn't want to force you to use a ton of custom nodes to make it useable all-in-one.These do still use one custom node for quality-of-life. Minimal quality-of-life that is; I promise there's nothing unnecessary or pointless.Dev WorkflowsLinking these separately to avoid cluttering the post attachment.These are the same as the normal workflows but without any quality-of-life nodes or 'helpful' stuff. Grab these if you want to copy the logic over to other workflows, or if you just an easier view of how it works without any clutter.I do not recommend using the dev workflows for actual gens because you will constantly forget to manually adjust stuff correctly.Dev Single ImageDev 2 ImageModelsMain Modelqwen_edit_2511_fp8or GGUF versionsImportant: the FP8 version of Qwedit is much higher quality than the Q8 GGUF, always use FP8 if you can. Only use the GGUFs if you need to use quants lower than Q8.FP8 is 22GB, so you'll need a combined ~26GB of RAM + VRAM to run itYou don't need 24GB of VRAM to run it thanks to ComfyUI's blockswapping, but the less VRAM you have the slower it'll runOnly use Q6 & lower quants if you absolutely have to; the quality will noticeably go downGoes in models/diffusion_modelsText EncoderUse only the normal FP8 text encoder with Qwedit; abliterated/GGUF encoders will reduce your output quality.qwen_2.5_vl_7b_fp8Goes in models/text_encodersVAEqwen_image_vaeGoes in models/vaeLorasYou can use them as normal, just load them however you normally would. I left out lora loader nodes to avoid cluttering the workflow.It's worth noting that many Qwen Image loras work with Qwen Edit too, but you'll need to test them individually to be sure.Lightning Loras - BADAll the lightning loras / distils for Qwedit (that I've tested) are shit and make your outputs look bad, so I'm not linking them here. The main issue is the same as with Klein Distilled: it makes people's skin look like plastic.But you can technically use them. Don't do it tho. But you can if you want. But don't.Alternative: if you want to cut your gen time down while testing prompts, just set it to 10 steps instead of 20, then go back to 20 once you're satisfied your prompt is correct. It'll still work fine, the quality just dips.Custom NodesLayerStyle - A set of handy nodes that manipulate images. We're just using this for its image scaling node which allows you to scale by an image's long edge while maintaining divisibility by 16. You can skip this if you want to use a different scaling method, but you'll need to fix the workflow switch for scaling if you do.SeedVR2 (OPTIONAL) - Only get this if you want to use the seedvr upscale workflow that's included.How To UseHow To Use Part 1 - Basic OptionsThere are instructions in the workflow as well, but there's more detail here. Read parts 2 & 3 as well, they're important.It works just like a normal Qwedit workflow, but has a couple of extra options available. This section just tells you what they are and how to use them.Enhance with Double RefThis is a switch that turns on double-ref mode. This feeds your input images in TWICE to the model, and generally produces much higher quality results. Downside? It takes about 50% longer to gen.I recommend leaving this on 100% of the time for single-image prompts, unless you're just messing around and want speed. It is ALWAYS better for single image prompts, and will improve everything from prompt adherence to output clarity.For multi-image prompts, it usually increases adherence but sometimes reduces it. So, if you're doing multi-image stuff I recommend switching this on/off as needed based on how it's going with your prompt.Input ScaleWhen off, your image doesn't get scaled (it still gets cropped to be divisible by 16). When on, the long edge of your image gets scaled to the number you put in the box. For example, if you feed in a 2560x1440 image and set the scale to 1920 it will scale your image to 1920x1080. That will then get cropped to 1920x1072 so it's divisible by 16.Custom Output SizeWhen the switch is off, your output image will be the same size as your input image (after it's been scaled). If you turn this switch on, it will instead output an image with the dimensions you specify.As a general rule, you should try to set your scales to be similar along at least one edge. For example, a 1920x1440 input image and a 1024x1440 input image are both suitable for a 1440x1440 output image. You can be more flexible with this if you know what you're doing.How To Use Part 2 - Multi-image Prompting RequirementThis section is not a prompting guide (that's further below). This is about an actual requirement for prompting multi-image stuff. It is NOT required for single-image prompts.You do multi-image prompts like normal, except you need to write a very basic description of your input images. Qwedit needs you to do this in order to know which image is which. I explain why in detail later.You may find this slightly annoying, but I guarantee you it's dramatically better than using Qwedit the normal way that other workflows do - and it's pretty easy.The format:At the start of your prompt, write an extremely simple description for each of your input images; one sentence for eachStart each sentence with "Picture 1:", "Picture 2:", etcYou must write it this way because Qwedit was trained on this exact formatAfterwards, write your actual prompt as usual; you can refer to your input images as "Picture 1" and so onThe model uses these descriptions to understand which input picture is which, and it works better with SIMPLE descriptions. You only need to help it know which one is which, it doesn't need a full rundown.ExamplesPicture 1: a man wearing a t-shirt. Picture 2: a top hat. Make the man in Picture 1 wear the top hat from Picture 2.Picture 1: a living room. Picture 2: a woman. Put the woman from Picture 2 into the living room in Picture 1.Picture 1: a man wearing a professional suit. Picture 2: a man wearing a superhero outfit. Make the man in Picture 1 wear the outfit from Picture 2.How To Use Part 3 - UpscalingBecause the qwen VAE tends to put a subtle halftone pattern over images (see limitations just below this section), I recommend downscaling and then re-upscaling your images afterwards. A big benefit of being able to work at high res with the edit model is that you rarely lose any detail doing this.This eliminates the halftone pattern if you're using something like seedvr, or at least reduces it if you're using other upscalers. I think seedvr is best for this, but it's very beefy and hard to run on older GPUs.Note: the workflow is set to do 20 steps of inference. It actually gives sharper results at 30 steps, but I don't bother with that because it takes longer and I down-upscale them afterwards anyway. If you aren't planning on down-upscaling them, you might consider doing 30 steps for the extra sharpness.Seedvr2 sometimes gives better output at 0.5x downscale, and other times 0.75, so the workflow is configured to run BOTH for you to pick which one turned out best.Normal upscalers are a bit different; a relatively small downsize to something like 1920p -> 1600p is usually reasonable, before then running the upscaler. Play around with it. The non-seedvr workflow has a longest_edge scale option so you can tweak the number specifically.My preferred regular upscaler is 4x Nomos2 HQ DAT2, but you can use whatever.Examples of upscaling:Here's the pic raw output of the robot-arm girl in a dress from the post: https://ibb.co/B5jhrsL9 (if you zoom in you'll see the qwen halftone pattern, it looks like a grid)Here's the pic after it's been run through seedvr after a 0.75x downscale: https://ibb.co/hJcn2f5tHere's the pic after it's been run through a regular Nomos2 upscale after a downscale to 1600p: https://ibb.co/Kc2YSbVcLimitations of Qwen EditLimitation 1The Qwen VAE will often put a subtle halftone grid pattern over your images. It's noticeable if you zoom in, and more noticeable at higher resolutions. This is a feature of pretty much every Qwen-based model, but it's particularly present with the Edit model.You can easily resolve this by downscaling your image a bit, then re-upscaling it again to your desired resolution. The section above explains this in better detail.It sounds like a big issue, but the downscale-upscale trick solves it easily and it's not always necessary either. The higher quality your input image, the less bad the halftone pattern will be.Limitation 2Qwedit struggles with complex multi-image stuff most of the time (it's just a limitation of the model). This workflow makes it much better, but it's still not great. You'll have to play around with it to know which things work and which things don't.I recommend using a different model if you want to do anything complicated with multi-image.Limitation 3It takes a while to gen stuff if not using the lightning loras. Very similar to the time it takes with Klein 9B base. The double-ref trick increases it by roughly 50%, and multi-image edits take a lot longer.For low res images (typical 1mpx size) it's pretty okay, around 50 seconds on a 5090 with the double-ref option turned on.But then there's high-res stuff. Gen time scales non-linearly as you go higher. Going from 1024x1024 (1 mpx) to 1440x1440 (2 mpx) takes around 2.5x as long. Going from 1 mpx to 3 mpx is around 4x as long. 5 mpx is 9.5x as long. In conclusion, stick to 2-3 mpx unless you're cool with long-ass gen times. Stick around 1-2 mpx for multi-image gens, or turn off the double ref switch.On the plus side, it's pretty reliable for single-image edits so you don't typically need to do many gens to get a good result.Examples using a 5090:Single-image edit @ 1024x1024 (1 mpx), double-ref OFF = 38 secondsSingle-image edit @ 1024x1024 (1 mpx), double-ref ON = 52 secondsSingle-image edit @ 1920x1088 (2 mpx), double-ref OFF = 91 secondsSingle-image edit @ 1920x1088 (2 mpx), double-ref ON = 131 secondsSingle-image edit @ 3072x1728 (5.3 mpx lol), double-ref ON = 550 secondsTwo-image edit @ 2560x1440 each, double-ref ON = serial killer behaviourThat's it for how-to! Read on for more tips & info, as well as an explanation of what the workflow is doing & why.Explanation - what is this garbage and why is it so good?There are three important things this workflow is doing that other workflows do not do. I'm going to call these The Comfy Problem, The VL Problem, and The Double Ref Enhancement.The Comfy ProblemComfy's native "TextEncodeQwenImageEditPlus" node is what most people use in their workflows. It handles your prompt and image inputs for you. It's pretty handy, except for the small problem that it's SHIT.Do you work at Comfy? If so: GET YOUR SHIT TOGETHER AND FIX THIS NODE, IT'S SO EASY. Much respect to u tho, thanks for making ComfyUI.The first issue is that this node resizes your image down to 1 megapixel, and you can't stop it from doing that. The second issue is that it does this with the AREA downscale method, which is so incredibly bad that I want to slap whoever implemented this node. The area downscale is what makes all of your output images blurry. The third issue is that it crops your dimensions to be divisible by 8, but they actually need to be divisible by 16.Specifically, ComfyUI does this:Calculates 1 megapixel as 1024x1024, which is 1,048,576 pixelsCalculates your new image dimensions to match that number of pixels, rounded to be divisible by 8Scales your image to those new dimensions using the AREA methodWhy is all this bad?It's completely unnecessary; Qwedit can easily handle images of varying size, all the way up to 3 megapixels (or even higher for simple edits)The area downscale method makes images extremely blurry, and this is the primary reason all ComfyUI qwen edits give blurry images out. Yes it's literally this dumb, this huge problem would easily be solved by changing the word "area" to "lanczos" in the code, it's a one-word fix.Not even MS paint uses area downscale, wtf is wrong with you Comfy devs (much respect)The Comfy Problem SolutionThis workflow bypasses the the Comfy node entirely, allowing you to size your images however you want. And using chad lanczos scaling instead of loser area scaling. Magic.Qwedit easily handles resolutions like 1440x1440 and 1600x1200. Every edit example in this post was done natively at 1920p, except for a few (which are labelled as such).Really high resolutions (3mpx) sometimes have trouble with anatomy, but usually you can just do multiple gens and one of them will turn out fine.If you're doing a simple in-place edit like changing an outfit, you can go VERY high. Here's an example edit done at 1728x3072, which is 5 megapixels: https://ibb.co/twCSWrjy (outfit change -> bikini top + short shorts)The VL ProblemEdit: I've been educated by someone that my interpretation of how the VL works here is not correct, so take this little VL section with a grain of salt until I reword it. My conclusion about it helping in this workflow still stands, but my explanation of exactly what's happening under the hood is a bit off. I'll update the info soon!In the background, Qwedit 2511 uses a vision-language model (VL model) to describe your images, then gives those AI-generated descriptions to the edit model. It also re-interprets your instructions with these descriptions. Ostensibly this helps the model understand your input images better, leading to better results.The problem? It doesn't lead to better results, it's bad. VL models aren't very good for this sort of thing because they don't know what to focus on. The VL describes your images in excruciating detail, totally overwhelming the edit model and leading to bad prompt adherence + weird outputs.The Qwen team's official python code does this, and the ComfyUI "TextEncodeQwenImageEditPlus" node copies it exactly. No disrespect to the Comfy team on this one, they're doing what the Qwen team officially recommended.The VL Problem SolutionSame solution as the previous problem: bypass the Comfy node entirely. This results in the VL step being completely ignored. No AI-generated descriptions get fed into the edit model.For single-image edits, this is a 100% complete and total victory. The model performs way better without the crappy VL interpretation.For multi-image edits, there's a small issue; this step is where the input images normally get labelled. Specifically, the VL outputs are fed into the model in the following exact format:> Picture 1: <shitty VL description> > Picture 2: <shitty VL description>Look familiar? This is why we manually have to type the descriptions in for multi-image edits - otherwise the model doesn't actually know which image is which.The upside is that the model works way better with simple descriptions, so cutting out the VL is still 100% the correct move. A 5 word description wins over whatever BS the VL model spews out, every time.The Double Ref EnhancementI really have no idea why this works so well, but basically if you feed in your reference images twice the model just works better. This was known back in 2509 days (hence the previous post linked at the top), and back then I didn't know why it worked either.For single image edits it's ALWAYS better. And it's not just the quality, for some reason it even helps with prompt adherence. The interesting thing is that the difference is really, really significant. Here's the full list of stuff it improves:Better prompt adherenceSharper output images / more visual clarityImproved consistency of objects & texturesBetter resemblance of characters at different anglesMore intelligent guesses, like what to add when outpainting or what's behind a removed objectFor multi-image edits it can sometimes confuse the model a bit, but most of the time it confers all the same benefits listed above. I recommend switching it on & off randomly when you're doing multi-image stuff, just in case.Note: there are a lot of different ways the input references can be handled. There are conditioning combine/concatenate nodes, you can pass the refs in a different order, you can change the negative conditioning input (read next section for that), etc. I A/B tested SIXTEEN different reference-handling combinations, and a bunch of smaller minor variations of those. Some of them worked, some of them didn't. Of those sixteen combinations, two of them gave the best results; both of them are in this workflow, and you switch between them by turning the double ref method on & off. So, don't fuck with the positive/negative conditioning & reference setup, it's very specific.Extra Info: The "Conditioning Zero Out"You may notice that the negative prompt input is the first reference image(s) and positive prompt fed into a "conditioning zero out" node.Feeding the input images into the model's negative conditioning is required (it's just how Qwedit works). The only question is whether to feed in the positive prompt zeroed-out too, and whether the double ref should get fed in.Through a lot of A/B testing, I can tell you that the way it's done here is the best. IDK why, it's just how it is. Some other combinations do technically work, but they degrade the output quality.Prompting AdviceOther than just following the instructions in the workflow, here's some extra stuff.Keep your prompts simple and directIf you need to, point out details the model is missing or be more specific about stuff you do/don't want to change. For example, when doing a simple outfit swap it helps to specify you don't want their pose to change.Using the robot arm girl, here's a prompt that doesn't follow this advice:Change her outfit to a bikini top and short shorts.While it sometimes does what we want, it tends to get confused by her robot arm and often changes her pose too: https://ibb.co/7dyKZttp (notice the human arm showing underneath the robot arm, and the pose change)Here's a better prompt that gives a correct result 99% of the time:Change her outfit to a bikini top and short shorts. Leave her robot arm and pose unchanged.Now it does the right thing every time: https://ibb.co/DP9gZHVvAvoid using fancy words or convoluted phrasingPretend you're talking to a child. The model will probably still understand you if you talk fancy, but why take the risk?As an example, imagine you have a pic of a table with some plates on it.Bad:Place a red apple on the table, ensuring it's in the center and removing the plate that was in the same spot.Good:Replace the middle plate with a red apple.Also good:Remove the plate from the center. Put a red apple there instead.If there's only one plate, this is even better:Remove the plate, replace it with a red apple.Adjusting LightingYou may want or need to adjust the lighting in an image. Aside from being helpful in general, there are situations where Qwedit may simply not realise that something needs to be lit in a particular way (or re-lit when moved).To do this, you need to know the magic word: relightThat is the actual magic word, you are 100% required to use it if you want to adjust lighting properly.Specifically, follow this format:Relight to <strength> <color> <direction>.Strength - bright, dim, etcColor - white, cool, warm, etcDirection - diffuse, frontlit, backlit, etcTip: for basic lighting, use "white diffuse".Examples:Make a new shot of the man sitting in a chair in a kitchen. Relight to white diffuse.Change the time of day to evening. Relight to warm backlit.You don't actually need anything else in the prompt, you can just change the lighting of a pic like this:Relight to bright cool frontlit.Other StuffEuler-simple and no ClownsharKSampler?No Clownshark this time. It reduces output quality quite a bit and doesn't confer any benefits. I also didn't find any sampler/scheduler combos that were better than euler/simple.So, this is just one of those classic times where the ol' euler-simple wins the day. Let me know if you happen to know a better combo.Image Quality in->outQwedit is very sensitive to the quality of your input image. If you feed in a grainy or blurry image, it will usually make your output image blurry or grainy too - even if it's an 'entirely new' shot with nothing copied over 1:1.So, make sure to use HQ images. You can optionally use the upscale workflows to bump up the sharpness/quality of poor input images before you feed them in.What about the flux super duper double resolution special VAE trick?Doesn't work for 2511, it destroys your image. TBH it never really worked for 2509 either, but I won't argue with you if you liked it for some reason.Making character referencesTip 1 - Make a nude ref (even for sfw stuff)Qwen is killer for making character references. Other than using similar prompts to the examples I posted, my advice is to make a nude reference shot instead of a clothed one like I did.I only made a clothed ref for the sake of propriety here, but a nude ref (or near-nude, like wearing plain white underwear) will be much easier to prompt into different outfits, and also gives Qwedit the maximum info needed to correctly size your character and know what they look like in clothing or doing different actions.You do not need any loras to do this if you're just using it as a reference; the 'sensitive' parts will lack detail but that doesn't matter for new shots you make. If you don't want them nude, just request plain white underwear and, if relevant, a strapless white bra.Nude ref = best ref.Tip 2 - Make multiple zoom levels, use the thighs-upwards one for most stuffThe example I showed was a little too zoomed out for normal reference stuff. I'd recommend making your reference slightly closer like this: https://ibb.co/Q33BJDLXStart at whatever zoom level your initial character pic is at, then make more references at different zoom levels. If you're starting zoomed out, then prompt the model to zoom in. If you start zoomed in, prompt it to zoom out.And, of course, different angles too.Examples:Zoom in on the person's upper body. The composition should frame their head and thighs.Zoom out to show more of the character. The composition should frame their head and thighs.Zoom out to a full body shot.Zoom in for a close up portrait.Once you've got references, you should usually use the head-to-thighs ref for making new iamges. Switch to the other refs as necessary; like if you want a close up, use the close up reference. Qwedit is really good at keeping likeness, so you can do 90% of your stuff with only a single input reference.I don't think there's a better open-weight model out there than Qwedit for making new shots of character without loras, right now.Enjoy!
Qwen-image TXT2IMG, IMG2IMG, controlnet (canny, depth), lora manager, caption, tiled and SeedVR2 upscale.
A pretty simple workflow with a few common features. There are some important (more or less) notes inside the workflow. It is designed for 'legacy' mode. For now, I still think that Nodes 2.0 are mostly buggy and slower garbage, even though they have a few cool things.I tried Fun-Control ControlNet. For some reason, I always get a black image with any model and any settings, even in the official template.
Nano Banana 2 Alternative
After extensive testing, I think Qwen Image Edit 2511 is finally getting close to becoming a real Google Nano Banana 2 alternative — especially for local and open-source workflows.The biggest issue with Qwen 2511 was always the overly plastic and smooth skin textures. The moment you changed poses or moved a subject, faces would suddenly turn doll-like and lose the original image texture and vibe.So I spent time testing different setups, LoRA combinations, and quantizations.
Brie's Advanced Character Fusion
Hey y'all!Use familiar SDXL prompting to generate a character and pose separately, then place and relight that posed character onto an input background, all while maintaining character consistency and style! All without a character lora!Special thanks to tori29umai (on Twitter) for the charaBG and image2body loras, dx8152 (on Twitter and Hugging Face) for the Fusion lora and @DigitalPastel for SDXL workflows which portions of this one is based on.I made this workflow because my previous workflow lost too much style after RePosing a character. Also my other Fusion workflow's compositing part was waaay too clunky.Video Tutorial will be made eventually. This one needs it.[bideo lonk]Usage:Keyframes for video generationPanels for comicsMaterial for training character lorasMemes! I dunno ~Notes:Read the notesDownload the models, which are all linked in the workflow.Set the Cache Full File Path, otherwise it won't work!Run the workflow ONE GROUP AT A TIME!READ THE NOTESHere's a map of the workflow and what each active part does.Workflow Map:Active Workflow Groups:SDXL Pose - Generates a dynamic character pose from prompts (optional with input pose image)SDXL Character - Generates an idle standing character from prompts (optional with input character image)Qwen Character Sheet - Generates a 4-view character sheet from the SDXL characterQwen Pose - Extracts the pose mannequin from SDXL PoseQwen RePose - Reposes character to the Qwen Pose using the Character SheetSDXL Refine - Refines with SDXL prompts and loras to reinject SDXL style (optional)Compositing - Removes background and places character onto background imageQwen Fusion - Relights and casts shadows for the character in the scene.See example below:Info Sheet:Character Sheet:All of these, except for the background image, are generated in this workflow.All intermediate images, ARE NOT, outputted by default. You can output the intermediate images by checking 'preview_only' to false.You can also fetch the most recent intermediate images from the cache folder.Version 1.2Added option and notes to load the Qwen Rapid All-in-One model, which can do NSFW stuff with the Qwen steps. Mostly works on the Qwen Fusion step.Version 1.1Added character and pose input images and switch. Allows you to skip either or both the SDXL pose and character groups.Made the SDXL Refine group optional. You can run Qwen RePose then Qwen Fusion, skipping SDXL Refine.Version 1.0;Initial Release!Takes prompts and a background image and generates a posed character onto that background image.Future variants & planned workflows:A variant with a pose input images, to aid generating the pose via Control Net.A pose input image, use pose ref images to control poses for your generated character.(ADDED)A character input image, use SDXL to generate poses for an existing character. (ADDED)Versions of all those above, with the compositing group removed. Similar to Lazy RePose.A separate bigger compositing workflow, for multiple character and prop placement.NSFW versions!This workflow is time-consuming to run, but once you've got a character going, you can pose that character in almost any way. The RePose part will fail if the pose is too complex. SDXL refine brings back a lot of the style that is lost in Qwen, but its also not perfect.Regardless, I achieved 90% of what I wanted, and it will serve as a great base for future workflow variants.Anyhow, enjoy the workflow. Please post if you have any questions or suggestions!Stay Cheesy Y'all!~Brie WensleydalePS. Please, if you see something that could be improved, please tell me! Much appreciated!
ComfyUI Image Workflows
Workflow for XL/Illustrious/NoobAI ModelsAny feedback would be appreciated.See the Article for a Guide:https://civitai.com/articles/17339 (probably doesn't work rn)https://civitai.red/articles/17339Key Features:LoRATIPO PromptingDetail EnhancersControlNet UpscaleIPAdapterHiresFixOpenPoseWildcardsControlNETColor MatchFBCNNRegional PromptingInpaintOutpaintWorkflows included:🟥Advanced: comprehensive text to image workflow🟨Basic: more basic version of the Standard workflow🟦Standard: more basic version of the Advanced workflow🟩Detailer: detailing previously generated or existing imagesExperiment and enjoy!
Qwen Image 2512 + SAM3 Face Inpainting Workflow
🎨 Qwen Image 2512 + SAM3 Face Inpainting Workflow> Generate stunning portraits with automatic face detail enhancement — using single prompts or batch scheduling.## ✨ What This Workflow DoesThis ComfyUI workflow combines Qwen Image 2512 (a powerful text-to-image model) with SAM3 Smart Inpainter to:1. 🖼️ Generate high-quality images from text prompts2. 👤 Automatically find and enhance faces (man + woman) using AI segmentation3. 🔁 Support single prompt OR batch prompt scheduling for creative variations4. ⚡ Speed up generation with CacheDiT acceleration5. 🎯 Use advanced sampling for better resultsPerfect for: character portraits, fashion shoots, story illustration, face refinement, and creative batch generation.## ✨ What's New in V1.1### 🔁 Auto-Reset Counter (No More Manual Reset!)- Replaced the old PrimitiveInt node with my new AutoResetInt node- ✅ Automatically resets to 0 after each generation- ✅ Also resets if you cancel a generation- ✅ Want to run multiple batches? Just wait for the first image to finish — the counter resets itself!- 🎯 No more forgetting to reset — just click "Queue" and go### ⚙️ Model Format Switch- Added a simple toggle to choose your model format: - GGUF — for lower VRAM / quantized setups - FP8 / BF16 — for full quality on powerful GPUs- 🎯 Pick what works best for your hardware — no need to edit the workflow## 📦 Required Custom NodesInstall these in your ComfyUI/custom_nodes/ folder:| Node Pack | Purpose | Link ||-----------|---------|------|| RES4LYF | Advanced sampler with better quality control | https://github.com/ClownsharkBatwing/RES4LYF || SAM3_SmartInpainter | Auto-detect & inpaint faces using SAM3 | https://github.com/majidfida/SAM3_SmartInpainter || ComfyUI-CacheDiT | Speed up Qwen Image generation | https://github.com/Jasonzzt/ComfyUI-CacheDiT || comfyui_auto_prompt_schedule | Run multiple prompts in one batch | https://github.com/ReinerBforartists/comfyui_auto_prompt_schedule |### Also Required:- ComfyUI-Easy-Sam3 or ComfyUI-RMBG (provides the sam3 Python package for SAM3_SmartInpainter)- Qwen Image model files: qwen_image_vae.safetensors, qwen_2.5_vl_7b_fp8_scaled.safetensors, and your chosen checkpoint (e.g., Qwen-Overcooked_fp8_V2.safetensors)- SAM3 model: Download sam3.pt from https://huggingface.co/Majidfida/Sam3/tree/main → place in ComfyUI/models/sam3/## 🔧 How to Use### Step 1: Install & Setup1. Clone all required node packs into custom_nodes/2. Install Python requirements (see each repo's README)3. Download models and place them in correct folders4. Restart ComfyUI### Step 2: Load the Workflow1. Drag the .json file into ComfyUI2. Wait for nodes to load (missing nodes will show red)### Step 3: Choose Your Mode#### 🎯 Single Prompt Mode (Default)- Set the toggle true=Single Prompt | false=Batch Prompts → true- Edit the prompt in the ⚡Single Prompt⚡ node- Click Queue Prompt#### 🔄 Batch Prompt Mode- Set the toggle → false- Edit prompts in the Input Prompts for Schedule node (format: "0": "prompt text", "30": "next prompt", ...)- Set max_frames in PromptSchedule node to control how many to generate- Click Queue Prompt — it will auto-run through your list### Step 4: Face Inpainting (Automatic)The workflow automatically:1. Finds the woman's face → enhances it with prompt: "woman, exquisite detailed face..."2. Finds the man's face → enhances it with prompt: "man with dark beard, exquisite detailed face..."3. Blends both back into the main image💡 Tip: To inpaint different subjects, chain more SAM_SmartInpainter nodes and change the prompt field (e.g., "hat", "eyes", "scarf").### Step 5: Adjust Key Settings| Parameter | Recommended | What It Does ||-----------|-------------|--------------|| min_mask_area | 0.001 | Allows tiny faces to be detected || crop_padding | 32–128 | Extra space around face for better inpainting || inpaint_resolution | 1280 | Higher = more detail in repainted face || denoise | 0.3–0.5 | Lower = subtle changes, higher = more creative || steps | 6 | Qwen Image works well with low steps || cfg | 1.0 | Keep low for Qwen Image |---## 💡 Pro Tips🔹 Small or distant faces not detected?→ Lower min_mask_area to 0.0005 and text_threshold to 0.25🔹 Inpainting looks blurry?→ Raise inpaint_resolution to 1280 or 1536🔹 Edges look harsh?→ Use blend_mode: feather and increase mask_blur🔹 Want to inpaint more than faces?→ Change the SAM prompt to "hat", "logo", "hands" — it will find and repaint that area🔹 Batch generation too slow?→ Enable CacheDiT and use fp8 models for faster runs## ⚠️ Important Notes🔸 This workflow is designed for Qwen Image 2512 — other models may need parameter adjustments🔸 I do not monitor this post — I may not reply to comments or issues. This is shared for the community to use and improve.🔸 Always test with one image first before running large batches---## 🙏 Credits- Qwen Image 2512 — Alibaba Cloud / Qwen Team- SAM3 Model — Meta AI Research / Majidfida (HuggingFace)- RES4LYF — @ClownsharkBatwing- SAM3_SmartInpainter — @majidfida- CacheDiT — @Jasonzzt- Auto Prompt Schedule — @ReinerBforartists> 🎨 Simple rule: Start with single prompt → get it perfect → then try batch mode for variations.> If faces aren't found → lower min_mask_area. If results look soft → raise inpaint_resolution.Happy creating! ✨👫🖼️
【QWEN】IMG to IMG
✨ QWEN — Image to Image — Simple WorkflowA clean, all-in-one QWEN image-to-image workflow built entirely with the UmeAiRT Toolkit for ComfyUI.Only 8 nodes. No spaghetti wires. Just load your model, write your prompt, and hit generate.⚠️ IMPORTANT — Nodes 2.0 RequiredThis workflow is built for the Nodes 2.0 (Vue) interface of ComfyUI. If you don't enable it, the workflow may have display problems.How to activate Nodes 2.0:Open ComfyUIGo to Settings (⚙️ icon, bottom-left)Find "Use Nodes V2 (Vue)" and toggle it ONRefresh the pageLoad the workflowIf you prefer the classic interface, check out my Legacy version of this workflow instead (link).🎯 FeaturesText-to-Image generationAutomatic download of models in auto versionLighning LoRA includedDetail Refiner samplerBuilt-in SeedVR2 upscaler — high-quality tiled upscaling (toggleable on/off) Slower than a classic upscaler, but significantly better qualityFull metadata embedding — your images are saved with all generation parameters, ready for online publishing and remixing3 LoRA slots — with individual on/off toggles and strength control and you can connect as many other lora modules to each other for as many LoRA as you want.📦 Custom Node RequiredOnly one custom node to install:👉 ComfyUI-UmeAiRT-ToolkitInstall via ComfyUI Manager (search "UmeAiRT") or use the UmeAiRT Auto-Installer.The Toolkit packages everything internally — upscaler, face detailer, metadata saver. No other custom nodes needed.📂 Files you need (in manual version)📂Files :For base versionModel : fp8, bf16in models\diffusion_modelsCLIP : qwen_2.5_vl_7b_fp8_scaled.safetensorsin models\clipFor GGUF versionModel : Q8, Q5, Q4in models\unetCLIP : Q8, Q6, Q5, Q4, Q3in models\clipCommon filesVAE : qwen_image_vae.safetensorsin models\vaeLoRA : Qwen-Image-Lightning-8steps-V2.0.safetensors or Qwen-Image-Lightning-4steps-V2.0.safetensorsin models\loras
【QWEN】TXT to IMG
✨ QWEN — Text to Image — Simple WorkflowA clean, all-in-one QWEN text-to-image workflow built entirely with the UmeAiRT Toolkit for ComfyUI.Only 8 nodes. No spaghetti wires. Just load your model, write your prompt, and hit generate.⚠️ IMPORTANT — Nodes 2.0 RequiredThis workflow is built for the Nodes 2.0 (Vue) interface of ComfyUI. If you don't enable it, the workflow may have display problems.How to activate Nodes 2.0:Open ComfyUIGo to Settings (⚙️ icon, bottom-left)Find "Use Nodes V2 (Vue)" and toggle it ONRefresh the pageLoad the workflowIf you prefer the classic interface, check out my Legacy version of this workflow instead (link).🎯 FeaturesText-to-Image generationAutomatic download of models in auto versionLighning LoRA includedDetail Refiner samplerBuilt-in SeedVR2 upscaler — high-quality tiled upscaling (toggleable on/off) Slower than a classic upscaler, but significantly better qualityFull metadata embedding — your images are saved with all generation parameters, ready for online publishing and remixing3 LoRA slots — with individual on/off toggles and strength control and you can connect as many other lora modules to each other for as many LoRA as you want.📦 Custom Node RequiredOnly one custom node to install:👉 ComfyUI-UmeAiRT-ToolkitInstall via ComfyUI Manager (search "UmeAiRT") or use the UmeAiRT Auto-Installer.The Toolkit packages everything internally — upscaler, face detailer, metadata saver. No other custom nodes needed.📂 Files you need (in manual version)📂Files :For base versionModel : fp8, bf16in models\diffusion_modelsCLIP : qwen_2.5_vl_7b_fp8_scaled.safetensorsin models\clipFor GGUF versionModel : Q8, Q5, Q4in models\unetCLIP : Q8, Q6, Q5, Q4, Q3in models\clipCommon filesVAE : qwen_image_vae.safetensorsin models\vaeLoRA : Qwen-Image-Lightning-8steps-V2.0.safetensors or Qwen-Image-Lightning-4steps-V2.0.safetensorsin models\loras
Qwen Image Edit 2511 Seamless Head Swap Redraw-and-Paste-Back Master Workflow
This workflow is designed for Qwen Image Edit 2511 seamless head replacement with redraw-and-paste-back control. Compared with a simple head-swap workflow, this version adds a more complete local editing structure: the target region is prepared, redrawn with Qwen Image Edit 2511, then pasted back into the original image area with stronger composition preservation. The goal is to replace the head identity naturally while keeping the original body, clothing, background, lighting, camera angle, and image layout stable.The workflow uses Qwen Image Edit 2511 as the main editing route, with qwen_image_vae, qwen_2.5_vl_7b_fp8_scaled, and Qwen 2511-related model / LoRA components. It is structured for controlled image editing rather than full-image regeneration. This is important because head replacement can easily break the original body proportion, neck connection, shoulder alignment, light direction, skin tone, and facial scale. This workflow is designed to reduce those problems through masking, reference preparation, and paste-back reconstruction.The key idea is “redraw first, then restore.” The workflow prepares the main image and reference image, uses background / subject extraction and mask processing to isolate the editable region, then expands and softens the mask with MaskGrow-style controls. After that, Qwen Edit configuration nodes prepare the image references, visual-language inputs, VAE images, masks, and prompt structure. The edited result is decoded, cropped with pad information, scaled back when needed, and restored into the original image through a paste-back / restore-crop-box route.This gives the workflow a more practical production behavior. Instead of allowing the model to reinterpret the entire image, the edit is concentrated on the face / head area. The original image remains the structural anchor, while the reference image provides the replacement identity. This makes the final result more suitable for portrait correction, character testing, fashion model previews, authorized identity replacement, cosplay transformation, AI influencer image editing, and Civitai / RunningHub showcase examples.The workflow also includes preview and comparison tools, such as mask preview, image comparer, image reel display, and before-after outputs. These are useful for checking whether the mask is correct, whether the reference identity is preserved, and whether the paste-back result blends naturally with the original image. For head replacement, this inspection step is important because the smallest mismatch in jawline, neck shadow, skin texture, or light direction can make the result look artificial.In short, this is a Qwen 2511 head-swap master workflow with a stronger redraw-and-paste-back pipeline. It is built for users who need more control than a basic face swap and want cleaner integration with the original image. If you want to see how the mask, reference image, Qwen Edit nodes, redraw stage, and paste-back reconstruction connect together, watch the full tutorial from the YouTube link above.⚙️ Try the Workflow Online👉 Workflow: https://www.runninghub.ai/post/2027381547103424514?inviteCode=rh-v1111Open the link above to run the workflow directly online and view the generation results in real time.If the results meet your expectations, you can also deploy it locally for further customization.🎁 Fan Benefits: Register now to get 1000 points, plus 100 daily login points — enjoy 4090-level performance and 48 GB of powerful compute!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you are in Mainland China or the Asia-Pacific region, you can watch the video below for workflow demos and a detailed creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1DpADznEo2/I will continue updating model resources on Quark Drive:👉 https://pan.quark.cn/s/20c6f6f8d87bThese resources are mainly prepared for local users, making creation and learning more convenient.⚙️ 在线体验工作流👉 工作流： https://www.runninghub.ai/post/2027381547103424514?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1DpADznEo2/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
Qwen Image Edit 2511 Seamless Head Swap Master Workflow
This workflow is designed for Qwen Image Edit 2511 seamless head replacement, focused on controlled, authorized portrait editing with strong identity transfer and natural blending. Its main purpose is to use one image as the body / scene reference and another image as the head / face identity reference, then generate a clean final result where the replacement looks integrated rather than pasted.The workflow is built around Qwen Image Edit 2511, using qwen_image_edit_2511_fp8mixed / qwen_image_edit_2511 model routing, qwen_2.5_vl_7b_fp8_scaled as the vision-language text encoder, and qwen_image_vae as the VAE. It also includes a dedicated head-swap LoRA route, such as bfs_head_v5_2511_merged_version_rank_16_fp16, to strengthen the model’s ability to perform realistic and stable head replacement. This makes the workflow more specialized than a normal image-editing graph.The core idea is not to regenerate the whole image. Instead, the workflow uses inpainting and reference control to focus the edit on the head area while preserving the body, clothing, pose, background, lighting, camera angle, and overall composition from the base image. This is important because many face-swap or head-swap attempts fail by changing the entire person, destroying the original body structure, or creating mismatched lighting and skin tone. This workflow is designed to reduce that drift.The workflow includes InpaintCropImproved and InpaintStitchImproved, which help crop the target region, process the edit, and stitch the result back into the original image. This gives the workflow a more professional local-editing structure. It also uses ReferenceLatent, FluxKontextMultiReferenceLatentMethod, CFGNorm, ModelSamplingAuraFlow, SamplerCustomAdvanced, and VAEDecode to improve reference stability and output quality.A key part of this workflow is mask and foreground preparation. BiRefNetRMBG is included for background removal and subject isolation, helping the system identify the usable human region more clearly. The workflow also uses ImageResizeKJv2, image concatenation, preview nodes, and comparison output, making it easier to check the body image, head reference, generated result, and before-after differences.The prompt logic is written specifically for seamless replacement. It asks the model to preserve the body image environment, framing, lighting, camera perspective, exposure, and composition while replacing only the head identity. It also emphasizes matching head size, face-to-body ratio, neck thickness, shoulder alignment, gaze direction, expression, skin texture, jawline blending, shadow contact, and sharpness consistency. These details are critical for making the final result look natural.This workflow is suitable for authorized portrait replacement, character concept testing, fashion model preview, AI influencer image editing, cosplay-style transformation, commercial mockups, and Civitai / RunningHub showcase examples. If you want to understand how the reference images, inpaint crop, LoRA route, and final stitch-back structure work together, watch the full tutorial from the YouTube link above.⚙️ Try the Workflow Online👉 Workflow: https://www.runninghub.ai/post/2027327275267530754?inviteCode=rh-v1111Open the link above to run the workflow directly online and view the generation results in real time.If the results meet your expectations, you can also deploy it locally for further customization.🎁 Fan Benefits: Register now to get 1000 points, plus 100 daily login points — enjoy 4090-level performance and 48 GB of powerful compute!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you are in Mainland China or the Asia-Pacific region, you can watch the video below for workflow demos and a detailed creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1DpADznEo2/I will continue updating model resources on Quark Drive:👉 https://pan.quark.cn/s/20c6f6f8d87bThese resources are mainly prepared for local users, making creation and learning more convenient.⚙️ 在线体验工作流👉 工作流： https://www.runninghub.ai/post/2027327275267530754?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1DpADznEo2/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
FireRed-Image-Edit-1.0 Dual-Image Editing Workflow
This workflow is designed for FireRed-Image-Edit-1.0 dual-image editing, mainly focused on transferring visual information from one image to another while keeping the target subject stable. A typical use case is: Image 1 provides the main person, pose, face, body structure, and composition, while Image 2 provides clothing, style, texture, or a reference object. The workflow then edits Image 1 according to Image 2 and the text instruction, making it suitable for outfit transfer, style replacement, character restyling, product reference editing, and controlled visual fusion.The workflow is built around FireRed-Image-Edit-1.0_fp8_e4m3fn.safetensors as the main editing model. It uses qwen_2.5_vl_7b_fp8_scaled.safetensors as the Qwen image text encoder and qwen_image_vae.safetensors as the VAE. It also applies Qwen-Image-Lightning-8steps-V2.0 as a model-only LoRA acceleration route, with the LoRA strength set lower than a full-force generation path. This makes the workflow practical for fast dual-image editing while still keeping enough model control for reference-based transformation.The core editing logic uses TextEncodeQwenImageEditPlus, which allows the prompt to be conditioned together with multiple image references. In this workflow, the prompt example is simple and direct: “the woman in Image 1 wears the clothes from Image 2.” This structure is very useful because the user does not need to describe every clothing detail manually. Image 2 provides the visual reference, while the prompt defines the editing relationship between the two images.The workflow also uses ImageScaleToTotalPixels to normalize both input images before editing. This helps keep the image references aligned and prevents unstable size differences from damaging the result. GetImageSize and EmptySD3LatentImage are used to match the latent canvas to the processed image dimensions, so the output remains consistent with the target image structure.A key part of this workflow is reference control. FluxKontextMultiReferenceLatentMethod is used with the index_timestep_zero method, helping the model understand how to use the visual references during generation. CFGNorm and ModelSamplingAuraFlow help stabilize the FireRed edit model, while KSampler runs an 8-step, low-CFG editing pass with Euler sampling and a simple scheduler. This setup is fast, direct, and suitable for repeated testing.The result is decoded through VAEDecodeTiled, which helps handle larger images more safely and reduces memory pressure. Image Comparer is included for before-and-after inspection, making it easy to compare the original target image and the edited output. This is especially important for dual-image editing, because the best result should preserve the identity and pose from Image 1 while accurately borrowing the clothing, style, or reference features from Image 2.This workflow is ideal for creators who want more control than ordinary text-only editing. It can be used for outfit transfer, fashion try-on concepts, character costume replacement, product-style reference transfer, visual identity remixing, social media cover creation, and Civitai example demonstrations. If you want to see how the dual-image references are connected, how the prompt should be written, and how FireRed handles reference fusion, watch the full tutorial from the YouTube link above.⚙️ Try the Workflow Online👉 Workflow: https://www.runninghub.ai/post/2024517003498754050?inviteCode=rh-v1111Open the link above to run the workflow directly online and view the generation results in real time.If the results meet your expectations, you can also deploy it locally for further customization.🎁 Fan Benefits: Register now to get 1000 points, plus 100 daily login points — enjoy 4090-level performance and 48 GB of powerful compute!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you are in Mainland China or the Asia-Pacific region, you can watch the video below for workflow demos and a detailed creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1R3ZBBrEik/I will continue updating model resources on Quark Drive:👉 https://pan.quark.cn/s/20c6f6f8d87bThese resources are mainly prepared for local users, making creation and learning more convenient.⚙️ 在线体验工作流👉 工作流： https://www.runninghub.ai/post/2024517003498754050?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1R3ZBBrEik/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
FireRed-Image-Edit-1.0 Single-Image Creative Divergent Editing Workflow
This workflow is designed for FireRed-Image-Edit-1.0 single-image editing with a more divergent and creative transformation style. Compared with the anti-drift version, this workflow is more suitable when you want to keep the core subject recognizable while allowing the scene, clothing, environment, atmosphere, and visual concept to change more aggressively. It is useful for turning one source image into a new cinematic concept, fantasy scene, sci-fi poster, commercial visual, character redesign, or creative image-editing result.The workflow uses FireRed-Image-Edit-1.0_fp8_e4m3fn.safetensors as the main image editing model, with qwen_2.5_vl_7b_fp8_scaled.safetensors as the Qwen image text encoder and qwen_image_vae.safetensors as the VAE. It also applies the Qwen-Image-Lightning 8-step LoRA route, making the generation process faster and more practical for repeated testing. The sampling configuration uses an 8-step, low-CFG editing route, which is useful for fast iteration when testing larger creative changes.The core of the workflow is built around TextEncodeQwenImageEditPlus, multi-reference image input, FluxKontextMultiReferenceLatentMethod, ReferenceLatent-style conditioning, CFGNorm, ModelSamplingAuraFlow, VAEEncode, KSampler, and VAEDecodeTiled. These nodes work together to let the model understand the input image, read the editing instruction, preserve key image information where needed, and still allow strong visual transformation.The included prompt example shows the intended use clearly. It asks the workflow to transform a half-body portrait into an astronaut performing an EVA spacewalk outside a spacecraft. The instruction keeps the person’s identity, facial structure, gaze, expression, head direction, and pose stable, while changing the clothing into a realistic white EVA spacesuit, adding a transparent helmet visor, replacing the background with outer space, Earth’s curve, spacecraft surfaces, and solar panels, and rebuilding the lighting with strong sunlight plus blue Earth-reflected fill light.This makes the workflow suitable for “controlled divergence”: it is not a simple color edit, and it is not a full random regeneration. It sits between both. The user can ask for major creative changes while still writing preservation rules for identity, posture, texture, lighting consistency, and edge blending. This is especially useful for AI portrait transformation, product concept editing, cosplay-style redesign, sci-fi conversion, fashion restyling, cinematic poster creation, and social media cover generation.ImageScaleToTotalPixels and GetImageSize help align the image scale and latent canvas with the source image. VAEEncode prepares the source image as a latent reference, while FluxKontextMultiReferenceLatentMethod helps control how the reference images affect the edit. CFGNorm stabilizes guidance, and VAEDecodeTiled helps decode the final result more safely. Image Comparer is included for before-and-after inspection, making it easier to check whether the edit changed the intended areas without completely losing the original subject.In short, this workflow is for creators who want stronger creative editing than a conservative anti-drift workflow, but still need enough control to keep the original subject readable. If you want to see how the prompt is structured, how the reference images are connected, and how the final divergent edit is produced, watch the full tutorial from the YouTube link above.⚙️ Try the Workflow Online👉 Workflow: https://www.runninghub.ai/post/2024514345513787393?inviteCode=rh-v1111Open the link above to run the workflow directly online and view the generation results in real time.If the results meet your expectations, you can also deploy it locally for further customization.🎁 Fan Benefits: Register now to get 1000 points, plus 100 daily login points — enjoy 4090-level performance and 48 GB of powerful compute!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you are in Mainland China or the Asia-Pacific region, you can watch the video below for workflow demos and a detailed creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1R3ZBBrEik/I will continue updating model resources on Quark Drive:👉 https://pan.quark.cn/s/20c6f6f8d87bThese resources are mainly prepared for local users, making creation and learning more convenient.⚙️ 在线体验工作流👉 工作流： https://www.runninghub.ai/post/2024514345513787393?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1R3ZBBrEik/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
Redraw Paste-Back Qwen 2511 Local Inpainting Workflow
已按最新格式处理：不写 Title:，不写 Description:，英文标题放在整段文案最后。Bilibili 链接已清理为公开播放链接：https://www.bilibili.com/video/BV1SK68BwE33/。RunningHub 已统一为 .ai 域名。解析依据：你上传的工作流包含 Qwen Image Edit 2511、Lightning 4steps LoRA、Qwen 2.5 VL 文本编码器、Qwen Image VAE、Mask 局部区域、QwenEditConfigPreparer、TextEncodeQwenImageEditPlusCustom、QwenEditOutputExtractor、CropWithPadInfo、RestoreCropBox、Image Comparer、ImageReel 等关键节点，核心是“局部重绘 → 裁剪编辑 → 贴回原图 → 对比输出”。This ComfyUI workflow is designed for Qwen Image Edit 2511 local inpainting, masked region editing, and redraw-and-paste-back image correction. The main purpose of this workflow is to let creators modify only a selected part of an image while preserving the original full-frame composition, camera angle, background, subject identity, and non-edited areas as much as possible. Unlike a full-frame image editing workflow, this workflow focuses on local precision. It does not ask the model to reinterpret the entire image unnecessarily. Instead, it extracts or prepares the target area, sends the selected region into Qwen Image Edit 2511 for controlled repainting, then restores the edited result back into the original image position. This makes it useful for fixing specific problems inside an image without destroying the rest of the picture. The workflow is built around Qwen Image Edit 2511. It uses qwen_image_edit_2511_bf16.safetensors as the main editing model, Qwen-Edit-2511-Lightning-4steps-V1.0-bf16.safetensors for faster generation, qwen_2.5_vl_7b_fp8_scaled.safetensors as the vision-language text encoder, and qwen_image_vae.safetensors as the VAE. This gives the workflow strong image understanding, instruction-following ability, and practical local editing speed. The core workflow logic is “redraw first, then paste back.” In normal image editing, the model may change too much of the original image. Faces may drift, backgrounds may shift, lighting may change, or the whole image may become a new generation. This workflow is designed to reduce that problem by limiting the edit to a selected region. The model can focus on the masked area, while the final RestoreCropBox stage places the corrected result back into the original canvas. This is especially useful when the original image is already good, but one part needs repair or replacement. For example, the workflow can be used to fix a hand, repair a face, change a clothing area, replace an object, adjust a small prop, correct a product detail, repaint part of the background, modify an accessory, or improve a broken AI-generated region. Instead of rerolling the whole image, you only repaint the problem area. The workflow begins with image and mask preparation. A source image is loaded, and the target region is defined with a mask. The mask tells the workflow where editing should happen. Mask quality is extremely important. A mask that is too small may not give the model enough room to blend the new content naturally. A mask that is too large may affect areas that should remain unchanged. This workflow includes mask preview and mask processing tools so users can check and adjust the selected region before generation. The workflow also includes MaskGrow and mask-to-image conversion tools. MaskGrow helps expand and soften the editing region. This is useful because hard mask edges often create visible seams after repainting. A slightly expanded and blurred mask usually gives the model more room to create natural transitions between the edited area and the untouched original image. ImageResizeKJv2 is used to resize the image and mask into a controlled working resolution. This helps keep the workflow stable across different input image sizes. The included setup shows high-resolution processing logic, with the image prepared at a large working size and aligned to a divisible-by value. This is useful for preserving detail while still keeping the generation pipeline manageable. QwenEditConfigPreparer is one of the core preparation nodes. It prepares the image, mask, reference configuration, visual-language input, crop mode, resize behavior, and longest-edge settings for Qwen Image Edit. This node helps define how the image is passed into the model and how the masked area is interpreted. In local editing workflows, this preparation stage is important because the model must understand both the original image and the target edit area. The central text-conditioning node is TextEncodeQwenImageEditPlusCustom. This node receives the image edit configuration, Qwen text encoder, VAE, and the user instruction. The workflow’s instruction logic is designed to describe the key features of the input image, such as color, shape, size, texture, objects, and background, then explain how the user’s text instruction should alter the image. This is useful because Qwen Image Edit 2511 benefits from clear editing instructions rather than vague prompts. The prompt should describe what needs to change and what must remain unchanged. For local inpainting, preservation language is very important. A good prompt does not only say “change this area.” It should also say “keep the original background, camera angle, lighting, face, pose, and non-masked regions unchanged.” This helps reduce unwanted global changes. The workflow uses ConditioningZeroOut for negative conditioning management. It also uses ModelSamplingAuraFlow and CFGNorm to stabilize the model behavior. CFGNorm is useful in image editing because too much guidance can over-transform the image. In local editing, the goal is not maximum creativity, but controlled correction. The edit should be strong enough to solve the target issue, but conservative enough to preserve the original image. The generation stage uses a KSampler setup with Qwen Image Edit 2511 and the Lightning 4-step LoRA. The included sampling route uses a short-step configuration, making it suitable for fast iteration. Local editing often requires multiple tests with different masks and prompts, so a faster sampling route is practical. Users can quickly test whether the masked region, prompt, and paste-back result are working before spending more time on final polishing. After generation, QwenEditOutputExtractor extracts the generated image, mask, main image, reference data, and related outputs from the Qwen editing result. This makes the workflow more transparent. Users can inspect the main image, edited region, mask, and final result instead of only seeing a single output. CropWithPadInfo is used to crop and manage the edited region with padding information. This is important because the model often works better when it receives a focused crop rather than the full original image. The crop gives the model more local detail and a clearer region to edit. The padding information records where that crop belongs in the original image, so the result can later be restored accurately. The RestoreCropBox node is the key paste-back component. After the selected area has been edited, RestoreCropBox places the edited crop back onto the original full-size image. This allows the final output to preserve the original full-frame layout while replacing only the corrected area. This is the main advantage of the workflow: local editing without unnecessary full-image drift. The workflow also includes optional object isolation and mask-related tools, including background removal and expansion logic. These can help prepare cleaner regions when the user wants to isolate a subject, prop, or object before local editing. This makes the workflow useful not only for simple repair, but also for controlled replacement and object-level local modification. Image Comparer is included for before-and-after inspection. This is important because local editing must be judged carefully. A result may look good at first glance but still have alignment issues, visible seams, changed colors, broken edges, or unwanted identity drift. The comparison view helps users check whether the edited region actually blends into the original image. ImageReel and ImageReelComposit are also included for presentation. They can show the main image, reference or intermediate image, mask, and result in a clear visual layout. This is useful for Civitai posts, RunningHub demonstrations, YouTube tutorials, and Bilibili showcases, because viewers can immediately understand the full editing logic instead of only seeing the final image. Main features: - Qwen Image Edit 2511 local inpainting workflow - Redraw-and-paste-back image editing pipeline - Mask-based local repainting - Uses qwen_image_edit_2511_bf16.safetensors - Uses Qwen-Edit-2511 Lightning 4-step LoRA - Qwen 2.5 VL 7B FP8 text encoder support - Qwen Image VAE support - QwenEditConfigPreparer for edit configuration - TextEncodeQwenImageEditPlusCustom for instruction-based editing - QwenEditOutputExtractor for output management - MaskGrow for mask expansion and edge control - CropWithPadInfo for local crop editing - RestoreCropBox for pasting the edited region back into the original image - CFGNorm for more stable edit guidance - Image Comparer for before-and-after checking - ImageReel layout for demonstration and publishing - Suitable for local repair, object replacement, clothing correction, face/hand fixes, and AI image post-production Recommended use cases: Local inpainting, masked region repainting, face correction, hand repair, clothing replacement, object replacement, accessory editing, product detail correction, background region repair, character detail modification, AI image post-production, social media cover correction, product mockup editing, commercial visual cleanup, Civitai workflow demonstration, RunningHub online workflow publishing, and before/after image editing showcases. Suggested workflow: Start by loading the main image. Choose an image where the overall composition is already acceptable and only one part needs to be edited. This workflow is most useful when the goal is targeted correction rather than full regeneration. Prepare the mask carefully. The mask should cover the exact area you want to repaint, but it should also include a little extra space around the target region for natural blending. For hand repair, cover the full hand and a small part of the wrist. For clothing edits, cover the full garment area and nearby edges. For object replacement, cover the object and some surrounding contact area. Use the mask preview to check the selected area before running the model. If the mask is too tight, expand it with MaskGrow. If the edge looks too hard, increase blur or soften the mask. Clean mask preparation is often more important than writing a long prompt. Write a direct edit instruction. A good prompt should clearly describe what should happen inside the masked area. For example: “repair the hand with natural fingers,” “replace the damaged object with a clean black leather handbag,” “change only the shirt into a white linen shirt,” or “repaint the broken background area with matching wall texture.” Add preservation rules to the prompt. For example: “Keep the original face, pose, background, lighting, camera angle, body shape, and all non-masked regions unchanged.” This kind of instruction helps the model understand that the task is local correction, not full-image redesign. Run the Qwen Image Edit 2511 generation stage. The Lightning 4-step route is useful for fast testing, so you can quickly compare multiple prompt and mask versions. If the result is too weak, make the prompt more direct or expand the mask. If the result changes too much, simplify the prompt or reduce the edited region. After generation, check the cropped result first. Make sure the local edit itself is correct before judging the full image. Look for correct shape, believable lighting, texture consistency, and whether the object or repaired area matches the original scene. Then check the paste-back result. The final image should preserve the original layout while replacing only the target region. Look carefully at the seam between the edited area and the original image. If the seam is visible, adjust mask blur, padding, or crop size. Use Image Comparer to inspect the before-and-after output. A successful local inpainting result should fix the selected region while keeping the original image identity stable. If the face, background, lighting, or camera angle changed too much, the edit is not controlled enough. Use ImageReel for public demonstrations. Showing the main image, mask, and final result makes the workflow easier to understand. This is useful for Civitai posts, RunningHub workflow pages, YouTube thumbnails, and Bilibili tutorial examples. For production use, test several versions. Local editing is usually an iterative process. The best result often comes from adjusting the mask, prompt, crop padding, and seed rather than relying on one generation. Keep the best version and compare it with the original before final export. This workflow is designed for creators who need practical local image correction inside ComfyUI. It is not only a Qwen 2511 editing demo; it is a complete local repainting pipeline with mask preparation, instruction-based editing, crop processing, paste-back restoration, and visual comparison. It is especially useful when you want to repair or replace one part of an image without damaging the rest. 🎥 YouTube Video Tutorial Want to know what this workflow actually does and how to start fast? This video explains what the tool is, how to launch the workflow instantly, and shares my core design logic — no local setup, no complicated environment. Everything starts directly on RunningHub, so you can experience it in action first. 👉 YouTube Tutorial: https://youtu.be/nlkrfEaScM0 Before you begin, I recommend watching the video thoroughly — getting the full context helps you understand the tool faster and avoid common detours. ⚙️ RunningHub Workflow Try the workflow online right now — no installation required. 👉 Workflow: https://www.runninghub.ai/post/2018213964676603905/?inviteCode=rh-v1111 If the results meet your expectations, you can later deploy it locally for customization. 🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance and 48 GB super power! 📺 Bilibili Updates (Mainland China & Asia-Pacific) If you’re in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown. 📺 Bilibili Video: https://www.bilibili.com/video/BV1SK68BwE33/ ☕ Support Me on Ko-fi If you find my content helpful and want to support future creations, you can buy me a coffee ☕. Every bit of support helps me keep creating — just like a spark that can ignite a blazing flame. 👉 Ko-fi: https://ko-fi.com/aiksk 💼 Business Contact For collaboration or inquiries, please contact aiksk95 on WeChat. 🎥 YouTube 视频教程 想了解这个工作流到底是怎样的工具，以及如何快速启动？ 视频主要介绍 工具定位、快速启动方法 和 我的构筑思路。 我们会直接在 RunningHub 上进行演示，让你第一时间看到实际效果。 👉 YouTube 教程： https://youtu.be/nlkrfEaScM0 开始前建议尽量完整地观看视频 —— 把握整体思路会更快上手，也能少走常见弯路。 ⚙️ 在线体验工作流 现在就可以在线体验，无需安装。 👉 工作流： https://www.runninghub.ai/post/2018213964676603905/?inviteCode=rh-v1111 打开上方链接即可直接运行该工作流，实时查看生成效果。 如果觉得效果理想，你也可以在本地进行自定义部署。 🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！ 📺 Bilibili 更新（中国大陆及南亚太地区） 如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。 📺 B站视频： https://www.bilibili.com/video/BV1SK68BwE33/ 我会在 夸克网盘 持续更新模型资源： 👉 https://pan.quark.cn/s/20c6f6f8d87b 这些资源主要面向本地用户，方便进行创作与学习。
Redraw Paste-Back Qwen 2511 Local Inpainting & Object Migration Workflow
This ComfyUI workflow is designed for Qwen Image Edit 2511 local inpainting, object transfer, reference-based editing, and redraw-then-paste-back image production. The main goal of this workflow is to let users modify only a selected area of the main image, transfer visual elements from a reference image, generate a clean edited result, and then paste the edited region back into the original image layout. This makes it much more practical than full-frame image editing, especially when the original image is already good and only one object, region, clothing area, product element, background section, or visual style needs to be changed.The workflow is built around Qwen Image Edit 2511, using qwen_image_edit_2511_fp8mixed.safetensors as the main editing model, Qwen-Edit-2511-Lightning-4steps-V1.0-bf16.safetensors as the acceleration LoRA, qwen_2.5_vl_7b_fp8_scaled.safetensors as the vision-language text encoder, and qwen_image_vae.safetensors as the VAE. This combination gives the workflow strong image understanding, instruction-following ability, reference-image support, and fast local editing performance.The core concept is “redraw and paste back.” In many image editing workflows, a model edits the whole image directly. That can be convenient, but it also creates a common problem: the face changes, the background shifts, the original composition drifts, the camera angle changes, or unrelated parts of the image are modified. This workflow is designed to reduce that problem. It crops or isolates the target region, sends that region into Qwen 2511 for local editing, then restores the edited result back into the original image using crop-box restoration logic.This makes the workflow suitable for “万物迁移” style editing. A reference image can provide an object, style, texture, material, design, color, or visual identity, while the main image provides the target scene and composition. The workflow can then transfer the reference element into the masked region of the main image. For example, it can transfer a product, clothing style, accessory, logo, material texture, decorative pattern, prop, or object identity into another image while keeping the original scene more stable.The first stage of the workflow prepares the input images and mask. The main image is loaded, resized, and prepared for editing. The mask defines the region that should be changed. The workflow includes mask processing tools such as MaskGrow, mask preview, and mask-to-image conversion. These tools help users check whether the selected area is correct before sending it into the model. A good mask is critical for local editing: if the mask is too small, the new element may not blend naturally; if it is too large, the edit may affect too much of the image.The workflow also includes background removal and object isolation logic through Image Rembg. This is useful when the reference object needs to be separated from its original background before being transferred. For example, if you want to migrate a product or subject from one image into another, background removal can help isolate the key object and reduce unwanted reference contamination. This makes the reference cleaner and easier for Qwen 2511 to interpret.TTP_Expand_And_Mask is used to expand and prepare the object region. This is useful when the edited area needs extra surrounding space for blending or when a transparent object/reference needs to be placed into a larger canvas. It helps create a more controlled input for local editing and avoids harsh boundary problems.The workflow uses QwenEditConfigPreparer to prepare reference configuration. This node manages how images are passed into Qwen Image Edit as reference images and vision-language images. It supports reference resizing, longest-edge control, crop behavior, and visual-language resizing. This is important because Qwen Image Edit 2511 needs to understand both the main image and the reference image. If the reference size or crop is poorly prepared, the model may misunderstand what should be transferred.The QwenEditAdaptiveLongestEdge node helps adapt the reference or main image size according to the image dimensions. This makes the workflow more flexible across different input resolutions. Instead of forcing every image into the same static size, it can adapt the longest edge to a controlled value, which is useful for high-resolution edits and different aspect ratios.The central text-conditioning node is TextEncodeQwenImageEditPlusCustom_lrzjason. It receives the Qwen text encoder, VAE, prepared image configs, and the user instruction. The instruction describes how the image should be edited. In this workflow, the instruction logic is designed to analyze the input image, describe key visual features such as color, shape, size, texture, objects, and background, then apply the user’s modification instruction while keeping consistency with the original image where appropriate. This is very suitable for reference-based migration and local transformation tasks.The workflow also uses ConditioningZeroOut for negative conditioning. This can help keep the generation process cleaner and reduce unwanted negative interference when using the Qwen edit conditioning structure. The model route then passes through ModelSamplingAuraFlow and CFGNorm. CFGNorm helps stabilize the guidance behavior, which is especially useful in image editing workflows where too much guidance can cause over-editing or identity drift.The sampling stage uses Qwen Image Edit 2511 with Lightning 4-step acceleration. The workflow is designed for fast local editing rather than long slow sampling. The included KSampler route uses a short-step setup with Euler-style sampling and a low CFG-style setting. This makes the workflow suitable for quick iteration. Users can test masks, references, and instructions quickly before finalizing the best result.After the Qwen edit result is generated, QwenEditOutputExtractor extracts the main image, mask, reference outputs, VAE images, and related edit outputs. QwenEditListExtractor and QwenEditAny2Image are used to extract reference-related images from the output list and convert them back into image format for preview or comparison. This makes the workflow transparent: users can inspect the main image, reference, mask, and generated result rather than only seeing the final output.The paste-back stage is one of the most important parts of the workflow. CropWithPadInfo and RestoreCropBox allow the workflow to return the edited crop back to the original image position. This means the model can work on a focused local area, while the final output still preserves the original full image. This is the practical advantage of a redraw-and-paste-back pipeline: it gives the model enough room to edit locally, but avoids unnecessary changes to the whole picture.The workflow also includes ImageReel and ImageReelComposit nodes. These create a visual comparison layout showing the main image, reference image, main image mask, and final result. This is very useful for Civitai, RunningHub, Bilibili, and YouTube demonstrations because the viewer can immediately understand what the workflow is doing. Instead of only showing a final result, the workflow can show the full logic: source image, reference, mask, and output.Image Comparer is also included for before-and-after checking. This helps users inspect whether the paste-back result is aligned, whether the edited region blends well, whether the object migration succeeded, and whether unrelated areas were preserved. This is important for practical editing because a visually impressive edit is not enough if it damages the original image.This workflow is especially useful for commercial and creator workflows. It can be used for product replacement, clothing transfer, accessory transfer, hairstyle modification, object migration, logo placement, prop transfer, material replacement, local style conversion, character detail correction, and social media cover editing. It is also useful for AI image post-production, where a generated image is mostly good but one section needs to be repaired or redesigned.Main features:- Qwen Image Edit 2511 local inpainting workflow- Redraw-and-paste-back editing structure- Uses qwen_image_edit_2511_fp8mixed.safetensors- Uses Qwen-Edit-2511 Lightning 4-step LoRA- Qwen 2.5 VL 7B FP8 text encoder support- Qwen image VAE support- Reference-based object and style transfer- Mask-based local editing- Image Rembg background removal for object isolation- MaskGrow for stronger local selection control- TTP_Expand_And_Mask for expanded object preparation- QwenEditConfigPreparer for reference and VL image configuration- TextEncodeQwenImageEditPlusCustom for instruction-based editing- QwenEditOutputExtractor for output management- CropWithPadInfo and RestoreCropBox for paste-back restoration- ImageReel comparison layout- Image Comparer before/after inspection- Suitable for “万物迁移”, object migration, and local image post-productionRecommended use cases:Local inpainting, object transfer, product replacement, clothing transfer, accessory migration, logo transfer, material transfer, reference-based redesign, masked region editing, AI image repair, background object replacement, character detail correction, product mockup creation, fashion image editing, commercial visual testing, social media cover correction, Civitai workflow demonstration, RunningHub online workflow publishing, and before/after editing showcases.Suggested workflow:Start by preparing the main image. This is the image you want to edit. Choose a source image with a clear composition and a visible target region. The workflow works best when the original image is already acceptable and only a local part needs to be changed.Next, prepare the reference image. The reference image should clearly show the object, texture, material, style, logo, or visual element that you want to transfer. If the reference contains too much background noise, the workflow can use background removal to isolate the subject. A clean reference usually gives better migration results.Create or load the mask on the main image. The mask should cover the area where the new element should appear. For object transfer, the mask should be slightly larger than the object area so the model has enough space to blend edges naturally. For precise edits, keep the mask tight but not too narrow.Use the mask preview and mask grow tools to check the selected area. If the mask edge is too hard, increase blur or grow the mask slightly. If the mask covers unrelated parts, reduce it. Mask quality strongly affects final quality.Write a clear editing instruction. A good instruction should explain what should be transferred and what should be preserved. For example: “Transfer the object style from the reference image into the masked area of the main image, while preserving the original background, camera angle, lighting, and subject identity.” For product edits, include material, placement, color, and scale. For clothing edits, include garment type, fabric, fit, and style.Keep preservation language in the prompt. Since this is a local editing workflow, it is important to tell the model what should not change. Mention that the original composition, face, pose, background, lighting direction, and non-masked regions should remain unchanged where appropriate.Run the Qwen 2511 edit stage. The Lightning 4-step route is designed for fast iteration, so it is practical to test several prompts and masks quickly. If the result is too weak, make the instruction more direct or expand the mask. If the result changes too much, simplify the instruction, improve the reference image, or reduce the edited region.After generation, inspect the output before paste-back. Check whether the transferred object or style matches the reference, whether the object fits the target region, and whether lighting and perspective are believable. If the local crop looks good, continue to the paste-back stage.Use the RestoreCropBox stage to paste the edited crop back into the original image. Check alignment carefully. The final image should preserve the original full-frame structure while replacing only the target region. If the pasted area has a visible seam, adjust the mask, blur, crop padding, or reference preparation.Use ImageReel and Image Comparer to evaluate the workflow result. The best demonstration should show the main image, reference image, mask, and final result together. This makes the workflow easier to understand for viewers on Civitai, RunningHub, YouTube, and Bilibili.For commercial use, test several references and prompts before choosing the final output. Some objects transfer better than others depending on shape, lighting, texture, and reference clarity. For logos or text-heavy objects, check readability carefully. For clothing, check body shape and fabric continuity. For product images, check material realism, reflection, and perspective.This workflow is designed for creators who need precise local editing rather than full-image regeneration. It is especially useful when the goal is not to create a completely new image, but to modify one important part while preserving everything else. With Qwen Image Edit 2511, Lightning LoRA acceleration, reference configuration, mask control, object isolation, crop editing, and paste-back restoration, this workflow provides a practical pipeline for local repainting and “all-things migration” style image editing.🎥 YouTube Video TutorialWant to know what this workflow actually does and how to start fast?This video explains what the tool is, how to launch the workflow instantly, and shares my core design logic — no local setup, no complicated environment.Everything starts directly on RunningHub, so you can experience it in action first.👉 YouTube Tutorial: https://youtu.be/nlkrfEaScM0Before you begin, I recommend watching the video thoroughly — getting the full context helps you understand the tool faster and avoid common detours.⚙️ RunningHub WorkflowTry the workflow online right now — no installation required.👉 Workflow: https://www.runninghub.ai/post/2017885601479004162/?inviteCode=rh-v1111If the results meet your expectations, you can later deploy it locally for customization.🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance and 48 GB super power!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you’re in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1SK68BwE33/☕ Support Me on Ko-fiIf you find my content helpful and want to support future creations, you can buy me a coffee ☕.Every bit of support helps me keep creating — just like a spark that can ignite a blazing flame.👉 Ko-fi: https://ko-fi.com/aiksk💼 Business ContactFor collaboration or inquiries, please contact aiksk95 on WeChat.🎥 YouTube 视频教程想了解这个工作流到底是怎样的工具，以及如何快速启动？视频主要介绍 工具定位、快速启动方法 和 我的构筑思路。我们会直接在 RunningHub 上进行演示，让你第一时间看到实际效果。👉 YouTube 教程： https://youtu.be/nlkrfEaScM0开始前建议尽量完整地观看视频 —— 把握整体思路会更快上手，也能少走常见弯路。⚙️ 在线体验工作流现在就可以在线体验，无需安装。👉 工作流： https://www.runninghub.ai/post/2017885601479004162/?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1SK68BwE33/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
Qwen3-TTS Preset Voice & Custom Voice Design Workflow
This ComfyUI workflow is designed for Qwen3-TTS voice generation, preset speaker testing, custom voice creation, and reference-based voice cloning. It combines three practical audio generation routes in one workflow: Voice Clone, Custom Voice, and Design Voice. Instead of only offering a single text-to-speech path, this workflow gives creators several ways to generate voices depending on whether they want to clone an existing reference voice, use a preset speaker, or design a new voice through descriptive character instructions.The workflow is built around Qwen3-TTS ComfyUI nodes and audio preprocessing tools. It includes FB_Qwen3TTSVoiceClone for reference-based voice cloning, FB_Qwen3TTSCustomVoice for preset speaker generation, FB_Qwen3TTSVoiceDesign for instruction-based custom voice design, MelBandRoFormer for vocal extraction, Whisper Large V3 for automatic transcription, LoadAudio for importing reference audio, PreviewAudio for quick listening, and SaveAudio for exporting final results.The first route is Voice Clone. This route is useful when you already have a reference audio sample and want Qwen3-TTS to generate new speech in a similar vocal style. The workflow loads the reference audio, separates the vocal track with MelBandRoFormer, transcribes the reference voice with Whisper, and then passes the cleaned reference audio and reference text into the Qwen3-TTS voice clone node. This makes the workflow suitable for voice imitation tests, narration style transfer, character voice reuse, AI dubbing, and digital human voice production.MelBandRoFormer is important in the cloning route because many reference audio samples are not perfectly clean. They may contain background music, room noise, ambience, or mixed sound effects. By extracting the vocal part before cloning, the workflow gives Qwen3-TTS a cleaner voice reference. This can improve speaker consistency, reduce unwanted background artifacts, and make the generated voice more stable.Whisper transcription is also important. Voice cloning works better when the reference audio and reference transcript match. The Apply Whisper node automatically transcribes the extracted vocal audio, so users do not always need to manually type the reference text. This is especially useful for longer reference clips or audio samples taken from existing videos. However, for production results, it is still recommended to check the transcript and correct errors before final generation.The second route is Custom Voice. This route uses a preset speaker option inside the Qwen3-TTS custom voice node. In the included workflow, the speaker setting uses a preset voice such as Eric, with Chinese language mode, bf16 precision, and a text prompt for direct speech generation. This route is useful when users do not want to upload a reference voice and only need a fast, stable TTS result from a predefined speaker style.The Custom Voice route is suitable for quick testing, short-form narration, rough audio drafts, social media voiceover, product explanation voice, tutorial speech, and stylized dialogue generation. Because it does not depend on an external reference audio file, it is easier to use for fast online generation. Users can change the text, speaker, language, seed, and instruction settings to test different vocal outputs.The third route is Design Voice. This route is instruction-based. Instead of cloning an existing voice or selecting a fixed preset speaker, the user writes a voice profile describing the character identity, vocal texture, age, emotional state, personality, background, and speaking style. The FB_Qwen3TTSVoiceDesign node then generates speech according to this voice design prompt.This is especially useful for original character creation. For example, the workflow can describe a senior strategic scientist with a deep, powerful, steady middle-aged male voice, strong emotional weight, national responsibility, and a serious speaking style. The TTS model then attempts to generate speech that matches the written character profile. This makes the Design Voice route valuable for AI short dramas, game NPC voice design, fictional character narration, audiobook prototypes, animation dubbing, and concept voice testing.A key benefit of this workflow is that it separates three common voice-generation needs. If you want to copy the tone of a reference sample, use Voice Clone. If you want a fast preset speaker, use Custom Voice. If you want to invent a new character voice from text instructions, use Design Voice. This makes the workflow more flexible than a single TTS graph.The workflow also includes PreviewAudio nodes so users can immediately listen to generated results inside ComfyUI. This is important for audio workflows because quality cannot be judged visually. Users need to hear pronunciation, rhythm, emotional tone, voice texture, pacing, and stability. PreviewAudio makes rapid iteration easier before saving the final audio.The SaveAudio node is used to export audio output for later use in video editing, digital human workflows, lip-sync workflows, podcast production, narration assets, or AI short film production. Generated voice can be used as a standalone audio track, or combined with video-generation workflows such as talking head animation, character dialogue, or image-to-video scenes.Main features:- Qwen3-TTS audio generation workflow- Three routes in one graph: Voice Clone, Custom Voice, and Design Voice- Reference-based voice cloning- Preset speaker text-to-speech generation- Instruction-based custom voice design- MelBandRoFormer vocal extraction- Whisper Large V3 automatic transcription- LoadAudio reference voice input- PreviewAudio for quick listening- SaveAudio for final export- Supports Chinese voice generation- Supports seed control and model choice settings- Suitable for AI video voiceover, dubbing, narration, and character voice designRecommended use cases:Voice cloning tests, preset speaker TTS generation, custom character voice design, AI short video narration, digital human voiceover, game NPC voice prototyping, audiobook sample generation, AI drama dialogue, product explanation audio, Bilibili voiceover production, YouTube narration, character concept testing, lip-sync audio preparation, voice style comparison, and ComfyUI audio workflow research.Suggested workflow:Start with the Voice Clone route if you have a reference audio sample. Upload a clear reference voice through LoadAudio. Use a clip with clean speech, stable volume, and minimal background noise. If the source contains background music or ambience, MelBandRoFormer can help isolate the vocal part, but a clean source is still better.After vocal separation, check the Whisper transcription. If the transcript is wrong, correct it before running the final voice clone generation. The reference text should match the reference audio as closely as possible. This improves stability and helps the model understand the reference speaker.Use shorter target text for early tests. One or two sentences are enough to check whether the cloned voice sounds stable. After you find a good seed and reference setup, you can test longer text. Long paragraphs may need to be split into smaller segments for better rhythm and fewer pronunciation problems.Use the Custom Voice route when you want a quick preset speaker result. Select a speaker, enter the target text, choose the language, and generate. This is useful for fast preview, simple narration, rough production drafts, and situations where voice identity does not need to match a specific person.Use the Design Voice route when you want to create an original voice. Write a detailed voice profile. Include role identity, age, gender presentation, vocal tone, speaking energy, personality, emotional state, and scene context. For example, a deep and authoritative scientist voice, a calm documentary narrator, a warm product presenter, or a sharp comedic character voice.For better Design Voice results, make the instruction specific but not overloaded. Describe the voice clearly, but avoid too many conflicting traits. If you ask for a voice that is both old and young, calm and explosive, soft and extremely powerful, the output may become unstable. A focused character profile usually gives better results.Use seed control for repeatability. If you get a good voice style, keep the seed fixed. If the voice feels unnatural, randomize the seed and test again. The same text and instruction can produce different delivery styles depending on the seed.Preview the generated audio before exporting. Listen for pronunciation, pacing, emotional expression, noise, clipping, and whether the voice matches the intended role. If the output is too flat, strengthen the emotion in the instruction. If it is too exaggerated, simplify the instruction and reduce dramatic wording.For production use, you can export the audio and then apply light post-processing such as loudness normalization, EQ, compression, or noise reduction. This can help the generated voice sit better inside videos, podcasts, short dramas, and lip-sync scenes.This workflow is designed for creators who need flexible Qwen3-TTS voice generation inside ComfyUI. It is not only a simple TTS template; it is a practical audio production workflow for cloning voices, testing preset speakers, and designing original character voices. It is useful for AI creators, video producers, game developers, digital human creators, audiobook testers, and anyone building voice assets for AIGC production.Responsible use note: only clone voices that you own, have permission to use, or are legally allowed to reproduce. Do not use this workflow to impersonate real people without consent, mislead audiences, commit fraud, or create deceptive audio. For public-facing content, it is recommended to disclose when a voice is AI-generated.🎥 YouTube Video TutorialWant to know what this workflow actually does and how to start fast?This video explains what the tool is, how to launch the workflow instantly, and shares my core design logic — no local setup, no complicated environment.Everything starts directly on RunningHub, so you can experience it in action first.👉 YouTube Tutorial: https://youtu.be/iHM2VOtUAZ0Before you begin, I recommend watching the video thoroughly — getting the full context helps you understand the tool faster and avoid common detours.⚙️ RunningHub WorkflowTry the workflow online right now — no installation required.👉 Workflow: https://www.runninghub.ai/post/2015365617691402241?inviteCode=rh-v1111If the results meet your expectations, you can later deploy it locally for customization.🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance and 48 GB super power!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you’re in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV132zxBsEAX/☕ Support Me on Ko-fiIf you find my content helpful and want to support future creations, you can buy me a coffee ☕.Every bit of support helps me keep creating — just like a spark that can ignite a blazing flame.👉 Ko-fi: https://ko-fi.com/aiksk💼 Business ContactFor collaboration or inquiries, please contact aiksk95 on WeChat.🎥 YouTube 视频教程想了解这个工作流到底是怎样的工具，以及如何快速启动？视频主要介绍 工具定位、快速启动方法 和 我的构筑思路。我们会直接在 RunningHub 上进行演示，让你第一时间看到实际效果。👉 YouTube 教程： https://youtu.be/iHM2VOtUAZ0开始前建议尽量完整地观看视频 —— 把握整体思路会更快上手，也能少走常见弯路。⚙️ 在线体验工作流现在就可以在线体验，无需安装。👉 工作流： https://www.runninghub.ai/post/2015365617691402241?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV132zxBsEAX/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
Fully automated pose transfer workflow (Qwen image edit)
More workflow parameter settings, please refer to my settings on rh: https://www.runninghub.ai/post/2048449837238063105/?inviteCode=rh-v1315Bind my invitation code: rh-v1315 to receive an extra 1000 RH coins. You can also earn 100 points by logging in daily, which can be used for generating images or videos.My AI models/apps are on RH now! Try them out & support my work — every click counts. Thank you!Thank you all for your support. Your support is the motivation for me to update more interesting and useful workflows. Thank you so much!!!Models used in the workflow:Qwen_image_edit_2511_fp8mixedQwen_image_edit_2509_fp8mixedQwen-Image-Edit-2511-Lightning-8stepsQwen-Image-Edit-2509-Lightning-8stepsqwen2511 alpha&qwen 2509v2Qwen_2.5_vl_7b_fp8_scaledQwen_image_vaeQwen3.5-9B-Uncensored-HauhauCS-Aggressivesd-pose modelCustom nodes used in the workflow： https://github.com/yolain/ComfyUI-Easy-Use https://github.com/ltdrdata/ComfyUI-Impact-Pack https://github.com/NewLouwa/ComfyUI-Model_preset_Pilot https://github.com/iacoposk8/ComfyUI-Fooocus-Inpaint-Wrapper https://github.com/Suzie1/ComfyUI_Comfyroll_CustomNodes https://github.com/lihaoyun6/ComfyUI-llama-cpp_vllm https://github.com/audioscavenger/ComfyUI-Thumbnails https://github.com/chflame163/ComfyUI_LayerStyle https://github.com/brucew4yn3rp/ComfyUI_SelectiveMetadata https://github.com/ltdrdata/was-node-suite-comfyui https://github.com/aining2022/ComfyUI_Swwan https://github.com/rgthree/rgthree-comfy https://github.com/bollerdominik/ComfyUI-load-lora-from-url https://github.com/TTPlanetPig/Comfyui_TTP_Toolset https://github.com/numz/ComfyUI-SeedVR2_VideoUpscalerhttps://github.com/JamePeng/llama-cpp-python
All-in-One. Z-Image, Ernie, Klein, t2i, i2i, controlNet, inpaint, outpaint, LLM, WAN, Qwen image, SDXL, Chroma
!!! Before installing custom nodes from manager you need to install this one from Git urlhttps://github.com/BigStationW/ComfyUI-ppm - negative prompts for Flux1, ZIT and AnimaFor example if you don't want to see glasses and tie on Anima, Zit and Flux1 generations add to fixed prompt window string (glasses, tie:-1)If manager fails to find LLM and lora nodes, install them from git url - https://github.com/willmiao/ComfyUI-Lora-Managerhttps://github.com/KingManiya/ComfyUI-LLM-text-processor (Automatic setup currently supports Windows x64 CUDA 13 only. Other platforms require manual setup of llama.cpp release binaries.)Since v.8.3 there is a separate version of workflow that can be used with Sage Attention enabled in Comfyui launch arguments (it can be easily installed from https://github.com/wildminder/AI-windows-whl?tab=readme-ov-file with no need to install compiler)This workflow supportZ-Image, Ernie image, MS Lens, Wan, Flux, Flux 2, Flux 2 Klein, Anima, Qwen Image, Qwen Image Edit, SDXL/Pony, Chroma and Lumina-Image 2.0txt2img, img2img, Inpaint functionality.Face swap (not that perfect atm)SeedVR2 upscalerLLM to describe images and enhance promptsControlNet only for Edit models in img2img group (just describe your image in prompt and add some instruction like "use input image as normal map reference")Outpainting in Edit/Kontext modesafetensor, gguf and svdq (Nunchaku) checkpointsPowerful multi Lora loader (Lora manager)Face After DetailerMulti or single images load for Edit/KontextLLM Models:VRAM 8 Gb:Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-Q8_0.ggufmmproj-Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-BF16.ggufVRAM 12+ Gb:Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-Q4_K_M.ggufmmproj-Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-f16.gguforGemma-4-E4B-Uncensored-HauhauCS-Aggressive-Q6_K_P.ggufmmproj-Gemma-4-E4B-Uncensored-HauhauCS-Aggressive-f16.ggufPut models and folder with system prompt here:📂 ComfyUI/ ├── 📂 models/ │ └── 📂 LLM/ │ ├── Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf │ ├── mmproj-Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-f16.gguf │ └── 📂 prompts/ │ ├── LLM-system.txt │ ├── LLM-system-SDXL.txtStarting from v8.0, it is possible to enhance T2I prompts on the fly using switches in the T2I group. You still need to manually copy and paste prompts from the LLM group to the I2I group, but you can directly describe Image 1 loaded in the I2I block.Follow the short block notes for more information.User prompts in both groups are now divided into two parts. The first part (Fixed) is never passed to the LLM block for processing and will always remain unchanged at the beginning of the prompt. The second part (User Prompt) can be passed to the LLM (if enabled) and improved.
Qwen 2511 2509 Swap Anyting (head, face, body, outfit, etc..)
About Native Workflow; FIRST OF ALL UPDATE YOUR COMFYUI !!!This has been the best node I've tried so far. In my tests, I saw that the 3.1 model made much better choices. Instead of person mask, I put the Sam3.1 text option where the prompt is entered. You can download the model from here and put it in the checkpoint folder; https://huggingface.co/Comfy-Org/sam3.1/tree/main/checkpoints____________________________________________________________About V3I use a different node that performs the same function in place of the original Sam3 node. yolain/ComfyUI-Easy-Sam3.Only python 3.12. If anyone runs this node on Python 3.13, can they share it with me?Please upgrade your photobuf and mediapipe for "personmask" ( python_embeded\python.exe -m pip install --upgrade mediapipe protobuf google-cloud-videointelligence )___________________________________________________________About V2I have added some spicy commands and their guide.. I removed the ToolYoloCropper node due to complaints. Instead, I added PersonMaskUltra V2 (Advance). I reverted to the defaults on some nodes. I simplified wf. Outfit swaps have become more consistent. But remember, when swapping outfits, you place the person whose outfit you want to copy into image2. So it's the exact opposite of what you do when swapping faces. Have fun!_____________________________________________________________About V1.2;In this version,I added a "PERSON SWAP". This is essentially a pose swap, and it works great.I add 3 variable latent. (Two vae encode and an empty latent) This way, you will be able to set the desired resolution and aspect ratio. It's better "select 1" But what is it?"Select 1" uses the aspect ratio of Img1 in the resulting output image."Select 2" uses the aspect ratio of Img2 in the resulting output image.In this case, Select 2 is "tool yolo cropper.""Select 3" Empty Latent Image. Opens a clean page. You determine the resolution.In other words, you are switching between "VAE encodes and empty latent".I updated the prompts, organized and simplified the nodes.About Outfit Swap: Outfit swap is tricky. I admit that the normal regular qwen edit workflow changes clothes better. Just type new prompt that i create. I'm providing it below. Just enter this prompt in the regular qwen edit wf;"Style the woman in image1, with every article of clothing in image2"☕ Buy Me a Coffee:🔗 https://buymeacoffee.com/ugurdoyduk***Qwen 2511 2509 Swap Anyting (head, face, body, outfit, etc..) ***In this workflow, you can easily swap anything you want with *100% CONSISTENCY** and without node complexity. The chance of facial mixing is ZERO..Here, I utilized the advanced SAM3 model. You can easily swap faces, heads, just hair, bodies, or a part of an object. Simply type what you want to mask into the SAM3 Text Prompt panel. For example, "woman's head" or "man's hat." Everything you type in the prompt will be masked.Just upload the relevant images to the corresponding "load image" section. You can use any checkpoint you like. I chose Phr00t and Qwen-Image-Edit-Rapid-AIO because they are very successful. (https://huggingface.co/Phr00t/Qwen-Image-Edit-Rapid-AIO). You can trust my Ksampler settings because I did a lot of testing.If you want exactly the same facial expression, Use FP2 LORA and adjust 0.5, 0.6https://huggingface.co/DiffSynth-Studio/Qwen-Image-Edit-F2P/tree/mainThanks DiffSynth-Studio
Edit workflow 2 in one - Flux Klein 9 + Qwen edit 2511 + SAM3
Workflow OverviewThis is a dual editing workflow for Qwen Edit 2511 and Flux Klein 9B; however, both model groups are self-contained. You can safely delete the entire group of whichever model you don't use without breaking the workflow. The remaining model group will work independently without any additional changes needed.Additionally, the workflow can function as a standard text-to-image workflow.Why a dual workflow?Flux Klein is my preferred editing model and I use it all the time; however, I like to have the Qwen model as a backup when Klein struggles with certain edits, and vice versa.Some personal notes:Qwen Edit has better character consistency, but it loses skin details; Flux Klein can sometimes affect character consistency, but it preserves details better.Two variants are provided in the zip file:A standard edit workflow (EditStandard.json)The same workflow as the first one, but with the extra feature of saving images with metadata compatible with CivitAI (EditCivitAI.json)Main featuresStandard text-to-image generationEdit images guided by a text prompt, using up to three reference source imagesIsolate and mask specific objects or subjects via SAM3 segmentation or hand-drawn masks (or both) to avoid pixel driftingCamera angle control (especially useful with Qwen + multi-angle LoRA)Basic pose transferExample use casesVirtual try-on / outfit transferObject / subject removalBackground / scene changeCharacter swap / face swapInpaintingMulti-angle viewPose changeCharacter sheetScene compositionImage fixingImage style changeDetailed instructions are contained in the workflow itself:Red nodes are instructions and useful notes.Yellow nodes are configurable elements you can adjust to your needs.
🌸KiraNugget's Multiview Character Sheet V1.0🌸
🌸 KiraNugget's Multiview Character Sheet V1.0 🌸Turn One Image Into a Complete Character Reference.Input a single image of your character and instantly generate a full multi-angle character sheet ready for art, animation, 3D modeling, or future generations.✨ Why You’ll Love It🖼️ Input just one image🔄 Automatically recreates your character from multiple angles😀 Includes multiple facial expressions and close-ups🎨 Keeps the design consistent across every view⚡ Minimal setup — no complicated prompting required⚡ What It CreatesFrom a single reference image, the workflow generates:Front viewsSide viewsBack viewsThree-quarter viewsFace close-ups with multiple expressionsBonus portrait imageBonus landscape / cinematic imageAll while preserving your character’s:HairstyleOutfitColorsFacial featuresOverall style and personality✏️ Fully EditableEvery section of the sheet can be customized individually with your own prompts.Want:A different pose for the side view?A different expression in one close-up?A unique outfit, background, or camera angle for the bonus images?Each panel can be edited separately while still keeping the overall character design consistent.🎭 Expressions IncludedYour character isn’t just recreated from different angles — they also come to life with multiple expressions such as:HappyAngrySadSurprisedSmugNeutralPerfect for making your character feel more complete and usable.🧠 Smart ConsistencyThe workflow is designed to maintain:Consistent proportionsMatching clothing detailsStable colors and accessoriesAccurate facial structure across every angleNo more manually trying to redraw the same character from scratch.🎨 Perfect ForCharacter reference sheetsOriginal charactersAnime and game design3D modeling referencesComics, VTubers, and visual novelsTraining datasets or future image generations🚀 Bonus ImagesIn addition to the main sheet, the workflow also creates:A polished portrait-style imageA cinematic landscape shot of your characterGreat for thumbnails, profile pictures, or showing off the final design.❤️ One Character. Every Angle.Drop in your favorite character image and let the workflow build the full reference sheet for you.Because one image should be enough.
🌸Real2Anime/4K RTX Upscaler Fast and Easy Workflow🌸 (Qwen edit 2509 with Nunchaku)
🌸 Ultimate Anime Waifu Creation Workflow 🌸Turn any real-life beauty into a stunning anime waifu — instantly!Need a brand-new anime girl OC for a project? Want to stylize a cosplayer into the perfect character?This workflow has you covered.💡 Designed to be extremely noob-friendly — plug it in, follow the prompts, and you’re creating waifus in minutes.Now with Nunchaku Loader - Extremely fast generationCredit to Nunchaku: https://nunchaku.tech/docs/nunchaku/index.html✨ Why This Workflow?Pros⚡ Fast & effortless character creation — perfect as a base before deeper edits🎯 Turn a Cosplayer into your favorite character without waiting for specific LoRA🎨 Fully editable prompts for poses, colors, expressions & more📚 Massive masterlists for inspiration and detailed customization💘 WAIFUS. Lots of waifus.Cons💔 Your waifu is still not real… but she’s closer than ever.🔍 What’s InsideThis workflow includes Qwen Image Edit–optimized prompts for both adult and younger characters.Just pick the appropriate prompt and the workflow will:Generate your character in the same pose as your referenceReproduce accurate hair, clothing, accessories & body proportionsMaintain a clean white backgroundKeep all interacting objects properly depictedYou’ll also get extensive masterlists covering:Realistic & anime hair colorsRealistic & anime eye colorsPosesExpressionsDetailed usage instructions so you always know how to plug customizations inWant a different artstyle?Swap the included flat-anime LoRA for any favorite artstyle LoRA — see the suggestion section for recommendations.⚠️ Important NoteThe base workflow is intended for SFW character generation.While NSFW might be possible, that is NOT the intended use.Create responsibly ❤️Loras needed:https://huggingface.co/lightx2v/Qwen-Image-Lightning/tree/main/Qwen-Image-Edit-2509 (Credits to lightx2v)https://civitai.com/models/1994714/breast-slider-qwen (Credits to keltuzad) (For best reference accurate results)Optional Artstyle Loras I used in base workflow:https://huggingface.co/autoweeb/Qwen-Image-Edit-2509-Photo-to-Anime/tree/main (Credits to Autoweeb)https://civitai.com/models/1952560/anime-flat-style (Credits to Agino)V1.1:Edited Prompt so the younger version are easier to generate when reference has a bigger chest.Added some note to facilitate usage of worflow.V1.2Further refined prompts.Changed encoder for one that seems to give bettrer results overall.V1.3I removed the randomization for hair color from the prompt as it was not really working and was just making the prompts too long. I instead made lists of colors for hair and eyes and the optimized way to implement it in the prompt as needed. The base prompts should recreate accurate hair colors.-Added notes for noob users like I was a couples of week ago. :)V1.4MASTERLISTS! Added an extensive list of colors and style for optional hair and eyes color change.V1.5Added Masterlists for poses and expressions.V2.0Changed to Nunchaku Loader.Now extremely faster.V3.0Complete Overhaul and Simplification of the workflow.Refined prompt.Added easy switches.Added 4K RTX upscaler. (Need a Nvidia RTX GPU to use.)Included Base Loras used in .zip file
Qwen 2511 (Turbo) Turnaround Workflow (Concept Art to 3D Views)
Turn a single concept image into full 4-view (front/left/right/top) renders using Qwen Edit 2511!This is a workflow using Qwen Edit 2511 to turn your concept art into 3d-rendered turnarounds. It's pretty simple, most of everything is sub-graphed, so you don't have to do much except for upload your image, hit run, and watch it go. The first step takes your sketch and transforms it into a 3-d rendered image (This is the heavy portion.)Then it renders out a back, right, and top view. (This part is faster with turbo mode enabled.)I'm using an 8-step LORA here, but feel free to use a 4-step.I'll improve the workflow as I go, I'm sure there are things I can add to make the workflow better. Watch this space for updates!
Multiple character angles
As the title says, this workflow using qwen2511 allows you to take one character and have qwen create multiple angles from a single photo. This workflow is built off of an existing template I found on the ComfyUI template page, I changed a few of the prompts so they respond better than what the original had as well as added a selector switch so you weren't forced to generate every angle if you just wanted 1 or 2 different images.Note that when qwen is giving you a difference angle it is basically regenerating the entire image while using your input image as a reference. things in the background my not be consistent throughout all images.
Qwen2511+ZIT A2R HD
This workflow using Qwen-Image-Edit 2511 and Z-Image Turbo to convert anime images to realistic style.Qwen-Edit A2R LoRA: https://civitai.com/models/2294036Qwen-Edit 2511 Lightning LoRA: https://huggingface.co/lightx2v/Qwen-Image-Edit-2511-Lightning此工作流使用 Qwen-Image-Edit 2511 和 Z-Image Turbo 模型来将动漫图像转换为逼真的写实照片.Qwen-Edit 动漫转写实 LoRA: https://civitai.com/models/2294036Qwen-Edit 2511 加速 LoRA: https://huggingface.co/lightx2v/Qwen-Image-Edit-2511-Lightning
Character Expression Editing Qwen Image Edit 2511
More workflow parameter settings, please refer to my settings on rh: https://www.runninghub.ai/post/2039239226247618561/?inviteCode=rh-v1315Bind my invitation code: rh-v1315 to receive an extra 1000 RH coins. You can also earn 100 points by logging in daily, which can be used for generating images or videos.My AI models/apps are on RH now! Try them out & support my work — every click counts. Thank you!Thank you all for your support. Your support is the motivation for me to update more interesting and useful workflows. Thank you so much!!!Models used in the workflow:[PixelSmile-preview](https://huggingface.co/PixelSmile/PixelSmile/blob/main/PixelSmile-preview.safetensors)[Qwen_image_edit_2511_fp8mixed](https://huggingface.co/Comfy-Org/Qwen-Image-Edit_ComfyUI/blob/main/split_files/diffusion_models/qwen_image_edit_2511_fp8mixed.safetensors)[Qwen-Image-Edit-2511-GGUF](https://huggingface.co/unsloth/Qwen-Image-Edit-2511-GGUF/tree/main)[Qwen-Image-Edit-2511-Lightning-4steps-V1.0-fp32](https://huggingface.co/lightx2v/Qwen-Image-Edit-2511-Lightning/blob/main/Qwen-Image-Edit-2511-Lightning-4steps-V1.0-fp32.safetensors)[Qwen_2.5_vl_7b_fp8_scaled](https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/blob/main/split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors)[Qwen2.5-VL-7B-Instruct-GGUF](https://huggingface.co/unsloth/Qwen2.5-VL-7B-Instruct-GGUF/tree/main)[Qwen2.5-VL-7B-Instruct-mmproj-BF16](https://huggingface.co/unsloth/Qwen2.5-VL-7B-Instruct-GGUF/blob/main/mmproj-BF16.gguf)[Qwen_image_vae](https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/blob/main/split_files/vae/qwen_image_vae.safetensors)<br/>Custom nodes used in the workflow：https://github.com/judian17/ComfyUI-PixelSmile-Conditioning-Interpolation.githttps://github.com/chrisgoringe/cg-use-everywherehttps://github.com/yolain/ComfyUI-Easy-Usehttps://github.com/city96/ComfyUI-GGUFhttps://github.com/ltdrdata/ComfyUI-Impact-Packhttps://github.com/bollerdominik/ComfyUI-load-lora-from-urlhttps://github.com/Suzie1/ComfyUI_Comfyroll_CustomNodeshttps://github.com/chflame163/ComfyUI_LayerStylehttps://github.com/aining2022/ComfyUI_Swwanhttps://github.com/rgthree/rgthree-comfyhttps://github.com/kijai/ComfyUI-KJNodes
Folder Image Captioner with Qwen-VL WF
Folder Image Captioner with Qwen-VLThis ComfyUI workflow allows you to batch caption entire folders of images quickly and efficiently.It loads images from a selected folder, resizes them if needed, generates high-quality detailed captions using Qwen-VL-Mod (Qwen3-VL-8B-Instruct-Abliterated), and saves both the original image and its corresponding caption file with the exact same filename (e.g., photo.jpg + photo.txt).Ideal for creating training datasets for LoRAs, character fine-tuning, or any project that requires consistent captions.Features:Batch processing directly from folderSaves image + caption with the same nameHigh detail and accuracy thanks to Qwen-VLMaintains the same pose, camera angle, lighting, and location from the original imageRequired Custom Nodes:ComfyUI Custom NodesQwen-VL-Mod (or Qwen3-VL-8B-Instruct-Abliterated)Resize Image v2Load Image Dataset from FolderSave Image and Text Dataset to FolderCreated by: bobgus39 Original profile: https://civitai.com/user/bobgus39Usage: Simply select your image folder and run the workflow. The captions will respect the original pose, camera angle, lighting, and background/location of each image, making them perfect for training consistent characters or scenes.
SDXL Portrait-to-Scene Master Workflow
This workflow is designed to maintain a consistent SDXL artistic style by generating characters and backgrounds independently.Key Features:Interactive Layout: Utilize a manual canvas to freely adjust character positioning and scaling.Spatial Awareness: Automatically extracts Depth and LineArt from the character to guide background synthesis, ensuring perfect spatial integration.AI-Powered Refinement: Leverages Qwen-VL for intelligent image blending, automated self-checking of anatomical/logical issues, and targeted inpainting repairs.It transforms a standard generation process into a professional, feedback-driven production pipeline.
elpideus Qwen Image Edit 2509 + Upscaling + Face Restoration + Hand Restoration
Workflow I did to learn Image Generation in ComfyUI
Qwen3-TTS Voice Design
QwenTTS Voice Design Workflow: Custom Voice Architect[Best for: Tech Podcasts, Professional Narrators, Multi-Accent Customization]🚀 FREE Cloud Computing (No VRAM Limits): High-quality voice generation requires stable compute power. Get 1,000 FREE credits by registering on RunningHub: 👉https://www.runninghub.ai/post/2037042533586440193/?inviteCode=rh-v1497📺 Full Video Tutorial: See how I fine-tuned the parameters to get the perfect "Australian Male Host" voice in my latest video: 👉 💡 Key Features: A precision-engineered workflow based on QwenTTS, designed for creators who demand high-end vocal texture.Precision Fine-Tuning: Define the narrator’s age, gender, speaking rate, chest resonance, and specific accent nuances.Information Dense: Ideal for long-form professional tech podcasts or news commentary with steady, logical delivery.
Sq Qwen Edit 2511 Subgraph
All your image editing workflows in one simple to use Subgraph. With the right prompting you can:Change styles of your image (realistic, anime, dystopian, cartoon, etc.)Add, remove, or edit elements in the image (remove unwanted people, change clothes, change hair, add someone to a photo, etc.)Create new images based on your reference images (3D figurines, model sheets, scene changes, etc.)Much more.A Subgraph workflow I made to make using Qwen Edit 2511 easy for myself. Information / Instructions are included with the workflow.With the models used this runs well locally on my NVIDIA GeForce RTX 4070 with 12 GB VRAM. Feel free to adjust models if you have a more or less able video card.BASIC USAGELoad your primary image in Load Image 1.Images 2 and 3 are optional.Subgraph values:image#_enable: true if using that image, else false (ignored).override_image_size: false to use Image 1 size. true will use the override_width and override_height size, adjusted by aspect_ratio.aspect_ratio: 'original' for the same aspect as Image 1. 'fixed' to use the override_width/height values. Otherwise use the aspect selected.The first prompt is Positive and the second is Negative (CFG must NOT be 1 to use).steps: 8 is a good default with the lightning lora used, but can go between 4 to 20 or more.Other values are per usual usage.Open the Subgraph to change model, lora, etc.If you're not familiar with editing you can request changes to items or styles of the image, insert one image element into another, or create whole new images using the characters or items in the reference pic(s).USED MODELSModel [Unet Loader (GGUF)]qwen-image-edit-2511-Q3_K_M.ggufhttps://huggingface.co/unsloth/Qwen-Image-Edit-2511-GGUF/tree/main→ models/diffusion_modelsLora [LoraLoaderModelOnly]Qwen-Image-Edit-2511-Lightning-4steps-V10-bf16.safetensorshttps://huggingface.co/lightx2v/Qwen-Image-Edit-2511-Lightning/tree/main→ models/lorasClip [Load CLIP]qwen_2.5_vl_7b_fp8_scaled.safetensorshttps://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/tree/main/split_files/text_encoders→ models/clipVAE [Load VAE]qwen_image_vae.safetensorshttps://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/tree/main/split_files/vae→ models/vae
2511 Seamless Face Swap Master Edition
🎥 YouTube Video TutorialWant to know what this workflow actually does and how to start fast?This video explains what the tool is, how to launch the workflow instantly, and shares my core design logic — no local setup, no complicated environment.Everything starts directly on RunningHub, so you can experience it in action first.👉 YouTube Tutorial: https://youtu.be/9l_xjeI881wBefore you begin, I recommend watching the video thoroughly — getting the full context helps you understand the tool faster and avoid common detours.⚙️ RunningHub WorkflowTry the workflow online right now — no installation required.👉 Workflow: https://www.runninghub.ai/post/2027327275267530754?inviteCode=rh-v1111If the results meet your expectations, you can later deploy it locally for customization.🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance and 48 GB super power!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you’re in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1DpADznEo2/☕ Support Me on Ko-fiIf you find my content helpful and want to support future creations, you can buy me a coffee ☕.Every bit of support helps me keep creating — just like a spark that can ignite a blazing flame.👉 Ko-fi: https://ko-fi.com/aiksk💼 Business ContactFor collaboration or inquiries, please contact aiksk95 on WeChat.🎥 YouTube 视频教程想了解这个工作流到底是怎样的工具，以及如何快速启动？视频主要介绍 工具定位、快速启动方法 和 我的构筑思路。我们会直接在 RunningHub 上进行演示，让你第一时间看到实际效果。👉 YouTube 教程： https://youtu.be/9l_xjeI881w开始前建议尽量完整地观看视频 —— 把握整体思路会更快上手，也能少走常见弯路。⚙️ 在线体验工作流现在就可以在线体验，无需安装。👉 工作流： https://www.runninghub.ai/post/2027327275267530754?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1DpADznEo2/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
VAE merger for comfyUI (anima)
This is a custom ComfyUI node created to merge VAE models directly inside ComfyUI.The node performs weighted merging of VAE state_dict parameters and allows saving the result as a new safetensors VAE file.Features:Merge VAE inside ComfyUIDesigned for Anima / Qwen workflowsShape check for safe mergingSave merged VAE directlyNo external script requiredThis tool was used to create the VAE released on this page.
Rebels Qwen Image 2512 TURBO
Generates images in 2 steps!workflow for gguf of qwen image 2512 + 2 step loraNOTE: text encoder is a different encoder than the original qwen image 2512. you MUST download the new encoder to run the workflow.you must adhere to the aspect ratio resolutions provided in the workflow or you will introduce major artifacts.Model is NSFW. Examples are provided.ggufhttps://huggingface.co/unsloth/Qwen-Image-2512-GGUF/tree/maintext encoderhttps://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/blob/main/split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensorsvaehttps://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/blob/main/split_files/vae/qwen_image_vae.safetensors2 step lorahttps://huggingface.co/Wuli-art/Qwen-Image-2512-Turbo-LoRA-2-Steps/blob/main/Wuli-Qwen-Image-2512-Turbo-LoRA-2steps-V1.0-bf16.safetensors
Flux.2-Klein-Base-9B — All-in-One Workflow by TreeMale
🐞 Send feedback about any bugs you find or potential improvements.A series of processes for generating high-quality images focused on realism use model Flux.2-Klein-Base-9B (for the test, I took one image and ran it through the entire creation process).⚠️ You can find a list of required models, instructions and a description of the process in the Workflow.For your convenience i have combined all processes into one workflow and made "user friendly" interface. Processes list:🤖 Flux.2-Klein-Base-9B — Mannequin Character by TreeMale🗓️ RELEASE DATE: [17.03.2026]📋 What does this workflow do?This workflow converts a photograph of a character into a stylized mannequin image.🚩 Result: A clean 3D mannequin with a crisp white surface on a black background.🤖 Flux.2-Klein-Base-9B — Camera Angle by TreeMale🗓️ RELEASE DATE: [18.03.2026]📋 What does this workflow do?This workflow changes the camera angle of the input composition image.🚩 Result: The input composition image is acquired from a different camera angle.🤖 Flux.2-Klein-Base-9B — SeedVR2 Upscaler by TreeMale🗓️ RELEASE DATE: [20.03.2026]📋 What does this workflow do?This workflow upscale the input image and removes random noise.🚩 Result: Upscaled and improved version of image.🤖 Flux.2-Klein-Base-9B — Qwen-3-VQA Prompt by TreeMale🗓️ RELEASE DATE: [26.03.2026]📋 What does this workflow do?This workflow creates a structured prompt from a set of user positive and negative tags.🚩 Result: Generate text prompt from a set of user tags.🤖 Flux.2-Klein-Base-9B — Text Label by TreeMale🗓️ RELEASE DATE: [31.03.2026]📋 What does this workflow do?This workflow creates a text label in bottom image border.🚩 Result: Image with a frame and text label.🤖 Flux.2-Klein-Base-9B — Realistic Skin by TreeMale🚫 Soon... Works in progress... Look to updates!🙏 Special thanks!Black Forest Labs — for the FLUX.2-Klein-9B model!Comfy-Org — for the optimized CLIP and VAE!Civit.AI — for the good community!You — for using!Created with ❤️ for the community. Share your results and feedback! Many thanks to the creators of the original models - give him some tips!
LTX 2.3 3-Image Sequence & Auto-Prompt(Qwen 3.5 / Gemini Driven)
🚀 No 24G VRAM? Run this directly in the Cloud (RunningHub):https://www.runninghub.ai/post/2034210033776463874/?inviteCode=rh-v1497(🎁 Bonus: Register via this link to get 1,000 FREE credits instantly + 100 daily login credits!)📺 Watch the Full Step-by-Step Tutorial Here:Welcome to the ultimate cinematic video generator by CottonByte! 、So what are the key upgrades of LTX 2.3?First, a significant boost in 9:16 vertical video generation quality — optimized for mobile screens, making it much more mobile-friendly.Second, sharper visuals, more accurate lip-sync and voice clarity, with noticeably improved background music and sound effects.Third, better prompt comprehension — no matter how complex your prompts are, LTX 2.3 handles them with ease.Fourth, more precise text rendering in videos. For example, if you want to generate a video with a logo, it accurately renders the logo text into the final output.
LTX 2.3 Keyframe Master: Start-to-End Frame Consistency
🚀 No 24G VRAM? Run this directly in the Cloud (RunningHub): https://www.runninghub.ai/post/2034201899603333121/?inviteCode=rh-v1497(🎁 Bonus: Register via this link to get 1,000 FREE credits instantly + 100 daily login credits!)📺 Watch the Full Step-by-Step Tutorial Here: Welcome to the ultimate cinematic video generator by CottonByte! So what are the key upgrades of LTX 2.3?First, a significant boost in 9:16 vertical video generation quality — optimized for mobile screens, making it much more mobile-friendly.Second, sharper visuals, more accurate lip-sync and voice clarity, with noticeably improved background music and sound effects.Third, better prompt comprehension — no matter how complex your prompts are, LTX 2.3 handles them with ease.Fourth, more precise text rendering in videos. For example, if you want to generate a video with a logo, it accurately renders the logo text into the final output.
Ultimate LTX 2.3 Master (Qwen 3.5 / Gemini Driven)
🚀 No 24G VRAM? Run this directly in the Cloud (RunningHub): https://www.runninghub.ai/post/2034150071498973185/?inviteCode=rh-v1497(🎁 Bonus: Register via this link to get 1,000 FREE credits instantly + 100 daily login credits!)📺 Watch the Full Step-by-Step Tutorial Here: Welcome to the ultimate cinematic video generator by CottonByte! 、So what are the key upgrades of LTX 2.3?First, a significant boost in 9:16 vertical video generation quality — optimized for mobile screens, making it much more mobile-friendly.Second, sharper visuals, more accurate lip-sync and voice clarity, with noticeably improved background music and sound effects.Third, better prompt comprehension — no matter how complex your prompts are, LTX 2.3 handles them with ease.Fourth, more precise text rendering in videos. For example, if you want to generate a video with a logo, it accurately renders the logo text into the final output.
QWEN Image Edit I2I, T2I, Edit Plus, Upscale, Seed VR2, Post Process
Qwen I2I, T2I, Image Edit, Detailers, and a hefty Post Production SuiteFor Qwen 2512, please use this workflow: https://civitai.com/models/2457117/qwen-image-2512-t2ii2i-controlnet-detailer-upscaler-workflowAllows up to 3 images to be loaded for referenceUltimate Upscaler packageFace, Hand, Eye, and Expression DetailersAdd lighting, color correction , and specicl effects after generation in the Post Production SuiteSeed VR2 UpscalerImproved Prescaler (Hi rez fix)Massive Post Production area.Instagram: https://www.instagram.com/synth.studio.models/This represents hundreds of hours of work. If you enjoy it, please 👍like, 💬 comment , and feel free to ⚡tip 😉
Universal Video-to-Prompt: Decode Camera Motion & Action with Qwen 3.5
[Description] Understanding motion is the hardest part of AI video. This workflow is designed specifically for creators using Kling, Runway, or SVD who want to "steal" the camera work and pacing from professional clips. Qwen 3.5 (9B) acts as your director, breaking down video frames into actionable cinematic prompts.🚀 Try the Cloud Version (Instant Access): https://www.runninghub.ai/post/2033408926225534977/?inviteCode=rh-v1497Tutorial：Core Highlights:Motion Capture: Specifically tuned to recognize lens movement (Pan, Tilt, Zoom) and subject action.Browser-Based Power: Decoding video is resource-heavy; skip the local lag by using the RunningHub Cloud Version.Bonus: Sign up via the link to claim 1,000 FREE credits and a 100-credit daily bonus.
Qwen 3.5 Reverse Image Prompting: Capture Styles & Details
[Description] Stop playing the "Prompt Gacha" game. If you've ever seen a stunning Midjourney or Flux image and wondered, "How did they write that?" — this workflow is your answer. Powered by the state-of-the-art Qwen 3.5 (9B), it deep-dives into pixels to extract lighting, textures, and composition logic.🚀 Try the Cloud Version (Instant Access): https://www.runninghub.ai/post/2033408926225534977/?inviteCode=rh-v1497Tutorial：Core Highlights:Precision Decoding: Not just "tagging," but understanding the cinematic logic of the image.Reverse-Test Proven: Successfully tested by regenerating prompts through Gemini/MJ APIs with near-perfect accuracy.0-Setup Cloud Option: Don't have a high-end GPU? You can run this instantly on RunningHub.Free Credits: Use the link below to get 1,000 FREE credits (enough for dozens of tests) plus 100 daily login credits.
AI Storyboard Assistant: Smart Text-to-Prompt & Scripting
Struggling with "Prompt Block"? This workflow turns your raw concepts into structured, logic-driven scripts optimized for high-end video AI. Instead of just adding random keywords, Qwen 3.5 builds a narrative flow, ensuring the AI understands the "Why" behind the scene.🚀 Try the Cloud Version (Instant Access): https://www.runninghub.ai/post/2033408926225534977/?inviteCode=rh-v1497Tutorial：Core Highlights:• Narrative Expansion: Turn "A cat in space" into a detailed, multi-dimensional cinematic prompt.• Cloud Ready: Experience the full reasoning speed of Qwen 3.5 on RunningHub without any environment setup.• New User Gift: Register via the link for 1,000 FREE credits and enjoy 100 daily credits just for logging in!
Qwen 3.5 Multi-Prompt Master: Image / Video / Text-to-Prompt (3-in-1)
[About This Workflow]Tired of manual prompt engineering? This master workflow leverages the power of Qwen 3.5 (9B) to help you "steal" and optimize prompts with precision. Whether you are a beginner or a pro, this is the ultimate tool to bridge the gap between your ideas and AI-generated masterpieces.[Try now！]Try it immediately on RunningHub Cloud Version (No GPU Needed): Qwen 3.5 Master Workflow (3-in-1): https://www.runninghub.ai/post/2033408926225534977/?inviteCode=rh-v1497🎁 Bonus: Registering through the link above will grant you 1,000 FREE credits—more than enough to experiment with dozens of prompts.💎 Pro Tip: Don't forget to check back daily! You’ll receive an additional 100 credits every day just for logging in, keeping your creative energy flowing without cost.[Watch the Tutorial]For a full deep dive and step-by-step setup, watch my YouTube guide here: [Key Features]• 3-in-1 Versatility: Switch seamlessly between Image-to-Prompt, Video-to-Prompt, and Text-to-Prompt modes.• Powered by Qwen 3.5 (9B): Superior logical reasoning and multi-modal understanding.• Fast Bypasser Logic: Integrated switch for lightning-fast mode selection (Yes/No toggle).• Cloud & Local Ready: Optimized for platforms like RunningHub but fully compatible with local ComfyUI setups.• Reverse-Verification Proven: Tested with Gemini and Midjourney API nodes to ensure maximum accuracy.[How to Use]1. Select Mode: Use the Fast Bypasser switch to choose your desired generator.2. Upload / Input: Upload your reference image/video or type your core concept.3. Run: Get the optimized, cinematic prompt in seconds.4. (Optional) Edit the Qwen-generated output to add your own personal flair.
Merge two characters in one image using Qwen
Hello CivitAI community, here is a simple ComfyUI workflow to merge 2 peoples from 2 images to one scene preserving faces consistencyDownload the model from : https://huggingface.co/Phr00t/Qwen-Image-Edit-Rapid-AIO/tree/main/v19Tested a lot of version, v19 seems to be the bestI also use these Lora :https://civitai.com/models/1834338/wan22-smartphone-snapshot-photo-reality-stylehttps://civitai.com/models/1972643/studio-realism
Qwen Image 2512 T2I/I2I/ Controlnet/ Detailer/ Upscaler Workflow
Note: I spend a lot of time on these to make sure they are top notch. If you have ANY issues, I'm very responsive. hit me up and I'll resolve them for you.For Qwen Image Edit, please use this workflow: https://civitai.com/models/2263383?modelVersionId=2781495A complete user and troubleshooting guide inclusing a list of common error messages can be found here https://civitai.com/articles/24678/how-to-use-my-workflows-a-comprehensive-breakdown也有中文说明Instagram: https://www.instagram.com/synth.studio.models/Buy me a☕ https://ko-fi.com/lonecatoneThis represents many of hours of work. If you enjoy it, please 👍like, 💬 comment , and feel free to ⚡tip 😉This is an all in one workflow:Controlnet: background remove, Canny & DepthI2I prompt assistQwenV2 Prompt enhancer.Florence and Qwen prompt assistUltimate Upscaler for prescaling and hi res fixDetailing Suite for face, hands, eyes, and expressionsSeed VR2 Upscaler to get 4k on low VRAMMassive Post Production SuiteSmart Noise scrubberFull tutorial available.也有中文说明
FireRed Image Edit 1.1 and Qwen Image Edit 2511 ComfyUI Workflow by Smittie
Smittie's FireRed-Image-Edit-1.1 and Qwen-Image-Edit-2511 WorkflowDescriptionWith this workflow you can edit images using FireRed-Image-Edit-1.1 or Qwen-Image-Edit-2511. It is kept simple and well structured.Install Instructions1. Install all missing custom nodes via the ComfyUI Manager2. Download all needed modelsDiffusion modelOriginal: FireRed-Image-Edit-1.1-transformer.safetensorsRecommended: FireRed-Image-Edit-1.1_fp8mixed_comfy.safetensorsLightning LoRAFireRed-Image-Edit-1.0-Lightning-8steps-v1.0.safetensorsText encoderqwen_2.5_vl_7b_fp8_scaled.safetensorsVAEqwen_image_vae.safetensorsAcknowledgementMost credits goes to the Qwen, FireRed, ComfyUI and CivitAI team, as well as kijai, Firetheft and cocorang. Thank you guys!
FireRed Image editor: Seamless Clothing Swap, Pose Change & Portrait Editing
How to Use:Upload your base photo & Prompt.Upload a reference image of your product.Hit generate and watch the magic happen!https://www.runninghub.ai/post/2030916473266376705/?inviteCode=rh-v1497Skip the complex setup! ✨ Sign up through my link for 1,000 FREE credits. You'll also get 100 bonus RH coins for daily login, making it super easy to experiment with styles for free!FireRed Image Edit | The Ultimate 0-Shift E-commerce EditorTired of AI distorting your model's face or body while changing clothes? 🛑Meet the FireRed Image Edit workflow—the most precise tool for fashion and e-commerce photography. Powered by the latest open-source model, it delivers professional-grade editing with zero identity shift.✨ Core Features:• 0-SHIFT Precision: Swap outfits, jewelry, or makeup while keeping the model’s identity and body shape 100% intact.• 1-CLICK Operation: No complex node-spaghetti. Just upload your images and hit generate.• Total Control: Easily change poses, backgrounds, or even edit text with high fidelity.🚀 Why You Need This:Perfect for Shopify, Amazon, and independent sellers. Scale your product content, create virtual try-ons, and save thousands on traditional photoshoots.⚠️ Important: Model Requirement To run this workflow locally, you will need the FireRed Image Edit model &lora 】.Easy Way: If you want to skip the massive downloads and environment setup, I highly recommend using the RunningHub link above—the model is pre-installed and ready to go!
ComfyUI Flash Head Workflow: Ultrafast Head Lip-Sync
Video Introdution: Click here to try workflow online:(Notice：Some nodes are biulding by Runninghub ,if you downloading and running offline , may not work!) Open Source Address: https://github.com/Soul-AILab/SoulX-FlashHead (Workflows can be downloaded via the links below—click the link and find the download button in the top right corner. Due to limited VRAM on my local machine, I haven't been able to test these myself. So if you're not familiar with running ComfyUI locally, it's best to run them online. The FlashHead node is built on RH.) Workflow: AA--Ultra-Fast Digital Human FlashHeadExperience Link: https://www.runninghub.ai/post/2030340894288781313/?inviteCode=rh-v1401 Workflow: AA--Emotion Control Digital Human - Ultra-Fast FlashHead + Index Voice Cloning (8 Emotion Controls)Experience Link: https://www.runninghub.ai/post/2030585203055398914/?inviteCode=rh-v1401 Workflow: AA--Preset Voice Ultra-Fast Digital Human - FlashHead + QwenTTS - One Image, 9 VoicesExperience Link: https://www.runninghub.ai/post/2030658043398070273/?inviteCode=rh-v1401 Workflow: AA--Fully Automatic Ultra-Fast Digital Human - FlashHead + Qwen Sound Design - Auto-Prompt from One Image - Digital Human Card Pull!Experience Link: https://www.runninghub.ai/post/2030589859588476930/?inviteCode=rh-v1401 ### Introduction to Flash Head Digital Human Workflows Flash Head is a digital human generation project running on ComfyUI, focused on speed. It achieves extreme video generation speed by only driving the head region for lip-sync, sacrificing dynamics in other parts of the body. #### Core Features: * Ultimate Speed: At 512p resolution, generating a 5-second video takes only about 30 seconds.* Two Models: Offers Pro and Light versions. The Light version is three times faster than Pro but compromises on quality, suitable for quick validation.* Image Requirement: Must use a facial close-up image; otherwise, the model cannot recognize the head and lips. #### Main Workflows: The following workflows are introduced to meet different application scenarios: 1. Basic Workflow * The simplest version, containing only 6 core nodes. 2. Voice Cloning Digital Human * Allows you to upload an image and reference audio to clone the voice and drive the digital human. 3. Voice Preset Digital Human * Similar to cloning, but uses pre-set voices within the workflow, eliminating the need for user uploads. 4. Sound Design Digital Human * Fully Automatic Workflow: You only need to upload an image. The model analyzes the image via a VQA prompting node, automatically generates a voice prompt, and then a TTS model designs and generates the sound based on that prompt. #### Summary: Overall, the Flash Head series of workflows performs well in scenarios that demand ultimate speed (such as real-time interaction, rapid prototyping) and are "worth trying out." However, there is still a gap in generation quality and stability compared to more mature solutions like Infinite Talk, so currently, they are "not recommended for productivity."
Xiaohongshu FireRed-Image-edit1.1 Image Editing (Makeup Transfer)
You can click on the link below and try it out directly. If the effect is good, you can deploy it locally https://www.runninghub.cn/post/2029575093277433858/?inviteCode=rh-v1121Fan benefits, register to get 1000 points, daily login 100 points, play 4090! Experience the super power of 48G!B-site video: https://www.bilibili.com/video/BV17uP1zUEJdKnowledge Planet: https://t.zsxq.com/7F90AGraphic tutorial: https://t.zsxq.com/lNK8vThe workflow is very simple to use, and I have uploaded a tutorial on how to use it on YouTube. You can try it out and follow the tutorial and Civitai's workflow to easily reproduce the results. Thank you. If you have any other questions, please leave a comment in the comment sectionThis tutorial is very important, not watching it will surely lead to a lot of problems! Watch the video tutorial before trying ！！！！ Very important! Especially a key point at the end!
FireRed Image Editor 1.1 Dual-Image Editing (Costume Swap)
You can click on the link below and try it out directly. If the effect is good, you can deploy it locally https://www.runninghub.cn/post/2029575073815863298/?inviteCode=rh-v1121Fan benefits, register to get 1000 points, daily login 100 points, play 4090! Experience the super power of 48G!B-site video: https://www.bilibili.com/video/BV17uP1zUEJdKnowledge Planet: https://t.zsxq.com/7F90AGraphic tutorial: https://t.zsxq.com/lNK8vThe workflow is very simple to use, and I have uploaded a tutorial on how to use it on YouTube. You can try it out and follow the tutorial and Civitai's workflow to easily reproduce the results. Thank you. If you have any other questions, please leave a comment in the comment sectionThis tutorial is very important, not watching it will surely lead to a lot of problems! Watch the video tutorial before trying ！！！！ Very important! Especially a key point at the end!
Xiaohongshu FireRed-Image-edit1.1 Single Image Editing (Portrait Consistency)
You can click on the link below and try it out directly. If the effect is good, you can deploy it locally https://www.runninghub.cn/post/2029575073815863298/?inviteCode=rh-v1121Fan benefits, register to get 1000 points, daily login 100 points, play 4090! Experience the super power of 48G!B-site video: https://www.bilibili.com/video/BV17uP1zUEJdKnowledge Planet: https://t.zsxq.com/7F90AGraphic tutorial: https://t.zsxq.com/lNK8vThe workflow is very simple to use, and I have uploaded a tutorial on how to use it on YouTube. You can try it out and follow the tutorial and Civitai's workflow to easily reproduce the results. Thank you. If you have any other questions, please leave a comment in the comment sectionThis tutorial is very important, not watching it will surely lead to a lot of problems! Watch the video tutorial before trying ！！！！ Very important! Especially a key point at the end!
FireRed-Image-edit1.1 + Dual Image Editing (Text Layout Reference)
You can click on the link below and try it out directly. If the effect is good, you can deploy it locally https://www.runninghub.cn/post/2029575105692573697/?inviteCode=rh-v1121Fan benefits, register to get 1000 points, daily login 100 points, play 4090! Experience the super power of 48G!B-site video: https://www.bilibili.com/video/BV17uP1zUEJdKnowledge Planet: https://t.zsxq.com/7F90AGraphic tutorial: https://t.zsxq.com/lNK8vThe workflow is very simple to use, and I have uploaded a tutorial on how to use it on YouTube. You can try it out and follow the tutorial and Civitai's workflow to easily reproduce the results. Thank you. If you have any other questions, please leave a comment in the comment sectionThis tutorial is very important, not watching it will surely lead to a lot of problems! Watch the video tutorial before trying ！！！！ Very important! Especially a key point at the end!
[AI Digital Human] how to make a EMO radio show: Flash Talk Digital Human shares three workflows, twice as fast as Infinite Talk!
For a detailed video introduction, results showcase, and step-by-step tutorial, please refer to the video below: Click to experience the workflow online. Open Source Address: https://github.com/Soul-AILab/SoulX-FlashTalk Workflow: AA--FlashTalk Digital Human - Twice the SpeedExperience Link: https://www.runninghub.ai/post/2030296318643544066/?inviteCode=rh-v1401 Workflow: AA--FlashTalk Digital Human + Qwen TTS Voice Presets - Generate Digital Anchor from One ImageExperience Link: https://www.runninghub.ai/post/2029760205487087617/?inviteCode=rh-v1401 Workflow: AA--FlashTalk Digital Human + Voice Cloning (Emotion Control)Experience Link: https://www.runninghub.ai/post/2029525701015117826/?inviteCode=rh-v1401 Workflow: AA--Chasing Closed-Source Consistency! Xiaohongshu FireRed-Image-Edit 1.1 Image Editing + Upscaling RestorationExperience Link: https://www.runninghub.ai/post/2029824095113711618/?inviteCode=rh-v1401 I built three ComfyUI workflows for different scenarios using the Flash Talk real-time digital human model, aiming to solve the problem of quickly generating high-quality lip-sync videos from text or audio. Here is a core introduction to these three workflows: Original Basic Workflow:This is the simplest version. The model is encapsulated within the pipeline, requiring no prompt. Just upload a reference image and driving audio, set the resolution (note that resolutions above 540p require VRAM adjustments), and directly generate a lip-sync video. Its running speed is about twice as fast as Infinite Talk, making it ideal for quickly validating results. Qwen TTS + Voice Preset Workflow:This combination suits scenarios needing quick voiceovers. I utilized Qwen TTS's nine preset voices (covering Chinese, English, Japanese, Korean, and dialects), eliminating the hassle of finding external voiceovers. Paired with Flash Talk, it enables rapid conversion from text to digital human video. F5-TTS + Voice Cloning + Emotion Control Workflow:This is my most frequently used advanced version. By uploading original audio for cloning and combining it with eight emotion parameters (such as adjusting Happy and Sad values), you can finely control the emotional tone of the output speech. This is particularly suitable for creating videos with delicate emotional radio styles, like the "Queen of the Kingdom of Women." Additionally, when creating the intro for Queen of the Kingdom of Women – Emotional Radio, I used the FireRed image editing workflow to restore and redraw the character from a blurry screenshot, ensuring consistency and clarity. Finally, the video was upscaled to 4K resolution using an upscaling workflow. The core advantages of this solution are: fast speed, no complex prompts required, simple nodes, and the ability to endow characters with rich emotional expression through voice cloning. As mentioned in the video, technological progress simplifies expression, but to create truly moving content, we still need a deep understanding of emotions and characters.
Multi-Model Workflow - Qwen 2511 + Wan 2.2
Here's an interesting little T2I/I2I tool for you guys...This is a multi-model workflow for Qwen ImageEdit 2511 and Wan 2.2 T2V.So, like... why? 😂This workflow is surprisingly useful. I picked Qwen 2511 and Wan 2.2 very purposefully.Qwen 2511 is highly responsive to both text prompts, edits, character positioning, camera positioning, character consistency, and perhaps most importantly, allows huge malleability of a reference image. 🧈Wan 2.2 T2V is exceptional at detailing and already functions as a 2-model generator, with high noise for composition and low noise for detail. So basically, we're just replacing the Wan high noise model with Qwen to take advantage of both model's strong suits.Qwen sets the composition up, Wan knocks it down. 🎳This can be used in a number of ways. It can be used to detail a pre-existing image, it can be used to set-up character or scene composition with follow through to Wan detail, it can be used as a creative tool to generate multiple-style iterations of one image (this is made easier with LoRAs).This workflow has a lot going on. I've set up an options panel to help turn on an off relevant features, as well as a lot of explanation in the workflow itself. To run you through some of the options:Reference Image - can be turned on and off for either I2I or T2IQwen/Wan Specific Prompting - target only the relevant model with certain detailsWan NAG - tell Wan what to focus onPatch SageAttention - if you're using Sage, Qwen does not like this and it needs to be turned off as a global setting and toggled in workflowReactor FaceSwap - to reinforce Wan's character consistencyI've also included notes in the workflow to help you achieve your desired result - how to modify settings to either achieve increased consistency or increased divergence.This one is pretty cool, I use it a lot. Great if you're not sure what you want and want to play with both minimalistic and high detail options. I think I'll make some variants with Pony/ZiT/SDXL as base also.(Note - I use Qwen ImageEdit 2511 w/Lightning Lora embed, which means if you're using a different base model of Qwen ImageEdit you may want to download the 4-step LoRA to keep the steps/speed consistent.)Enjoy! 😊🥳
QWEN3 TTS-Voice Cloning/Creating Sound-Integrated Version
Video Tutorial:How to use:Select the corresponding mode and fill in the corresponding contentWe recommend trying it out online before downloading; it's completely free!Online experience address:https://www.runninghub.ai/post/2028677768417906689/?inviteCode=rh-v1058
Qwen Image 2512 workflow+ SageAttention + GGUF + MultiGPU
My TG Channel - https://t.me/StefanFalkokAIMy TG Chat - https://t.me/+y4R5JybDZcFjMjFiHi! I introduce my working workflows with Qwen Image 2512 generating image for ComfyUI.I have included workflow with Qwen Image modelsYou need to have Qwen Image Model (https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/tree/main/split_files/diffusion_models), clip (https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/tree/main/split_files/text_encoders) and vae (https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/tree/main/split_files/vae)GGUF Qwen Image 2512- https://huggingface.co/unsloth/Qwen-Image-2512-GGUF/tree/mainIf you need 4steps and 8steps loras for Qwen Image - https://huggingface.co/lightx2v/Qwen-Image-Lightning/tree/mainAlso 4steps and 8steps lora from https://civitai.com/models/2046883?modelVersionId=2318885download my ComfyUI Build https://huggingface.co/datasets/StefanFalkok/ComfyUI_portable_torch_2.10.0_cu130_cp313_sageattention_triton/tree/main, also u need download and install CUDA 13.0 (https://developer.nvidia.com/cuda-13-0-0-download-archive) and VS Code (https://visualstudio.microsoft.com/downloads/)Leave comments if you have trouble or you found the problem with workflows
How to Create Multilingual Shopping Videos with Digital Avatars? Complete Workflow pack
These workflows are the tools I use to create multilingual e-commerce digital human promotion videos. The entire production process is online, free, and easy to follow. For details on the sequence and specifics of using these workflows, please refer to the detailed introduction in the video below: Brief Tutorial: 1. Use Zimage to generate model images and scene images that match the appearance of people in foreign regions, as well as scene images that fit the product and setting.2. Use Qwen image edit 2511 to merge the model, product, and scene images to generate the base image for the digital human.3. Use Qwen TTS to generate advertising voiceovers in foreign languages.4. Use the infinite talk digital human workflow to generate lip-sync videos of the digital human.5. Upscale the resulting digital videos to obtain 4K ultra-high-definition lip-sync videos of the digital human. The storyboard design workflow uses Gemini Flash to generate scene prompts and image editing prompts.The image processing workflow is used to upscale images to 4K to add details. Tutorial: How to Get Digital Human Avatars for Free Log in to Runninghub and get 1,000 RH credits. Enter the invite code rh-v1401 to get another 1,000 RH credits. Because RH standard mode consumes 12 credits per minute when running, generating one image with Zimage costs only 12 RH credits, and a 30-second voice clip also costs around 12 credits. Using Qwen2511 for image editing takes about 3 minutes per image and costs around 40 credits. Running a 30-second digital human video takes 30 minutes and consumes about 360 credits. Upscaling a 30-second video in segments takes about 35 minutes and consumes about 420 credits. When generating digital human videos, write good prompts. Simple character movements and dynamic background descriptions usually work on the first try, and upscaling also generally works on the first try (for long videos over 20 seconds, there is a very small chance that a few seconds may be cut off, so it's recommended to generate in 15-second segments for upscaling). Cost for creating a four-scene English digital human shopping video (final length: 50 seconds):Zimage image generation: 1 model image, 4 scene images; approximately 60 credits.Qwen2511 image editing: 6 times; approximately 216 credits.Voice generation: 4 segments, about 4 minutes total; approximately 50 credits.Digital human videos: 4 segments totaling 50 seconds; approximately 600 credits.Upscaling 4 video segments: approximately 720 credits.Total around 1,650 credits. The 2,000 free credits are more than enough! If you're just practicing with digital humans, you can skip upscaling the videos, which allows you to experiment many more times!!
Extreme HD Repair - Add Details - Eliminate AI Qwen2511+Flux kontext
Low memory 12G can be run, below 12G click the connection below to run directly, and 12G video memory operation tutorial is includedYou can click on the link below and try it out directly. If the effect is good, you can deploy it locallyhttps://www.runninghub.ai/post/2022083211601387522?inviteCode=rh-v1500Fan benefits, register to get 1000 points, daily login 100 points, play 4090! Experience the super power of 48G!
Teleport
🌀 Teleport – Move Your Selfie Through Space, Time & StyleTeleport is a ComfyUI node that lets you move your selfie (or any character image) into different locations, eras, and cinematic styles — while preserving identity, pose, and composition.Upload your photo.Choose a destination.Select a style or era.Model Stack: Optimized for Flux.2 Klein 9B Distilled (Base) and Qwen3-VL-4B-Instruct (Vision Analysis).· Z-Image Resolution Picker: This workflow requires the Z-Image Resolution PickerEnrico's Nodes: This workflow requires the ComfyUI-enricos-nodes suite for advanced processing.Precision Control: Powered by CompositorV3 for perfect subject placement.🛠 How to Use (Interactive Step):This workflow features an interactive pause to give you full artistic control over the composition:Initial Run: Start the queue. The process will automatically analyze your image and prepare the cinematic prompt.The Pause: The workflow will stop at the CompositorV3 node.Manual Composition: * Adjust the position and scale of your subject on the canvas.This is crucial for Outpainting — you decide exactly where the character stands in the new cinematic environment.Resume: Click the "Continue" button at the bottom of the CompositorV3 node to finish the generation.6. And your character will be seamlessly integrated into a new world. From ancient temples to cyberpunk streets.From film noir to VHS retro.From realistic photography to stylized aesthetics.✨ What It DoesChanges the background onlyKeeps your character’s identity and facial structurePreserves pose, framing and perspectiveAllows lighting to naturally interact for realistic integrationCan change facial expressionCan apply lens distortion (fisheye strength)Supports cinematic enhancement modesThis is not a simple background swap.It builds a structured transformation instruction for Qwen → Flux to generate coherent, physically believable results.🎛 Settings Explained📍 DestinationSelect a location from locations.json.You can:Expand the list by editing data/locations.jsonOr override everything using Custom BackgroundExamples:Ancient temple corridorsAbandoned amusement parkCyberpunk alleyDesert western townRetro 80s VHS interior🧠 Mode (Very Important)This setting controls how aggressively the environment is interpreted.1️⃣ Replace ExactlyStrict background replacementNo major new elements addedBest for controlled resultsMinimal artistic interpretationUse this if you want predictable output.2️⃣ Adapt CreativelyUses the location as a baseAdds subtle atmospheric detailsSlight creative expansionStill respects original descriptionBalanced mode for most use cases.3️⃣ Cinematic EnhanceAdds dramatic lightingAtmospheric depthParticles, haze, spatial contrastStronger cinematic feelingLighting can naturally interact with the character for seamless integration.Best for:Movie-poster styleEpic fantasyHigh-drama scenes🎭 EmotionReplace the character’s facial expression while preserving identity.Examples:HappySadFuriousMischievousThoughtfulShySuspiciousAwkwardThis only affects expression — not structure.🎬 Style EraApply predefined visual aesthetics:Film Noir (1940s)Disco 70sVHS 80sCyberpunk FutureSpaghetti WesternPolaroid VintageAnime StyleIf no era is selected, default mode uses realistic full-frame digital photography.📷 Fisheye StrengthControls lens distortion:0.0 → No distortion0.3 → Subtle wide-angle0.6 → Noticeable fisheye1.0 → Extreme GoPro-style curvatureOnly affects perspective — not pose.🛠 Custom BackgroundOverride the selected location completely.Write your own environment description here.Example:"Narrow corridor with towering stone columns covered in ancient glyphs, golden dust in the air, dramatic top lighting"➕ Extra InstructionsAdd advanced tweaks:Depth of field adjustmentsColor grading preferencesSpecific lighting behaviorComposition hints💡 Use CasesTeleport your selfie into historical erasCreate cinematic key visualsGenerate fantasy character scenesMake retro VHS memoriesTurn casual photos into movie stillsTest lens distortion effectsEmotion-swapping portraits🔧 For Advanced UsersAdd new locations inside data/locations.json🚀 SummaryTeleport is a structured prompt-engineering node designed for:Character preservationControlled environment replacementCinematic enhancementStyle-era transformationsSelfie-to-scene teleportationMove through space.Move through time.Stay yourself. InstallationDownload the ZIP file from this page.Unzip the downloaded file.Move the ComfyUI-Teleport folder into your ComfyUI/custom_nodes/ directory.Restart ComfyUI.(No external dependencies or pip install required).