ComfyUI Workflows
Browse real AI workflows from Flux, SDXL, Pony, ControlNet, video generation and commercial AI creators.
Z-Image Split-Sigma Workflow: ZiB to ZiT
Z-Image 5.0 EmberFrame Split-Sigma Workflow GuideFirst of all, a quick thank you again to the people behind ComfyUI, CivitAI, and all of the creators making nodes, checkpoints, LoRAs, workflows, tools, and helper scripts, then sharing them for everyone else to use. Without that community, this kind of workflow would not be possible, and it definitely would not be anywhere near as fun to build.This is the 5.0 EmberFrame update to my Z-Image Dual-Stage workflow. It is still very much the same basic idea: use ZiB for the first image pass, then use ZiT as the refiner. The difference with 5.0 is mostly cleanup, usability, and making the public release line up properly, but I also want to be clearer about how the dual-stage part is actually working.So this is not a huge new experimental release. It is a small, practical update. The workflow is cleaner, the helper nodes now live on GitHub, and the version jump to 5.0 fixes a mismatch between my local files, the GitHub repo, and the CivitAI version naming.Older 4.0 and 4.6 versions may still be available on this page for reference, but 5.0 is the version I recommend using now. It is the one I expect to build on going forward.The short version is this· You can still use the workflow very simply by writing a normal prompt.· ZiB / Base handles the first composition pass.· ZiT / Turbo handles the lower-sigma refinement pass.· This is a split-sigma workflow, not just a normal two-KSampler chain.· The main controls are gathered into the Global subgraph.· The Prompt / Prompt Builder can be used as a plain prompt box or as a wildcard prompt system.· Prompt enhancement is still available using Gemma4.· Image-to-image, ControlNets, post-processing, and upscaling are still optional.· The EmberFrame helper nodes are now in a public GitHub node pack instead of being treated as loose local extras.What this workflow doesThis workflow includes:· Text-to-image generation· Image-to-image· A dual-stage ZiB to ZiT generation path· Built-in ControlNet sections· A Prompt / Prompt Builder subgraph· Sequential and random wildcard prompt tools· Prompt enhancement using Gemma4· Prompt enhancement from a loaded image· Optional LG Noise Injection· Optional post-processing· Optional upscaling with SeedVR2 and Ultimate SD Upscale· CivitAI-friendly image saving and metadataImportant first step: custom nodes requiredThere are still a few custom node packs required. The easiest route is to install what you can through ComfyUI Manager. For the EmberFrame helper nodes, use the GitHub repo directly if it has not appeared in your Manager list yet.· EmberFrame Nodes: https://github.com/emberframe/emberframe-nodes - required for the new helper nodes· rgthree-comfy: https://github.com/rgthree/rgthree-comfy· ComfyUI-Image-Saver: https://github.com/alexopus/ComfyUI-Image-Saver· ComfyUI-Impact-Pack: https://github.com/ltdrdata/ComfyUI-Impact-Pack· ComfyUI-LG_SamplingUtils: https://github.com/LAOGOU-666/ComfyUI-LG_SamplingUtils· ComfyUI_essentials: https://github.com/cubiq/ComfyUI_essentials· ComfyUI-Easy-Use: https://github.com/yolain/ComfyUI-Easy-Use· ComfyUI-KJNodes: https://github.com/kijai/ComfyUI-KJNodes· RES4LYF: https://github.com/ClownsharkBatwing/RES4LYF· ComfyUI ControlNet Aux: https://github.com/Fannovel16/comfyui_controlnet_aux· ComfyUI-DepthAnythingV3: https://github.com/PozzettiAndrea/ComfyUI-DepthAnythingV3· ComfyUI-SeedVR2_VideoUpscaler: https://github.com/numz/ComfyUI-SeedVR2_VideoUpscaler· ComfyUI-Unload-Model: https://github.com/SeanScripts/ComfyUI-Unload-Model· ComfyUI_UltimateSDUpscale: https://github.com/ssitu/ComfyUI_UltimateSDUpscale· WAS Node Suite - Revised: https://github.com/ltdrdata/was-node-suite-comfyui· comfyui-vrgamedevgirl: https://github.com/vrgamegirl19/comfyui-vrgamedevgirlFor a manual install, custom nodes normally go here:ComfyUI/custom_nodes/For a Windows portable install, that is usually:ComfyUI_windows_portable/ComfyUI/custom_nodes/After installing or updating node packs, restart ComfyUI before loading the workflow.Model setupYou will still need to point the workflow at your own ZiB and ZiT model files. The Global subgraph is the first place to check after loading the workflow.· Base Model: your Z-Image Base / ZiB model· Turbo Model: your Z-Image Turbo / ZiT model· CLIP/text encoder and VAE: set these to the files used by your local Z-Image setup· Gemma4 text encoder: only needed if you use the Prompt Enhancer· ControlNet, SeedVR2, and upscale models: only needed if you use those optional sectionsThe cleaned example path style is:Z-Image\Base\z_image_base_model.safetensorsZ-Image\Turbo\z_image_turbo_model.safetensorsMain workflow overviewThe main graph is meant to be readable from left to right. The big pieces are deliberately grouped so you can start simple, then enable the extras only when you need them.· Image Saver· ZiB LoRA Loader· ZiT LoRA Loader· Prompt From Image· Prompt / Prompt Builder· Prompt Enhancer· Negative Prompt· Image 2 Image· Latent Switch· Global· LG Noise Injection· Post Processing· Optional ControlNets· Optional upscaling sectionsA note on the split-sigma approachThis is the part that is easy to miss if you only look at the workflow from a distance. It can look like a standard double KSampler setup, where one model makes an image and a second model refines it afterward. That is close enough to understand the broad idea, but it is not quite what this graph is doing.In this workflow, the scheduler creates the sigma schedule for the generation, then the workflow splits that schedule into two parts. The higher-sigma portion is used for the early, noisier part of the image formation, and the lower-sigma portion is used for the later refinement part. So instead of running two totally separate sampling jobs, the two Z-Image models are sharing one staged denoising plan.Practically, I think of it like this: ZiB gets the first pass where the image is still loose and structural, then ZiT takes over when the latent is already partly formed and the remaining work is more about polish, detail, and coherence. The handoff is controlled by the sigma split rather than just by sending a finished image or finished latent into a second sampler.In the Global subgraph, the chain is:1. BasicScheduler creates the full sigma schedule.2. SplitSigmas separates that schedule into high-sigma and low-sigma portions.3. Sigmas Resample reshapes the first-stage sigma path used by the ZiB pass.4. The first SamplerCustomAdvanced pass uses the ZiB model and the resampled high-sigma section.5. The second SamplerCustomAdvanced pass uses the ZiT model and the low-sigma section, using the latent from the first pass.I do not want to oversell the theory here. I am explaining it in practical workflow terms rather than pretending this is a finished technical paper. The important point for users is that the workflow is trying to let ZiB and ZiT do different parts of the same denoising journey, rather than simply stacking a generic refiner on top of a completed first result.Why use this approach? In my testing, it gives a useful blend of the two models' strengths. ZiB tends to be the more inventive composition model, so I like letting it handle the early, looser part of the generation. ZiT tends to bring the cleaner finish, so I like letting it take over for the later, lower-sigma refinement work.The main drawback is that you are still asking ComfyUI to work with two models instead of one. If your system can handle that, the tradeoff has been worth it for me: the workflow can run at a relatively low step count, not as low as a straight ZiT-only workflow, but usually not as high as I would expect from a ZiB-only setup either.As always, I am standing on the shoulders of much smarter people here. This workflow is my practical way of wiring the pieces together, not a claim that this is the only correct way to use these models.1. Global subgraphMost of the controls I reach for first are in the Global subgraph. This is where you set the two model paths, sampler, scheduler, seed behavior, aspect ratio, orientation, megapixels, batch size, and the core generation values.In simple terms:· ZiB / Base handles the first composition pass.· ZiT / Turbo handles the lower-sigma refinement pass.· Split-sigma generation lets the two stages do different jobs inside one staged denoising path.For a first run, I would keep the defaults close to the included setup and only change the model paths, aspect ratio, and orientation. Once you know it is running, then start playing with the rest.2. Prompt and negative promptYou can use the workflow with a normal prompt. You do not have to use wildcards, prompt enhancement, image prompting, or any of the extra toys on your first run. A plain positive prompt plus the included negative prompt is enough to test that everything is connected.The negative prompt is deliberately fairly general. It is not magic, but it gives the workflow a useful starting bias against low-quality image problems, text, logos, watermarks, and obvious anatomy issues.3. Prompt / Prompt Builder subgraphThe Prompt / Prompt Builder subgraph can be used in two ways: as a simple prompt box, or as a wildcard prompt builder. If you just want to type a prompt and run the workflow, use it like a normal prompt field. That is still completely valid.The wildcard side is there for people who like building batches of related ideas without rewriting the whole prompt every time. The system is made from three main parts:· Wildcard Rule Builder nodes· Wildcard Config Combiner· Wildcard Prompt AssemblerEach Rule Builder controls one source file and one token name. The Config Combiner gathers those rules together, and the Prompt Assembler turns the final resolved values into a usable prompt.A basic token prompt might look like this:{subject} in {location}, {lighting}, {camera_angle}A good way to test it is to enable one Rule Builder, run once, check the preview, then add another rule. Build it up slowly and it becomes much easier to see what each wildcard source is doing.4. Prompt From Image and Prompt EnhancerThe Prompt Enhancer is still built around Gemma4. You can feed it a written idea, an image-guided idea, or a simple subject, then let it expand that into a more complete prompt structure.The flow is basically:6. Write a simple prompt or build one with the Prompt Builder.7. Optionally load an image into Prompt From Image.8. Let Gemma4 enhance the prompt or image-guided idea.9. Send the enhanced text onward to the positive prompt conditioning.The Prompt Enhancer subgraph includes a Gemma4 CLIP loader, the TextGenerate node, the main enhancement instruction, and preview output for the generated prompt. The required Gemma4 text encoder should be placed in your ComfyUI text encoder folder.ComfyUI/models/text_encoders/gemma4_e4b_it_fp8_scaled.safetensors5. LoRAsThe workflow still includes two LoRA loaders in the main graph: one for Z-Image Base and one for Z-Image Turbo. You can leave them empty for a first test. Once the base workflow is generating correctly, add LoRAs back in one at a time so it is obvious what changed.6. Image-to-Image subgraphThe Image-to-Image section is still there for img2img. Load your source image, then use the Latent Switch to choose the image latent path instead of the empty latent path.Important:· Input 1 is the img2img path.· Input 2 is the empty latent path.· When using img2img, reduce the denoise value in the Global subgraph.For anime-to-realistic conversion, I usually start by loading the same reference into Image 2 Image and Prompt From Image, then let the enhancer build a more descriptive prompt. After that, the useful work is mostly denoise, prompt wording, and deciding whether ControlNet is helping or getting in the way.7. ControlNet subgraphThe ControlNet subgraph is optional. If you are doing a simple text-to-image test, leave it alone at first. Once the main workflow is behaving, you can enter the ControlNet subgraph and enable the groups you want to use.The built-in options include:· Depth Anything V3· Depth Anything V2· Canny· OpenPoseFor anime-to-realistic conversions, I usually try depth first. It gives the model structural guidance without forcing every line of the original image to survive.8. Post Processing subgraphThe post-processing subgraph is intentionally simple. It adds a little sharpen and a little film grain. Some images benefit from that. Some images look better without it. Treat it as a finishing option, not a required part of the generation.9. Upscaling optionsThere are still two optional upscaling paths. I would not turn either of them on for the very first test run. Make sure the base workflow works first, then add upscaling when you are happy with the generation path.SeedVR2 UpscaleSeedVR2 can be a very nice final enlargement path, but it has its own model requirements and settings. Use it after the base image is already working.Ultimate SD UpscaleUltimate SD Upscale is the second optional upscale route. It is still useful when you want a more traditional tiled upscale pass with familiar controls.10. Image saving and metadataThe Image Saver setup is one of the parts I care about most, because it makes the workflow easier to share, debug, and revisit later. The saved data is set up to include the important generation details where possible.That includes:· Checkpoints used· LoRAs used· Prompt-related information· Image dimensions· Sampler and scheduler information· The seed usedRecommended starting workflowIf you are loading the workflow for the first time, I would suggest this order:1. Install the required custom node packs.2. Install or update EmberFrame Nodes from GitHub, or from ComfyUI Manager if it is already listed for you.3. Restart ComfyUI.4. Load the workflow.5. Set your ZiB and ZiT model paths in the Global subgraph.6. Set your CLIP/text encoder and VAE paths.7. Leave LoRAs empty for the first test.8. Set aspect ratio, orientation, and megapixels in the Global subgraph.9. Write a simple prompt in Prompt / Prompt Builder.10. Run a basic text-to-image test first.After that, enable extras one at a time:· Prompt Enhancer· Prompt From Image· Wildcard Prompt Builder nodes· Img2Img· ControlNets· Post Processing· SeedVR2 Upscale· Ultimate SD UpscaleA few final notesThere are a lot of moving parts in this workflow. I know that. The best way to learn it is to start with the plain text-to-image path, then add one section at a time as you need it.Version 5.0 is mainly about making that experience cleaner and easier to maintain. The workflow naming is now lined up, the helper nodes are on GitHub, and this should be a more sensible starting point for anyone downloading it fresh.Most importantly: please enjoy it. I hope it helps you make cool things.If you use it, remix it, break it, improve it, or make something interesting with it, I would genuinely love to see what you make.Thanks for checking it out, and happy generating.
ZIB-GGUF-dAIver-v1.5
Optimized Low-VRAM Workflow for Z-Image-Base (GGUF)Refined version of my ZIT workflow, modified to work with ZIBThis workflow delivers the full power of Z-Image-Base in GGUF format, specially optimized for GPUs with less than 8 GB VRAM, like my RTX 4050 with only 6 GB. The pre-selected Z-Image-Fun-LoRA provides a noticeable speed boost with almost no quality loss. Two intelligent upscaling stages, optional automatic trigger-word integration via the Super LoRA Loader, and an extended save node complete this elegant setup.Version 1.5 (previous versions were only my personal use) brings significant improvements in speed, usability, and upscaling quality — while remaining extremely VRAM-efficient (tested on RTX 4050 with only 6 GB).What’s new in v1.5:selectLatentSizePlus — intuitive aspect-ratio and resolution selector with beautiful presets (including 7:12 Tall Vista and other golden-ratio-friendly options) plus easy orientation swapFull SEEDVR2 Video Upscaler Subgraph — powerful DiT-based (.safetensors or GGUF) high-end upscaler that delivers stunning 4K+ results with intelligent resolution handling, Lab color correction, and temporal settings. Works exceptionally well on still images too, producing superior detail and coherenceCLIP switched to the abliterated Qwen3-4B-Instruct-2507.Q5_K_S.gguf (lumina2 type)Improved workflow organization, expanded notes, and more robust saving optionsRequired Custom Nodes (updated for v1.5):ComfyUI-GGUF - https://github.com/city96/ComfyUI-GGUF - UnetLoaderGGUF + CLIPLoaderGGUFnd-super-nodes - https://github.com/HenkDz/nd-super-nodes - NdSuperLoraLoader with tags, trigger words & beautiful UIsave-image-extended-comfyui - https://github.com/thedyze/save-image-extended-comfyui - Advanced saving with metadata & dynamic filenamesComfyUi-MzMaXaM - https://github.com/MzMaXaM/ComfyUi-MzMaXaM - selectLatentSizePlusComfyUI-SeedVR2_VideoUpscaler - https://github.com/numz/ComfyUI-SeedVR2_VideoUpscaler - SEEDVR2 Video Upscaler SubgraphModels & Downloads (exact paths)The following list explains the base models I am most frequently using with this workflow. The list as well explains where to put each file after you downloaded it.1. Main Model (Z-Image-Base GGUF):File: Juggernaut_Z_V1_by_RunDiffusion_q6_k-004.ggufDownload: https://huggingface.co/RunDiffusion/Juggernaut-Z-Image/resolve/main/Juggernaut_Z_V1_by_RunDiffusion_q6_k-004.ggufTarget folder: ComfyUI/models/diffusion_models/2. Text Encoder (CLIP)File: Qwen3-4B-Instruct-2507-abliterated.Q5_K_S.ggufDownload: https://huggingface.co/mradermacher/Huihui-Qwen3-4B-Instruct-2507-abliterated-GGUF/resolve/main/Huihui-Qwen3-4B-Instruct-2507-abliterated.Q5_K_S.ggufTarget folder: ComfyUI/models/text_encoders/ (or clip/)3. VAEFile: ae.safetensors (~335 MB)Download: Usually included with Z-Image-Turbo setups or available here: https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/vae/ae.safetensorsTarget folder: ComfyUI/models/vae/4. Upscalers4× Upscaler: 4xLSDIRplusN.pth (variant of 4x-UltraSharp) → https://civitai.com/models/116225/4x-ultrasharp1× Skin-Contrast Upscaler: 1xSkinContrast-High-SuperUltraCompact.pth Download: https://huggingface.co/notkenski/upscalers/blob/main/1xSkinContrast-High-SuperUltraCompact.pthTarget folder: ComfyUI/models/upscale_models/5. Pre-loaded LoRA (included in workflow):File: Z-Image-Fun-Lora-Distill-2603_UDCAI_ComfyUI.safetensorsDownload: https://civitai.com/models/2362961/z-image-fun-distill-lora (UDCAI Edit version)Target folder: ComfyUI/models/loras/6. SEEDVR2 Models (for the new high-end upscaler – optional but recommended):DiT Model: seedvr2_ema_3b-Q8_0.ggufVAE: ema_vae_fp16.safetensorsDownload from the official ComfyUI-SeedVR2_VideoUpscaler repository or Hugging Face and place in the folders required by the custom node.Key Nodes & Their FunctionsNdSuperLoraLoader with automatic trigger-word detection and clean tag interfaceselectLatentSizePlus → effortless aspect-ratio and resolution controlKSamplerAdvanced with proven settingsTwo-stage classic upscaler (4× LSDIR + 1× Skin-Contrast) in its own subgraph, orNew SEEDVR2 Video Upscaler Subgraph → for ultimate quality (optional, easily bypassed)SaveImageExtended with full metadata and dynamic filenamesRecommended Settings (already set in the workflow)Sampler: euler, dpmpp_sde or res_multistepScheduler: simple, normal, beta or ddim_uniformSteps: 8–12 (with Z-Image-Fun-LoRA)CFG Scale: 1.0–1.5Shift: 4–7Resolution: 1024×1536 (portrait) – perfectly balanced for the golden ratio and typical ZIB outputs - the upscaler will automatically upscale by a factor of 4How to Use the WorkflowInstall all required custom nodesLoad the workflowEnter your positive prompt (the pre-loaded Fun-LoRA trigger words are handled automatically)Adjust negative prompt if neededChoose aspect ratio and resolution via the Size SelectorGenerate (Euler + Simple recommended)Optionally upscale with SEEDVR2 for cinematic resultsDone — beautifully saved with metadataSpecial thanks to @LumaRift who provided the SeedVR2 subworkflow and some good advise on optimizing my setup.
Lord Zolaris No Upscale Needed workflow
This is the workflow I have been using to generate images on z-Image models. Both Turbo and Base.Includes a custom node from my experimental pack on github:https://github.com/TripleHeadedMonkey/ComfyUI-Zlycoris/tree/mainModels used:https://civitai.com/models/2655305/lord-zolaris?modelVersionId=2983642https://civitai.red/models/2644869/solordz?modelVersionId=2969766https://civitai.com/models/2650573/thmuncz-imagete?modelVersionId=2976257https://civitai.com/models/2662833/zimagetesolordz?modelVersionId=2990083
【Z-IMG】INPAINT
✨ Z-IMG — Inpaint — Simple WorkflowA clean, all-in-one Z-IMG inpaint 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).🎯 FeaturesImage-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)📂for "base" version :Model : z_image_turbo_bf16.safetensorsin ComfyUI\models\diffusion_modelsCLIP : qwen_3_4b.safetensorsin ComfyUI\models\clip📂for GGUF version :Model : Q8, Q6, Q5, Q4, Q3in ComfyUI\models\unetCLIP : Q6in ComfyUI\models\clip📂CommonVAE : ae.safetensorsin ComfyUI\models\vae
【Z-IMG】IMG to IMG
✨ Z-IMG — Image to Image — Simple WorkflowA clean, all-in-one Z-IMG 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)📂for "base" version :Model : z_image_turbo_bf16.safetensorsin ComfyUI\models\diffusion_modelsCLIP : qwen_3_4b.safetensorsin ComfyUI\models\clip📂for GGUF version :Model : Q8, Q6, Q5, Q4, Q3in ComfyUI\models\unetCLIP : Q6in ComfyUI\models\clip📂CommonVAE : ae.safetensorsin ComfyUI\models\vae
Prompt and seed travel
I used to use A1111 extension called prompt-travel. I did not find anything like that for ComfyUI so decided to make it myself.This workflow allows for probably any image generation model since you can just toggle cfg and negative prompt on and off. Some models do not work without cfg and negative prompt while others can or do work without. Turning it on doubles time however.Settings:on top below the giant wall of text you have the CFG settings and to the left of it settings to video framerateThen there is positive and negative prompt. I decided that 4 is a good count there. More could be possible but I decided it to be unneeded. It would be possible to even make procedural but even now I have a headache thinking just thinking about the mathNext is timing setting below that. It is the meat of this. It allows 8 different curves and 4 time compression settings. From off to pow(5).Below that is image count. Self explanatery. It is set to 300 which can take a LOOONG time. Over an hour on a 5090. There should be ETA in console.To the left of it is Adaptive module settings. For the first time I recommand keeping it off. Turning it on requires at least 41 images. (did not test with less then 60 images)Below you can set seed for each point. Be carefull some curves just do not work with noise and will result in grey.Also on top you can select model and even exchange nodes when using model requiring more CLIP models or exchange it entirly for a checkpoint loader when you want to use SDXL or simillar.The right side:Be carefull. The right side can be laggy after generationg. On top you have your video output, on the bottom you have all imades it generates (the source of lag) and in the middle you have a graph which shows you the density of output. Does not show exactly reality when adaptive stages are off becouse the circullar corners.
【Z-IMG】TXT to IMG
✨ Z-IMG — Text to Image — Simple WorkflowA clean, all-in-one Z-IMG 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)📂for "base" version :Model : z_image_turbo_bf16.safetensorsin ComfyUI\models\diffusion_modelsCLIP : qwen_3_4b.safetensorsin ComfyUI\models\clip📂for GGUF version :Model : Q8, Q6, Q5, Q4, Q3in ComfyUI\models\unetCLIP : Q6in ComfyUI\models\clip📂CommonVAE : ae.safetensorsin ComfyUI\models\vae
Moody ZIB (Zimage Base) + ZIT (Zimage Turbo) Simple Workflow
I just made a new Telegram Group and Discord group drop in for chats, questions, feedbacks or just to share your work. 欢迎加入纸飞机和Discord群组。RH体验地址：https://www.runninghub.ai/post/2054639764392947713/?inviteCode=cptpovgo喜欢我的作品吗？🎨 我会持续免费分享更多优质内容✨，欢迎支持一下，请我喝杯咖啡吧！💖 Like my work? 🎨 I share all my work for free ✨, so support me by buying me a coffee 💖!Ko-fi: ko-fi.com/catlover1937Fanvue: https://www.fanvue.com/catlover1937Please do not redistribute without consent. 如非作者同意禁止搬运。2026-05-12🔥 ZIB + ZIT Simple Workflow V7.0 🔥V7.0 is finally here — and just like V6.0, this update shifts the balance in a BIG way. 🚀In this release, I’ve reworked the workflow to become significantly more ZIB-dominant, moving away from the older ZIT-heavy approach. Why?Because we now have a much more refined and sophisticated undistilled ZIB base — Moody Wild V4.0 — and the results are honestly stunning. 🌌✨The new balance allows the workflow to preserve far more natural detail, structure, lighting, and overall visual quality while remaining extremely flexible and easy to use.✨ What’s New in V7.0🔹 Solid preparation for the upcoming Moody Wild V4.0 release🔹 Noticeably improved body anatomy consistency (greatly because of wild V4)🔹 New manual detail anything node group🔹 Slightly faster generation times ⚡🔹 Multiple QoL improvements and workflow cleanup🔹 Better organization inspired by my newer workflow releases🔹 Overall smoother generation experienceThis update may look “simple” on the surface, but internally the workflow has been heavily refined for future scalability and image quality improvements. 👀Thank you all for the incredible support as always ❤️Have a fantastic day generating!🔥 ZIB + ZIT Simple Workflow V7.0 🔥ZIB 主导时代正式开启 ⚡V7.0 终于来了！这次更新和 V6.0 一样，再次大幅改变了整个工作流的平衡方式。🚀在这个版本中，我进一步重构了工作流，让它从过去偏向 ZIT 的结构，转变为更加 以 ZIB 为核心 的方案。为什么这么做？因为现在我们拥有了更加成熟、更加高级的未蒸馏 ZIB 基底 —— Moody Wild V4.0。而它带来的生成效果真的非常惊艳。🌌✨新的平衡方式能够在保持极高灵活性与易用性的同时，更好地保留自然细节、画面结构、光影表现以及整体视觉质量。✨ V7.0 更新内容🔹 为即将到来的 Moody Wild V4.0 做好了充分准备🔹 人体结构稳定性明显提升（很大程度归功于 Wild V4）🔹 全新手动涂抹Detail Anything 节点组🔹 生成速度略微提升 ⚡🔹 大量 QoL（体验优化）与工作流整理🔹 参考新版工作流进行更好的结构优化🔹 整体生成体验更加顺滑这个版本表面上看起来或许变化不算特别巨大，但实际上内部已经进行了大量深度优化，为未来的扩展性以及更高的画质提升打下了非常重要的基础。👀一如既往，非常感谢大家一直以来的支持 ❤️祝大家生成愉快！2026-04-20🔥 ZIB + ZIT Simple Workflow V6.0V6.0 is here — and this update shifts the balance in a big way.In this version, I’ve adjusted the workflow to be more ZIB-dominant instead of the previous ZIT-heavy approach. Why? Because we now have a much more sophisticated undistilled ZIB base — Moody Wild x EASON V3.0 — and it delivers absolutely insane results.This change brings:🎨 Richer color tones🧬 Higher image diversity✨ More expressive and dynamic outputsOn top of that, this update includes:⚙️ Minor quality-of-life improvements🐛 Bug fixes⚡ Optimized generation timeOverall, V6.0 is cleaner, faster, and pushes visual quality further while fully leveraging the strength of the ZIB base.Good luck and happy generating 🚀Your support means a lot ❤️🔥 ZIB + ZIT 简易工作流 V6.0V6.0 来了，这次是一次比较明显的方向调整。在这个版本中，我将整体工作流调成了更偏向 ZIB 主导，而不再像之前那样偏向 ZIT。原因很简单——现在我们有了更成熟、更强大的未蒸馏 ZIB 基底：Moody Wild x EASON V3.0，实际效果已经非常让人满意。这样的调整带来了：🎨 更丰富的色彩层次🧬 更高的图像多样性✨ 更有表现力和变化的画面输出同时，这次更新还包含：⚙️ 一些体验优化（QoL）🐛 Bug 修复⚡ 生成速度优化整体来说，V6.0 更干净、更高效，也更好地发挥了 ZIB 基底的潜力。祝您玩得开心，出图愉快 🚀感谢支持 ❤️2026-04-09V5.0 ZIB+ZIT Dual Ksampler EnhancedMinor updates on the newer techniques that delivery better results, also consistent with my other newer workflows.New upscaler used for further attempt to eliminate the side effect of SD upscaleQoL updates as my other newer workflows小更新：采用了一些较新的技术，可以带来更好的效果，同时也与我其他较新的工作流程保持一致。使用了新的放大器（SD upscaler），进一步尝试消除 SD 放大带来的副作用 （莫名的水渍，鸡皮疙瘩，油光）加入了一些使用体验上的优化（QoL），整体流程更加顺手，并与我其他较新的工作流保持一致2026-03-19V4.0 ZIB+ZIT Dual Ksampler EnhancedV4 Dropped. Some optimization and QoL update for Moody Wild V2 better support.Preview images generated by Moody Wild V2 BASE 45 STEPS undistilled + Moody Porn Mix V10V4 版本已发布。针对 Moody Wild V2 进行了一些优化和体验改进，以更好地支持 V2。预览图由 Moody Wild V2 BASE 45 STEPS undistilled + Moody Porn Mix V10 生成。V3.0 ZIB+ZIT Dual Ksampler EnhancedDownload URLs 下载地址:------------------------------------------------------------ZIB Model / Base 模型:https://civitai.com/models/2384856?modelVersionId=2714022（Warning！ Do not download the distilled model, look for the undistilled model）（注意！ 不要下错成蒸馏模型了 - 要下载未蒸馏模型 undistilled version)------------------------------------------------------------Avaible ZIT Models / 可选 Turbo 模型：https://civitai.com/models/620406/moody-porn-mixhttps://civitai.com/models/621441/moody-real-mixhttps://civitai.com/models/2384856/moody-wild-mix?modelVersionId=2681752------------------------------------------------------------upscaler model / 放大模型：4x-ultrasharp:https://openmodeldb.info/models/4x-UltraSharp1xSkinContrast-SuperUltraCompact:https://openmodeldb.info/models/1x-SkinContrast-SuperUltraCompactBefore getting into V3, here’s a quick recap of the foundation:The core idea behind this workflow has always been combining Zimage Base (ZIB) and Zimage Turbo (ZIT) to leverage the strengths of both models.Why ZIB + ZIT?ZIB excels at prompt adherence, composition stability (especially at low resolution), and seed-to-seed variation. It’s ideal for locking in structure and layout early.ZIT is significantly faster and strong at refining details, making it perfect for the finishing pass.Both models share the same CLIP, VAE, and prompt ecosystem, allowing seamless switching without reloading major components.Evolution So FarV1 used a two-pass KSampler setup:ZIB for base composition → ZIT for refinement.V2 upgraded to dual Advanced KSamplers, reducing generation time by ~33% while maintaining composition strength and prompt accuracy.The result is a workflow that balances:Strong structure (ZIB)Fast refinement (ZIT)Efficient pipeline integrationV3 builds on this foundation.V3 – 简要回顾 & 为什么选择 ZIB + ZIT在介绍 V3 之前，先快速回顾一下整个工作流的核心思路。Before diving into V3, here’s a quick recap of the foundation behind this workflow.整个流程的核心一直都是将 Zimage Base（ZIB） 与 Zimage Turbo（ZIT） 结合使用，发挥两者各自的优势。The core idea has always been combining Zimage Base (ZIB) and Zimage Turbo (ZIT) to leverage the strengths of both models.为什么 ZIB + ZIT？ZIB 的优势：更强的提示词遵循度低分辨率下更稳定的构图能力不同种子之间更高的变化多样性→ 非常适合在第一阶段锁定整体结构与画面布局ZIT 的优势：生成速度更快擅长细节刻画与高分辨率精修→ 适合在第二阶段完成细节和质感强化同时，ZIB 与 ZIT 共享相同的 CLIP、VAE 与提示词体系，可以无缝衔接，无需反复加载大型组件，使整个流程更加轻量、高效、统一。版本演进V1：使用双 KSampler 两段式流程ZIB 负责构图 → ZIT 负责细化V2：升级为双 Advanced KSampler 架构在保持构图与提示词强度的前提下，生成时间缩短约 33%整体思路始终围绕三点：ZIB 负责结构ZIT 负责速度与细节两者结合实现效率与质量的平衡V3 正是在这个基础上继续优化与进化。V2.0 Dual KsamlersThis is a classic dual Advanced KSampler build using ZIB + ZIT. Huge thanks to the AI artists in the group for sharing their experiments — based on that, I’ve reworked the flow to use a dual Advanced KSampler instead of the old two-pass KSampler setup.The good stuff 🔥V2 cuts generation time by roughly ~33%Still hits strong prompt adherence while keeping most of ZIT’s speedImage composition is still on par with V1The trade-offs ⚖️Preview is now a noisy latent only (V1 let you see rough image details on the first pass)You’ll get a bit more ZIT and slightly less ZIB detail compared to V1, but ZIB’s composition strengths are still very much thereAlso tossed in some misc fixes to clean up a few of my earlier mistakes 😅这是一个经典的双 Advanced KSampler 架构，使用 ZIB + ZIT。非常感谢群里的 AI 艺术家们分享他们的实验成果——在此基础上，我重新调整了流程，改用双 Advanced KSampler，取代了旧的两次 KSampler方案。亮点 🔥V2 将生成时间缩短了大约 33%在保持 ZIT 速度优势的同时，依然具备很强的提示词贴合度图像构图效果仍然与 V1 持平取舍 ⚖️预览现在只显示带噪的 latent（V1 在第一遍就能看到大致的图像细节）相比 V1，ZIT 成分稍多、ZIB 细节略少，但 ZIB 在构图方面的优势依然非常明显另外还顺手修了一些杂项问题，清理了我之前的一些小失误 😅V1.0 2 Ksampler PassesAll showcased result images are arranged as: ZIB only | ZIB + ZIT | ZIT only.I’ve been experimenting with a new workflow lately. With the release of Zimage Base (ZIB), it introduces several advantages that Zimage Turbo (ZIT) doesn’t currently offer—most notably stronger prompt adherence, better compositional stability at low resolutions, and significantly higher variation when iterating with different seeds. These strengths make ZIB especially well-suited for defining the overall structure and layout of an image early in the pipeline.However, until a properly optimized speed-up LoRA for ZIB is released, generation times remain relatively slow—unless you’re running on very high-end hardware. Because of that limitation, I developed a workaround:I first generate a low-resolution pass using ZIB to quickly establish the base composition. This allows for faster iteration, broader variety across seeds, and more reliable prompt interpretation. Once the foundation is set, I switch to Zimage Turbo (ZIT) to refine and complete the finer details at higher resolution.Why ZIB instead of SDXL, Pony, or other base models?You can absolutely use those, but the key advantage of ZIB is ecosystem consistency. ZIB and ZIT can share the same CLIP, prompt structure, and VAE, which means everything integrates smoothly without needing to reload or re-import large components. This keeps the workflow lightweight, efficient, and cohesive.The drawbacks are fairly obvious as well—ZIB cannot directly generate explicit (XXX) content. While private parts can be redrawn or enhanced during the second pass using specialized models like Moody Porn Mix, poses in those scenarios are still hit-or-miss. That said, with more XXX-focused LoRAs for ZIB expected in the near future, the lack of XXX poses should be resolved shortly.This is an experimental workflow, and likely a temporary one until an official speed-up LoRA for ZIB is released—but it’s a practical and effective approach that’s absolutely worth trying in the meantime.所有展示的结果分为三类：仅 ZIB | ZIB + ZIT | 仅 ZIT。最近我一直在尝试一种新的工作流程。随着 Zimage Omni Base（ZIB）的发布，它带来了一些 Zimage Turbo（ZIT）目前尚未具备的优势——最显著的是更强的提示词遵循度、在低分辨率下更好的构图稳定性，以及在使用不同种子迭代时显著更高的变化多样性。这些优势使 ZIB 特别适合在流程早期用来确立图像的整体结构和布局。然而，在针对 ZIB 发布真正优化好的加速 LoRA 之前，生成时间仍然相对较慢——除非你使用的是非常高端的硬件。正因如此，我设计了一个变通方案：先用 ZIB 生成一轮低分辨率图像，快速确立基础构图。这样可以实现更快的迭代、在不同种子间获得更广泛的多样性，以及更可靠的提示词理解。一旦基础构图确定，我就切换到 Zimage Turbo（ZIT），在更高分辨率下精炼并完成细节。为什么选择 ZIB 而不是 SDXL、Pony 或其他基础模型？当然可以使用那些模型，但 ZIB 的核心优势在于生态系统一致性。ZIB 和 ZIT 可以共享相同的 CLIP、提示词结构和 VAE，这意味着整个流程无需反复加载或导入大型组件，就能实现无缝衔接，从而保持工作流程的轻量、高效和连贯。缺点也很明显——ZIB 无法直接生成显式（XXX）内容。虽然在第二轮精炼时可以使用 Moody Porn Mix 等专用模型对私密部位进行重绘或增强，但这类场景下的姿势仍然不够稳定。不过，随着未来不久将推出更多专注于 XXX 内容的 ZIB LoRA，姿势问题应该很快得到解决。这是一个实验性工作流程，很可能只是暂时的——直到官方的 ZIB 加速 LoRA 发布为止。但在目前阶段，这是一个实用且高效的方法，绝对值得一试。
Moody Zimage Base Simple Workflow
I just made a new Telegram Group and Discord group drop in for chats, questions, feedbacks or just to share your work. 欢迎加入纸飞机和Discord群组。喜欢我的作品吗？🎨 我会持续免费分享更多优质内容✨，欢迎支持一下，请我喝杯咖啡吧！💖 Like my work? 🎨 I share all my work for free ✨, so support me by buying me a coffee 💖!Ko-fi: ko-fi.com/catlover1937Fanvue: https://www.fanvue.com/catlover1937Please do not redistribute without consent. 如非作者同意禁止搬运。2026-05-09🌴✨ Hola everyone! ✨🌴I just realized… I NEVER actually posted my ZImage Base workflow 🤣💀But honestly… understandable LOL. I guess nobody can survive the generation times with those crazy high steps 😭🔥HOWEVER… 👀👀👀As you probably guessed already, I’ve been cooking up some HUGE improvements for the next big release of Moody Wild V4 🐆⚡And yes…Generating images with ZIB can now look EVEN BETTER than ZImage Turbo 🤯🚀Yeah… absolutely insane.The showcase images in the carousel were generated using the upcoming Moody Wild V4 💥(Release coming soon™️ 👀)For now, feel free to experiment with Wild V3 if you want to try the workflow out early ✨Huge thanks to @salttaro ❤️for lending me his NVIDIA RTX 6000 🖥️⚡so I could spend only DAYS (not YEARS with my slow GPU) doing endless testing, optimization, trial & error, and workflow tuning to squeeze out the best possible quality with the LEAST amount of compute time 🔥🧪The goal was simple:✨ Better quality⚡ Faster generation🧠 Smarter workflow🔥 Maximum visual insanityYou know the drill 😎Stay safe & have fun generating! 🫶✨🌴✨ 大家好呀！✨🌴我突然发现……我居然从来没有发过我的 ZImage Base 工作流 🤣💀不过仔细想想也挺合理 哈哈毕竟那个超高步数生成时间……估计没几个人顶得住 😭🔥但是！！！👀👀👀正如你们猜到的那样，下一次 Moody Wild V4 大版本更新里，我给你们准备了一个超级大的惊喜 🐆⚡而且是那种离谱级别的提升：✨ 使用 ZIB 生成的图，现在甚至可以比 ZImage Turbo 还要更好看 🤯🚀真的疯狂。轮播里的展示图，都是使用即将发布的 Moody Wild V4 生成的 💥（很快就会发布啦 👀）这段时间如果想提前体验，可以先用 Wild V3 试试看 ✨特别感谢芋头哥 @salttaro ❤️借给我他的 NVIDIA RTX 6000 🖥️⚡让我能够连续好几天疯狂测试、调参、试错、优化工作流，最终做到：✨ 更高质量⚡ 更快生成🧠 更聪明的工作流🔥 绝美视觉效果目标就是用最少的算力，榨出最极致的效果 🧪🔥老规矩 😎注意安全，玩得开心，继续疯狂出图吧！🫶✨
Berts Eyebrow's Workflow (Z-image)
This is a workflow that I cobbled together for Z-Image. It uses Z-Image base for composition, Z-Image Turbo for detailing and then a Z-Image Turbo skin detailing pass. I don't claim this workflow is the most efficient or well thought out but it has given me good results, YMMV.You can use LORAs in the generation stage and detailing stages. I set the ModelSamplingAuraFlow to 7-9 for composition. In the detailing stage use denoise as required but I usually have it between .28-.34. For the skin detailing pass I leave it at .32 and it gives me good results.If you don't like LORA Manager you can swap out those nodes. Also, I use subgraphs to help organize the workflow but feel free to unpack those to your liking!
Prompt generator idea + Z-Image Turbo/Base mix
Mix Z-Image Turbo with Z-Image Base in a generative prompt environment to create various charactersI find z-image base with then-injected noise and re-sampling via z-image turbo the most efficient way to use z-image. Plus modular prompting, ie. mixing partial prompts to guide the details of a given scene/character with some variables to add even more flavour.
ZImage Base Essentials workflow
Note: If you have ANY issues with nodes not downloading, read the notes or reach out. There's nothing that special about any of them that aren't core modules.⛔⚠️🛑✋ Read the notes completly before using. Most common install and node problems are listed in the directions也有中文说明A 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-breakdownWhat's IncludedCheckpoint/ GGUF/ and basic model loaderworks on Turbo as well as base. Prompt generationLora LoaderDraft mode for making corrections without having to rerun the workflow.Ultimate Upscaler for prescaling and hi res fixSimple Detailing Suite for face & eyesSeed VR2 Upscaler to get 4k on low VRAMEssential Post Production SuiteSmart Noise scrubberFull tutorial available.也有中文说明Optical realism (BETA): Give me feedback on this please. It works well, but not so sure it helps everything.Adds depth to imagesDepth enhanced film grainWraps the background over the egdes of the image to eliminate the cardboard cutout lookLook, it's just badass. Mess with it, see if you like it and let me know.Check out my other models:My NSFW version with Anantomy Correctors:https://civitai.com/models/2270894/zit-nsfw-anatomy-correctorFor ultimate Anime detail and control: https://civitai.com/models/2220766/zimage-ultra-anime-detail-workflow-get-the-most-out-of-your-generationsFor Realism and variety over the base model: https://civitai.com/models/2231181/z-image-ultra-real-workflowInstagram: https://www.instagram.com/synth.studio.models/Buy me a☕ https://ko-fi.com/lonecatoneThis represents hundreds of hours of work. If you enjoy it, please 👍like, 💬 comment , and feel free to ⚡tip 😉
TAKATO Simple Workflow for Z-IMAGE Base,ZIB
追記：よりリアル寄りにしたい場合は、Raz's SceneSmithのLoRA https://civitai.com/models/1714609/razs-scenesmith-fluxzit?modelVersionId=2618747 も試してみてください。ZIT用ですが、ZIBでも効き目がありそうです。ウェイトは 0.3〜0.5 前後、stepsは12前後 にすると、かなりリアルな質感になりやすいです。もちろん、画像や好みによって合う設定は変わるので、そこはお好みで微調整してください。今日は、画像生成を始めてちょうど3年になりました。その記念として、このワークフローを一度公開してみようと思います。まだ完成度が高いワークフローではなく、改善できる部分も多いですが、今の自分の記録も兼ねて載せてみます。少しでも参考になる部分があればうれしいです。このワークフローは、基本的に Z-IMAGE Base で使える前提で組んでいます。必要に応じて LoRAで調整 したり、数値・sampler・scheduler などを好みに合わせて変更 して使ってください。また、このワークフローを作るにあたって、尊敬している好きな方々のプロンプトを一部参考にさせていただきました。まだ発展途中の内容ですが、3年続けてきた記念として公開します。少しでも誰かの参考になればうれしいです。Additional note: If you want a more realistic look, you can also try Raz’s SceneSmith LoRA:https://civitai.com/models/1714609/razs-scenesmith-fluxzit?modelVersionId=2618747It is made for ZIT, but it seems to work quite well with ZIB too.Using a weight around 0.3 to 0.5 and setting the steps to around 12 tends to produce a much more realistic texture.Of course, the best settings may vary depending on the image and your personal preference, so feel free to fine-tune them as needed.Today marks exactly 3 years since I started image generation, so I wanted to share this workflow as a small anniversary release.It is not a highly polished workflow yet, and I know there is still a lot of room for improvement, but I wanted to upload it as a record of where I am right now after 3 years of learning and experimenting.This workflow is mainly built for Z-IMAGE Base.Please feel free to adjust it with your own LoRAs, and also change the values, sampler, scheduler, and other settings to match your preferences.Also, while making this workflow, I referenced and borrowed some ideas from prompts created by people I respect and admire.It is still a work in progress, but I wanted to share it as a small milestone.I hope some part of it can be useful to others.
Ray Z-image base T2I Workflows
This is a very simple Z-image base T2I workflow. If you need LoRa or other features, please add them yourself.
My Z-Image Turbo/Base Workflow
This is a modified workflow i found online, I added detailer's and it makes images that are light-years better then the other ones i tried. You can't make the Latent image size bigger then W/H 480, H/W 640 because it would mess up the image but at the end you'll get a image of W/H 2880 X H/W 3840. Below are the things i use in this workflow, Enjoyqwen_3_4b_uncensored_aggressiveComfyUI-Impact-Packperson_yolov8s-segface_yolov8s
z-image-base basic workflow, with high-definition restoration
The basic workflow of z-image, using Tile ControlNet for high-resolution restoration
Flux Klein + Z Image Turbo Workflow in ComfyUI to Beat Nano Banana 2
Hey i have built this workflow to beat nano banana 2 and here is the video also i made to fully understand how it works give it a watch as well and all the necessary links are in the description of the video
🚀 Z-Image Base Workflow (+ Portrait Master) 🚀
This ComfyUI workflow gives you maximum control when generating with the Z-Image Base and Turbo models. It features a flexible setup that allows you to seamlessly switch between the Lora Manager and classic Lora Stacks. Perfect for high-end photography (Portrait Master is included!) or stunning artworks (like the Spanish Comic Lora). ⚡ Turbo Boost: If you want to use the Turbo version, grab this one, it's even faster: https://civitai.com/models/2371047 💡 Special Feature: No need to hunt for the other models! Murph has integrated all direct download links (Z-Image Base, VAE, etc.) right into the workflow notes. 📺 Video Tutorial & Tips: Make sure to watch the companion YouTube video. Ziggy explains how to customize the workflow, clean up subnodes, and get the absolute best results. (Stick around until the end for an exclusive AI music teaser! 🤘) 👉 Watch the video here: https://youtu.be/twAKy2cg89A
DIVERSITY - ZIB + ZIT Refine (+ Simple Prompt Helper)
This workflow is designed to preserve the diversity of the Base model while benefiting from the photorealism of the Turbo model.If the Turbo model is given too much freedom during the refinement step, the faces tend to lose the diversity provided by the Base model.This workflow therefore limits the influence of the Turbo model so it only adds a subtle layer of texture and realism to the images.How to use the random prompt helper (Blue nodes on the left) :Check “Yes” for every Style, Framing, Mood, or Angle you want to include.One option will be randomly selected and applied to your prompt.If only one option is checked, it will be applied every time.In the “Prompt” group, you can disable these random prompt helpers by unchecking them in the yellow nodes.
Z-image base: simple workflow for high quality realism + tips & info + separate inpainting/refiner workflow
What is this?This is a simple workflow for Z-image base that produces high quality, extremely realistic images at high native resolution. There are also tips for using Z-image base below and some general info you might find helpful. There's now also a separate inpainting workflow.The sampler settings are geared towards sharpness and clarity, but you can introduce grain and other defects through prompting.All the images attached to the image-gen post were generated directly with this workflow with no further editing.Read on for update details and all the info you need to get started.Update 2026-05-26: Updated inpainting workflowI've been finding the second Ksampler in the inpainting workflow to be more trouble than it's worth. It tends to cause problems at low-ish denoise (starting from about 0.5 and below), and I'd rather not confuse everyone by saying it should be bypassed sometimes.So, I've updated the workflow to exclude the second KSampler entirely. If you were having difficulty with low-denoise edits before, this will help you a lot.Update 2026-03-29: Upgraded inpainting workflowI found a node that blends images together with semi-transparent masks. This means we can blend the old image with the inpainted area much more effectively now, completely eliminating the seams that would sometimes be visible around the masked area.TLDR: download the inpainting workflow again, and get the LayerStyle nodes from the list below. You'll get much better results than before.Update 2026-03-12: New inpainting workflowTurns out Z-image base is really good at inpainting! I've put together an easy-to-use workflow for it, which is an expansion of the original workflow. It's not as simple to follow what it's doing though, so start with the normal workflow first if you're new. You can also use the workflow for localised refinement (e.g. fixing faces or hands), or as a simple image-to-image generator too.All the same info below applies, and the workflow itself has a bunch of info inside about how it works and how to use it. Long story short, I've made some small improvements to how inpainting is done (with any model) in ComfyUI, so you can apply the same methodology to other models if you want improvements there. Read the text boxes to see what it's doing!You'll also need an additional custom node set to use it (more info below): ComfyUI_essentialsHere's an album of all the separate example images attached to the post, including the masks I used for them: g-driveIf you want examples/info on unsafe-for-work ways to use this, I made a post on reddit about it: https://www.reddit.com/r/unstable_diffusion/comments/1rrt0ny/degenerate_inpainting_with_zimage_base_workflow/Nodes & ModelsCustom Nodes:RES4LYF - A very popular set of samplers & schedulers, and some very helpful nodes. These are needed to get the best z-image base outputs, IMO.RGTHREE - (Optional) A popular set of helper nodes. If you don't want this you can just delete the seed generator and lora stacker nodes, then use the default comfy lora nodes instead. RES4LYF comes with a seed generator node as well, I just like RGTHREE's more.ComfyUI GGUF - (Optional) Lets you load GGUF models, which for some reason ComfyUI still can't do natively. If you want to use a non-GGUF model you can just skip this, delete the UNET loader node and replace it with the normal 'load diffusion model' node.ComyUI Essentials - (Inpaint workflow only) Adds a bunch of very helpul nodes. We're using it specifically for its number comparison node so we can switch between the image-to-image and inpainting modes automatically.ComfyUI LayerStyle - (Inpaint workflow only) Adds a ton of nodes for image transformations, similar to the tools in photoshop. We're using this for its image blending node, which allows us to blend two images using a semi-transparent mask.Models:Main model: Z-image base GGUFs - BF16 recommended if you have 16GB+ VRAM. Q8 will just barely fit on 8GB VRAM if you know what you're doing. Q6_k will fit easily in 8GB. Avoid using FP8, the Q8 gguf is better.Text Encoder: Normal | gguf Qwen 3 4B Text Encoder - Grab the biggest one that fits in your VRAM, which would be the full normal one if you have 10GB+ VRAM or the Q8 GGUF if you have less than 8GB VRAM. Some people say text encoder quality doesn't matter much & to use a lower sized one, but it absolutely does matter and can drastically affect quality. For the same reason, do not use an abliterated text encoder unless you've tested it and compared outputs to ensure the quality doesn't suffer.If you're using the GGUF text encoder, swap out the "Load CLIP" node for the "ClipLoader (GGUF)" node.VAE: Flux 1.0 AEInfo & TipsSampler SettingsI've found that a two-stage sampler setup gives very good results for z-image base. The first stage does 95% of the work, and the second does a final little pass with a low noise scheduler to bring out fine details. It produces very clear, very realistic images and is particularly good at human skin.CFG 4 works most of the time, but you can go up as high as CFG 7 to get different results.This is all with shift 1. If you don't know what that is, don't worry - it's the default!Stage 1:Sampler - res_2sScheduler - betaSteps - 22Denoise: 1.00Stage 2:Sampler - res_2sScheduler - normalSteps - 3Denoise: 0.15ResolutionsHigh res generationOne of the best things about Z-image in general is that it can comfortably handle very high resolutions compared to other models. You can gen in high res and use an upscaler immediately without needing to do any other post-processing.(info on upscalers + links to some good ones further below)Note: high resolutions take a long time to gen. A 1280x1920 shot takes around ~95 seconds on an RTX 5090, and a 1680x1680 shot takes ~110 seconds.Different sizes & aspect ratios change the outputDifferent resolutions and aspect ratios can often drastically change the composition of images. If you're having trouble getting something ideal for a given prompt, try using a higher or lower resolution or changing the aspect ratio.It will change the amount of detail in different areas of the image, make it more or less creative (depending on the topic), and will often change the lighting and other subtle features too.I suggest generating in one big and one medium resolution whenever you're working on a concept, just to see if one of the sizes works better for it.Good resolutionsThe workflow has a variety of pre-set resolutions that work very well. They're grouped by aspect ratio, and they're all divisible by 16. Z-image base (as with most image models) works best when dimensions are divisible by 16, and some models require it or else they mess up at the edges.Here's a picture of the different resolutions if you don't want to download the workflow: imgbb | g-driveYou can go higher than 1920 to a side, but I haven't done it much so I'm not making any promises. Things do tend to get a bit weird when you go higher, but it is possible.I do most of my generations at 1920 to a side, except for square images which I do at 1680x1680. I sometimes use a lower resolution if I like how it turns out more (e.g. the picture of the rat is 1680x1120).Realism Negative PromptThe negative prompt matters a lot with z-image base. I use the following to get consistently good realism shots:3D, ai generated, semi realistic, illustrated, drawing, comic, digital painting, 3D model, blender, video game screenshot, screenshot, render, high-fidelity, smooth textures, CGI, masterpiece, text, writing, subtitle, watermark, logo, blurry, low quality, jpeg, artifacts, grainyPrompt StructureYou essentially just want to write clear, simple descriptions of the things you want to see. Your first sentence should be a basic intro to the subject of the shot, along with the style. From there you should describe the key features of the subject, then key features of other things in the scene, then the background. Then you can finish with compositional info, lighting & any other meta information about the shot.Use new lines to separate key parts out to make it easier for you to read & build the prompt. The model doesn't care about new lines, they're just for you.If something doesn't matter to you, don't include it. You don't need to specify the lighting if it doesn't matter, you don't need to precisely say how someone is posed, etc; just write what matters to you and slowly build the prompt out with more detail as needed.You don't need to include parts that are implied by your negative prompt. If you're using the realism negative prompt I mentioned earlier, you don't usually need to specify that it's a photograph.Your structure should look something like this (just an example, it's flexible):A <style> shot of a <subject + basic description> doing <something>. The <subject> has <more detail>. The subject is <more info>. There is a <something else important> in <location>. The <something else> is <more detail>. The background is a <location>. The scene is <lit in some way>. The composition frames <something> and <something> from <an angle or photography term or whatever>.Following that structure, here are a couple of the prompts for the images attached to this post. You can check the rest out by clicking on the images at the top.The ballet womanA shot of a woman performing a ballet routine. She's wearing a ballet outfit and has a serious expression. She's in a dynamic pose. The scene is set in a concert hall. The composition is a close up that frames her head down to her knees. The scene is lit dramatically, with dark shadows and a single shaft of light illuminating the woman from above.The rat on the fence postA close up shot of a large, brown rat eating a berry. The rat is on a rickety wooden fence post. The background is an open farm field.The woman in the waterA surreal shot of a beautiful woman suspended half in water and half in air. She has a dynamic pose, her eyes are closed, and the shot is full body. The shot is split diagonally down the middle, with the lower-left being under water and the upper-right being in air. The air side is bright and cloudy, while the water side is dark and menacing.The space capsuleA woman is floating in a space capsule. She's wearing a white singlet and white panties. She's off-center, with the camera focused on a window with an external view of earth from space. The interior of the space capsule is dark.UpscalingZ-image makes very sharp images, which means you can directly upscale them very easily. Conventional upscale models rely on sharp/clear images to add detail, so you can't reliably use them on a model that doesn't make sharp images.My favourite upscaler for NAKED PEOPLE or human face close-ups is 4xFaceUp. It's ridiculously good at skin detail, but has a tendency to make everything else look a bit stringy (for lack of a better word). Use it when a human being showing lots of skin is the main focus of the shot.Here's a 6720x6720 version of the sitting bikini girl that was upscaled directly using the 4xFaceUp upscaler: imgbb | g-driveFor general shots you can use something like 4xNomos2.Alternatively, you can use SeedVR2, which also has the benefit of working on blurry images (not a problem with z-image anyway). It's not as good at human skin as 4xFaceUp, but it's better at everything else. It's also very reliable and pretty much always works. I have a basic workflow for it here: https://pastebin.com/9D7sjk3zClownShark Ksampler - what is it?It's a node from the RES4LYF pack. It works the same as a normal sampler, but with a few differences:"ETA". This setting basically adds extra noise during sampling using fancy math, and it generally helps get more detail out of generations, and makes them more stable. It's insane how much of a difference this makes with some models. This is the main reason we're using the clownshark sampler. A value of 0.5 is usually good, but I've seen it be good up to 0.7 for certain models (like Klein 9B), and lower is good for some models too. You can range it from 0.1 to 0.9 for interesting results."bongmath". This setting turns on bongmath. It's some kind black magic that improves sampling results without any downsides - but only if you're using the sampler a certain way. It does nothing in these particular workflows. Someone tries to explain what it is here: https://www.reddit.com/r/StableDiffusion/comments/1l5uh4d/someone_needs_to_explain_bongmath/It has access to a ton of alternative samplers/schedulers (we're not using anything interesting in these workflows, just res_2s)It has some funky mechanisms for doing follow-on/continuous generation by chaining multiple samplers together (this is what the "sampler_mode" setting is for). We're not using that in these workflows though.You don't need to use this sampler if you don't want to; you can use the res_2s/beta sampler/scheduler with a normal ksampler node as long as you have RES4LYF installed. But seeing as the clownshark sampler comes with RES4LYF anyway we may as well use it - the ETA setting is awesome.Effect of CFG on outputsLower than 4 CFG is bad. Other than that, going higher has pretty big and unpredictable effects on the output for z-image base. You can usually range from 4 to 7 without destroying your image. It doesn't seem to affect prompt adherence much.Going higher than 4 will change the lighting, composition and style of images somewhat unpredictably, so it can be helpful to do if you just want to see different variations on a concept. You'll find that some stuff just works better at 5, 6 or 7. Play around with it, but stick with 4 when you're just messing around.Going higher than 4 also helps the model adhere to realism sometimes, which is handy if you're doing something realism-adjacent like trying to make a shot of a realistic elf or something.Base vs Distil vs TurboThey're good for different things. I'm generally a fan of base models, so most workflows I post are / will be for base models. Generally they give the highest quality but are much slower and can be finicky to use at times.What is distillation?It's basically a method of narrowing the focus of a model so that it converges on what you want faster and more consistently. This allows a distil to generate images in fewer steps and more consistently for whatever subject/topic was chosen. They often also come pre-negatived (in a sense) so that you can use 1.0 CFG and no negative prompt. Distils can be full models or simple loras.The downside of this is that the model becomes more narrow, making it less creative and less capable outside of the areas it was focused on during distillation. For many models it also reduces the quality of image outputs, sometimes massively. Models like Qwen and Flux have god-awful quality when distilled (especially human skin), but luckily Z-image distils pretty well and only loses a little bit of quality. Generally, the fewer steps the distil needs the lower the quality is. 4-step distils usually have very poor quality compared to base, while 8+ step distils are usually much more balanced.Z-image turbo is just an official distil, and it's focused on general realism and human-centric shots. It's also designed to run in around 10 steps, allowing it to maintain pretty high quality.So, if you're just doing human-centric shots and don't mind a small quality drop, Z-image turbo will work just fine for you. You'll want to use a different workflow though - let me know if you'd like me to upload mine.Below are the typical pros and cons of base models and distils. These are pretty much always true, but not always a 'big deal' depending on the model. As I said above, Z-image distils pretty well so it's not too bad, but be careful which one you use - tons of distils are terrible at human skin and make people look plastic (z-image turbo is fine).Base model pros:Generally gives the highest quality outputs with the finest details, once you get the hang of itCreative and flexibleBase model cons:Very slowUsually requires a lengthy negative prompt to get good resultsCreativity has a downside; you'll often need to generate something several times to get a result you likeMore prone to mistakes when compared to the focus areas of distilse.g. z-image base is more likely to mess up hands/fingers or distant faces compared to z-image turboDistil pros:Fast generationsGood at whatever it was focused on (e.g. people-centric photography for z-image turbo)Doesn't need a negative prompt (usually)Distil cons:Bad at whatever it wasn't focused on, compared to baseUsually bad at facial expressions (not able to do 'extreme' ones like anger properly)Generally less creative, less flexible (not always a downside)Lower quality images, sometimes by a lot and sometimes only by a little (depends on the model, the specific distil, and the subject matter)Can't have a negative prompt (usually)You can get access to negative prompts using NAG (not covered in this info page)
Eason ZimageBASE workflow
BASE workflow
Rebels Kandinsky + Z-Image
UPDATED TO VERSION 2VERSION 2 is a complete upgrade to quality and resolution. utlizing the FULL step count of both models with a small upscaler "realESERGAN2x" on the FIRST pass to ensure that the second pass of Z-Image is handed the latent data at 2048x2048 for use of full resolution capabilities of the model. this gives the final result a large resolution for minimal vram requirement and a highly detailed and refined image.2 pass image workflow for detail refinement.1st Model - Kandinsky Image Lite (base image)2nd Model - Z-Image BASE gguf (detail refiner)essentially it goes:Kandinsky generates at 1024x1024 (maximum). you generate at 100 steps and then into the upscaler to bring the resolution to 2048x2048 (maximum). this passes the latent data to Z-Image to run 50 steps at the now larger resolution.Z-Image base can NATIVELY GENERATE AT 2048X2048!(this means after the low grade upscale, theres no artifacting because Z-Image corrects it in the second pass!)REQUIRES KJNODESREQUIRES realESERGAN2xREQUIRES UNLOAD MODELS NODEALL CAN BE FOUND IN THE COMFY MANAGERnote: generation times will increase significantly. this is not about speed, this is about QUALITY.__________________________________________________________________________________________________2 pass image workflow for detail refinement.1st Model - Kandinsky Image Lite (base image)2nd Model - Z-Image BASE gguf (detail refiner)VERSION 1the kandinsky model acts as your first pass model, generating 20-30 steps to create an overall composition. the z-image base model runs on the second pass as a detail refiner, running 20-30 steps. this when playing with the denoise on the SECOND PASS can provide some interesting results!Supported ResolutionsSquare: 1024 × 1024Landscape: 1280 × 768Portrait: 768 × 1280NOTE: i chose to run the gguf model for the zimage base. this workflow is vram intense and using the gguf takes some of the pressure off. the regular load diffusion model node is in the workflow and can be unbypassed and strung in if youd like to try using the full weight model instead!CUSTOM NODE REQUIREMENTS:Kiko Tools NodesEasy Use NodesUnload Model Nodesthe unload model node is NECESSARY to prevent OOM. all custom nodes can be found in the comfyui custom nodes manager!Kandinsky files can be downloaded in comfyui templates!
Famegrid Revolution Hybrid Workflow (ZIB + ZIT)
🔥 FameGrid Revolution — Hybrid Workflow (ComfyUI)If you’re using FameGrid Revolution and you want the cleanest realism + strongest adherence, this is the workflow. It’s built for ultra-realistic influencer photography: natural skin, modern lighting, clean composition, minimal artifacts.This workflow is designed to run FameGrid Revolution on Z-Image Base + Z-Image Turbo as a hybrid pipeline. You can run it normally — but Hybrid is where it becomes premium.✅ What This Workflow DoesLocks in realism (skin texture + believable faces)Improves composition (less random framing, more “intentional shot”)Reduces artifacts (less texture junk / weird noise)Keeps the influencer aesthetic without plastic/overprocessed driftPerforms better with long prompts (this model loves detail)🧠 Recommended Baseline Settings (Start Here)Use the default settings in the workflow. You can play around with the Lora weights, CFG, and denoise on the 2nd Ksampler, adjusting to your liking.✍️ Prompting (This Matters More Than People Admit)This workflow rewards structure. Lazy prompts get average output.FameGrid Revolution likes long prompts — 800+ characters is ideal when you want maximum control.Start prompts like:“iPhone photo of an IGMODEL …”Then add: clothing, setting, lighting, camera angle, lens vibe, mood, composition, background details.🎛️ Optional: Pick Your “Flavor” Checkpoint (available at the Discord)These are alternate variants if you want different bias:ZIB_Bold — stronger baked-in look, better-looking subjectsZIB_More_Bold — even stronger IG look (max stylization)ZIB_ExtraReal — nicer backgrounds, more average/believable subjectsZIB_UltraReal — extremely nice backgrounds, great prompt adherence, minimal artifacts🧩 What’s IncludedHybrid workflow file (ComfyUI)Recommended settings baked into the nodesTuned specifically for FameGrid Revolution📌 Get Updates + Extra Checkpoints (UltraMuse Discord)The workflow gets updates, and some checkpoints/resources won’t be posted publicly.👉 https://discord.gg/4zbGm5j6jW
BitDance 14B FP8 Custom Node - Workflow
REQUIREMENTS:ComfyUI-BitDance Custom Node RepositoryBitDance Main Model FP8 (Place in models/diffusion_models)BitDance Text Encoder FP8 (Place in models/text_encoders)BitDance VAE FP16 (Place in models/vae)DESCRIPTION:Low VRAM: Utilizing the built-in "model to offload" setting in the Text Encode node, this workflow automatically shifts the massive main model off your GPU right before heavy text encoding begins, saving your hardware from crashing.Speed: By quantizing to fp8 e4m3fn scaled and routing the main model to the graphics card, generation times are drastically reduced on standard consumer hardware.Quality: You retain near-full visual fidelity. Using the required euler_maruyama sampler, it flawlessly decodes the massive Binary Tokenizer (2^256 states) to generate perfect micro-details, accurate text (like reading a newspaper), and complex material blending from simple everyday English prompts.LINKS: Video Tutorial: https://www.youtube.com/watch?v=4O9ATPbeQyg | Full Documentation: https://aistudynow.com/how-to-fix-the-generic-face-bug-in-bitdance-14b-optimize-speed/Models: https://huggingface.co/comfyuiblog/BitDance-14B-64x-fp8-comfyui/tree/mainGithub: https://github.com/aistudynow/Comfyui-bitdance
ComfyUI beginner friendly Z-Image Base Seamless Fast Outpainting with LORAs Workflow by Sarcastic TOFU
This is a very simple ComfyUI beginner friendly Seamless Fast Outpainting workflow that will work with the Z-Image Base All-in-One model (with baked in Text Encoder & VAE) that work well with my 8GB AMD GPU or anything similar from Nvidia. But if you have a much weaker GPU (4GB or lower) you can find other low grade hardware supported GGUF models from these HuggingFace Repos -Jayn7's HuggingFace repo for Z-Image Base GGUF models ---------------------------------------------------------https://huggingface.co/jayn7/Z-Image-GGUF/tree/mainUnsloth's QWEN3 GGUF Text Encoders for Z-Image Base GGUF models -------------------------------------------------------------------https://huggingface.co/unsloth/Qwen3-4B-GGUF/tree/mainThe newly released Z-Image Base model, developed by Tongyi-MAI (Alibaba's AI lab), is the non-distilled foundation of the Z-Image family, designed to serve as the "raw" starting point for the open-source community. Unlike Z-Image Turbo, which is a distilled version optimized for ultra-fast performance and sub-second inference using as few as 8 steps, the Base model is a full-weight checkpoint that requires more steps (typically 28 to 50) but offers significantly higher diversity in poses, compositions, and artistic styles. While Z-Image Turbo is specifically tuned for immediate photorealism and consumer-grade GPU efficiency, Z-Image Base is built for researchers and creators who need a more flexible model for fine-tuning, training LoRAs, or achieving complex prompt adherence that the faster, more "opinionated" Turbo variant might miss. Compared to other specialized versions like Z-Image Edit, which is fine-tuned specifically for instruction-based image-to-image tasks, the Base model provides the most neutral and unrefined data distribution, making it the superior choice for those looking to push the 6-billion parameter S3-DiT architecture into new, creative territories beyond just speed.You can download your necessary Z-Image Base ALL-in-One model and LORA used on this workflow from CivitAI (Details are mentioned below). Make sure you have latest enough ComfyUI installation and install any necessary nodes for this workflow using ComfyUI manager and place the correct files (AIO model, LORAs etc) in correct places. Also check out my other workflows, LORAs, toolsets, datasets for SD 1.5 + SDXL 1.0, Pony, Illustrious, WAN 2.1, WAN 2.2, MagicWAN Image v2, QWEN, HunyuanImage-2.1, HiDream, KREA, Chroma, AuraFlow, lumina 2, Z-Image Base & Turbo, Flux.2 Klein and Flux.1 Dev. Feel free to toss some yellow Buzz on stuffs you like.How to use this -#1. Just select your desired Z-Image Base All-in-One model (or swap it with GGUF model files) file first and now#2. one or multiple matching LORA(s) from LORA Manager to start#3. then load your reference image to write your positive and negative prompts for output.#4. select how many images you want (Change the number besides the "Run" button)#5. select image sampling methods, CFG, steps etc. settings#6. finally press the run button to generate. That's it..** LORA usage for this workflow is optional you can use it without any LORAs, use with 1 or 2 or any other number of LORAs, too add new LORAs press the L button on top to lunch LORA Manager on a new tab find your LORA and if you want to use that LORA just click the upward Kite button. You can use the mask editor node for ComfyUI (masking works very well with it) or any of your favourite image editor for masking. But I would strongly recommend you to use ComfyUI's mask editor node because sometimes ComfyUI fails to use masked image created using external tool.Required Files===============### Download Link for Z-Image Base ALL-in-One Model (FP8) -------------------------------------------------------------https://civitai.com/models/2173571?modelVersionId=2637423### Download Link for Z-Image-Aesthetic LORA (Base v1) -----------------------------------------------------------https://civitai.com/models/2214707?modelVersionId=2635939
ComfyUI beginner friendly Z-Image Base Seamless Fast Inpainting with LORAs Workflow by Sarcastic TOFU
This is a very simple ComfyUI beginner friendly Seamless Fast Inpainting workflow that will work with the Z-Image Base All-in-One model (with baked in Text Encoder & VAE) that work well with my 8GB AMD GPU or anything similar from Nvidia. But if you have a much weaker GPU (4GB or lower) you can find other low grade hardware supported GGUF models from these HuggingFace Repos -Jayn7's HuggingFace repo for Z-Image Base GGUF models ---------------------------------------------------------https://huggingface.co/jayn7/Z-Image-GGUF/tree/mainUnsloth's QWEN3 GGUF Text Encoders for Z-Image Base GGUF models -------------------------------------------------------------------https://huggingface.co/unsloth/Qwen3-4B-GGUF/tree/mainThe newly released Z-Image Base model, developed by Tongyi-MAI (Alibaba's AI lab), is the non-distilled foundation of the Z-Image family, designed to serve as the "raw" starting point for the open-source community. Unlike Z-Image Turbo, which is a distilled version optimized for ultra-fast performance and sub-second inference using as few as 8 steps, the Base model is a full-weight checkpoint that requires more steps (typically 28 to 50) but offers significantly higher diversity in poses, compositions, and artistic styles. While Z-Image Turbo is specifically tuned for immediate photorealism and consumer-grade GPU efficiency, Z-Image Base is built for researchers and creators who need a more flexible model for fine-tuning, training LoRAs, or achieving complex prompt adherence that the faster, more "opinionated" Turbo variant might miss. Compared to other specialized versions like Z-Image Edit, which is fine-tuned specifically for instruction-based image-to-image tasks, the Base model provides the most neutral and unrefined data distribution, making it the superior choice for those looking to push the 6-billion parameter S3-DiT architecture into new, creative territories beyond just speed.You can download your necessary Z-Image Base ALL-in-One model and LORA used on this workflow from CivitAI (Details are mentioned below). Make sure you have latest enough ComfyUI installation and install any necessary nodes for this workflow using ComfyUI manager and place the correct files (AIO model, LORAs etc) in correct places. Also check out my other workflows, LORAs, toolsets, datasets for SD 1.5 + SDXL 1.0, Pony, Illustrious, WAN 2.1, WAN 2.2, MagicWAN Image v2, QWEN, HunyuanImage-2.1, HiDream, KREA, Chroma, AuraFlow, lumina 2, Z-Image Base & Turbo, Flux.2 Klein and Flux.1 Dev. Feel free to toss some yellow Buzz on stuffs you like.How to use this -#1. Just select your desired Z-Image Base All-in-One model (or swap it with GGUF model files) file first and now#2. one or multiple matching LORA(s) from LORA Manager to start#3. then load your masked image to write your positive and negative prompts for output.#4. select how many images you want (Change the number besides the "Run" button)#5. select image sampling methods, CFG, steps etc. settings#6. finally press the run button to generate. That's it..** LORA usage for this workflow is optional you can use it without any LORAs, use with 1 or 2 or any other number of LORAs, too add new LORAs press the L button on top to lunch LORA Manager on a new tab find your LORA and if you want to use that LORA just click the upward Kite button. You can use the mask editor node for ComfyUI (masking works very well with it) or any of your favourite image editor for masking. But I would strongly recommend you to use ComfyUI's mask editor node because sometimes ComfyUI fails to use masked image created using external tool.Required Files==================### Download Link for Z-Image Base ALL-in-One Model (FP8) -------------------------------------------------------------https://civitai.com/models/2173571?modelVersionId=2637423### Download Link for Z-Image-Aesthetic LORA (Base v1) -----------------------------------------------------------https://civitai.com/models/2214707?modelVersionId=2635939
ZImageBase - Simple WorkFlow With Upscaling
A simple workflow with the best settings i have tested so far. If you can get better results with other settings please let me know.
R-Image-Base (Model Workflow)
workflow for the MODEL, not the checkpoint
Rebels Z-Image+DetailDaemon
VERSION 1: (experimental)REQUIRES DETAIL DAEMON CUSTOM NODES.REQUIRES FOOLHARDY 4X UPSCALE MODEL.(both are available through the comfyui manager)VERSION 2: Better ResultsREQUIRES DETAIL DAEMON CUSTOM NODES.REQUIRES REALESERGAN 2X UPSCALE MODEL.REQUIRES KIKO TOOLS CUSTOM NODES.(both are available through the comfyui manager)version 2 is a 2-pass BASE model workflow focusing the detail daemon nodes pushing detail noise on the second pass as opposed to running it in 1 pass. this gives the model a chance to develop core details and overall composition BEFORE pushing the noise sampling detail daemon on top to finish the generation. This version works ALOT better! you can adjust both samplers step counts to whatever you feel necessary, the workflows values are just the best ive found so far after testing for a few hours. i do suggest keeping the second sampler lower than 15 steps. it creates TOO much detail when you give it too much room to throw noise over the image. trust me 10 to 12 is enough. if you want slightly more detail, lower the "start" value in the daemon node to .40 and lower the "end" value to .60 to begin filling the image with noise earlier which forces the model to clean it up with more detail. give it less steps if you do.Alternative settings:Detail Daemon Nodestart - .40 (currently .60)end - .60 (currently .80)the workflow is set to run the detail sampling at the end of each step, these alternate settings begin with natural sampling then in the middle of the step it introduces noise with the daemon node then stops early to give the model a chance to smooth it back out. you can also drop the "start" to .20 and the "end" to .40 if you want to begin the sampling step with extra noise but i recommend the middle or end settings as it gives a better output.PROMPTING MATTERS!whatever you want to be scaled in detail, you need to prompt for. if you want their eyes to be detailed, PROMPT FOR DETAILED EYES. the sampler and model work together to focus prompt specific details.works best on close-up portraits. characters far from focal point are sometimes blurry. this is an experimental workflow for adding detail daemon and foolhardy 4x upscaler to z-image base workflow.WARNING: this workflow is extremely violatile and doesnt always yield the best results. that being said, running multiple seeds can sometimes bring out some AMAZING generations. helps if you just want an extra pop in some iterations and recompression back onto smartphones hides the artifacting that occurs in some generations because of the upscaler. (youll see)dont come for me, i warned you its not perfect lol
ComfyUI beginner friendly Z-Image Base (or Z-Image Turbo) ALL-in-One Txt2Img with LORAs Workflow by Sarcastic TOFU
This is a very simple ComfyUI beginner friendly Text-to-Image workflow that will work with Z-Image Base (or Z-Image Turbo) ALL-in-One model that has both Text Encoder & VAE baked into just a single file (I used FP8 models on this but if you have 12 or 16GB VRAM you can try the BF16 or FP16 models available from same source). These models aren't GGUF models or individual regular unquantized models for which you have to match & manage multiple files (if you have very low garde 6GB or 4GB VRAM GPU you can see my Z-Image GGUF workflows, these ALL-in-One models will not work on anything less than 8GB VRAM). Also for this workflow I have included my recent Easy Prompt Saver tweak so unlike my previous Z-Image workflows this one helps you to save your Text to Image Generation Data into a human readable .txt file. Also, if you want, with the Image Saver Simple node used on this you may embed the metadata with workflow itself or save the workflow for your work seperately.The newly released Z-Image Base model, developed by Tongyi-MAI (Alibaba's AI lab), is the non-distilled foundation of the Z-Image family, designed to serve as the "raw" starting point for the open-source community. Unlike Z-Image Turbo, which is a distilled version optimized for ultra-fast performance and sub-second inference using as few as 8 steps, the Base model is a full-weight checkpoint that requires more steps (typically 28 to 50) but offers significantly higher diversity in poses, compositions, and artistic styles. While Z-Image Turbo is specifically tuned for immediate photorealism and consumer-grade GPU efficiency, Z-Image Base is built for researchers and creators who need a more flexible model for fine-tuning, training LoRAs, or achieving complex prompt adherence that the faster, more "opinionated" Turbo variant might miss. Compared to other specialized versions like Z-Image Edit, which is fine-tuned specifically for instruction-based image-to-image tasks, the Base model provides the most neutral and unrefined data distribution, making it the superior choice for those looking to push the 6-billion parameter S3-DiT architecture into new, creative territories beyond just speed.You can download your necessary Z-Image ALL-in-One model and LORA(s) used on this workflow from CivitAI (Details are mentioned below). Make sure you have latest enough ComfyUI installation and install any necessary nodes for this workflow using ComfyUI manager and place the correct files (AIO model, LORAs etc) in correct places. Also check out my other workflows, LORAs, toolsets, datasets for SD 1.5 + SDXL 1.0, Illustrious, WAN 2.1, WAN 2.2, MagicWAN Image v2, QWEN, HunyuanImage-2.1, HiDream, KREA, Chroma, AuraFlow, Z-Image Base & Turbo, Flux.2 Klein and Flux.1 . Feel free to toss some yellow Buzz on stuffs you like.How to use this -#1. Just select your desired Z-Image Base (or Z-Image Turbo) ALL-in-One model file first and now#2. set your desired image dimensions to start#3. then input your positive and negative prompts.#4. select how many images you want (Change the number besides the "Run" button)#5. select image sampling methods, CFG, steps etc. settings#6. finally press the run button to generate. That's it..** LORA usage for this workflow is optional you can use it without any LORAs, use with 1 or 2 or any other number of LORAs, to add new LORAs press the L button on top to lunch LORA Manager on a new tab find your LORA and if you want to use that LORA just click the upward Kite button. Just a fair warning for LORA usage for Z-Image AIO models - if you do use LORA(s) make sure that your each selected LORA is supported on the All-in-One Z-Image model (Base or Turbo) you have choosen. Normally Z-Image Turbo LORAs don't work with Z-Image Base Model, similarly Z-Image Base LORAs don't work for Z-Image Turbo Model. Only a few LORAs work on both Z-Image Base & Z-Image Turbo Models. The LORA(s) I used on the example images support both Z-Image Base & Z-Image Turbo but that may not be the case for anything else you may find on CivitAI or elsewhere.Required Files==================### Download Link for Z-Image Base ALL-in-One Model (FP8) -------------------------------------------------------------https://civitai.com/models/2173571?modelVersionId=2637423### Download Link for Z-Image Turbo ALL-in-One Model (FP8)-------------------------------------------------------------https://civitai.com/models/2173571?modelVersionId=2448013### Download Link for Jibify Z-image Base -Skin/Realism Detailer LORA -------------------------------------------------------------------------https://civitai.com/models/2346302/jibify-z-image-base-skinrealism-detailer### Download Link for Amateur Nude Selfies LORA (Only for Last two images)-----------------------------------------------------------------------------https://civitai.com/models/2271699?modelVersionId=2639244
R-Image-Base (Checkpoint Workflow)
workflow for the R-Image-Base checkpoint!Link to Checkpoint:https://civitai.com/models/2390650?modelVersionId=2688019Link to Step Counter (optional):https://civitai.com/models/2387088/rebels-step-monitor-nodeWORKFLOW INCLUDES MATRIX STEP COUNTER!!!!(the step counter node is NOT needed, If you dont want it just delete the node and save the workflow without it)
Pornmaster Z Image Base t2i Workflow
CFG:1，4 steps.CFG:1.5，5~8 steps.--下載Lora Manager.下載Z-Image-Fun-Lora-Distill-4-Steps-2602-ComfyUI lora.將Z-Image-Fun-Lora-Distill-4-Steps-2602-ComfyUI lora的lora權重設定為1。(此工作流"ZImageTurbo"也可使用，只需不要添加Turbo lora.)--Download Lora Manager.Download Z-Image-Fun-Lora-Distill-4-Steps-2602-ComfyUI lora.Set the lora weight of Z-Image-Fun-Lora-Distill-4-Steps-2602-ComfyUI lora to 1.(This workflow "ZImageTurbo" can also be used, just don't add Turbo Lora.)--先安裝缺失的節點.下載face_yolov9c.pthttps://huggingface.co/Bingsu/adetailer/blob/main/face_yolov9c.pt將face_yolov9c.pt放入以下路徑:ComfyUI\models\ultralytics\bbox--下載sam_vit_b_01ec64.pth將sam_vit_b_01ec64.pth放入以下路徑:ComfyUI\models\sams--如果面部修復功能沒有生效時，需提高"guide size"的數字."FaceDetailer"安裝有錯誤時，請參考"ComfyUI-Impact-Pack"中的教學.--四倍放大圖片需下載4x-Ultrasharp,並將它放入以下路徑: ComfyUI\models\upscale_models“放大 latent 空間”中的”去雜訊強度”數字越大整題構圖的改變將會越大。--Install the missing nodes first.Download face_yolov9c.pthttps://huggingface.co/Bingsu/adetailer/blob/main/face_yolov9c.ptPlace face_yolov9c.pt in the following path:ComfyUI\models\ultralytics\bbox--Download sam_vit_b_01ec64.pthPlace sam_vit_b_01ec64.pth in the following path: ComfyUI\models\sams--If the facial repair function is ineffective, the "guide size" number needs to be increased.--If you encounter an error during the installation of "FaceDetailer", please refer to the tutorial in "ComfyUI-Impact-Pack".--To enlarge an image by 4x, download 4x-Ultrasharp and place it in the following path: ComfyUI\models\upscale_models.The larger the "denoise" number in "Enlarge the latent space", the greater the change in the overall structure of the problem.text_encodersqwen_3_4b.safetensorsdiffusion_modelsz_image_bf16.safetensorsvaeae.safetensors我的其他檢查站和Lora。My other checkpoints and Lora.
Smooth Workflow (txt2img)
SMOOTH WORKFLOWUPDATE: Smooth Workflow v.4!With support for ZImage! Choose between a a workflow for ZImage or Illustrious/Pony generations.Made it very simple to use. ^^Both Workflows have to option to save with or without Adetailer/Face Detailer.All images on the Showcase used the new SmoothMix Ultimate ZImage Turbo model!ZImage parameters are very different for the Illustrious/Pony models. Make sure to not confuse them.ComfyUI updates can be very frustating as they can make many extension incompatible/stop working - So I tried to use as little extensions as possible on v.4.Shout out to Martin Rizzo (@Photographer) for making probably the best 'Save Image' Node compatible with CivitAI. It just works! =DUsed extensions on v.4:ComfyUI-KJNodesComfyUI-mxToolkitComfyUI-Easy-Usergthree-comfyComfyUI-ZImagePowerNodesComfyUI-Impact-Pack and ComfyUI-Impact-SubpackSmall UPDATE: Smooth Workflow v3.1!Facedetailer should work much better.Used extensions:Save image with generation metadataComfyUI-Easy-UseComfyUI-Image-SaverComfyUI-Impact-Pack and ComfyUI-Impact-SubpackComfyUI-mxToolkitAdvanced CLIP Text Encode (needed for workflow v.2 and v.3)ComfyUI-FBCNN (Flexible Blind JPEG Artifacts Removal) (needed for workflow v.3)UPDATE: Smooth Workflow v3!Changed the node used for Positive Prompts: and ComfyUI-Impact-SubpackThe image results should propely show the effect of adding weights to LoRAs now!The Positive Node is now divide 2 areas. Your Positive Prompt goes on the top one!Use "Select to add LoRA" to open a list of all LoRAs you have installed!Lets say you select "Smooth Booster v.3" on your list. It will show up on the prompt as <your_Loras_path\Smooth_Booster_v3>.Remember to set the Weight you want on the end!! <your_Loras_path\Smooth_Booster_v3:0.5> Like so!Check the showcase image for an example!The bottom area is used to show the prompt that was executed. Its used for adding Wildcards and see the prompt used in the execution. If you don't use Wildcards don't worry about it. For more info check its Github on the Wildcards Nodes section.Added extension: Flexible Blind JPEG Artifacts RemovalAdded Compression Removal to the "(7) SAVE IMAGE" section.This is a JPEG artifact/compression removal tool to help clean up the image a little. Its set to "auto detect - enable" by default!More advanced users can disable it an modify the compression level as they wish.For an indepth look at it check this Github!For a even more indepth look check the Paper on Cornell Unniversity.===========================================================UPDATE: Smooth Workflow v2!Added extension: Advanced CLIP Text EncodeAdded CLIP to the "(4) IMAGE PARAMETERS" section.If you are having dificulties recreating images from CivitAI or tools using A1111, like Forge, selecting "stop_at_clip_layer" to -2 will help to recreate images similar to those more easily.For more advanced users the Advanced CLIP Text Encode was added to the "(3) PROMPTS" sectionWith these you can change both "weight_interpretation" between A1111, ComfyUI and others. For more info check the GitHub.==========================================================Its divided in sections to make it easy for begginers to navigate.(1) CHECKPOINT / VAEVery straight foward. Just choose your Checkpoint and VAE.(2) IMAGE SIZESelect your image size. I can freely adjust it. By default its set at 832x1216 since its the one used by CivitAI.(3) PROMPTSType your positive and negative promtps here.To use embeddings use this format: embedding:your_embedding_hereExample: embedding:SmoothNoob_QualityTo use Loras, use this format: <lora:you_lora_here:weight_number_here>Example: <lora:Smooth_Booster:0.5>Doing so will make sure CivitAI can detect your embeddings and loras propely.(4) IMAGE PARAMETERSSet the number of STEPS and CFG. By default its set as STEPS: 30 and CFG: 5.0.Choose the Sampler and the Scheduler. By default its set as Sampler: DpmPP2M and Scheduler: KarrasYou can also set the SEED in this section.(5) HIRESFIXHere the HiresFix will Rescale your image. You can choose the method: "by percentage", "set Width/Weight" or "to longer side - maintaing aspect". By default its set "By percentage" at 40. This will upscale your image to 1328x1944 if you set image size as 832x1216. Choose your favorite method.Make sure to set your favorite upscaler model as well. By default its set to REMACRI.(6) ADETAILERHere you can adjust the face detailer. The default settings work really well already but feel free to make any adjustments.(7) SAVE IMAGEHere you can choose the path your images will be saved.I installed ComfyUI using StabilityMatrix, so I set my path to F:\StabilityMatrix\Images\Text2Img\%date.Remember to change this in accordance to your PC settings. I recommend leaving the %date since it create a folder with the date the images were created. It helps keeping everything in order.=======================================================Other recommend extensions:ComfyUI-ManagerComfyUI_essentialsComfyUI_PNGInfo_SidebarComfyUI-Custom-ScriptsComfyUI-Autocomplete-Plus
Zimage-base-MeanCache
You can refer to the video above for a demonstration of this workflow, including its effects and usage instructions.Try Oline：https://www.runninghub.ai/post/2019376193790808065/?inviteCode=rh-v14015 mode Compare oline:https://www.runninghub.ai/post/2019746684867190785/?inviteCode=rh-v1401When the MeanCache node is enabled for acceleration starting from step 3, it can achieve up to 100% speed increase, though this may lead to a decline in detail construction. Generally, the Balanced mode is recommended to balance quality and speed. The turbo mode is suitable for quickly testing ideas, while custom is better suited for users already familiar with this node.For an optimal experience, I recommend applying caching acceleration between 10% and 90% of the steps. This provides a good balance of speed while keeping quality within an acceptable range!Next is a detailed analysis of the parameters.：Core Principle of MeanCache: By caching and reusing results from similar denoising steps, it reduces the number of model computations, thereby significantly accelerating the generation process.cache_device: The device used to store the speed cache.- cpu: Saves VRAM. There is a slight data transfer overhead between CPU and GPU, which may slightly affect generation speed.- cuda: Graphics Processing Unit. Offers faster access speed but consumes more VRAM.rel_l1_thresh:This parameter is the relative L1 threshold for skip decisions. Simply put, at each denoising step, the MeanCache node needs to decide whether to skip computation.If the current latent variable is sufficiently similar to the cached latent variable, allowing the cached result to be reused, then it skips; otherwise, it proceeds with computation.A lower threshold results in higher quality latent variables. A higher threshold leads to more noticeable speed improvements but may cause quality degradation.The recommended value is around 0.2 to 0.4.skip budget:This is the maximum number of steps that can be skipped throughout the entire denoising process. For example, with a 50-step process and a skip budget of 10 steps, up to 10 steps can be skipped.A larger value means faster speed (e.g., 0.3 indicates a 30% speed boost), but it will inevitably lead to a drop in quality.start step:The step at which the acceleration node begins to take effect.end step:The step at which the node stops caching and accelerating.These two parameters control the range of steps during which the node is active, including the steps eligible for skipping. Therefore, it is advisable to apply caching acceleration primarily during the middle steps.For instance, in a 50-step process, accelerating between steps 5 and 40 is ideal. The early steps should be preserved as much as possible to ensure the integrity of the image's basic structure.The final few steps are for refining details. For high-quality portraits, sacrificing some speed in these later steps is acceptable.If the generated images lack detail, you can configure the node to stop caching acceleration earlier.enable pssp Simple Threshold Mode:Used to filter or smooth certain noise or artifacts during image generation. For those prioritizing generation speed or the original effect, this can be turned off.peak threshold:This is the threshold for the filtering mentioned above. A higher value produces "cleaner" images but at the cost of lost details.adaptive k:An adaptive parameter k is a coefficient that controls the degree of algorithm adaptation, used to dynamically adjust the intensity of smoothing or processing.debug:Debug mode. Enable this when troubleshooting image anomalies or studying the node's principles. It outputs additional intermediate step information to the console.
Z-image+Base+Meancache Accelerated Edition V1
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/2019064610539704322/?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/BV15ffZBNE39 Knowledge 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!
Z-image+Base+Distilled Accelerated Edition
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/2019063606171668481/?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/BV15ffZBNE39 Knowledge 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!
Z-Image-Base+ Comparison of Multiple Acceleration Methods
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/2019021050884923393/?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/BV15ffZBNE39 Knowledge 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!
ComfyUI beginner friendly Z-Image Base (or Z-Image Base GGUF) Image-to-Image with LORAs Workflow by Sarcastic TOFU
This is a very simple ComfyUI beginner friendly Image-to-Image workflow that will work with Z-Image Base model or it's GGUF model equivalents. I made this workflow to use Z-Image Base's awesome AI photograpic capabilities to convert flat vector or anime images to AI photos. It currently does not transfer poses just concepts based on slightly altered prompts out of original with denoise value between 0.70 to 0.90 (but I will create a Controlnet pose tranfer version once ControlNet Union will be available for ZiB). I used BOTH full non-distilled base model and a Q6 Model to test this workflow and both work well with my 8GB AMD GPU. But if you have a weaker GPU you can find other low grade hardware supported GGUF models from these HuggingFace Repos -Jayn7's HuggingFace repo for Z-Image Base GGUF models ---------------------------------------------------------https://huggingface.co/jayn7/Z-Image-GGUF/tree/mainUnsloth's QWEN3 GGUF Text Encoders for Z-Image Base GGUF models -------------------------------------------------------------------https://huggingface.co/unsloth/Qwen3-4B-GGUF/tree/mainThe newly released Z-Image Base model, developed by Tongyi-MAI (Alibaba's AI lab), is the non-distilled foundation of the Z-Image family, designed to serve as the "raw" starting point for the open-source community. Unlike Z-Image Turbo, which is a distilled version optimized for ultra-fast performance and sub-second inference using as few as 8 steps, the Base model is a full-weight checkpoint that requires more steps (typically 28 to 50) but offers significantly higher diversity in poses, compositions, and artistic styles. While Z-Image Turbo is specifically tuned for immediate photorealism and consumer-grade GPU efficiency, Z-Image Base is built for researchers and creators who need a more flexible model for fine-tuning, training LoRAs, or achieving complex prompt adherence that the faster, more "opinionated" Turbo variant might miss. Compared to other specialized versions like Z-Image Edit, which is fine-tuned specifically for instruction-based image-to-image tasks, the Base model provides the most neutral and unrefined data distribution, making it the superior choice for those looking to push the 6-billion parameter S3-DiT architecture into new, creative territories beyond just speed.You need a Hugging Face account to download your necessary files (Details are mentioned below). Make sure you install GGUF addon for ComfyUI using ComfyUI manager or any other missing nodes you may have and place the correct files in correct places. Also check out my other workflows for SD 1.5 + SDXL 1.0, WAN 2.1, WAN 2.2, MagicWAN Image v2, QWEN, Flux.2 Klein, HunyuanImage-2.1, HiDream, KREA, Chroma, AuraFlow, Z-Image Turbo and Flux. Feel free to toss some yellow Buzz on stuffs you like.How to use this -#1. Just select your desired Z-Image Base (or swap it with GGUF model files) model files first and now#2. now select a Reference Image for Image-to-Image process#3. then input your positive and negative prompts.#4. select how many images you want (Change the number besides the "Run" button)#5. select image sampling methods, CFG, steps etc. settings#6. finally press the run button to generate. That's it..** LORA usage for this workflow is optional you can use it without any LORAs, use with 1 or 2 or any other number of LORAs, too add new LORAs press the L button on top to lunch LORA Manager on a new tab find your LORA and if you want to use that LORA just click the upward Kite button. For both example Images I used this Z IMAGE PAWG LORA for Z-Image Base from here - https://civitai.com/models/2262743/z-image-pawg?modelVersionId=2639737Required Files==================### Download Link for Z-Image Base full non-distilled Checkpoint --------------------------------------------------------------------https://huggingface.co/Comfy-Org/z_image/resolve/main/split_files/diffusion_models/z_image_bf16.safetensors### Download Link for Text Encoder of Z-Image Base full non-distilled Checkpoint ------------------------------------------------------------------------------------https://huggingface.co/Comfy-Org/z_image/resolve/main/split_files/text_encoders/qwen_3_4b_fp8_mixed.safetensors### Download Link for VAE of Z-Image Base (or Z-Image Base GGUF) Checkpoint -------------------------------------------------------------------------------https://huggingface.co/Comfy-Org/z_image/resolve/main/split_files/vae/ae.safetensors### Download Link for Z-Image Base Q6 GGUF Checkpoint ---------------------------------------------------------https://huggingface.co/jayn7/Z-Image-GGUF/resolve/main/z_image-Q6_K.gguf### Download Link for Text Encoder of Z-Image Base Q6 GGUF Checkpoint -------------------------------------------------------------------------https://huggingface.co/unsloth/Qwen3-4B-GGUF/resolve/main/Qwen3-4B-Q6_K.gguf
Jibs double Turbo ZIB to ZIT Workflow
This workflow can make a decent 1024x102 image in 12 seconds, a good image in 30 seconds and a really good one in 60 seconds, depending on how long you want to let it cook!VAE Tech Merge Node: https://civitai.com/models/2231351?modelVersionId=2638152ZIB 8 step lora: https://civitai.com/models/2362961?modelVersionId=2657506
ComfyUI beginner friendly Z-Image Base (or Z-Image Base GGUF) Text-to-Image with LORAs Workflow by Sarcastic TOFU
This is a very simple ComfyUI beginner friendly Text-to-Image workflow that will work with Z-Image Base model or it's GGUF model equivalents. I used BOTH full non-distilled base model and a Q6 Model to test this workflow and both work well with my 8GB AMD GPU. But if you have a weaker GPU you can find other low grade hardware supported GGUF models from these HuggingFace Repos -Jayn7's HuggingFace repo for Z-Image Base GGUF models ---------------------------------------------------------https://huggingface.co/jayn7/Z-Image-GGUF/tree/mainUnsloth's QWEN3 GGUF Text Encoders for Z-Image Base GGUF models -------------------------------------------------------------------https://huggingface.co/unsloth/Qwen3-4B-GGUF/tree/mainThe newly released Z-Image Base model, developed by Tongyi-MAI (Alibaba's AI lab), is the non-distilled foundation of the Z-Image family, designed to serve as the "raw" starting point for the open-source community. Unlike Z-Image Turbo, which is a distilled version optimized for ultra-fast performance and sub-second inference using as few as 8 steps, the Base model is a full-weight checkpoint that requires more steps (typically 28 to 50) but offers significantly higher diversity in poses, compositions, and artistic styles. While Z-Image Turbo is specifically tuned for immediate photorealism and consumer-grade GPU efficiency, Z-Image Base is built for researchers and creators who need a more flexible model for fine-tuning, training LoRAs, or achieving complex prompt adherence that the faster, more "opinionated" Turbo variant might miss. Compared to other specialized versions like Z-Image Edit, which is fine-tuned specifically for instruction-based image-to-image tasks, the Base model provides the most neutral and unrefined data distribution, making it the superior choice for those looking to push the 6-billion parameter S3-DiT architecture into new, creative territories beyond just speed.You need a Hugging Face account to download your necessary files (Details are mentioned below). Make sure you install GGUF addon for ComfyUI using ComfyUI manager or any other missing nodes you may have and place the correct files in correct places. Also check out my other workflows for SD 1.5 + SDXL 1.0, WAN 2.1, WAN 2.2, MagicWAN Image v2, QWEN, Flux.2 Klein, HunyuanImage-2.1, HiDream, KREA, Chroma, AuraFlow, Z-Image Turbo and Flux. Feel free to toss some yellow Buzz on stuffs you like.How to use this -#1. Just select your desired Z-Image Base (or swap it with GGUF model files) model files first and now#2. set your desired image dimensions to start#3. then input your positive and negative prompts.#4. select how many images you want (Change the number besides the "Run" button)#5. select image sampling methods, CFG, steps etc. settings#6. finally press the run button to generate. That's it..** LORA usage for this workflow is optional you can use it without any LORAs, use with 1 or 2 or any other number of LORAs, too add new LORAs press the L button on top to lunch LORA Manager on a new tab find your LORA and if you want to use that LORA just click the upward Kite button.Only on the first example Image I used this Flat Color Style LORA from here - https://civitai.com/models/1132089?modelVersionId=2637093Required Files==================### Download Link for Z-Image Base full non-distilled Checkpoint --------------------------------------------------------------------https://huggingface.co/Comfy-Org/z_image/resolve/main/split_files/diffusion_models/z_image_bf16.safetensors### Download Link for Text Encoder of Z-Image Base full non-distilled Checkpoint ------------------------------------------------------------------------------------https://huggingface.co/Comfy-Org/z_image/resolve/main/split_files/text_encoders/qwen_3_4b_fp8_mixed.safetensors### Download Link for VAE of Z-Image Base (or Z-Image Base GGUF) Checkpoint -------------------------------------------------------------------------------https://huggingface.co/Comfy-Org/z_image/resolve/main/split_files/vae/ae.safetensors### Download Link for Z-Image Base Q6 GGUF Checkpoint ---------------------------------------------------------https://huggingface.co/jayn7/Z-Image-GGUF/resolve/main/z_image-Q6_K.gguf### Download Link for Text Encoder of Z-Image Base Q6 GGUF Checkpoint -------------------------------------------------------------------------https://huggingface.co/unsloth/Qwen3-4B-GGUF/resolve/main/Qwen3-4B-Q6_K.gguf
Z Image Base / Z Image Turbo Workflow -Includes Lora Stack / Face Detailer / SeedVR2 Upscale
Super simple, super quick workflow.Contains a Lora Stack and SeedVR2 UpscaleWould love to hear any feedback and see you share your images!I ran some simple prompts from some of my favorite Z-image posts found here for samples.
Z-image-based dual sampling
Get the latest global workflows on my RunningHub homepage and run them free on RTX4090.Sign Up with my invite link for 1000 bonus points + 100 daily:https://www.runninghub.ai/?inviteCode=rh-v1221Run Directly:https://www.runninghub.ai/post/2016379330103873538/?inviteCode=rh-v1221Please credit and link to my homepage if reprinting.国内合作加Q：549791525It ensures both the speed of output and the details of the base，The first sampling uses the base model, and the second sampling uses the turbo model，Finally, use the Klein model for high-definition magnification
Z-Image Base w/ SeedVR2 Upscaler, GGUF and Safetensors model supported
I converted my Z-Image Turbo Workflow to a Z-Image Base Workflow. Still has a Lora Loader and the ability to choose between using the gguf or safetensors version of the model (just bypass the one you don't want to use).
Z Image Base with Merged Clip (Qwen3 4b & Qwen3 4b Thinking)
My workflow for Z Image Base with merged clip (Qwen3 4b & Qwen3 4b Thinking). Using ClownsharKsampler and a fixed seed. I find the optimal result is 30 steps and cfg 5.5 with res2m/beta or res2s/bongtangent and to generate at 1440x1440 (or equivalent rations). Negative prompt based on a Qwen negative that I found on civitai ages ago.You can find that Thinking clip here: https://huggingface.co/Qwen/Qwen3-4B-Thinking-2507
Z image clown shark sampler
Upgraded basic Z-Image process. Based on the custom clown shark sampler. This is a more refined and high-quality render setting than the basic sampler. Lora AntiBlur and Upscale have been added.If the clown shark sampler is not working, then you need to update everything. You need more up-to-date libraries today. For example, llama, NumPy, and Numba. There may be conflicts between libraries when loading ComfyUI. This needs to be fixed, and the neural network will write the code for you.Lora - Anti-blur Flux Lora - v1.0 | Flux LoRA | CivitaiUpscale models - 4x-UltraSharpV2.safetensors · Kim2091/UltraSharpV2 at mainPs: MultiGPU works well with the llama_cpp_python library version 0.3.4 . But first, you will need to update pip and PyTorch. Then, check for any startup errors and resolve them. Remember to back up your libraries.
Z-Image (base) t2i workflow
zimage (base) t2i workflowMore workflow parameter settings, please refer to my settings on rh:https://www.runninghub.ai/post/2016446913859559425/?inviteCode=rh-v1315Click the link to download the workflow. After registering and binding my invitation code: rh-v1315, you can get an extra 1000 RH coins. You can also get 100 points by logging in daily. It can be used to generate images or videos.
Z-Image Base + LoRA Simple Workflow
Z-Image Base + LoRA With Upscaler Simple Workflow
Friendly Z-Image Base
Welcome to my 💫🌄 Friendly Z-Image Base✨ Less mess, more magicZ-Image - The foundation model behind Z-Image-Turbo. Z-Image focuses on high-quality generation, rich aesthetics, strong diversity, and controllability, well-suited for creative generation, fine-tuning, and downstream development. It supports a wide range of artistic styles, effective negative prompting, and high diversity across identities, poses, compositions, and layoutsUse this worflow in the future with fine-tuning models based on Z-Image!Generate 2048 x 2048 images even with 6GB VRAM! I offer my workflow, with which you can generate on a full BF16 model without sacrificing quality and at the same speed. This requires at least 6GB of video memory and at least 32GB of RAM. No Tritons or SageAttention! Simply configure the required amount of video memory dumped to RAM in the "VRAM Optimizer+" node.The latest version of ComfyUI is required.💻 System requirements:Minimum system requirements:RTX 3000-s, 6GB VRAM, 32GB+ RAM, 8-core processor, SSD, latest ComfyUI📌 Detailed tips and links to models in the workflow✨ Workflow features:Extremely user-friendly interfaceMaximum performance even on low-end systems with at least 6GB of VRAMEnchance details parameterSupport for up to 3 LoraDetailed tooltips, including recommended samplers for best qualityManual random seed for complete control over generations🤗🙏🏼 Thanks to Tongyi-MAIOriginal repo — GitHub
Z-Image Base T2I Compact
A simple and compact txt2img workflow for Z-Image Base with GGUF and Safetensor support.
Z-Image GGUF with Detail Daemon
Hello, handsome people!This is just a simple Z-Image workflow using GGUF models, and Detail Daemon.I use Qwen3-4B-UD-Q8_K_XL, and z_image_Q8_0, but you can use whatever is best for you.The links are below. Also, I use res_2s as a sampler, but it's slower than anyone else, so you might want to try others; however, res_2s is a beast!. CFG 3-4 is good. I use 30 steps, but 25 is alright, Alpha 0.5 Beta 0.7 gives good contrast. If you don't have res_2s, you need the RES4LYF node (link below).Resource linksText EncodersQwen3-4B-UD-Q8_K_XL.ggufDiffusion Modelsz_image-Q8_0.ggufVAE (flux)flux_ae.safetensors (A.K.A. ae.safetensor)GGUF Custom NodeComfyUI-GGUFDetail Deamon Custom NodeComfyUI-Detail-DaemonRES4LYF Custom NodeRES4LYFNOTE: This is a Z-Image workflow, not Z-Image-Turbo, but Civitai doesn't have the category yet. BTW, if you can donate some BUZZ, so I can make LoRAs for Z-Image and Chroma, I would appreciate it.
Rebels Z-Image BASE
condensed workflow for the base model of z-image. (NOT TURBO MODEL)Modelhttps://huggingface.co/Comfy-Org/z_image/tree/main/split_files/diffusion_modelsGGUF version (if your vram cant handle bf16)https://huggingface.co/jayn7/Z-Image-GGUF/tree/maintext encoderhttps://huggingface.co/Comfy-Org/z_image/tree/main/split_files/text_encodersvaehttps://huggingface.co/Comfy-Org/z_image/tree/main/split_files/vaeyou have to play with the steps and cfg.steps - 25-50cfg - 4.0 - 6.0if your using the GGUF model, youll need to swap out the diffusion model loader node with the gguf loader node.