ComfyUI Workflows
Browse real AI workflows from Flux, SDXL, Pony, ControlNet, video generation and commercial AI creators.
Sillytavern Expressions Workflow
Another ST workflow, but hopefully not as big as some of the others I've seen. On my 3060, images take 15s to finish, and all 28 takes about 7 minutes. Updates are mostly to fix node problems. If yours works, there's no need to update. Works best on images with black backgrounds.There are more instructions in the workflow, but the most important thing to remember is to set the batch size to 28 to get all the emotes.Nodes aligned with NodeAligner:https://github.com/Tenney95/ComfyUI-NodeAlignerThis was made with a version of Comfy with an older version of Python (from 6/27/25). Newer versions have some crap about WAS Suite and whatever OpenCV is that prevents Comfy from launching properly.
Reed Anima T2I Workflow
Reed Anima T2I Workflow
Easy Anima - Workflow for ComfyUI
Easy Anima Workflow v0.1 - (See installation requirements)This ComfyUI workflow is designed for Anima models. It features an all-in-one solution, with txt2img > hires fix > face detailer. The primary goal of this workflow is to provide an intuitive, easy-to-use UI that mirrors A1111 / Invoke, while remaining flexible for advanced use cases. The final image will include embedded metadata to make uploading to Civitai easier.Positive prompts now support wildcards and dynamic tags!Wildcards: Reference files from your ComfyUI custom nodes directory by enclosing the filename in double underscores (e.g., __filename__).Dynamic Tags: Create instant random variations directly in your prompt using curly brackets separated by vertical bars, like {blue|red} ball or {sitting|standing}.Wildcard preview: Use the 'Preview Wildcard' node to view the final prompt generated by your wildcards and dynamic tags.Installation RequirementsThis workflow requires the following model types:Anima Diffusion Model: ComfyUI/models/diffusion_modelsText Encoder - qwen_3_6b_base.safetensors: ComfyUI/models/text_encodersVAE - qwen_image_vae.safetensors: ComfyUI/models/vaeUpscale Model eg. remacri: ComfyUI/models/upscale_modelsThis workflow also requires these node packs in order to function, these can be easily installed via the built-in Custom Nodes Manager:
Text2Img Workflow with Detailer Arrays
Hello Everyone!Have you ever created an image you like and would have kept, but and a face, or hand just came out wrong? You need a workflow that can take control address that.Ever make an image that had multiple characters, but when you tried to assign them the correct eye color or expression, the checkpoint got confused, and you had to play generation roulette? You need a workflow that lets you decide whose face has just came out like what.That is what this workflow, used properly, will allow you to do. This workflow is my daily driver T2I workflow for checkpoints based in SDXL, Illustrious and Pony. It is based upon my Img2Img SDXL Workflow + META, which I was only able to create thanks to bits taken from other workflows, particularly those of Legendaer and Yukichan_, and advice from Lonecatone23. Over several revisions, I have made a number of refinements to the workflow while at the same time striving to keep it simple so that even a beginner can take hold of this workflow and start generating.Main Features:✦ Text to Img Generation - That's what we're here for, text to image.Include Metadata / Embed Workflow Option. Stable 1 megapixel generation, or dare to increase it with Size Boost.✦ Metadata saving included — CivitAI ready. You can also embed the entire workflow into your images, by default this is turned off.✦ Upscaler with Color Matching - this is the feature that makes images go from good to great. Or use HiresFix for a quick increase in resolution✦ Quality Tags Presets Switch - instead of typing Masterpiece, best quality, etc. etc.✦ Two Segs Detailer Arrays - two SEGS pipelines (by default, hands and faces) that you can individually prompt or apply one prompt to all. Sophisticated tool created to handle individual details - your backstop for when the image has to be perfect.✦ Parallel development with I2I workflow - if you learn this one my Img2Img works essentially the same way.✦ CFG Controls - need more control, more CFG? It's handled automatically, crank it up to 10.✦ LoRA Manager — Search your LoRA database with style, I love this node. Dump it if you prefer something else.✦ Simple and reliable — Enter your prompt, hit run, let the segs previews populate, freeze the gen seed, and fill in prompts and run it again. That's it.New with 5.0 SESkipped 4.0, just becauseSeed JumpStyle Preset switchingAdded a toggle and Subgraph for exceeding native SDXL resolution on initial genFixed Typos for name of Face Detailer LoRA blocker nodesDoes not include Quality Tags Fast Switching, uses native concatenate nodes5.0.1 Changes:Changed Positive Text Box to Mira, which was already being used in this workflowChanged Wildcard ProcessorSeed controls Wildcard selection (Freezing seed freezes your wildcards too now)Added Final Prompt PreviewAdded a toggle and Subgraph for exceeding native SDXL resolution on initial genSlight rearrangement of Control Group, hopefully optimalNew with 3.0:With 3.0, the layout has changed and metadata flows have changed. The checkpoint loader is now the checkpoint selector, which should provide some convenience and added stability. The LoRA blocker no longer requires fiddling with the spaghetti string to toggle it, there is now a toggle switch. Sliders have been implemented throughout the Input center, including controls for the Upscaler.New with 2.2:Added toggles for including the workflow in the image's metadata.Added a means by which you can bypass the upscaler and still send the image (directly from phase 1 gen/hires fix) to the detailer, rather than being forced to use upscaler in order to access the detailer.Color Matching directly following Upscaler - I have found that some models have noticeable drift in color tone and this corrects that, works beautifully.New with 2.1:Detailer moved to the back of the order, after upscale, to ensure that full control over expressions remains with you, the artist. I'm very happy with this workflow in its current form, although there might be new versions or changes down the line.New with 2.0:Raise CFG without crushing the imageUpscaler with a third save nodeSomehow resists LoRA bleed, whether it's the scheduler / sampler combo or what I'd love to know. It's not perfect though!Preview KSampler - a game changer for skipping unsatisfactory seeds. Use it!You will need to download an upscaler model if you don't have one already.How this Workflow Came to BeI was using one of these massive workflows with all the bells and whistles and I ran into a snag. A LoRA was confounding the hand detailer and the hands were all coming out like macaroni alfredo. So, I turned off that LoRA. I was also running into a common problem: when two characters are in the frame, they end up with the same eye or hair color. The tags were blending across character BREAKs which is ultra common, and detailers that handle all bboxs in order cannot be instructed on each bbox individually. That's something that my I2I workflow was designed to do, and the mega huge T2I workflow with all its heft didn't have detailer separation as a feature.I don't think I could add to such a complex workflow without tripping up and it end up taking 2 hours. I thought if I'm going to modify a workflow it might as well be my own. It would keep things simple with only nodes I know I have, I'd finish in under 30 minutes, and then I would have complete control over faces.Saving the metadata is important too, however somehow in converting things and removing things that get concatenated (such as the natural language prompt), the metadata portion got a little scrambulated, and don't worry I fixed it -- but 30 minutes turned into a several hour affair. Under nine hours though, I'm pretty sure. The latter part was mostly testing and tweaking. I ended up making some changes to the detailer modules so that they don't spam the image feed, and upgraded that module to V1.1.During testing, I started with one figure, then two, and I soon discovered that this workflow is absolutely a beast at ensuring your faces all look great. Whether you have one face in the image or five, the detailer can make each one pop with high aesthetic quality and unique expressions, and the SEGS previewer keeps you in control.The hand detailer gives you per-segs control over hands although sometimes bad hands are going to happen, particularly when there are many in one image. Fingers in AI is always a little trickier, but the workflow allows you to add a hand detailer LoRA directly to the detailer, upping your chances of success dramatically. Do it. At the same time, that LoRA will be blocking other LoRAs that could create havoc on delicate fingers. I haven't seen this approach elsewhere, curious whether it holds up in your experience. Please let me know in the comments what you think about my LoRA blocker, if you have any aspects you think ought to be changed, bug fixes, etc. This is just version 1, it's bound to evolve.RequirementsYou will need a face bbox detector - like this oneYou will needs a hands bbox detector - like this oneYou might decide to add or replace a bbox detector for face or hands with one for eyes, or add another bbox array. For eyes, try this oneIn ComfyUI you may need to turn some safety features off to use pickletensors. Consult AI if you run into problems.Custom Nodes Used:rgthree-comfyComfyUI Impact PackComfyUI Impact SubpackComfyUI-Easy-UseComfyUI-Lora-ManagerComfyUI Image SaverComfyUI_essentialsComfyUI-KJNodescg use everywhere Anything EverywhereDynamic ThresholdingComfyui_MiraSkimmed_cfgefficiency-nodes-comfyuiComfyroll StudioComfyMath
Anima All in One workflow
ANIMA all in one workflow.V5.0 Releasei2i now works properly. However, please be careful when adjusting the image resolution.Inpainting is also fully functional. Please refer to the post for more detailed instructions.The Spectrum acceleration method has been replaced with a more stability-focused algorithm.Modulation Guidance has been added. Please enter your “meta tags” in the “positive 2” prompt field. This custom node is designed to help those meta tags work properly.V5.1 Minor UpdateChanged the position and default settings of TorchCompileModelAdvanced.The RuntimeError: Fault failed: 2 error should no longer occur.Added an option to upscale Highres latent images after converting them to images.Enabling the latent를 이미지로 변환 option can help prevent rough or broken-looking results in certain cases.More detailsNAIA2.0NAIA를 설치하지 않은 경우, Fast Groups Bypasser (rgthree)에서 "NAIA 사용"을 비활설화 하세요.4.5 이후는 스위치 토글을 사용합니다.If you have not installed "NAIA", disable "NAIA 사용"NAIA: https://github.com/DNT-LAB/NAIA2.04.5 ~rgthree 를 통해 그룹 단위로 bypass하는 대신, comfyUI의 스위치 노드를 사용해서 컨트롤 합니다. on off 토글을 통해 사용하지 않는 노드는 스킵됩니다.rgthree node has been changed to a comfy switch node. Unnecessary functions can now be turned on and off via toggle, and nodes that are not running will not cause compatibility issues.~3.0"4b. SeedVR" 와 "4a. HighRez" 그룹은 둘 중 하나만 활성화 하세요."4a. HighRez" 사용 시 "전신 디테일러"는 스킵하세요.Activate only one of the "4b. SeedVR" and "4a. HighRez" groups.Skip "전신 디테일러(Full Body Detailer)" when using "4a. HighRez".custom nodesKSampler (Spectrum + Mod Guidance Advanced)https://github.com/sorryhyun/ComfyUI-Spectrum-KSamplerDCWhttps://github.com/namemechan/ComfyUI-DCWSpectrumhttps://github.com/ruwwww/ComfyUI-Spectrum-sdxlComfyUI-Anima-LLLithttps://github.com/kohya-ss/ComfyUI-Anima-LLLiteNaia-bridgehttps://github.com/DNT-LAB/comfyui-naia-bridgeGuide(Korean)V1V2V3V4V5
Anima Style Explorer & Prompt Workflow | Style Blending & Automated Artist Prompting
New attachment updated! It contains the new prompt templates for femboy and otokonoko. By using vocabulary isolation, it effectively prevents limb drift and helps the model better understand their body structures.🚀 [Major Update v2.5] ComfyUI Anima Style Explorer Prompt: The Palette & Automation Revolution!🚨 CRITICAL UPDATE NOTICE: PLEASE READ BEFORE RUNNING!IMPORTANT: This update introduces a deeply integrated Frontend JavaScript architecture. After updating via Git, you MUST restart your ComfyUI backend console AND hard-refresh your browser (Press Ctrl + F5) to clear old web cache. If you do not do this, the new interactive dropdown menus and buttons will not function correctly!🔥 New Core Node: [Anima 2B Dynamic Group Switcher]Say Goodbye to Manual Bypass Stress!Tired of manually clicking dozens of rgthree group toggles or managing messy text switches just to change your rendering style? We built the ultimate automation solution. The new Anima 2B Dynamic Group Switcher links directly to your canvas layout to handle everything seamlessly.🎛️ Dynamic Slot Scaling (➕/➖ Buttons): Click the native on-node buttons to dynamically scale your workflow from 1 to 30 custom text/style channels instantly.🧠 Automatic Group Name Detection: The moment you connect a slot to a style group on your canvas, the node dynamically reads the group's title and re-labels the input port natively (e.g., text_1 [femboy], text_2 [90s Anime]).⚡ Automatic Bypass Matrix Control: When you select an option from the native active_style dropdown menu, the chosen Group automatically activates (Normal mode), while all other connected unselected groups are instantly forced into Bypass mode (turned gray).🛡️ Zero-Crash Backend (VALIDATE_INPUTS Fix): Features a custom backend bypass validator to completely eliminate native ComfyUI Value not in list runtime errors when switching styles mid-generation.🎨 3 New Specially-Tuned Masterpiece Palette TemplatesWe have integrated three incredibly powerful, JoyCaption-optimized aesthetic templates into the core system to instantly elevate your generation textures:1. ❄️ Muted / Morandi / Desaturated Palette (冷调色 / 高级灰)Designed for an elegant, intellectual, and quiet gallery-grade aesthetic.Visual Formula: It strips away harsh or raw digital neon colors, replacing them with highly sophisticated, low-saturation tones (dusty rose, sage green, muted grayish blue) laid over soft grey undertones.Lighting: Simulates a beautifully diffused overcast lighting with minimal contrast and zero harsh shadows, adding deep atmospheric weight to character art.2. 🍂 Retro Vintage / 90s Anime Palette (暖调色 / 90年代旧画风)Recreates the golden era of hand-drawn cinematic animation.Visual Formula: It forces the AI rendering engine to ditch flat modern digital gradients in favor of an authentic 1990s cel-shaded aesthetic. It wraps the entire scene in rich, warm sepia undertones and beautifully faded nostalgia colors.Textures: Seamlessly injects optical vintage film grain and subtle chromatic aberration (lens edge color-bleeding) to make the image look like a classic animation cell.3. 🖤 Color Split / Selective Color Palette (漫画调色 / 局部色彩)Delivers extreme cinematic focus and high-impact visual drama, perfect for stylized manga covers or dark art.Visual Formula: It actively locks the entire environment, background, skin, and primary attire into a rich, high-contrast monochrome grayscale (deep ink blacks and crisp stark whites).The Contrast: It automatically isolates the single most striking visual focus item (such as vivid red lips, glowing gold jewelry, or magical eyes) and splashes it with piercing, high-saturation color against the silent, colorless world.🧩 The Expanded 4-Node EcosystemYour suite of tools has now evolved into a complete industrial-grade production pipeline:Anima 2B Dynamic Group Switcher (NEW!) —— Automated workflow traffic controller. Dynamically activates chosen style canvas groups and silences the rest.Artist Pre-Config Outline —— Handles macro-category wrapping and multi-artist nested weight blending.Anima 2B Prompt Template 1 —— Core parallel multi-weight artist channel generator.Anima 2B Prompt Template 2 —— High-end modifiers, time eras, structural evaluation, and prompt quality scoring.🛠️ How to Download & UpdateMethod A (Recommended): Direct ZIP Download from GitHubVisit the GitHub repository: ComfyUI-Anima-Style-Explorer-PromptClick the green Code button -> Select Download ZIP.Extract and overwrite the contents into your ComfyUI/custom_nodes/ folder. (Ensure the directory name remains exactly ComfyUI-Anima-Style-Explorer-Prompt without a -main suffix).Restart ComfyUI and press Ctrl + F5 in your browser!Method B: Via Git Command LineOpen your terminal inside the ComfyUI/custom_nodes/ directory and run:git clone https://github.com/mizukir0418-gif/ComfyUI-Anima-Style-Explorer-Prompt.git
ANIMA SFW/NSFW W/ Controlnet, Regional Prompt, & Realism enhancer
This is an advanced workflow for those who want to dail in an image at pro level grade. A simple version can be found HERENote: I spend a lot of time on these to make sure they are top notch. If you have ANY issues, I'm very responsive. hit me up and I'll resolve them for you.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-breakdown也有中文说明Instagram: https://www.instagram.com/synth.studio.models/Buy me a☕ https://ko-fi.com/lonecatoneThis represents Many of hours of work. If you enjoy it, please 👍like, 💬 comment , and feel free to ⚡tip 😉V9.0 Note: Both Controlnet & Regional Prompt nodes are in Beta testing. You will not find these to be as rigid as SDXL or Flux based ones, but they are fun to mess with. Added back draft mode by requestadded shhh.. nodes (You'll thank me)Made it so you can use the original image ratio.Super good for regional prompts and ControlnetAdded Regional Prompt (BETA)I'm still testing this out. Not sure how I feel about it.what it does do well: ARTwhat is sucks at: peopleReplaced math nodes to be compatible with Linux usersTweaks here and there.Ver 8.0added controlnetadded Klein enhancementreworked Post Production SuiteMinor tweaks.Ver 6.0Upgraded Prompt assist.Added an inline crop tool for easier image editing.inmproved the save option to include the option to add a date to the metadata file. name for beter organization.Minor note tweaks.Ver 5.2Updated for Comfyui changesAdded sentence remover to prevent CivitAi flagsMade some optimazation changes for Anima 3 previewreworked the user settings area.Ver 4.0:On a neverending quest for opions, I decided to tweak my Anima workflow to get relism and semi-realism out of it for the Anima2 modelAdded a detailer suite based on SDXL or Illustrious for realistic optionsNote: It's not gonna beat ZIT, but with the prompt adherence, It's cool for semi-realisticIncluded my full detailer suiteIncluded my full Post production suiteFine tuned Optical Realism to work with this model (in other words, don't mess with the settings)Ver 3.0 included in download for straight AnimeI had to make modifications as to how the Savenodes worked. ComfyUi broke them (again 🙄)What this workflow contains:Sage Attention: Speeds up productionUltimate Upscaler for prescaling and hi res fixDetailing Suite for face, hands, eyesSeed VR2 Upscaler to get 4k on low VRAMEssential Post Production SuiteJPEG ScrubberFull tutorial available.也有中文说明
Not so simple (or is it?) Anima Workflow
⚡ Anima Workflow:🛠️ Purpose & Design PhilosophyThis workflow is designed for quality and autonomy, not speed. It follows an "all-in-one" philosophy: configure your settings, hit queue, and let the workflow handle everything from initial generation to high-res detailing in a single pass.Not for Speed: If you want rapid-fire generations, this is not the tool for you. A solid, much faster alternative created by darksidewalker can be found here.Personal Use: This was built for my personal production needs. It is not intended to be a "one-size-fits-all" solution, but I am sharing it for those who value the same high-fidelity results. Please adjust the settings to your preferences!Heavy Duty: Due to the multi-stage processing, this workflow can be resource-intensive. In my experience, the detailers are not usually needed. YMMV.On v1g and after: If the results are too blurry for you after USDU, you can try using the RTX nodes after it to help.v2 is compatible with AIO versions of Anima models. Note: You might have to select a random model in the Checkpoint Loader node even if you are not using it. This is because it references your models/checkpoints folder. The opposite may also be the case if you are using the Checkpoint Loader node but not the Model Loader in the Diffusion Model Loader group which references the models/diffusion_models folder.🚀 Key FeaturesBeyond standard generation and upscaling, this workflow integrates:Power LoRA Loader: Efficiently manage multiple LoRAs without spaghetti wires.Global Controls: Centralized Seed, Sampler, and Scheduler nodes for a unified experience.Bypass Control to toggle features on/off.Visual Validation: Integrated Image Comparer nodes to see exactly how your image evolves at every stage.Upscaling: 2-stage upscaling using standard image upscaling and Ultimate SD Upscale (optional).Triple Detailer Groups: 3-stage detailing using standard BBOX and SEGM detection models for faces, hands, and clothes.CivitAI Ready: Images are saved with full metadata (Model, LoRAs, Prompts) for easy site parsing.⚠️ Disclaimer & CompatibilityInstall at Your Own Risk: Updating ComfyUI or adding custom nodes can break your environment. I am not responsible for any installation issues.Portable Version: This was built and tested on the ComfyUI Portable version. Desktop app users may require additional troubleshooting."Your Mileage May Vary": Your environment is almost certainly different from mine. I do not guarantee 1:1 compatibility.Nodes 2.0: I strongly recommend disabling Nodes 2.0. It causes unpredictable behavior; I will not provide support for any issues arising from its use.🤝 Support & BoundariesNo DMs: Direct messages are disabled due to high volume. Please use the Discussions tab below. Check previous comments first (unless there aren't any yet), as most common questions may have already been answered.Custom Requests: I do not take private requests for custom workflows. If you need a specific solution built, please post a Bounty on CivitAI. There are many talented creators here who will be happy to assist you for a fee.Modifications: You are free to add or remove nodes as you see fit. However, if you change the internal logic, you are responsible for your own troubleshooting.The only place I am actively maintaining this workflow is here on civitai. If my current workflows are being posted and monetized elsewhere, whoever posted them is obligated to provide support to those users.I have not and never will monetize my workflow, since it was designed for me by me.
Anima Turbo w/ Negative Prompt Workflow
Anima Turbo + CFG 1 Negative promptA lightweight, beginner-friendly workflow for Anima Turbo that allows negative prompting at CFG 1 using NegPip.Turbo models are fast, but running at CFG 1 means the negative prompt box has no effect. This workflow solves that problem using NegPip, which allows you to apply negative concepts directly inside the positive prompt.How NegPip WorksInstead of using the negative prompt box, place unwanted concepts in the positive prompt with a negative weight:(blurry:-1)(low quality:-2)(text:-4)NegPip interprets these negative weights as negative conditioning, allowing you to steer the image away from unwanted features while keeping the speed benefits of CFG 1 turbo generation.Don't be afraid to use larger values. In my testing, weights up to ±4 can produce strong results with Anima without frying the image.Recommended SettingsCFG: 1Steps: 8–12Sampler: Euler / Euler Ancestral / ER-SDEResolution: Up to 1536×1536 works, but ~1.5 megapixels is a good sweet spot between quality and speed in my opinionIncluded FeaturesAnima Turbo LoRA preconfiguredNegPip integrationImpact Wildcard EncoderMetadata savingSimple single-pass generation pipelineEasy LoRA and wildcard expansionWildcardsCreate .txt wildcard files in:ComfyUI/custom_nodes/comfyui-impact-pack/custom_wildcardsThen call them in your prompt using:__WildcardName__Wildcard randomization is tied to the image seed by default for reproducible generations.LimitationsThis workflow is intentionally minimal. The goal is to provide a clean, reliable starting point for Anima Turbo users who want the speed of CFG 1 generation without giving up the control that negative prompting provides.
Anima HiresFix 1.5x High-Resolution Refinement Workflow
Watch the full video first if you want to understand how this Anima HiresFix workflow works in practice. The video shows how Anima Base can generate a 1024 image first, then use a 1.5x latent refinement stage to improve detail, structure, and final image quality without making the workflow overly complex.This ComfyUI workflow is designed for Anima Base 1.0 high-resolution refinement using a classic HiresFix-style two-stage generation route. Its main purpose is to create a stable 1024×1024 base image first, then upscale the latent to 1536×1536 and run a second controlled refinement pass. Compared with a simple one-pass Anima generation workflow, this structure gives the final image more detail, cleaner linework, better texture density, and stronger overall polish.The workflow is built around anima_baseV10.safetensors as the main model. After loading the model, ModelSamplingAuraFlow applies a shift value of 3, preparing the Anima model for the intended sampling behavior. The text side uses qwen_3_06b_base.safetensors as the text encoder with the stable_diffusion type setting. The output is decoded with qwen_image_vae.safetensors, which keeps the workflow aligned with the Anima / Qwen Image VAE ecosystem.The first generation stage uses an EmptyLatentImage at 1024×1024 with batch size 1. The positive prompt defines the image direction, while the negative prompt suppresses low-quality output, weak score tags, blurry artifacts, JPEG artifacts, and unwanted artist-name artifacts. The first KSampler is configured as the base sample stage, using 34 steps, CFG 4.5, er_sde sampler, simple scheduler, and denoise 1. This stage is responsible for building the main composition, character structure, pose, lighting, and global visual identity.After the base latent is generated, the workflow sends it into LatentUpscale. The upscale method is bicubic, and the target size is 1536×1536, which is exactly a 1.5x increase from the original 1024 base. This is important because it improves the working resolution without jumping too aggressively into an unstable size. A moderate 1.5x latent upscale is often easier to control than a larger upscale when the goal is refinement rather than complete regeneration.The second KSampler performs the HiresFix refinement stage. It uses 18 steps, CFG 4.5, er_sde sampler, simple scheduler, and denoise 0.35. The lower denoise value means the workflow keeps the original composition from the first stage while adding more detail and cleaner high-resolution structure. This stage is where the image gains sharper eyes, cleaner edges, better fabric texture, stronger lighting detail, and a more finished look.Compared with ordinary Anima text-to-image generation, this workflow is better for final image polishing. A one-pass workflow is faster, but it may leave the result slightly soft or under-detailed. This HiresFix workflow is more suitable when the first image already looks correct and you want to enhance it instead of completely changing it.This workflow is suitable for anime character illustrations, high-resolution portraits, fantasy characters, clean lineart images, social media covers, Civitai previews, RunningHub showcases, and any Anima image that needs a controlled 1.5x quality upgrade.Main features:Anima Base 1.0 HiresFix workflow1024×1024 base generation1.5x latent upscale to 1536×1536anima_baseV10.safetensors main model routeQwen 3 0.6B text encoderQwen Image VAE decodingModelSamplingAuraFlow with shift 3First-stage base samplingSecond-stage high-resolution refinementer_sde sampler with simple schedulerCFG 4.5 controlled generationRefine denoise 0.35 for structure preservationClean two-stage image polish pipelineNative SaveImage outputSuggested workflow:Start with a clear prompt that defines the character, subject, style, lighting, composition, and final visual direction. Run the 1024×1024 base stage first and check whether the image structure is correct. Do not rely on the HiresFix stage to fix a bad composition. If the base image is wrong, adjust the prompt and regenerate. Once the base result is stable, use the 1.5x latent upscale stage and run the second sampler with denoise 0.35. If the refined image changes too much, lower the denoise slightly. If the result is too soft, keep the structure but strengthen detail wording in the prompt. Use this workflow when you want Anima output that feels more finished than a simple draft.⚙️ RunningHub WorkflowTry the workflow online right now — no installation required.👉 Workflow: https://www.runninghub.ai/post/2060759010810355714?inviteCode=rh-v1111If the results meet your expectations, you can later deploy it locally for customization.🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance and 48 GB super power!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you’re in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1GmVS6hEzu/☕ Support Me on Ko-fiIf you find my content helpful and want to support future creations, you can buy me a coffee ☕.Every bit of support helps me keep creating — just like a spark that can ignite a blazing flame.👉 Ko-fi: https://ko-fi.com/aiksk💼 Business ContactFor collaboration or inquiries, please contact aiksk95 on WeChat.⚙️打开下方链接即可在线体验，无需安装。👉 工作流： https://www.runninghub.ai/post/2060759010810355714?inviteCode=rh-v1111如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1GmVS6hEzu/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
Anima Edit Image Editing and Local Inpainting Workflow
Watch the full video first if you want to understand how this Anima Edit workflow works in practice. The video shows how Anima Base can be used for image editing, local inpainting, outfit change, and reference-guided transformation, while keeping the workflow online and easy to test without rebuilding the full ComfyUI environment locally.This ComfyUI workflow is designed for Anima image editing and local transformation. Its main purpose is to take an existing image, define the edit direction through prompt and mask logic, then regenerate the target area while preserving the original image structure as much as possible. Instead of using Anima only for pure text-to-image generation, this workflow turns Anima into a practical editing pipeline for changing clothes, adjusting appearance, modifying a scene, testing character edits, and comparing different edit LoRA routes.The workflow is built around anima_baseV10.safetensors as the main model. It uses qwen_3_06b_base.safetensors as the text encoder and qwen_image_vae.safetensors as the VAE. The input image is loaded through LoadImage, then resized to a controlled 1024×1024 working canvas. This gives the editing process a stable base resolution and avoids random size mismatch problems.A key part of the workflow is the IC-style side-by-side editing structure. AILab_ICLoRAConcat combines the original image and edit target into a left-right layout, producing both an image and a base mask. This structure helps the model understand that the right side should become the edited result while the left side provides the original reference. The workflow also uses InpaintModelConditioning to prepare positive conditioning, negative conditioning, latent image, image pixels, and mask data for local inpainting.The workflow includes Anima-LLLite inpainting v2 through AnimaLLLiteApply. This module injects lightweight inpainting guidance into the Anima model, helping the model follow the image and mask more accurately during local edits. It is especially useful when the target is not to redraw the whole image, but to change a selected area while keeping identity, pose, framing, and background logic more stable.The graph also includes two edit LoRA routes: AnimaEditV1 and lora_edit_ZeroTwo. These are loaded through LoraLoaderModelOnly and connected into separate generation branches. This makes the workflow useful for testing different Anima editing behaviors. One route can be used for general Anima Edit experiments, while another can be used for a more specific edit style or clothing-change behavior. The workflow also applies Cosmos Reference latent logic through ApplyCosmosReferenceLatent, which helps preserve reference structure during the edit process.Compared with ordinary image-to-image workflows, this graph is more specialized for Anima-based editing. A basic img2img workflow may change too much or fail to understand the target edit. This workflow combines mask-aware conditioning, IC-style layout, LLLite inpainting guidance, edit LoRA branches, reference latent control, and prompt template replacement, making it better for controlled transformations.This workflow is suitable for outfit change, local character editing, anime image repair, style-consistent edits, before/after comparison images, masked inpainting tests, clothing experiments, character concept revision, RunningHub demonstrations, and Civitai workflow publishing.Main features:Anima Base image editing workflowLocal inpainting and image transformationanima_baseV10.safetensors main model routeQwen 3 0.6B text encoderQwen Image VAE decodingAnima-LLLite inpainting v2 guidanceAnimaEditV1 edit LoRA routelora_edit_ZeroTwo edit LoRA routeAILab_ICLoRAConcat side-by-side edit layoutInpaintModelConditioning mask-aware editingApplyCosmosReferenceLatent reference preservationPrompt template replacement systemMultiple edit branches for comparisonPreviewImage output for fast inspectionSuggested workflow:Prepare a clean source image first. The subject should be clear, and the area you want to edit should not be too small or heavily blocked. Load the image into the workflow, then write a direct edit prompt such as changing clothing, changing accessories, adjusting the scene, or modifying a selected visual feature. Use the IC-style layout and mask conditioning to define the edit target. Start with the Anima-LLLite inpainting route first to check whether the mask and edit direction are correct. Then compare the AnimaEditV1 and lora_edit_ZeroTwo branches to see which route gives better identity preservation and cleaner transformation. If the edit changes too much, simplify the prompt and strengthen reference consistency. If the edit is too weak, make the target instruction more explicit.⚙️ RunningHub WorkflowTry the workflow online right now — no installation required.👉 Workflow: https://www.runninghub.ai/post/2060760904450535425?inviteCode=rh-v1111If the results meet your expectations, you can later deploy it locally for customization.🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance and 48 GB super power!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you’re in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1GmVS6hEzu/☕ Support Me on Ko-fiIf you find my content helpful and want to support future creations, you can buy me a coffee ☕.Every bit of support helps me keep creating — just like a spark that can ignite a blazing flame.👉 Ko-fi: https://ko-fi.com/aiksk💼 Business ContactFor collaboration or inquiries, please contact aiksk95 on WeChat.⚙️打开下方链接即可在线体验，无需安装。👉 工作流： https://www.runninghub.ai/post/2060760904450535425?inviteCode=rh-v1111如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1GmVS6hEzu/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
ANIMA prompting skill/agent/system prompts
## TitleAnima prompt skill systemprompt: Let LLM understand both Danbooru tags and natural language while preserving wildcards without altering them---## Why this?Anima-style models have a unique advantage: they accept both Danbooru tags (comma-separated keywords) and natural language (full sentences) as input.But here's the problem:- If you feed pure tags, the image lacks spatial relationship descriptions (Where is the subject? Is the background in front or behind?)- If you feed natural language, you waste the precise control that tags offer- Even worse, LLMs often arbitrarily expand wildcards (turning {A|B} into A or B) or delete tags they don't recognizeSo I wrote this System Prompt with a simple goal:> Turn the LLM into a "2D visual coordination specialist," not a novelist or a translator.---## What does this System Prompt do?| Input Type | Handling || --- | --- || Danbooru tags (e.g., 1girl, solo, classroom, desk) | Preserve all tags, add "position within the frame" and "spatial relationships between elements" || Natural language (e.g., "a teacher teaching in front of a blackboard") | Transform into structured English descriptions, automatically derive appropriate Danbooru elements || Wildcards (e.g., {standing, sitting}) | Preserve completely, no expansion, no selection, no deletion |---## Core Rules (Simplified)1. No image generation (text output only)2. Tag priority (user's tags remain unchanged)3. Only reinforce position and spatial relationships (no weather, lighting, or clothing texture details)4. Output as a single English paragraph (no markdown, parentheses, or prefacing text)5. Full wildcard support (original syntax untouched)---## ExampleInput (Danbooru tags + wildcard):1girl, {standing, sitting}, classroom, desk, {morning, evening}Output:> masterpiece, 1girl, {standing,| sitting}, in the center of a classroom, positioned in front of a desk, with {morning,| evening} lighting implied by the scene context.---## Who is this for?- People using Anima / NovelAI / Stable Diffusion who are accustomed to mixing tags and natural language- People tired of LLMs messing up wildcards or adding unnecessary novel-like details- People who want LLM output that can be directly copy-pasted as image generation prompts---## Full System Prompt## System Prompt Role & Goal You are a precise 2D visual coordination specialist. You handle two input types:1. Danbooru tag input → Preserve all tags, reinforce spatial relationships and visual flow.2. Natural language input (e.g., "a teacher teaching in front of a blackboard") → Convert description into structured English scene narrative, automatically inferring appropriate Danbooru-style elements.Input Detection - Comma-separated English terms → Danbooru tag input → follow tag preservation workflow. - Chinese or full sentence description → Natural language input → follow language conversion workflow.Core Rules 1. Never generate images. 2. Tag priority: User-provided Danbooru tags are absolute core — preserve all, never delete or arbitrarily replace. 3. Spatial reinforcement only: Add subject position (center, foreground, background) and spatial/interaction relationships (standing in front of, surrounded by). 4. No over-expansion: Do not add weather, lighting, or irrelevant fabric details unless originally mentioned. Keep concise. 5. Format: Output as a single smooth English paragraph (but split into two lines: line 1 = Danbooru tags, line 2 = natural language). No Markdown, parentheses, or prefixes. 6. Wildcard handling: - Preserve raw wildcard syntax {A,|B,|C} or {A,B}_nounor {1-3$$ A,|B,|C} — never expand, never choose, never replace. - For positional wildcards → use neutral descriptions (e.g., on either side, relative position to be determined). - For attribute wildcards → process spatial relationships normally. - Never rewrite {A|B} as A or B. - Never delete or ignore wildcards.Workflow A (Danbooru tags) Output two lines: Line 1: Original quality + base + subject + action + background tags Line 2: Natural language describing subject position + interaction + background relationship Workflow B (Natural language) Extract subject/action/scene → infer logical elements → output: Line 1: Danbooru tags (masterpiece, best quality, 1girl/1boy, relevant clothing, expression, action, visible scene elements) Line 2: Smooth English scene description with spatial clarity---## ANIMA Model Skill ProfileSkill Name: spatial_tag_coordinatorDescription: Converts Danbooru tag lists or natural language prompts into ANIMA‑friendly two‑line outputs: raw tags + spatial natural language. Preserves all user tags, adds only positional/interaction relationships. No image generation.Input Format Examples:```1girl, knight, charging, riding horse, battlefield``````a wizard casting a spell in a library```Output Format (two lines, no markdown):```[line1: Danbooru tags][line2: Natural language spatial description]```Example Output for ANIMA:```1girl, knight, armor, charging, riding_horse, horse, battlefield, dust, spear, shield, actionA young female knight in armor charges on horseback across a battlefield, holding a spear and shield, with dust rising around her as she rides forward through the center of the scene.```Key Constraints for ANIMA Compatibility: - Flat text only (no JSON, no parentheses wrapping tags) - First line = pure Danbooru comma list - Second line = natural English, no tags inside - Wildcards {A,|B,|C, or {1-3$$ A,|B,|C,} passed through unchanged - Never generate images — only transform text---Feel free to use, modify, or share 🙌---
Broken Anima
Basically, it generates in low resolution first and then finishes at the original resolution. It uses just the basic latent upscale, so it's a bit unstable—but hey, it's just for my own use anyway.
简单自用Anima文生图
简单自用Anima文生图流程
![Anima 1.0 Base for the PC master race [SFW / NSFW] - Image to prompt + Turbo mode + ControlNet + 4k upscaler + CivitAI medatada](https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/7c25c964-3a46-45ee-8400-9b70e64e2891/original=true/132152599.jpeg)
Anima 1.0 Base for the PC master race [SFW / NSFW] - Image to prompt + Turbo mode + ControlNet + 4k upscaler + CivitAI medatada
⚠️ Custom Node DisclaimerThis workflow heavily utilizes custom nodes. Manually modifying the ComfyUI file system or improperly managing node installations can break your ComfyUI environment. Ensure you understand how to manage dependencies and custom node directories before altering the underlying installation to prevent conflicts or system instability.If you are unfamiliar with these topics, please make a backup of your ComfyUI setup before installing new nodes or updating ComfyUI. For beginners, I generally recommend using the portable version, as it is easier to maintain and back up.May 31 2026 Hotfix:Added missing negative prompt and seed to the CivitAI metadataCorrected the prefix of the saved file to "Anima"Added the node file field to the sub-graph top viewWorkflow OverviewThis is a structured, production-ready ComfyUI workflow designed exclusively for the Anima (2B-parameter) non-photorealistic text-to-image model. It provides an organized pipeline utilizing color-coded, modular groups to manage the entire generation process from initial input to post-processing. The layout allows for complex routing without requiring manual graph adjustments for each run.Core FeaturesDual Resolution Modes: Toggle between precise advanced image sizing and straightforward dimension presets using a fast group bypasser.Multi-Method Prompting: Choose between standard manual text input or an automated Image-to-Prompt generator that extracts concepts from reference images.Turbo Mode Integration: A single-click toggle that loads the Anima Turbo LoRA and overrides base sampling parameters, reducing iteration steps from 30+ to approximately 10. The workflow includes a turbo LoRA fix to avoid unwanted sweat skin.Dedicated ControlNet Routing: Independent image inputs for structural guidance (pose, depth, canny) separated from the prompt generation reference.Post-Processing Modules: Optional, toggleable groups for Nvidia Video Super Resolution upscaling and basic color correction (brightness, contrast, saturation).Stacked LoRA Management: Streamlined application of multiple styling LoRAs via a single Power LoRA Loader.Native Metadata Embedding: Automatically writes CivitAI-compatible generation parameters into the final image file.LicenseThe Anima model is licensed under the CircleStone Labs Non-Commercial License. The model and derivatives are only usable for non-commercial purposes. Additionally, the model constitutes a "Derivative Model" of Cosmos-Predict2-2B-Text2Image, and therefore is subject to the NVIDIA Open Model License Agreement insofar as it applies to Derivative Models.If you would like a commercial license, please email [email protected] SECURITY ERROR FIXIf you hit a security level error while trying to install the ComfyUI_NVIDIA_RTX_Nodes custom node from ComfyUI manager:You need to temporarily change the security level in your config.ini file.If you are using the portable version, the file is located at:ComfyUI/user/__manager/config.iniChange the value to:security_level = weakIMPORTANT:Do NOT forget to restore it back to its original value. It is safe to temporarily lower the security level for this node since it comes directly from the ComfyUI team, but you should restore it afterward for safety reasons.Perform a hard restart of ComfyUI (completely stop it, then start it again).Install the custom node using ComfyUI manager and double-check that it is the official one. The link should point to:https://github.com/Comfy-Org/Nvidia_RTX_Nodes_ComfyUIRestart ComfyUI and check if the node works properly.If you encounter a dependency error, keep reading...Once everything is working properly, restore your original security level in the config.ini file and restart ComfyUI one final time.COMFYUI PORTABLE DEPENDENCY ERROR FIXIf you hit a dependency error trying to run the Nvidia VSR node, try this:Create a backup of your embedded Python directory, in case you break something you can restore your embedded directory later using this backup. Then run the following command: python -m pip install -U --no-build-isolation nvidia-vfx --index-url https://pypi.nvidia.com Run the command in your embedded Python directory.For example: D:\ComfyUI\python_embeded.This command is designed for CMD. If you're using PowerShell, replace the initial python with .\python.exe.Detailed instructions are contained in the workflow itself:Red nodes are instructions and useful notes.Yellow nodes are configurable elements you can adjust to your needs.
Anima can edit images! | Testing Anima-Edit loras and IC-methods
This is a workflow for testing image editing with the Anima.UPDATE: As it turned out in my tests I mixed up the names, keep in mindResources used:Anima-LLLiteAnimaEdit-experimentalAnima Edit [Nude Filter + Clothes Change + More]ComfyUI-Cosmos-Reference
Comfyui - Anima Experimental Fast Training Node
ComfyUI-AnimaFastTrain is an experimental ComfyUI custom node for quickly training in-memory Anima reference context tokens from one or more reference images.Instead of creating a LoRA, checkpoint, embedding, or safetensors file, AnimaFastTrain learns temporary context tokens during the workflow and injects them into the Anima model’s cross-attention at generation time. This makes it useful for experiments with character consistency, visual reference influence, and style transfer without writing trained weights to disk.Main features: - Trains lightweight reference context tokens directly inside ComfyUI - Keeps everything in memory, no model files are saved - Patches the Anima model during sampling - Supports up to 3 reference images - Adjustable token count, training steps, learning rate, dtype, and runtime strength - Useful for quick style/identity experiments before committing to a full LoRA training run Included nodes: - AnimaFastTrain - Train Context Tokens - AnimaFastTrain - Patch ModelRecommended workflow: Checkpoint Loader -> LoRA Loader -> AnimaFastTrain Patch Model -> KSamplerTrain the context tokens from a reference image, patch the final model after other model/LoRA patches, and then generate normally.Install by cloning the repo or using comfy manager.GitHub:https://github.com/quinteroac/ComfyUI-AnimaFastTrain
📜 DaSiWa Image Workflows | T2I | I2I 📜
My ComfyUI workflows for T2I/I2IThey are optimized for my checkpoints and created of my latest knowledge to enhance the outcome. "If this workflow leveled up your day, I'd purr-eciate a like! 😻"Versions & Information👇👇👇👇👇👇👇👇👉 Please read below and the file descriptions "About this version" for more info's.⚠️ Do not use the workflows with the "Nodes 2.0 beta" from ComfyUi or it will mess up things.👇👇👇👇👇👇👇👇What you get from the comfy workflows:♨️ Easy controls✅ As less as possible dependencies🪧 Detailed documentation⛓️ Highly automatic logic✨ Optimized results🎬 Fully automated resolution logic🔖 Bookmark-Shortcuts with number keysTypes of workflowsDarkArts C-TI2I🖼️ T2I and I2I🧩 Automatic aspect‑ratio calculation and fitting✨ Multiple Upscalers + RefinersUpscale with Model (additional detail, high quality)RTX Super Resolution (+ Deblur + Denoise)⏩ Multiple sampler passes💎 Easy Quality prompt option🎰 Multiple Radomierzers & Wildcards🔖 Bookmark-Shortcuts - with number keys🩻 Known issues and advice'snone🫶 Special thanks @TrashAI for the idea and help with the wildcards!All older Versions are available inside my GitHub Repo.YOU are responsible for outputs as always! If you make ToS violating content and I get aware I WILL report this.
![[Anima] Hand & Foot Detail Fixer (Manual Workflow)](https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/6c30ea51-61bb-42ee-8e61-1dfa15b54900/original=true/131766932.jpeg)
[Anima] Hand & Foot Detail Fixer (Manual Workflow)
🚀 Run this Workflow for FREE on Runninghub: https://www.runninghub.ai/post/2058483000282800129/?inviteCode=rh-v1497(🎁 Tip: Sign up via the link above to get 1,000 free credits + 100 daily login points!) 🎥 Watch the Full Video Tutorial: 🌟 OverviewWhile the Anima model produces breathtaking, human-level artistic styles, it can occasionally struggle with rendering perfect hands and feet. This targeted workflow is designed specifically to fix those anatomical issues and save your beautiful generations!🚀 Key Features & LimitationsTargeted Detail Fix: Specializes in correcting and refining messy hands and feet generated by the base model.Manual Prompting Required: This workflow does not include an LLM auto-prompter; you must input your text prompts manually.No Built-in Upscaler: Focused purely on generation and anatomy fixing. You will need to route the output to a separate upscaler if you need high-resolution images.
![[Anima] Auto-Prompt & Perfect IP Recreation (with Upscaler)](https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/f99dadd7-6177-4302-bc58-607be02d8c4f/original=true/131753481.jpeg)
[Anima] Auto-Prompt & Perfect IP Recreation (with Upscaler)
🚀 Run this Workflow for FREE on Runninghub: https://www.runninghub.ai/post/2058477642734723074/?inviteCode=rh-v1497 (🎁 Tip: Sign up via the link above to get 1,000 free credits + 100 daily login points!) 🎥 Watch the Full Video Tutorial: 🌟 OverviewBuilt specifically for the new Anima official model, your fully automatic solution for stunning anime art. It uses an LLM to take your simple ideas and automatically expand them into highly detailed, SDXL-style tags that Anima loves.🚀 Key FeaturesAuto-Prompting: Just type a simple concept, and the LLM does the heavy lifting.Safe & Accurate: Built-in LLM guardrails prevent NSFW generation and perfectly recreate known anime IPs.Built-in Klein Upscaler: Just tweak your aspect ratio and let the upscaler handle the high-res output.Artist Style Selector: Easily apply styles by copying and pasting popular artist strings directly from the included markdown list!
anima-inpaint simple 8steps Workflows
anima-inpaint Workflows使用官方遮罩编辑器工具+controlnet-lllite实现的简单八步重绘工作流。
✦ NIGHTRUN ✦ Overnight Forge Randomizer (Low-High VRAM Optimized)
Installation — Folder StructureNightrun expects to live two levels inside your WebUI folder:YOUR-WEBUI-FOLDER\ workflows\ nightrun_2.0\ ← hereA local Forge companion that builds structured prompt batches, runs them overnight, manages metadata-based upscales, and watches the boring parts, so your PC can do the babysitting for once.GitHub Link: https://github.com/Bluenochian/NightrunNightrun 2.0A portable workflow system for Stable Diffusion Forge / Forge-Neo.Nightrun is not a model. Not a LoRA. Not a prompt file.It is the system that sits beside Forge and turns a messy local generation setup into something that can run for hours — without you checking whether the prompt broke, a LoRA name was wrong, the GPU is cooking itself, or the batch slowly turned into repetitive garbage.Nightrun is built for long local runs, editable prompt pools, metadata replay, LoRA-aware batching, upscale workflows, cooldown logic, and controlled randomness.Set the run. Leave it alone. Come back to something organized, replayable, and worth sorting through.What Nightrun DoesNightrun builds prompts from editable pools, sends jobs through Forge, writes replayable metadata, manages upscales, scans LoRAs, protects long runs with cooldown logic, and keeps the whole thing configurable.It is made for people who want big batches without babysitting every prompt, every upscale, every progress bar, and every local setup problem.Controlled chaos. Not prompt soup.What's New in 2.0Modular ArchitectureThe engine was split from a single ~2055-line file into a purpose-driven module system:nightrun_runtime.py — core shared logicprompt_engine.py — pool loading, weighted selection, prompt compositionlora_resolver.py — LoRA scanning, aliases, tag constructionforge_client.py — Forge API, txt2img/img2img, GPU guardsrunner.py — execution modes, output folders, cleanup, cooldowns, summariesvalidation.py — pool validation and dry-run reportingnightrun_engine.py — compatibility shim for older importsnightrun.py is now a thin entrypoint that only imports and runs main().Three-Engine Prompt RoutingPrompt generation now routes through three distinct builders, selected by weighted random choice (default 1/1/1). You can control routing via prompt_engines in config, or disable it to revert to legacy behavior.Classic — Standard pool sampling with optional world module support. Closest behavior to 1.5, cleaned and optimized.Combinator — Builds on Classic output by applying controlled transformations:Internal combinators modify existing elements (e.g. a katana becomes a burning variant)External combinators inject additional motifs (e.g. surreal, cursed, steampunk layers)At least one combinator fires per prompt, capped at twoFranchise — Forces generation through a world module from franchise_catalog.py. All visual elements are sourced from that universe for full coherence. Falls back to Classic if no module is available.Prompt Cleanup PipelineA post-processing stage via prompt_cleanup now runs after every engine:Duplicate comma-tag removalConcept-family repetition controlCompact SD-style prompt length enforcementRemoval of redundant subject labels when already impliedConsistent prompt structure across all three engines.Prompt Pools — Full RebuildAll pools were rebuilt for SD-style tag efficiency. Verbose prose entries are gone. 2.0 uses compact, composable tags throughout.Pool 1.5 lines 2.0 lines concept.txt 3,601 154 genre.txt 8,317 2,743 glamour_pose.txt 4,241 761 clothing.txt 1,801 721 character.txt 701 44 body_focus.txt 1,105 481 sensual_layer.txt 1,921 529 hair.txt 2,401 790Removed: character_loras.txt, style_loras.txt, lora_sampler_speed.txtAdded: genre_core.txt (69 lines), genre_root.txt (475 lines)A curated weighting system (curated_top_count / curated_top_weight) boosts high-quality early entries without removing full pool diversity.Franchise Catalog — Expanded to ~50 ModulesModules now use full dictionary-based definitions instead of compact tuples. The catalog grew from ~24 to ~50 entries:alien batman bloodborne cowboy_bebop doctor_who dungeons_and_dragons elder_scrolls evangelion final_fantasy_vii game_of_thrones halloween hellraiser lord_of_the_rings matrix midsommar nier_automata nightmare_on_elm_street persona pokemon scream shadowrun terminator the_conjuring the_ring the_witcher vampire_masquerade x_men zeldaNew filtering controls without touching the catalog directly: enabled_module_ids / disabled_module_ids / enabled_source_names / disabled_source_names / label_allow_patterns / label_block_patternsSelection Memory — RetunedLarger recent history windows across categoriesReduced repetition penalties for more natural variationAdded concept-level tracking via recent_conceptsDedicated repetition control for candle/light motifsCurated weighting for high-quality biasingConfig Changesfixed_suffix updated for sharper SD output:Before: clean coherent design, sharp focus, high detail, polished lightingAfter: crisp linework, sharp focus, high detail, clear material texture, clean color separation, high micro-contrast, dramatic lightingbase.steps_min raised: 30 → 50Output folders now use nightrun-YYYYMMDD-HHMMSS--reset-local now preserves local tuning instead of overwriting itNew VRAM Profile — tuned_6gbSetting Value Base resolution 1344×768 Upscale resolution 3072×1728 Steps 65 LoRA cap 7 VRAM threshold ≤ 6144 MBNSFW — sensual_layer ExpansionThe sensual_layer pool was fully reworked to properly support explicit NSFW generation paths. It now includes dedicated nudity, erotic styling, body exposure, lingerie, fetish-adjacent aesthetic, and mature composition tags designed to integrate cleanly with the SD-style prompt system and franchise/genre routing.NSFW is enabled by default. For SFW-only runs, open config.json, find sensual_layer, and set "chance": 0.Core FeaturesEditable Prompt PoolsNightrun uses plain .txt prompt pools instead of one giant hardcoded prompt. Edit pool files to change how the whole machine behaves:concept genre genre_core genre_root character hair eyes clothing pose glamour_pose expression sensual_layer body_focus equipment environment lighting camera effects style_layer color_grade composition detail and more.The world module system is deeper and optional. The pool system is where most users should start.Controlled RandomnessNightrun is random — but not stupid-random. You can configure:Pool chances and pick countsBoosted chances when related pools appearBase step rangesLoRA chances, weight ranges, and capsWorld module and fusion mode chancesCooldown timing and upscale behaviorSelection memory and color repetition penaltiesYou are not stuck with one fixed behavior. Tune the machine until it behaves like your machine.Selection MemoryDuring a run, Nightrun remembers recent choices and penalizes repetition. It tracks broad color families — blue, cyan/teal, red, purple, green, gold, black, silver, iridescent — so if a batch starts drifting into one color mood, it pushes away.This is one of the main reasons long runs do not feel repetitive.Rare Detail ControlSome details are only interesting when they stay rare. Heterochromia, for example, can be configured so it does not take over an entire batch. Special things stop feeling special when every third image has them.LoRA HandlingNightrun does not blindly slap LoRAs into the prompt. It can:Scan local LoRA folders and resolve namesUse aliases and soft-match ugly filenamesSkip missing LoRAs without crashing the runRandomize LoRA weights within configured rangesAdd synergy/support LoRAs with controlled chancesCap random and total LoRA countsDeduplicate final LoRA tagsLocal LoRA libraries are messy. Civitai names, trigger names, and local filenames are often all slightly different. Nightrun tries to survive that.Coherent World ModulesThe normal prompt pools are the everyday engine. The world module system is the coherence layer.When it activates, Nightrun selects a world/franchise module and pulls related elements from the same source family — character archetypes, outfits, environments, props, factions, technology, magic systems, lighting mood, color palettes, cinematic language, and more.The goal is not to replace the normal pools. The goal is to give the run a coherent themed spine when that system activates.Fusion mode allows controlled cross-world combinations while reducing risky generic pool chances — unusual mixes without the machine blindly dumping everything into one prompt.Long-Run SafetyGPU temperature pause — Nightrun watches GPU temperature through nvidia-smi. If the GPU gets too hot, it pauses. When it cools below the configured resume threshold, it continues.VRAM chill sessions — For big batches, Nightrun schedules random cooldown sessions. Run several generations, pause for a few minutes, let the system settle, continue. Small thing. For overnight workflows it matters.Between-image rest — Configurable sleep time between images so the machine is not being tortured nonstop.Nightrun is not just "go fast." It is "keep going safely."Upscale WorkflowNightrun supports both scheduled upscale workflows and metadata-based re-upscaling. It can:Generate base images and upscale during full runsAuto-upscale low-step base rendersAccept drag-and-drop metadata for upscalingRecreate from metadata if neededWrite new metadata beside upscaled imagesChain later upscales from previous upscale metadataA fast base image does not have to stay a rough base image. Low-step renders can go through upscale logic and recover quality without burning time on every single base generation.Metadata ReplayEvery image gets a metadata sidecar with everything important saved:seed prompt negative prompt selected pools selected modules LoRAs LoRA sources base steps status upscale reason errors GPU infoSo later, you are not staring at a good image wondering what made it. The answer is saved beside it.Validation and Dry-RunValidation mode checks pool loading, prompt formatting, LoRA resolution, missing resources, alias behavior, and generation logic — before wasting GPU time.Dry-run mode writes prompt and metadata without calling Forge image generation. See what the run would do before committing to a real batch.Measure twice. Render once.Prompt Variety AnalyzerA built-in tool scans old metadata and reports:Repeated categories, colors, and modulesRepeated prompt fragmentsWorld module usage patternsExact prompt repeatsSigns that a batch is becoming too sameyTune the machine instead of guessing.OutputsEach run creates an organized folder:nightrun-YYYYMMDD-HHMMSS/ 0001_base_WIDTHxHEIGHT.png 0001_upscale_WIDTHxHEIGHT.png 0001_prompt.txt 0001_metadata.json summary.json run.log contact_sheet.jpg validation_report.json Exact files depend on the run mode.Basic UsageTask Launcher First-time setup Run First Setup.bat Validate before rendering Run Validation Stress Test.bat Preview one prompt, no render Run One DRY TEST.bat Base generation only Run Nightrun BASE only.bat Full run with scheduled upscale Run Nightrun FULL upscale.bat Re-upscale from metadata Run 2x Upscale From Metadata.bat Analyze prompt variety Run Prompt Variety Analyzer.bat Stop cleanly after current image Stop After Current Image.batMetadata Upscalingpy upscale_from_metadata.py "D:\renders\0001_metadata.json" --config config.local.json --scale 2 Force base image as source:py upscale_from_metadata.py "D:\renders\0001_metadata.json" --config config.local.json --source base --scale 2 Use an existing upscale as source:py upscale_from_metadata.py "D:\renders\0001_metadata.json" --config config.local.json --source upscale --scale 2 Dry-run the upscale:py upscale_from_metadata.py "D:\renders\0001_metadata.json" --config config.local.json --scale 3 --dry-run First-Run SetupThe setup script creates a local config for your machine. It will:Ask for your Forge / Forge-Neo path and output folderDetect your VRAM profileCheck checkpoint path and count local LoRAsCheck for Ultimate SD UpscaleCreate config.local.json and desktop shortcutsThe shared config.json stays portable. Your private machine paths stay in config.local.json. The project can be shared without leaking your local folder structure.Recommended ForNightrun is useful if you:Use Forge / Forge-Neo locallyGenerate in large batches or overnight runsWant editable prompt pools instead of hardcoded promptsUse many LoRAs and want safer handlingWant metadata replay for good imagesWant upscaling tied to generation metadataWant less repetitive outputs over long runsWant a workflow you can tune instead of replaceNotesNightrun does not include checkpoints, LoRA files, model files, Forge itself, or Ultimate SD Upscale. You need your own local Forge / Forge-Neo setup and your own models.Franchise and world module names are used as inspiration labels for prompt organization only.License / Usage noteUse this with models, LoRAs, and extensions you are allowed to use.Franchise module names are inspiration labels for prompt organization. Nightrun does not include model files, LoRA files, checkpoints, or copyrighted assets.Nightrun is a prompt generation and batch runner tool. It does not include, distribute, or download any AI models, checkpoints, LoRAs, extensions, or generated images. All content is produced locally using software and model files you have installed yourself.You are solely responsible for the models and extensions you use, the content you generate, compliance with third-party license terms, and compliance with the laws of your jurisdiction.NSFW generation is supported. It is your responsibility to ensure any explicit content you generate is legal in your jurisdiction. This tool must not be used to generate illegal content of any kind.Nightrun is provided as-is, without warranty of any kind.Built for the dark hours between intention and discovery.
illu(t2i) + anima(i2i) Workflow
Anima is relatively new, its LoRA library is significantly more limited than that of Illustrious.Retraining or converting existing Illustrious LoRAs for Anima may not efficient. Therefore, I intend to implement a workflow using Illustrious (txt2img) , followed by Anima (img2img) processing.This approach retains the diverse variety of Illustrious LoRAs while leveraging Anima's superior rendering quality.
Zoda Anima Workflow
Example Workflow for ComfyUI beginner with ZodaPlus Lora Detailer.Model link is inside the workflowDocker Compose Config to run ComfyUIservices: comfyui: image: yanwk/comfyui-boot:cu130-slim-v2 container_name: comfyui-cu130 pull_policy: missing ports: - "8188:8188" environment: - CLI_ARGS= - HF_TOKEN=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 volumes: - ./storage-cache/dot-cache:/root/.cache - ./storage-cache/dot-config:/root/.config - ./storage-nodes/dot-local:/root/.local - ./storage-nodes/custom_nodes:/root/ComfyUI/custom_nodes - ./storage-models/models:/root/ComfyUI/models - ./storage-models/hf-hub:/root/.cache/huggingface/hub - ./storage-models/torch-hub:/root/.cache/torch/hub - ./storage-user/input:/root/ComfyUI/input - ./storage-user/output:/root/ComfyUI/output - ./storage-user/user-profile:/root/ComfyUI/user - ./storage-user/user-scripts:/root/user-scripts deploy: resources: reservations: devices: - driver: nvidia count: all capabilities: [gpu]
Anima Base 1.0 Enhancement Node A/B Comparison Workflow
Watch the full video first if you want to understand the workflow logic quickly.This ComfyUI workflow is designed for Anima Base 1.0 enhancement node comparison, A/B testing, and practical model behavior analysis. Instead of only showing a finished generation result, this workflow is built like a controlled experiment. Each group contains an A baseline and a B experimental branch, making it easier to judge whether a specific node actually improves quality, stability, speed, or low-step performance.The workflow uses anima_baseV10.safetensors as the base model, qwen_3_06b_base.safetensors as the Qwen Image CLIP text encoder, and qwen_image_vae.safetensors as the VAE. The testing rule is strict: within each A/B group, the model file, CLIP, VAE, positive prompt, negative prompt, seed, resolution, sampler, scheduler, steps, and CFG are kept the same. The A branch is the clean baseline. The B branch only adds the target test node. This makes the comparison much more meaningful than casual prompt testing.The workflow is divided into two major categories. The D series is designed for distilled / Turbo low-step testing, using steps=8 and CFG=1. This section is useful for testing Turbo LoRA, NAGuidance, and CFGNorm under fast generation conditions. Turbo LoRA is tested to see whether it can compensate for low-step quality loss. NAGuidance is tested to see whether negative concepts can still affect generation when CFG is very low. CFGNorm is tested to observe whether it improves color, edge stability, and composition under Turbo-style sampling.The N series is designed for non-distilled standard CFG workflows, using steps=40 and CFG=4. This section tests CFGZeroStar, RescaleCFG, and SageAttention. CFGZeroStar is useful for observing changes in classifier-free guidance behavior. RescaleCFG is tested to see whether it can reduce overexposure, over-saturation, color damage, or CFG-related image collapse. SageAttention is treated mainly as a performance and attention acceleration test rather than a style enhancement node. The correct result should stay close to the baseline while improving speed or memory behavior.This workflow is especially useful for creators who want to understand which nodes are actually worth using in Anima Base production. It helps separate real improvement from psychological bias. By comparing A and B under controlled conditions, users can decide whether a node is useful for Turbo workflows, standard workflows, low-step generation, quality recovery, stability control, or performance acceleration.Main features:- Anima Base 1.0 enhancement node comparison workflow- Strict A/B testing structure- Unified model, CLIP, VAE, prompt, seed, size, sampler, scheduler, steps, and CFG- Distilled / Turbo low-step test group- Standard non-distilled CFG test group- Turbo LoRA comparison- NAGuidance comparison- CFGNorm comparison- CFGZeroStar comparison- RescaleCFG comparison- SageAttention performance comparison- Designed for practical workflow decision-makingSuggested workflow:Start with one test group only. Compare the A baseline and B experimental output side by side. Do not change prompt, seed, resolution, or sampler while testing. For Turbo-style workflows, focus on the D series. For standard higher-step workflows, focus on the N series. Judge the results by image quality, prompt adherence, anatomy stability, color control, edge quality, generation speed, VRAM usage, black image risk, and whether the output drifts too far from the baseline.⚙️ RunningHub WorkflowTry the workflow online right now — no installation required.👉 Workflow: https://www.runninghub.ai/post/2057387404713095169?inviteCode=rh-v1111If the results meet your expectations, you can later deploy it locally for customization.🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance and 48 GB super power!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you’re in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1jALa6xEqH/☕ Support Me on Ko-fiIf you find my content helpful and want to support future creations, you can buy me a coffee ☕.Every bit of support helps me keep creating — just like a spark that can ignite a blazing flame.👉 Ko-fi: https://ko-fi.com/aiksk💼 Business ContactFor collaboration or inquiries, please contact aiksk95 on WeChat.⚙️ 在线体验工作流现在就可以在线体验，无需安装。👉 工作流： https://www.runninghub.ai/post/2057387404713095169?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1jALa6xEqH/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
Anima Prompt Structure and Wildcard Workflow
Moving to Anima from Illustrious has been daunting due to its reliance upon lengthy natural language descriptions to get quality results, whereas with Illustrious you relied upon tags with highly defined concepts as "building blocks".I created this workflow to help me navigate this, by using a prompting category approach to get a better defined structure.Part of the reason for this approach is that I enjoy tweaking prompts and using wildcards, but natural language is easy to entangle. So by keeping it all segmented, it allows me to adjust segments of the prompt freely and without issue.Included with the workflow are some example wildcard files, which contain a small number of random prompts I generated with ChatGPT to help get started.v1.1 adds a number of automated and manual prompting toggles to improve user experience. Most notably, category headers no longer appear unless that category is actually used, and you can now disable them entirely if you want more manual control.
Anima + lora + upscaler + resizer
anima + multilora loader + upscaler + comfy resizeEnable the disabled purple node to enable the upscaler
Characters generator | Generation of detailed сharacter for full body
V1.3 - Anima:- This is the version for Anima, I really liked it.- Added frame control to prevent the character from moving out of the frameV1.2:- Added "ControlNet"Some people have asked me to share my character workflow.Why not?So I refined it and added a randomizer, enjoy!WARNING! This method does not work well with V-Pred models.
Guy90's Very Simple and Easy to use Anima Workflow for ComfyUI
💰Style Model Commissions Here: ko-fi.com/guy90/commissions📧Join my Discord -> CLICK ME⭐If you like my work, leave a review. I really appreciate them!☑️I decided to make a workflow after switching to ComfyUI. I’m sure a lot of people are in the same situation when it comes to Anima, or maybe want to switch but don’t want to deal with learning how to use ComfyUI. Because of that, I created a SIMPLE and EASY to use workflow.This workflow includes Image Generation (txt to img), LoRAs, Upscaling, and Adetailer. There will be a large note titled “Start Here!!!!☝🤓” at the beginning. Please make sure to read it and follow the instructions. As long as you follow everything, it should work smoothly.Make sure your ComfyUI is updatedFeel free to comment below if you have any questions or run into any issues with the workflow. Also let me know what I can improve to make it even more user friendly.Issues and FixesTypeError: (intermediate value).reduce is not a functionThis error occurs because something is not updated. Please make sure your ComfyUI, LoRA Manager, and ComfyUI Manager are updated (yes, you need to update these all individually).Red nodes or missing node pop-upsMake sure to download missing nodes from the ComfyUI Manager and also make sure all nodes are up to date.Metadata doesn't load onto CivitaiHit the Refresh and Fetch button in the LoRA Manager. Then refresh ComfyUI.Planned FeaturesInpaintingImage Metadata Reader (Seed, prompts, maybe even a transfer button)Regional PromptingAnd More...
Anima Base Text-to-Image + ControlNet Face & Hand Repair Workflow
This workflow is designed for Anima Base text-to-image generation with ControlNet-style structure guidance, face refinement, and hand repair. Its main purpose is to let creators start from a pure text prompt, generate an anime-style image through Anima Base, use structure control to improve composition stability, and then automatically refine the most failure-prone areas: the face, eyes, hands, and fingers.Unlike a basic text-to-image workflow, this setup is not only a single-pass image generator. It combines Anima Base generation, Qwen image text encoding, empty latent creation, prompt-guided sampling, NAG guidance, ControlNet / Anima LLLite-style structure control, latent upscaling, second-pass refinement, face detection, SAM-assisted facial repair, hand detection, hand segmentation, FaceDetailer repair, preview nodes, and final image output. This makes it more suitable for creators who want a complete anime image production pipeline rather than a simple prompt-to-picture graph.The workflow uses anima_baseV10.safetensors as the main Anima Base model route, qwen_3_06b_base.safetensors as the text encoder, and qwen_image_vae.safetensors as the VAE. It also includes CLIPSetLastLayer, EmptyLatentImage, CLIPTextEncode, NAGuidance, AnimaLLLiteApply, AIO_Preprocessor, ClownsharKSampler_Beta, LatentUpscaleBy, VAEDecode, FaceDetailer, SAMLoader, UltralyticsDetectorProvider, and multiple preview nodes. This structure shows that the workflow is built for text-to-image generation plus controlled refinement.The first generation stage starts from an empty latent image, meaning the image is created from text rather than from an existing input image. The positive prompt describes the target anime key visual, such as an adult anime beauty, celestial fantasy scene, long platinum hair, luminous skin, floating palace balcony, colossal sky dragon, clouds, wind, sunlight, and cinematic scale contrast. The negative prompt suppresses low quality, blurry output, JPEG artifacts, low resolution, and other weak-image problems.A key part of the workflow is the ControlNet-style guidance route. The workflow includes a reference image path, DepthAnything preprocessing, and Anima LLLite application. This allows the creator to use structure guidance while still generating from text. In practical terms, this helps the output keep stronger pose, depth, silhouette, layout, and spatial composition instead of relying only on the text prompt. For anime key visual generation, this is useful because pure text-to-image generation can easily drift in character placement, body structure, perspective, or background scale.The workflow also includes NAG guidance. This helps strengthen prompt adherence and improve control during generation. For complex fantasy scenes with one character, giant creatures, dramatic lighting, and large-scale composition, stronger guidance is useful because the model needs to balance subject clarity, background scale, and visual style.The generation process uses a staged route. After the first sampling pass, the workflow can upscale the latent result and run a second refinement pass. This is important because a first-pass image may already have a strong composition, but details such as face clarity, clothing texture, hair, lighting, and background sharpness often need extra refinement. Latent upscaling improves the image before final decoding, making the result more suitable for high-quality preview images, cover images, and Civitai showcases.The face repair section uses face detection and SAM-based refinement to locate the face area and run a focused detail pass. This helps improve facial symmetry, eye clarity, expression quality, and local anime rendering detail without forcing the whole image to regenerate. For character images, this is one of the most important post-generation steps because a strong full-body image can still fail if the face is soft or distorted.The hand repair section uses dedicated hand detection and segmentation models, including hand_yolov8s and PitHandDetailer-style segmentation. This part is designed to identify hand regions, crop them, refine them, and paste the repaired result back into the full image. It helps reduce extra fingers, fused fingers, malformed hands, weak palms, broken gestures, and unclear finger structure.This workflow is ideal for anime text-to-image generation, Anima Base character creation, fantasy key visual production, ControlNet-style composition control, high-quality anime illustration generation, automatic face repair, automatic hand repair, Civitai preview images, RunningHub demos, and social media cover artwork.If you want to see how Anima Base, text-to-image generation, ControlNet-style structure guidance, NAG enhancement, latent upscale refinement, face detection, hand detection, SAM refinement, and final detail repair work together in one practical pipeline, watch the full tutorial from the YouTube link above.⚙️ Try the Workflow Online👉 Workflow: https://www.runninghub.cn/post/2055109924421685249?inviteCode=rh-v1111Open the link above to run the workflow directly online and view the generation results in real time.If the results meet your expectations, you can also deploy it locally for further customization.🎁 Fan Benefits: Register now to get 1000 points, plus 100 daily login points — enjoy 4090-level performance and 48 GB of powerful compute!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you are in Mainland China or the Asia-Pacific region, you can watch the video below for workflow demos and a detailed creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1ci5q65EGQ/I will continue updating model resources on Quark Drive:👉 https://pan.quark.cn/s/20c6f6f8d87bThese resources are mainly prepared for local users, making creation and learning more convenient.⚙️ 在线体验工作流👉 工作流： https://www.runninghub.cn/post/2055109924421685249?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1ci5q65EGQ/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
Anima Base Image-to-Image + ControlNet Face & Hand Repair Workflow
This workflow is designed for Anima Base image-to-image generation with ControlNet-style structure guidance, face refinement, and hand repair. Its main purpose is to take an existing image as the visual reference, regenerate it through Anima Base, preserve the original composition and style direction, and then automatically improve the most error-prone areas: the face, eyes, hands, and fingers.Unlike a basic image-to-image workflow, this setup is not only a simple restyle or redraw process. It combines Anima Base generation, image scaling, latent encoding, prompt-guided reconstruction, optional ControlNet / LLLite-style guidance, face detection, SAM-assisted refinement, hand detection, hand segmentation, FaceDetailer repair, and final preview / export logic. This makes it more suitable for creators who want a complete anime image polishing pipeline instead of a one-pass redraw.The workflow uses anima_baseV10.safetensors as the main Anima Base model route, qwen_3_06b_base.safetensors as the text encoder, and qwen_image_vae.safetensors as the VAE. It also includes image_scale_pixel_v2, VAEEncode, VAEDecode, CLIPTextEncode, NAGuidance, AnimaLLLiteApply, AIO_Preprocessor, FaceDetailer, SAMLoader, UltralyticsDetectorProvider, and multiple preview nodes. The structure shows that this workflow is built for controlled regeneration plus automatic detail correction.The image-to-image section first receives the input image, scales it to a suitable working resolution, encodes it into latent space, and uses the prompt to guide the Anima Base redraw. This helps preserve the main layout while giving the model enough freedom to improve detail, color, lighting, anime rendering quality, and character styling. The positive prompt route defines the target anime key visual style, while the negative prompt suppresses common failures such as low quality, blurry faces, bad anatomy, extra fingers, malformed hands, duplicated characters, cropped bodies, text, and watermark artifacts.A key part of this workflow is the ControlNet-style guidance section. The workflow includes a depth preprocessor route and Anima LLLite application logic. This can help preserve the structural relationship of the input image, such as pose, depth, silhouette, body placement, and large composition. In image-to-image generation, this kind of control is important because a pure prompt-based redraw can easily change the subject too much. The control route helps keep the image more stable while still allowing the Anima model to enhance style and quality.The face repair section uses face detection and SAM-based refinement to locate the face area and run a focused detail pass. This is useful because anime image generation often produces a good full-body composition but leaves the face slightly soft, distorted, asymmetrical, or lacking detail. By isolating the face area, the workflow can improve eyes, facial clarity, expression, and local rendering quality without forcing the whole image to regenerate.The hand repair section uses dedicated hand detection and segmentation models, including a hand bounding-box detector and a hand detail segmentation route. This is one of the most practical parts of the workflow because hands are still one of the most common failure points in AI image generation. The hand detail pass is designed to identify the hand region, crop it, refine it, and paste the improved result back into the full image. This helps reduce extra fingers, fused fingers, broken hands, malformed palms, and weak hand detail.This workflow is especially useful for anime creators who already have a promising base image but want a cleaner publish-ready version. It can be used for image-to-image restyling, anime character enhancement, composition-preserving redraw, ControlNet-style structure retention, automatic face repair, automatic hand repair, Civitai preview images, RunningHub demos, and social media cover artwork.If you want to see how Anima Base, image-to-image generation, ControlNet-style guidance, face detection, hand detection, SAM refinement, and final detail repair work together in one practical pipeline, watch the full tutorial from the YouTube link above.⚙️ Try the Workflow Online👉 Workflow: https://www.runninghub.cn/post/2055116648939565057?inviteCode=rh-v1111Open the link above to run the workflow directly online and view the generation results in real time.If the results meet your expectations, you can also deploy it locally for further customization.🎁 Fan Benefits: Register now to get 1000 points, plus 100 daily login points — enjoy 4090-level performance and 48 GB of powerful compute!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you are in Mainland China or the Asia-Pacific region, you can watch the video below for workflow demos and a detailed creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1ci5q65EGQ/I will continue updating model resources on Quark Drive:👉 https://pan.quark.cn/s/20c6f6f8d87bThese resources are mainly prepared for local users, making creation and learning more convenient.⚙️ 在线体验工作流👉 工作流： https://www.runninghub.cn/post/2055116648939565057?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1ci5q65EGQ/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
【Anima】Simple T2I Workflow with Upscale, Detailers and ControlNet
V3.1:This version ensures that Civitai can properly parse the required metadata in saved images, which includes all model info. Thanks to @Boerboer for the suggestion and help.V3.0:Add a bypassable ControlNet (Anima-LLLite). Big thanks to Kohya for creating this ControlNet.Introduction Page: https://huggingface.co/kohya-ss/Anima-LLLiterecommended CN file: anima-lllite-any-test-like-v2.safetensors https://huggingface.co/kohya-ss/Anima-LLLite/tree/maincustom node: https://github.com/kohya-ss/ComfyUI-Anima-LLLiteV2.0:Add face detailer and hand detailer.V1.0:A simple text-to-image workflow for Anima with Ultimate SD Upscale. (If you notice unnatural transitions when increasing the denoise value in the Upscale node, try increasing the tile_padding, such as 256.)Resources:Anima (Base model): https://civitai.com/models/2458426/anima?modelVersionId=2945208CLIP: https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/text_encodersVAE: https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/vaeThe following are just my personal preferences. Feel free to swap them out for files that suit you better.Upscale Model: https://huggingface.co/FacehugmanIII/4x_foolhardy_Remacri/tree/mainface_yolov8n: https://huggingface.co/Bingsu/adetailer/blob/main/face_yolov8n.pthand_yolov8s: https://huggingface.co/Bingsu/adetailer/blob/main/hand_yolov8s.ptPitHandDetailer-v2-Test-v9c: https://huggingface.co/iahhnim/adetailer_collection/blob/main/PitHandDetailer-v2-Test-v9c.ptRecommended Resources：Anima Turbo LoRA: https://civitai.red/models/2560840/anima-turbo-loraAnima Highres/Aesthetic Boost: https://civitai.red/models/2540444/anima-highresaesthetic-boostIf you're focusing on furry character generation, you might want to consider the following detailer scheme:https://civitai.red/models/1296534/fdetailer-example-workflowhttps://civitai.red/models/1228695/fdetailer-adetailer-for-furriesand a furry Anima model (While the model has some minor flaws, like slight overfitting and a lack of data for certain furry species, it’s still fantastic for generating furry characters—especially for that muscular bara physique!):https://civitai.red/models/2575487/indigo-void-furry-anima-furry-animaThanks to @UmeAiRT and @CptCharname! I’ve learned so much from studying their workflows.
Anima LoRA Trainer for ComfyUI
A ComfyUI custom node pack for training LoRA / LoKR / LoHA models on the Anima architecture using kohya-ss/sd-scripts. Train Anima LoRAs directly from ComfyUI with zero command-line knowledge required.InstallationDownload and extract the zip into your ComfyUI/custom_nodes/ folder, then restart ComfyUI. The nodes will appear under the "Anima Training" category.FeaturesOne-click model downloading from HuggingFace (always fetches the latest Anima base model)Two workflow modes: Wizard (questionnaire-driven auto-config) and Manual (full control over all parameters)Automatic sd-scripts installation and dependency managementTOML config generation with inline annotations explaining every settingSmart VRAM presets (12GB, 16GB, 24GB) with automatic memory optimizationLive config preview inside ComfyUI before launching trainingCross-platform support (Windows, Linux, macOS)Zero pip dependencies — pure Python stdlibIncluded NodesModel Downloader — Downloads Anima models from HuggingFaceSD-Scripts Manager — Installs/updates kohya-ss/sd-scriptsModel Selector — Finds and validates model filesTraining Wizard — Questionnaire-driven auto-configurationBasic/Advanced/Memory Params — Manual parameter controlTraining Launcher — Generates config and launches trainingConfig Preview — Shows annotated TOML config inlineRequirementsComfyUI (any recent version)Python 3.10+kohya-ss/sd-scripts (auto-installed by the SD-Scripts Manager node)Anima base model files (auto-downloaded by the Model Downloader node)Dataset PreparationYour training images should follow the sd-scripts convention: place images in a folder named <repeats>_<concept> inside your train_data_dir, with matching .txt caption files for each image.
Anima Ultimate SD Upscale workflow
a workflow that use Ultimate SD Upscale to upscale with anima model.this is just a example plz feel free to replace node to the version u are using.the result isn't great since it lost the advantage of DIT due to tiling.(mainly mess up mesh texture in the image)About Paramsu can use the prompt of original image.denoise should be 0.2-0.4.steps should be 6-18 base on denoise.tile base size should be 1200 for AP3 1520 for Anima Baseeular sampler will add grain like texture to background, avoid if u upscale image with large background area.tile upscale with Anima Base still mess up with fishnet/fence texture. plz let me know if any one knows how to fix this, thank u.26/04/12: 1.1 updated to preview326/04/18: added lora loader and adjusted params in 1.1, examples are in review26/05/14: 1.2 added following QoL node: auto tile sizing, comparsion, save with metedata
Anima (Preview) Workflow
Workflow for AnimaAny feedback would be appreciated.Downloadsofficial HF repo: https://huggingface.co/circlestone-labs/Animacivitai model page: https://civitai.com/models/2458426/anima-officialDetectors (YOLO/SEG):Hands (hand_yolov9c): https://huggingface.co/Bingsu/adetailer/blob/main/hand_yolov9c.ptFace (face_yolov9c): https://huggingface.co/Bingsu/adetailer/blob/main/face_yolov9c.ptEye (Eyeful_v2-Paired or Eyeful_v2-Individual): https://civitai.com/models/178518/eyeful-or-robust-eye-detection-for-adetailer-comfyuiNSFW (ntd11_anime_nsfw_segm_v5-variant1): https://civitai.com/models/1313556/anime-nsfw-detectionadetailer-all-in-oneTag order[quality/meta/year/safety tags] [1girl/1boy/1other etc] [character] [series] [artist] [general tags]Generation Settings30-50 Steps4-5 CFG1MP Resolution e.g. 1024x1024 or 896x1152er_sde, euler_a or dpmpp_2m_sde_gpuKey Features:LoRADetailersHiresFixWildcardsI2IWorkflows included:🟥Anima: Standard workflow version🟨AnimaStandard: more simplistic version of the workflow🟩AnimaBasic: more simplistic version of the standard workflow that lacks some of the advanced features for metadata and suchCustom NodesComfyUI-Manager (by Comfy-Org)https://github.com/Comfy-Org/ComfyUI-ManagerComfyUI-Impact-Pack (by ltdrdata)https://github.com/ltdrdata/ComfyUI-Impact-PackComfyUI-Impact-Subpack (by ltdrdata)https://github.com/ltdrdata/ComfyUI-Impact-Subpackrgthree-comfy (by rgthree)https://github.com/rgthree/rgthree-comfyComfyUI-Image-Saver (by alexopus)https://github.com/alexopus/ComfyUI-Image-SaverComfyUI-KJNodes (by kijai)https://github.com/kijai/ComfyUI-KJNodesComfyUI-Lora-Manager (by willmiao)https://github.com/willmiao/ComfyUI-Lora-ManagerComfyUI-Easy-Use (by yolain)https://github.com/yolain/ComfyUI-Easy-UseComfyUI_UltimateSDUpscale (by ssitu)https://github.com/ssitu/ComfyUI_UltimateSDUpscaleExperiment and enjoy!
Anima Preview3 | Tiled Upscale & High-Resolution Detail Refinement Workflow
This workflow is designed for Anima Preview3 tiled upscaling and high-resolution anime detail refinement. Its main purpose is to take an existing anime-style image, enlarge it, split it into manageable tiles, refine each local region with Anima Preview3, and then reconstruct the image into a cleaner high-resolution result. Instead of relying only on a basic pixel upscaler, this workflow combines traditional upscale, tile processing, prompt-assisted regeneration, VAE encoding / decoding, and final image assembly to produce a more polished output.The workflow uses anima-preview3-base.safetensors as the main generation model, qwen_3_06b_base.safetensors as the text encoder, and qwen_image_vae.safetensors as the VAE. It also includes an optional LoRA route, with Gundam_RX78_Flux used in the example setup, showing that the workflow can be adapted for mecha, armor, character, or specific style enhancement tasks. The initial enlargement stage uses 4x_NMKD-Siax_200k.pth, which helps expand the image before the Anima refinement stage begins.The core structure is a tiled enhancement pipeline. The source image is loaded, upscaled, resized to a larger total pixel target, and then divided into tiles through TTP_Tile_image_size and TTP_Image_Tile_Batch. Each tile becomes an independent local region that can be encoded, refined, decoded, and later reassembled. This is important because large anime images often lose detail or become unstable when processed in one full-frame pass. Tiled processing lets the workflow focus more generation attention on each region, improving local texture, line clarity, small mechanical details, clothing folds, hair strands, and background elements.The workflow also uses WD14 tagging. The tile or image content can be analyzed automatically, and the extracted tags can be used to help guide the refinement prompt. This makes the workflow useful when the creator wants the upscaler to understand what is in the image, not just enlarge it mechanically. For anime and illustration work, this can help keep character features, objects, and style direction more coherent during the enhancement stage.The sampling route uses ClownsharKSampler_Beta with a controlled denoise value. This is critical for tiled upscale work. If denoise is too high, each tile may redraw too aggressively and break consistency across the image. If denoise is too low, the image may only become larger without gaining meaningful detail. This workflow is set up for refinement rather than full redesign: preserve the original composition, improve the local details, and reconstruct the final image with better finish.After tile refinement, the workflow uses tiled VAE decoding and TTP_Image_Assy to rebuild the full image. Padding and overlap handling help reduce obvious seams between tiles. This makes the final result more suitable for Civitai previews, RunningHub demos, social media covers, high-resolution anime character art, mecha illustration enhancement, and final polish before publishing.This workflow is ideal for creators who already have a good anime image but want a sharper, larger, more detailed final version. If you want to see how Anima Preview3, tile splitting, WD14 tagging, local latent refinement, and final tile reconstruction work together, watch the full tutorial from the YouTube link above.⚙️ Try the Workflow Online👉 Workflow: https://www.runninghub.ai/post/2042211302168338434?inviteCode=rh-v1111Open the link above to run the workflow directly online and view the generation results in real time.If the results meet your expectations, you can also deploy it locally for further customization.🎁 Fan Benefits: Register now to get 1000 points, plus 100 daily login points — enjoy 4090-level performance and 48 GB of powerful compute!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you are in Mainland China or the Asia-Pacific region, you can watch the video below for workflow demos and a detailed creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1MEQbBsE5z/I will continue updating model resources on Quark Drive:👉 https://pan.quark.cn/s/20c6f6f8d87bThese resources are mainly prepared for local users, making creation and learning more convenient.⚙️ 在线体验工作流👉 工作流： https://www.runninghub.ai/post/2042211302168338434?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1MEQbBsE5z/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
Anima Preview3 | Local Redraw & Inpainting Workflow
This workflow is designed for Anima Preview3 inpainting and local redraw, focusing on controlled partial editing inside an existing anime image. Its main purpose is to let creators mask a specific region, describe the desired change, and regenerate only that local area while keeping the rest of the image stable, coherent, and visually consistent.The workflow uses anima-preview3-base.safetensors as the main generation model, qwen_3_06b_base.safetensors as the text encoder, and qwen_image_vae.safetensors as the VAE. It combines image loading, mask input, mask expansion, blurred mask blending, latent encoding, Differential Diffusion, prompt generation, sampling, VAE decoding, preview, and final saving into one practical local-editing pipeline. Compared with a normal image-to-image workflow, this setup is more suitable when you do not want to redraw the whole image. You only want to modify one selected area.The key part of this workflow is the inpainting mask route. The source image is loaded with its mask, then passed through image_scale_pixel_v2 to normalize the image and mask into a model-friendly size. The mask is then processed by GrowMaskWithBlur, which expands and softens the selected area. This is important because hard mask edges often create visible seams, while a slightly expanded and blurred mask gives the model enough transition space to blend the new content naturally into the original image.The workflow then uses VAEEncode to convert the source image into latent space and SetLatentNoiseMask to restrict the repainting area. This means the sampler is guided to focus on the masked region instead of freely changing the entire frame. The DifferentialDiffusion node further strengthens this local redraw behavior, making the workflow more suitable for precise anime inpainting, object replacement, facial adjustment, hairstyle changes, clothing fixes, prop replacement, and small composition corrections.Another important feature is the built-in prompt-generation logic. The workflow includes an LLM API prompt engineer node, using a role instruction specialized for anime inpainting and partial redraw. It analyzes the reference image, understands the overall composition, art style, lighting, pose, environment, and visual tone, then converts the user’s short edit instruction into a single English prompt line. This is useful because local redraw prompts need to be more careful than normal generation prompts. They must describe the new area while preserving the rest of the image.The example instruction in the workflow is “only modify the specified region: change the head into a pineapple.” This shows the intended usage clearly: the user provides a short local edit request, the LLM rewrites it into a model-friendly inpainting prompt, and Anima Preview3 performs the masked redraw while maintaining the original image identity, style, lighting, and surrounding details.The sampling stage uses ClownsharKSampler_Beta with a 30-step beta57 setup, CFG around 3, and a full local redraw denoise setting. This gives the masked region enough freedom to change while the unmasked area remains protected by the mask and latent structure. The negative prompt suppresses common anime-generation issues such as low quality, blurry output, bad hands, malformed limbs, duplicated elements, deformed faces, text, and watermark artifacts.This workflow is ideal for creators who need a clean Anima Preview3 local redraw setup for fixing details, replacing objects, editing character parts, modifying accessories, repairing small visual problems, testing funny local transformations, or preparing polished Civitai preview images. If you want to see how the mask, LLM prompt rewrite, Differential Diffusion, Anima Preview3 model, and final inpainting output work together, watch the full tutorial from the YouTube link above.⚙️ Try the Workflow Online👉 Workflow: https://www.runninghub.ai/post/2042205589152997377?inviteCode=rh-v1111Open the link above to run the workflow directly online and view the generation results in real time.If the results meet your expectations, you can also deploy it locally for further customization.🎁 Fan Benefits: Register now to get 1000 points, plus 100 daily login points — enjoy 4090-level performance and 48 GB of powerful compute!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you are in Mainland China or the Asia-Pacific region, you can watch the video below for workflow demos and a detailed creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1MEQbBsE5z/I will continue updating model resources on Quark Drive:👉 https://pan.quark.cn/s/20c6f6f8d87bThese resources are mainly prepared for local users, making creation and learning more convenient.⚙️ 在线体验工作流👉 工作流： https://www.runninghub.ai/post/2042205589152997377?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1MEQbBsE5z/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
Anima Based Flow (2 Node packs)
A super simplistic, 2 Node pack workflow for Anima.
Anima Preview2 Tiled Upscale Workflow
This workflow is designed for Anima Preview2 tiled upscaling and detail enhancement. Its main purpose is to take an existing image, enlarge it to a higher usable resolution, then refine the enlarged result tile by tile with Anima Preview2 so the final image looks cleaner, sharper, and more detailed without relying on simple interpolation alone. It is especially useful for creators who already have a good base image but want a higher-quality final version for publishing, preview display, cover images, or showcase output.The workflow uses anima-preview2.safetensors as the main refinement model, qwen_3_06b_base.safetensors as the text encoder, and qwen_image_vae.safetensors as the VAE. It also includes a LoRA loader in the graph, showing that the upscaling route can be combined with an additional style or detail bias when needed. The overall design is not just “make the image bigger.” Instead, it builds a multi-stage pipeline: upscale first, split into tiles, describe the tiles, refine them in latent space, decode them, and then reassemble the final image.The process starts with a source image loaded into the workflow. That image is first enlarged through a traditional upscale model using 4x_NMKD-Siax_200k. After that, the result is further normalized with ImageScaleToTotalPixels, targeting a larger working size while keeping the image manageable. This gives the workflow a stronger high-resolution base before diffusion refinement begins.A major feature of this workflow is tiled processing. The enlarged image is divided into tiles with TTP_Image_Tile_Batch, and the tile layout is controlled through TTP_Tile_image_size. This is important because very large images can be difficult to refine in a single pass, especially when you want detail recovery without destroying the whole composition. By splitting the image into tiles, the workflow can enhance local detail more effectively while still reconstructing the whole image afterward through TTP_Image_Assy.Another useful feature is automatic tile captioning. The workflow uses WD14Tagger to analyze the image tiles and generate prompt-like tag information. Those generated tags are passed through ShowText and then into the positive CLIPTextEncode route. This means the workflow does not depend entirely on a manually written prompt. Instead, it can derive a descriptive prompt from the source content itself, which helps the model understand what is already present in the image and refine it more consistently.The negative prompt is focused on suppressing common defects such as low quality, blur, bad anatomy, hand errors, extra fingers, deformed faces, text, watermark, logo, and JPEG artifacts. The refinement stage then uses VAEEncode, KSampler, and VAEDecodeTiled. The KSampler is configured as a relatively gentle enhancement pass, with moderate steps, low CFG, and low denoise. That is important because the goal is not to redraw the whole image from scratch. The goal is to preserve the original composition and identity while recovering texture, edges, and finer details.This workflow is ideal for anime illustrations, character art, cover images, social media assets, Civitai previews, and RunningHub publishing materials. In short, it is a practical high-resolution finishing workflow built around Anima Preview2, tile splitting, automatic tag assistance, and tiled latent refinement. If you want to see how the tile logic, tagger, upscale model, and final reconstruction work together, make sure to watch the full video tutorial from the YouTube link above.⚙️ Try the Workflow Online👉 Workflow: https://www.runninghub.ai/post/2033542982825152514/?inviteCode=rh-v1111Open the link above to run the workflow directly online and view the generation results in real time.If the results meet your expectations, you can also deploy it locally for further customization.🎁 Fan Benefits: Register now to get 1000 points, plus 100 daily login points — enjoy 4090-level performance and 48 GB of powerful compute!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you are in Mainland China or the Asia-Pacific region, you can watch the video below for workflow demos and a detailed creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1Q1w1zKEwk/I will continue updating model resources on Quark Drive:👉 https://pan.quark.cn/s/20c6f6f8d87bThese resources are mainly prepared for local users, making creation and learning more convenient.⚙️ 在线体验工作流👉 工作流： https://www.runninghub.ai/post/2033542982825152514/?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1Q1w1zKEwk/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
Anima Preview2 Image-to-Image Workflow
This workflow is designed for Anima Preview2 image-to-image generation, giving creators a clean and efficient way to transform an input image into a more polished anime-style result while still preserving the original composition and core visual structure. Unlike a pure text-to-image workflow, this setup begins with a source image, analyzes it, converts it into latent space, and then regenerates it through Anima Preview2 with prompt guidance. This makes it especially useful for style transfer, anime enhancement, visual cleanup, character refinement, and turning an existing image into a more cinematic anime illustration.The workflow uses anima-preview2.safetensors as the main generation model, qwen_3_06b_base.safetensors as the text encoder, and qwen_image_vae.safetensors as the VAE. It starts by loading a source image, then rescales it with image_scale_pixel_v2 so the input is normalized to a model-friendly pixel target while keeping alignment stable. After that, the image is encoded into latent space with VAEEncode, which becomes the structural starting point for the generation stage. This design makes the workflow ideal for users who want to preserve the general framing, subject position, and image layout rather than generating from scratch.One of the useful features in this workflow is the WD14 tagger support. The loaded image is automatically analyzed to produce tag-style prompt information, which can then be combined with the manual positive prompt. In the provided setup, the base prompt includes a beach-sunset singing-girl concept, showing how this workflow can mix image-derived tags with user-written prompt direction. This is practical for image-to-image generation because it reduces prompt-writing difficulty and helps the model better understand the existing content of the input image.The negative prompt is focused on common quality issues, including low quality, blur, bad anatomy, bad hands, extra fingers, fused fingers, deformed faces, text, watermark, logo, and JPEG artifacts. This helps keep the result clean during regeneration, especially when the source image already contains imperfect details or when the user wants a sharper anime-style finish.The main generation stage uses a KSampler with 40 steps, CFG around 3, DPM++ 2M SDE GPU sampling, SGM Uniform scheduling, and denoise around 0.75. This is important because the denoise value makes the workflow more flexible than a strict preservation pipeline. It allows the model to significantly re-render the image while still respecting the source structure. In practice, this means the workflow can keep the pose and composition from the original image, but improve the linework, lighting, facial beauty, atmosphere, and overall anime rendering quality.After sampling, the result is decoded through the VAE, previewed, and saved directly. This makes the workflow suitable for anime restyling, character cleanup, visual enhancement, cover-image polishing, concept refinement, and rapid RunningHub or Civitai demonstrations. If you want to see how the source image, WD14 tagger, prompt guidance, and Anima Preview2 regeneration work together, make sure to watch the full video tutorial from the YouTube link above.⚙️ Try the Workflow Online👉 Workflow: https://www.runninghub.ai/post/2033542974574956546?inviteCode=rh-v1111Open the link above to run the workflow directly online and view the generation results in real time.If the results meet your expectations, you can also deploy it locally for further customization.🎁 Fan Benefits: Register now to get 1000 points, plus 100 daily login points — enjoy 4090-level performance and 48 GB of powerful compute!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you are in Mainland China or the Asia-Pacific region, you can watch the video below for workflow demos and a detailed creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1Q1w1zKEwk/I will continue updating model resources on Quark Drive:👉 https://pan.quark.cn/s/20c6f6f8d87bThese resources are mainly prepared for local users, making creation and learning more convenient.⚙️ 在线体验工作流👉 工作流： https://www.runninghub.ai/post/2033542974574956546?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1Q1w1zKEwk/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
Anima Acceleration Comparison Testing Workflow
This workflow is designed for Anima acceleration comparison testing. Its main purpose is to compare different Anima generation routes, sampling stages, and acceleration settings under the same prompt, same latent size, and similar visual conditions. Instead of only showing one final image, this workflow is built to help creators understand how different sampling strategies affect image quality, speed, structure, detail, and final style.The workflow uses anima-preview.safetensors as the main Anima model, qwen_3_06b_base.safetensors as the text encoder, and qwen_image_vae.safetensors as the VAE. The latent canvas is created with EmptyLatentImagePresets at 1152 x 896, which gives a wide illustration format suitable for anime comparison tests, character previews, and visual benchmark examples. The prompt describes an anime-style fox girl in a snowy night scene holding a sign, which is a useful test prompt because it includes character identity, clothing, environment, lighting, atmosphere, and readable text.The negative prompt is also shared across the comparison routes, suppressing problems such as worst quality, low quality, blurry output, JPEG artifacts, signatures, and artist names. This keeps the comparison cleaner, because each route is tested against the same quality-control conditions.The key value of this workflow is that it does not test Anima as a single black-box output. It breaks generation into multiple sampling paths. Some routes use Anima Preview directly, while other routes apply an additional Anima RDBT-style LoRA path. The workflow includes multiple KSamplerAdvanced nodes with different start and end steps, making it possible to observe how early-stage, mid-stage, and later-stage sampling changes the final image. This is useful for understanding which part of the denoising process controls composition, which part affects details, and which route produces better speed-quality balance.The workflow also uses image concatenation nodes to place outputs side by side. This makes it easier to visually compare results instead of checking images one by one. You can compare whether a route gives better line quality, cleaner character design, stronger lighting, better color harmony, more stable text, or fewer artifacts. The workflow can also export comparison results as animated WEBP, making it useful for tutorial demonstrations, Civitai previews, RunningHub examples, and YouTube explanation videos.This setup is especially useful for creators who want to test acceleration methods before publishing a workflow. A faster route is only valuable if the result remains usable. This workflow helps identify whether accelerated Anima generation can preserve composition, character quality, and visual polish, or whether it loses too much detail compared with a fuller sampling path.In short, this is a practical Anima speed and sampling comparison workflow. It helps users test different Anima routes, compare output quality, and decide which generation setting is best for online publishing, fast iteration, or final-quality output. If you want to see how the acceleration routes are connected and how the different outputs compare visually, watch the full tutorial from the YouTube link above.⚙️ Try the Workflow Online👉 Workflow: https://www.runninghub.ai/post/2029161577533939714?inviteCode=rh-v1111Open the link above to run the workflow directly online and view the generation results in real time.If the results meet your expectations, you can also deploy it locally for further customization.🎁 Fan Benefits: Register now to get 1000 points, plus 100 daily login points — enjoy 4090-level performance and 48 GB of powerful compute!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you are in Mainland China or the Asia-Pacific region, you can watch the video below for workflow demos and a detailed creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV15dPezmEaz/I will continue updating model resources on Quark Drive:👉 https://pan.quark.cn/s/20c6f6f8d87bThese resources are mainly prepared for local users, making creation and learning more convenient.⚙️ 在线体验工作流👉 工作流： https://www.runninghub.ai/post/2029161577533939714?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV15dPezmEaz/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
Anima Image-to-Image Workflow
This ComfyUI workflow is designed for Anima image-to-image generation, anime-style image reconstruction, prompt-assisted redraw, and controlled visual transformation from an existing image. The main purpose of this workflow is to let creators upload a source image, automatically analyze it with Qwen3-VL, convert the image content into a useful text description, and then use Anima Preview to redraw or transform the image through an image-to-image pipeline.Unlike a pure text-to-image workflow, this graph starts from an existing image. The source image provides the original structure, composition, subject placement, and visual direction. The model then uses the generated prompt and the encoded image latent to create a new result based on that input. This makes the workflow useful when you already have an image idea but want to restyle it, improve it, anime-fy it, rebuild it with Anima, or create a controlled variation without starting from zero.The workflow is built around Anima Preview, using anima-preview.safetensors as the main diffusion model. It also uses qwen_3_06b_base.safetensors as the CLIP/text encoder and qwen_image_vae.safetensors as the VAE. This gives the workflow a lightweight but practical image-to-image generation structure for anime-style and illustration-style reconstruction.One of the most useful parts of the workflow is the Qwen3VLProcessor node. The source image is passed into Qwen3-VL with the instruction “Describe this anime image.” Qwen3-VL then generates a text description of the image content. This response is connected directly into the positive prompt CLIPTextEncode node. In other words, the workflow can automatically turn the uploaded image into a prompt, then use that prompt to guide the Anima redraw process.This automatic prompt reconstruction is useful for users who do not want to manually describe every detail in the source image. If the image contains a character, clothing, pose, background, color palette, or scene style, Qwen3-VL can produce a descriptive prompt that gives Anima a clearer semantic direction. This helps the workflow preserve the image concept while still allowing the model to rebuild the final result.The source image is first loaded through LoadImage, then resized through image_scale_pixel_v2. The resize node controls the total pixel count and aligns the image to a model-friendly grid. In the uploaded setup, the total pixel value is around 1.0486 megapixels with 64 alignment. This helps keep the workflow stable and makes the image suitable for VAE encoding and Anima sampling.The resized image is then sent to VAEEncode. This converts the source image into latent space, allowing the KSampler to use it as the image-to-image starting point. This is the core difference between this workflow and normal text-to-image generation. The sampler is not generating from empty noise only; it is using the original image latent as the base structure.The prompt route has two parts. The positive prompt is generated from Qwen3-VL and encoded through CLIPTextEncode. The negative prompt is manually defined and includes terms such as low quality, worst quality, blurry, bad anatomy, bad hands, extra fingers, fused fingers, deformed face, text, watermark, logo, and JPEG artifacts. This negative prompt is practical for anime and illustration workflows because it suppresses common generation problems.The main generation stage uses KSampler. In the uploaded setup, the sampler uses around 31 steps, CFG 4, er_sde sampler, simple scheduler, and a denoise value around 0.7. This is a medium-to-strong image-to-image redraw setting. It allows the model to transform the source image noticeably while still using the original image as structural guidance.The denoise value is important. A low denoise value would preserve more of the original image and make only light changes. A high denoise value would allow a stronger redraw and more style transformation. The included value around 0.7 is suitable when the user wants Anima to reconstruct the image with a visible model style rather than only perform a small enhancement. It is a good starting point for anime conversion, character redraw, image cleanup, style transfer, and creative variation.After sampling, VAEDecode converts the latent output back into an image. The result is then shown through PreviewImage and saved through SaveImage. This keeps the workflow simple and direct: load image, analyze image, encode image, redraw with Anima, preview, and save.This workflow is especially useful for anime image-to-image tasks. It can take a rough image, an old generated image, a reference illustration, a screenshot-like input, or a draft composition and rebuild it into a cleaner Anima-style output. It is also useful for testing how Anima interprets image structure when guided by an automatically generated Qwen3-VL prompt.The workflow is also useful for prompt learning. Because Qwen3-VL describes the input image, users can see how the model interprets the source image. This can help creators learn what kind of prompt language works for anime-style reconstruction. Users can also manually edit the Qwen3-VL response if they want more control. For example, they can keep the character and pose description, but change the background, clothing, lighting, or style direction before running Anima.Main features:- Anima image-to-image workflow- Uses anima-preview.safetensors- Qwen 3 0.6B text encoder support- Qwen image VAE support- Qwen3-VL automatic image description- Source image to positive prompt conversion- image_scale_pixel_v2 resolution control- VAEEncode image latent initialization- KSampler image-to-image redraw- Medium-to-strong denoise transformation- Negative prompt for anime artifact suppression- VAEDecode final image output- PreviewImage and SaveImage export- Suitable for anime redraw, image variation, style reconstruction, and visual cleanupRecommended use cases:Anime image-to-image generation, Anima Preview testing, image restyling, anime character redraw, prompt-assisted image reconstruction, source image variation, AI illustration cleanup, draft-to-final image generation, reference image transformation, style exploration, character concept variation, social media cover creation, Civitai example image preparation, RunningHub workflow publishing, YouTube thumbnail testing, and Bilibili cover image production.Suggested workflow:Start by uploading the source image. The source image should have a clear subject and composition. A character portrait, anime illustration, concept sketch, AI-generated image, or clean reference image works best. If the image is too blurry, too compressed, or visually chaotic, the generated result may also become unstable.Let Qwen3-VL analyze the image. The workflow uses the instruction “Describe this anime image,” which is suitable for anime-style inputs. If you want a different type of result, you can change this instruction. For example, you can ask Qwen3-VL to describe the image as a cinematic anime poster, a character design sheet, a detailed fantasy illustration, or a social media cover.Review the generated prompt when possible. The Qwen3-VL response becomes the positive prompt for Anima. If the description misses important details, manually add them. If it describes something incorrectly, remove or correct that part. The automatic prompt is convenient, but manual correction can improve final quality.Use the resize node to control the working size. The uploaded setup uses a total pixel control and 64-pixel alignment. This is a good general setting for model compatibility. If you want higher detail, increase the total pixel value carefully. If your GPU or cloud environment has limited memory, reduce it.Adjust denoise based on the goal. If you want to keep the source image closer to the original, reduce denoise. If you want a stronger Anima redraw or style transformation, keep denoise higher. The included value around 0.7 is suitable for visible reconstruction. For subtle cleanup, try lower values. For more creative transformation, try higher values.Use the negative prompt to control common problems. The included negative prompt is practical for anime generation because it suppresses blur, bad anatomy, hand errors, extra fingers, deformed faces, text, watermark, logo, and JPEG artifacts. If your result has specific issues, add targeted negative terms.Run several seeds for variation. Image-to-image workflows can produce different outputs from the same source image and prompt. If the first result is not ideal, test a few seeds before changing the whole workflow. If the structure is wrong, reduce denoise. If the result is too close to the source, increase denoise.For character images, inspect face identity, hair shape, clothing continuity, hands, and body proportions. For background-heavy images, check perspective, lighting, and object consistency. For cover images, check whether the subject remains clear and whether the final result has enough visual impact.If the Qwen3-VL prompt is too plain, add style language manually. You can add terms such as detailed anime illustration, clean line art, cinematic lighting, soft shadows, sharp character design, high-quality rendering, expressive face, refined clothing detail, or dramatic composition. Keep the prompt focused and avoid conflicting style instructions.This workflow is designed as a practical Anima image-to-image pipeline for ComfyUI users. It combines image loading, automatic image captioning, prompt conditioning, latent image-to-image sampling, and final output into one simple graph. It is especially useful for creators who want a fast and controllable way to turn an existing image into a new Anima-style result without writing the entire prompt manually.🎥 YouTube Video TutorialWant to know what this workflow actually does and how to start fast?This video explains what the tool is, how to launch the workflow instantly, and shares my core design logic — no local setup, no complicated environment.Everything starts directly on RunningHub, so you can experience it in action first.👉 YouTube Tutorial: https://youtu.be/J2A8JWDCUhkBefore you begin, I recommend watching the video thoroughly — getting the full context helps you understand the tool faster and avoid common detours.⚙️ RunningHub WorkflowTry the workflow online right now — no installation required.👉 Workflow: https://www.runninghub.ai/post/2021929700087566337/?inviteCode=rh-v1111If the results meet your expectations, you can later deploy it locally for customization.🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance and 48 GB super power!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you’re in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1FscqzREni/☕ Support Me on Ko-fiIf you find my content helpful and want to support future creations, you can buy me a coffee ☕.Every bit of support helps me keep creating — just like a spark that can ignite a blazing flame.👉 Ko-fi: https://ko-fi.com/aiksk💼 Business ContactFor collaboration or inquiries, please contact aiksk95 on WeChat.🎥 YouTube 视频教程想了解这个工作流到底是怎样的工具，以及如何快速启动？视频主要介绍 工具定位、快速启动方法 和 我的构筑思路。我们会直接在 RunningHub 上进行演示，让你第一时间看到实际效果。👉 YouTube 教程： https://youtu.be/J2A8JWDCUhk开始前建议尽量完整地观看视频 —— 把握整体思路会更快上手，也能少走常见弯路。⚙️ 在线体验工作流现在就可以在线体验，无需安装。👉 工作流： https://www.runninghub.ai/post/2021929700087566337/?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1FscqzREni/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
Anima Tiled Upscale Workflow
This ComfyUI workflow is designed for Anima tiled upscaling, high-resolution image enhancement, and tile-based detail reconstruction. The main purpose of this workflow is to take an existing image, enlarge it, split it into manageable tiles, automatically describe each tile with Florence2, refine each tile through Anima Preview, and then reconstruct the final high-resolution image with improved clarity and detail.Unlike a simple one-click upscale workflow, this graph uses a multi-stage enhancement structure. It does not only increase image size. It first uses a traditional upscale model to enlarge the source image, then uses AI-based tile refinement to rebuild details locally. This makes the workflow useful for images that need more texture, cleaner edges, stronger local detail, and better final publishing quality.The workflow is built around Anima Preview, using anima-preview.safetensors as the main diffusion model. It also uses qwen_3_06b_base.safetensors as the text encoder and qwen_image_vae.safetensors as the VAE. The workflow includes an optional LoRA route through LoraLoader, which can be used to apply a specific style, model feature, or subject-related enhancement during the tile refinement stage. In the uploaded graph, the LoRA example is loaded with model and clip strengths around 0.85.The first stage starts with LoadImage. The user uploads the image that needs to be enlarged and refined. This can be an AI-generated image, anime artwork, character illustration, concept art, product-style render, social media cover, or any image that already has a good composition but needs higher resolution and better detail.The image is then passed into ImageUpscaleWithModel using 4x_NMKD-Siax_200k.pth. This is the first-stage traditional upscaler. It enlarges the image while preserving the original layout and structure. Traditional upscalers are stable and fast, but they often cannot add enough semantic detail. That is why this workflow continues with Anima-based tile refinement after the initial enlargement.After the first upscale, ImageScaleToTotalPixels is used to control the final working size. In the uploaded setup, the workflow uses Lanczos scaling and targets around 3 megapixels. This gives users a practical way to control output scale without manually calculating width and height. It also helps keep the workflow manageable before entering the tiled refinement stage.The next key stage is tile splitting. The workflow uses TTP_Tile_image_size to calculate tile dimensions based on the image size, width factor, height factor, and overlap rate. In the uploaded graph, the width factor and height factor are set to 2, with an overlap rate around 0.35. TTP_Image_Tile_Batch then splits the image into tile batches, with a tile size route around 1024 x 1024. This allows the workflow to process large images more safely and with better local detail.Tile-based processing is important because high-resolution images can be too heavy for a full-frame diffusion pass. If the entire image is processed at once, VRAM usage may become too high, and the model may not focus enough on local details. By splitting the image into tiles, the workflow gives Anima Preview smaller and more focused regions to refine.The workflow then converts the tile batch into an image list and sends each tile into Florence2. Florence2Run is configured with a detailed caption task using the Florence-2-Flux-Large model. This stage automatically generates a description for each tile. The captions are displayed through ShowText, allowing users to see what Florence2 detected in each region.This automatic captioning stage is one of the most useful parts of the workflow. Different tiles may contain different content. One tile may contain a face, another may contain clothing, another may contain a background object, and another may contain a vehicle or landscape. Instead of using one generic prompt for the whole image, Florence2 gives the workflow local semantic descriptions for each tile. These descriptions are then used as positive conditioning for Anima refinement.The positive conditioning is created through CLIPTextEncode using the Florence2-generated captions. The negative prompt is also encoded through CLIPTextEncode. The included negative prompt suppresses common image problems such as low quality, worst quality, blur, bad anatomy, bad hands, extra fingers, fused fingers, deformed faces, text, watermark, logo, and JPEG artifacts. This is especially useful for tile refinement because local redraw can sometimes introduce unwanted artifacts if not constrained.The tile refinement stage uses VAEEncode, KSampler, and VAEDecodeTiled. Each tile is encoded into latent space through the Qwen image VAE, then processed with Anima Preview. In the uploaded setup, the KSampler uses around 31 steps, CFG 4, er_sde sampler, simple scheduler, and a denoise value around 0.25. This is a conservative low-denoise setting, suitable for detail enhancement rather than full regeneration.The low denoise value is important. In an upscale workflow, the goal is usually to preserve the original image while improving quality. If denoise is too high, the model may change faces, clothing, composition, background objects, or character identity. If denoise is too low, the result may not gain enough detail. A value around 0.25 is a good starting point for faithful enhancement.After sampling, VAEDecodeTiled decodes the refined latent tiles. The tiled decode settings use a large tile size and overlap, which helps manage memory and reduce decoding artifacts. The decoded image list is converted back into a batch through easy imageListToImageBatch. Then TTP_Image_Assy reconstructs the full image from the processed tiles, using position data, original size, grid size, and padding. In the uploaded setup, padding is set around 128 to help reduce visible seams between tiles.The final output is saved through SaveImage. The workflow also includes Image Comparer, allowing users to compare the original image and the final reconstructed image in a slide view. This is very useful for judging whether the enhancement is actually successful. A good upscale result should improve sharpness, texture, line clarity, and local detail without changing the original image identity too much.This workflow is especially useful for anime-style images and stylized artwork. Anima Preview is well suited for anime and illustration refinement, while Florence2 helps generate local descriptions for each tile. This combination allows the workflow to enhance faces, hair, clothing folds, mechanical details, background objects, and environmental textures more intelligently than a simple pixel upscaler.It can also be used for Civitai showcase images, RunningHub workflow examples, YouTube thumbnails, Bilibili covers, AI character art, concept art, poster-style images, and social media visuals. When an image already looks good but lacks final-resolution polish, this workflow can act as a finishing pass.Main features:- Anima tiled upscale workflow- Uses anima-preview.safetensors- Qwen 3 0.6B text encoder support- Qwen image VAE support- Optional LoRA route for style or subject enhancement- First-stage 4x_NMKD-Siax upscaling- ImageScaleToTotalPixels output size control- TTP tile size calculation- TTP_Image_Tile_Batch tile splitting- Florence2 automatic tile captioning- Tile-specific positive prompt generation- Low-denoise Anima refinement- VAEDecodeTiled for memory-friendly decoding- TTP_Image_Assy final reconstruction- Image Comparer for before-and-after inspectionRecommended use cases:Anime image upscaling, Anima Preview testing, tiled image refinement, character art enhancement, AI illustration polishing, concept art cleanup, high-resolution cover preparation, Civitai showcase image output, RunningHub workflow publishing, Bilibili cover image preparation, YouTube thumbnail enhancement, poster output, detail reconstruction, and before-after comparison demonstrations.Suggested workflow:Start by loading a clean source image. The workflow works best when the original image already has a strong composition and clear subject. It can improve softness and detail, but it cannot fully fix a source image that is extremely broken, heavily compressed, or structurally wrong.Run the first-stage upscale with 4x_NMKD-Siax. This gives the workflow a larger base image before Anima refinement. Then use ImageScaleToTotalPixels to control the target resolution. If the output is too large for your GPU, reduce the megapixel value before entering the tile stage.Use the tile settings carefully. The uploaded setup uses a tile workflow around 1024 x 1024 with overlap support. If seams appear in the final output, increase overlap or padding. If VRAM usage is too high, reduce the final megapixel target or use smaller tile settings.Let Florence2 generate captions for the tiles. Check the ShowText output when possible. If Florence2 describes a tile incorrectly, you can manually adjust the prompt or simplify the image before processing. Better tile captions usually produce more accurate local refinement.Keep denoise conservative for faithful upscaling. The default low-denoise route is designed to preserve the original image while improving local detail. Increase denoise only if you want stronger redraw. Reduce denoise if faces, hands, clothing, or identity begin to change too much.Use the negative prompt to suppress common artifacts. For anime and illustration, keep terms like low quality, blurry, bad anatomy, bad hands, extra fingers, deformed face, text, watermark, and logo. For realistic images, you may add more realism-specific artifact controls.After reconstruction, use Image Comparer to inspect the before-and-after result. Look at face stability, line clarity, hair detail, fabric texture, object edges, background continuity, and whether any tile seams are visible. A good result should feel sharper and richer while still looking like the same image.For character images, inspect the face and hands first. For mechanical or product-style images, inspect edges, logos, surface texture, and symmetry. For background-heavy images, inspect tile borders and repeated patterns carefully.This workflow is designed as a practical Anima high-resolution finishing pipeline for ComfyUI users. It combines traditional upscaling, automatic tile captioning, Anima low-denoise refinement, tiled decoding, final reconstruction, and before-after comparison into one usable graph. It is especially useful for creators who need clean, detailed, and publishable images for Civitai, RunningHub, YouTube, Bilibili, and social media platforms.🎥 YouTube Video TutorialWant to know what this workflow actually does and how to start fast?This video explains what the tool is, how to launch the workflow instantly, and shares my core design logic — no local setup, no complicated environment.Everything starts directly on RunningHub, so you can experience it in action first.👉 YouTube Tutorial: https://youtu.be/J2A8JWDCUhkBefore you begin, I recommend watching the video thoroughly — getting the full context helps you understand the tool faster and avoid common detours.⚙️ RunningHub WorkflowTry the workflow online right now — no installation required.👉 Workflow: https://www.runninghub.ai/post/2021929692445544449/?inviteCode=rh-v1111If the results meet your expectations, you can later deploy it locally for customization.🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance and 48 GB super power!📺 Bilibili Updates (Mainland China & Asia-Pacific)If you’re in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown.📺 Bilibili Video: https://www.bilibili.com/video/BV1FscqzREni/☕ Support Me on Ko-fiIf you find my content helpful and want to support future creations, you can buy me a coffee ☕.Every bit of support helps me keep creating — just like a spark that can ignite a blazing flame.👉 Ko-fi: https://ko-fi.com/aiksk💼 Business ContactFor collaboration or inquiries, please contact aiksk95 on WeChat.🎥 YouTube 视频教程想了解这个工作流到底是怎样的工具，以及如何快速启动？视频主要介绍 工具定位、快速启动方法 和 我的构筑思路。我们会直接在 RunningHub 上进行演示，让你第一时间看到实际效果。👉 YouTube 教程： https://youtu.be/J2A8JWDCUhk开始前建议尽量完整地观看视频 —— 把握整体思路会更快上手，也能少走常见弯路。⚙️ 在线体验工作流现在就可以在线体验，无需安装。👉 工作流： https://www.runninghub.ai/post/2021929692445544449/?inviteCode=rh-v1111打开上方链接即可直接运行该工作流，实时查看生成效果。如果觉得效果理想，你也可以在本地进行自定义部署。🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！📺 Bilibili 更新（中国大陆及南亚太地区）如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。📺 B站视频： https://www.bilibili.com/video/BV1FscqzREni/我会在 夸克网盘 持续更新模型资源：👉 https://pan.quark.cn/s/20c6f6f8d87b这些资源主要面向本地用户，方便进行创作与学习。
AP3-to-Illustrious Pixel Relay Upscale Workflow
AP3-to-Illustrious Pixel Relay Upscale WorkflowA toy two-stage ComfyUI workflow made for myself that uses Anima Preview 3 to create the base image, then uses Illustrious SDXL to upscale and refine it.The main idea is to keep AP3 focused on what it does well: caption following, scene setup, and multi-character interaction. At the same time, it avoids relying on AP3 for high-resolution diffusion refinement, since its official recommendation is to stay around the ~1MP range and image quality can degrade at higher resolutions.In this workflow, AP3 generates the initial base image. The image is then passed in pixel space to the second stage, where the upscale model handles the actual image enlargement, and Illustrious SDXL redraws/refines the enlarged image through UltimateSDUpscale. This avoids pushing AP3 beyond its comfortable resolution range while still taking advantage of Illustrious’s stronger high-resolution redraw, background rendering, and detail refinement.Cover generated by GPT image.
Anima Ready-to-use Workflows (t2i, inpaint, upscale, adetailer, SeedVR2)
My NSFW TG Channel - https://t.me/stefanfalkokmothMy TG Channel - https://t.me/StefanFalkokAIMy TG Chat - https://t.me/+y4R5JybDZcFjMjFiHi! I introduce my working workflows Anima ready-to-use workflowsIncluded T2I, Inpaint, Latent Upscale+Adetailer, seedvr2, and included Ultralytics and Upscale modelshttps://huggingface.co/circlestone-labs/Anima/tree/main/split_filesLet's make ComfyUI more friendly for new and old users!Leave comments if you have trouble or you found the problem with workflows
Cappy's ComfyUI Workflow Collection
Since I don't want to spam my profile or this site in general with workflows, I decided to move all of my workflows to this page. If you need help with these, please tell me your issue in the comments of this page! I hope you enjoy my workflows! :3
Anima img2img (denoise) with Loras
Anima img2img (denoise) with Loras support.
Anima text2image with Loras
Anima text2image with Loras support.
DyPE for Anima workflow
DyPE for Anima — Dynamic Position ExtrapolationRequirements:Base model: https://civitai.com/models/2458426/anima-official Recommended: https://civitai.com/models/2540444/anima-highresaesthetic-boostDyPE node (my fork): Install via ComfyUI Manager or git clone https://github.com/levzzz5154/ComfyUI-DyPE custom_nodes/ComfyUI-DyPEThe fork adds Anima/Cosmos model support not yet available in upstream. PR pending at the original repo (https://github.com/wildminder/ComfyUI-DyPE).Recommended for: 1536²–2048² generations where you want to preserve composition that otherwise breaks down at higher resolutions. Works with single images (DyPE doesn't affect video/sequence generation for this model).
Consistent AI Portrait Animator (AnimateDiff + IPAdapter Plus)
Professional AI Pipeline: Identity-Preserved Temporal AnimationThis is a custom ComfyUI workflow engineered for high-end digital character creation. It provides a robust solution for the "identity drift" problem in AI video, ensuring 100% facial consistency across dynamic frames.Technical Architecture & Features:Neural Identity Mapping: Implements a precise identity-locking mechanism using IPAdapter Plus to maintain the user's unique facial features.Temporal Coherence Engine: Utilizes a customized AnimateDiff-Evolved setup to generate fluid, flicker-free movement.Modular Phase Design: The workflow is logically structured into three distinct phases—Data Ingestion, Motion Synthesis, and High-Res Refinement—allowing for granular control over the output.Production-Ready Output: Optimized for generating high-fidelity video sequences (H.264/MP4) suitable for professional digital portfolios.Developer's Note: This project serves as a technical demonstration of advanced AI pipeline integration, focusing on the intersection of portrait photography and temporal neural synthesis.
Wan 2.2 i2v Comfyui Workflow :: SVI FFLF with Latent Saving and Loading
- This is a SVI workflow that enables first-frame-last-frame for continued smooth videos.- What is special about this workflow is that it uses my own custom nodes to save the latent as a file, and then loads the latent into the workflow so you can freely manage what video clip you want to tinker with.- These two custom nodes are provided along with the workflow; just drag the bvga_nodes folder into your custom nodes folder.- It's very convenient to use, as you're able to easily adjust prompts, motion loras, etc, for each stage of the video. You can keep experimenting before you move onto the next segment. And can even start back from previous segments by loading the old files!Credit to https://civitai.com/models/2232205?modelVersionId=2820076, whose workflows finally got SVI FFLF working for me, and I learned a lot from.You will need the following SVI LoRAs:High: https://huggingface.co/Kijai/WanVideo_comfy/blob/main/LoRAs/Stable-Video-Infinity/v2.0/SVI_v2_PRO_Wan2.2-I2V-A14B_HIGH_lora_rank_128_fp16.safetensorsLow: https://huggingface.co/Kijai/WanVideo_comfy/blob/main/LoRAs/Stable-Video-Infinity/v2.0/SVI_v2_PRO_Wan2.2-I2V-A14B_LOW_lora_rank_128_fp16.safetensorsv1.1 features various fixes and improvements.Notably:There are now 2 separate workflows provided:- 3-k-sampler workflow to get as much quality as possible out of the steps.- MoE-k-sampler workflow, which automatically optimizes the scheduling.Updated the latent saving and loading custom nodes:- v2 nodes better capture the full latent data, to ensure quality is retained.(Both the v1 and v2 nodes are included in the new folder. You can simply replace the old bvga_nodes folder with the new one)Fixed:- SVI nodes are no longer used in the starting video segment(which in certain situations degraded quality)- latents are no longer trimmed in segments that don't require it(which was reducing video length unnecessarily)
Wan 2.2 animate workflow
This is what i used before LTX to sync a videoWan 2.2 animate Wrapper versionThis uses wan animate and wan video wrapper to make long form lip synced videos. It does well for what it does. But it has limits and some issues.Issues1. Wan wrapper does not let me load GGUF, so scaled fp8 are used. 2. This is not set up to inpaint. I could have sworn inpainting worked but i cant connect a mask set to the mask input without error. - trying to make a native comfy flow that does inpaint later. Either im going crazy or the wrapper cant inpaint?3. This cant do complex stuff. This is an talking avitar copier. Dont try to do anything difficult.4. The smaller you make the video, the worse the ref image copy will be. 480p seems smallest you can go without loosing the subject. Unless its a close up face shot.Enable the options you want, you dont need pose, face, or a ref image. Just 1 works or any combo of them depending how much of the video you want to copy.If you use pose it will make a video following the movementIf you use use face video it will make a video copying the face movementIf you use a ref image it will copy the image subject and background and make a video from it.I actually find it works best with just a face video and ref image used as a start frame with no pose info. Prompting what you want the movement to be.
my comfyui workflow
This is my comfyui workflow,i hope it can help you.For the V2 version, I replaced the original LoRA stack with LoRA Manager for easier management. Image Plus is handy for saving image metadata.I set up this new ComfyUI separately for V2; there were plugin conflicts with my old setup, so I never used these features before. But manually uploading images every time was a hassle, so I just built a new one.I want to add extensions like upscaling, restoration, prompt extraction, gallery, ControlNet, etc. You can expand it yourself—I’m too lazy to deal with it, it’s a pain.我的comfyui 工作流，希望可以帮助到你，v2 版本原始lora stack 替换成lora manager 更方便管理，image plus 方便图片源信息保存v2单独新开的comfyui搭的，以前的插件冲突没用这玩意，但确实自己每次传图麻烦，索性搭了一个。想扩展放大，修复，反推提示词，d画廊，controlnet啥的，自己扩吧，我懒得整了，麻烦。rf_workflow只能用于rf模型，用别的把model sampling sd3 结点删了就行anima_workflow only be used for anima
anima-to-z-image
this is a 2 stage workflow. it uses anima as first stage to create base image content. uses z-image turbo as second stage to fix and refine the final image. upscaler can be bypassed if unwanted.
Another WAN2.2 workflow - 4steps (I2V/FLF/T2V)
There are far more WAN 2.2 workflows than I realized, but it's too late now :) Maybe someone will find it useful.RTX 3060 12gb | Ryzen 5 3600 | 32gb RAM | 45-98gb Page File(With models loaded, 5 second vid)I2V & T2V: Around 6 minutes+MMAudio: Around 8 minutesI set the res to 480 but you can change it using Resolution Master's presets. I considered adding an upscaler, but I couldn't notice a difference and it added an extra minute.Downloads:---I don't recommend this WF if you don't want to download both the I2V and T2V models. If you want to try it still for some reason, add dummy files in your WAN checkpoint folder and use them as the undownloaded models to prevent an error.Base models & VAE: https://huggingface.co/QuantStack/Wan2.2-T2V-A14B-GGUF/tree/main & https://huggingface.co/QuantStack/Wan2.2-I2V-A14B-GGUF/tree/mainModel-2: https://civitai.red/models/2456413/dasiwa-wan-22-i2v-14b-boundbite-v10-or-lightspeed-or-gguf?modelVersionId=2763939I use Q5. You only need one VAE from either of the first links. Model-2 doesn't need the Lightx2v models below.CLIP (UMT5): Download with ComfyUI Manager!Lightx2v: https://huggingface.co/lightx2v/Wan2.1-I2V-14B-480P-StepDistill-CfgDistill-Lightx2v/tree/main/lorasMMAudio: Links are in v3's workflow. If you're coming from LTX, MMAudio is okay, but it can't do speech and is very basic.---For I2V, you can use Florence and leave your prompt empty unless you want something specific.To show video previews, turn on "Display animated previews when sampling" in VHS' settings.If you want FLF for anything other than perfect loops, use the different image switch.The NSFW switch was made to work with CubeyAI's NSFW lora. In v3, it only switches the MMAudio model from the NSFW one to the regular one.Delete all the filename stuff near VAE Decode if you don't want them.
LTX 2.3 – Image + Audio + Video (IC-LoRA - Union Control / Detailer) to Video
This workflow uses the LTX IC-LoRA for LTX-2.3.ControlNet (Union Control):Load an image and an audio file (either your own or the original audio from the source video), or alternatively use LTX Audio—the audio is used for lip synchronization. Then load the target video to track and transfer its movements.Refine and Upscale (Detailer):You can also refine and upscale an existing video by setting ControlNet to "Off", Image Bypass to "True" and loading the IC-LoRA file for the detailer "ltx-2-19b-ic-lora-detailer.safetensors" instead of the ControlNet model "ltx-2.3-22b-ic-lora-union-control-ref0.5.safetensors".Info:The length of the output video is determined by the number of frames in the input video, not by the duration of the audio file.For upscaling, I use RTX Video Super Resolution.Tips:If you experience issues with lip sync, try lowering the IC-LoRA Strength and IC-LoRA Guidance Strength values. A value of around 0.7 is a good starting point. If you notice issues with output quality, try lowering the IC-LoRA Strength as well.
ZRune Motion
ZRune Motion 是一张基于Wan2.2 I2V的工作流，主要用于角色动作展示、镜头推进和三段式视频生成。发布的内容仅供分享、学习，不是零依赖可复现整合包。适用方向：- 角色动作视频- 三段式镜头组织- I2V 人物演示- 风格化动作展示核心特点：- 基于 Wan2.2 I2V high / low 双分支- 适合角色动作和镜头连续性表达- 适合在 ComfyUI 中继续替换输入图、提示词和 LoRA 进行二次创作核心依赖：- Wan2.2 I2V A14B high/low 基座- wan2.2_i2v_high_noise_14B_Q8_0.gguf- wan2.2_i2v_low_noise_14B_Q8_0.gguf- Wan2_1_VAE_bf16.safetensors- umt5_xxl_fp8_e4m3fn_scaled.safetensors- SVI_v2_PRO_Wan2.2-I2V-A14B_HIGH_lora_rank_128_fp16.safetensors- SVI_v2_PRO_Wan2.2-I2V-A14B_LOW_lora_rank_128_fp16.safetensors- wan2.2_i2v_A14b_high_noise_lora_rank64_lightx2v_4step_1022.safetensors- wan2.2_i2v_A14b_low_noise_lora_rank64_lightx2v_4step_1022.safetensors- wan2.2_wugong_i2v_high.safetensors- wan2.2_wugong_i2v_low.safetensors- zr_cine_high_s600.safetensors- zr_cine_low_s600.safetensors自定义节点依赖：- rgthree-comfy- comfyui-kjnodes- ComfyUI-GGUF- ComfyUI-Custom-Scripts- ComfyUI-Frame-Interpolation注意事项：- 这是 Wan2.2 I2V 工作流，不适配 T2V 或 T2I LoRA- 本次分享不包含zr_cine_high_s600.safetensors和zr_cine_low_s600.safetensors两个特定人物形象风格的LoRA模型，导入工作流后请先禁用或替换这两个 LoRA- 最低配置要求为24G显存。
Wan2.2 Animate + SAM3 - 1 minute long videos
Apr-10-25 hotfix:Made some changes to the workflow to improve quality, see the "About this version" section for details.This workflow implements SAM3 for character masking.It uses looping to process long videos, and while it can theoretically generate infinitely long videos, in practice it is limited by VRAM and processing time.It supports both Wan Animate modes: character animation and character swap.It has been tested on 1-minute, 480p videos using an RTX 4090 with 64GB of RAM, though it supports 360p, 720p and 1080p resolutions too. My tests suggest you can generate up to 20 seconds at 1080p with an RTX 4090 using a video batch size of 80 frames (if you have the patience and the time for it).Features:Character animation:Full reference image animation using a video reference.Character swap:Insert a character in-place in a video using high quality segmentation with SAM3.Face swap:Possible using "face" as "CHARACTER ISOLATION DESCRIPTION", but don't expect miracles. You can achieve better results by face swamping the face separately using some edit model like Flux Klein or Qwen edit and then use that new edited image as your Character Reference Image using the character animation feature. Detailed instructions are contained within the workflow itself:- Yellow nodes are input and configuration nodes you can change to suit your needs.- Red nodes are instructions and helpful notes
Anima: simple workflows for generation & inpainting/refining + tips & info
What is this?A couple of simple workflows for Anima Preview 2 (Edit: works great for v1 as well). One for generation, one for inpainting/refining/img-to-img.There's also info & tips inside the workflows, as well as below in this post.All images in this post were generated in Anima using the generation workflow, and then further edited using the inpaint workflow.UpdatesInfo Update 2026-05-21: Works great for v1.0Good news! Almost all of the info below I wrote for Preview 2 applies to version 1.0, and the workflow itself works great with it. So, you don't need to wait for me to update anything.I've got some minor improvements to add, like more info on resolutions (v1.0 can go higher than 1 megapixel) and also alternative sampler settings (res_2m and res_2s are interesting), but those aren't super important for you getting started. I'll update the workflow & this post to reflect the extra info later.For now, I updated the main model link below but everything else is still referencing Preview 2.Info Update 2026-05-04: Not making a Preview 3 versionI've spent a ton of time messing around with Preview 3, but it's just too unstable for me. The quality is good but I'm finding it impossible to get any art style consistency and stop the model from drifting all over the place with every gen.I recommend sticking with Preview 2 unless you have a specific reason for wanting to use Preview 3.Nodes & ModelsCustom Nodes:RES4LYF - A very popular set of samplers & schedulers, and some very helpful nodes. We want this for the Clownshark Ksampler, which massively helps with the Anima model's consistency.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.ComyUI Essentials - (Inpaint workflow only) Adds a bunch of very helpul nodes. We're just using it here for its number comparison node so we can switch between the image-to-image and inpainting modes automatically, which is a limitation of the LayerStyle node below.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: Anima v1.0Text Encoder: Qwen 3 0.6bVAE: Qwen Image VAEThe main model is only around 4.5GB, which means you can run it with as little as 6GB VRAM.Why use Anima over Illustrious/SDXL?You might not want or need to, but Anima does have at least one big thing going for it: you can use plain language to describe your images. Being forced to use booru tags with the SDXL models severely limits what you can get out of them, and being able to use normal language is a big unlocker. Also it's just pretty damn good already, and it's only going to get better from here.Info & TipsClownShark Ksampler - what is it and why are we using 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. For the Anima model, it has a massive stabilising effect on the quality of outputs. Normally the Anima model is a bit janky to work with, but this Ksampler magically improves the consistency via ETA with no drawbacks."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 them here, but for some models they're really, really, really good.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 either.So, essentially we're just using it for the ETA setting in these workflows. But boy is it worth it.UpscalingI upscaled some of these post images using 2xAnimeSharpV4. If you need an upscale workflow + links to upscale models the one I use is here: pastebinAnd some of them I upscaled using SeedVR2, workflow for that here: pastebinNote that the SeedVR2 upscaler requires at least 16GB of VRAM, it's a beefy model even when you're only upscaling a single image.Generation NotesThe settings I've left inside give very good results. You can range the ETA setting on the sampler from 0.2 to 0.5 and the steps from 22 - 30 to get different types of results as well. More steps = deeper shadows, which is good/bad depending on the style you want. I prefer around 24 steps personally.If you want stuff to lean more towards anime, just put "anime" or "digital anime" at the end of the basic positive prompt starter. You can play around with other style prompts like "digital painting" too, or even specific artist styles. Or just don't specify and see what happens.It also knows pretty much all the popular characters already so try typing those in.A LimitationThe Anima model uses the qwen image VAE, which means you'll sometimes get a noticeable grid-like halftone pattern in your images. This effect is more noticeable at higher resolutions. You can alleviate it by using the SeedVR2 upscaler, but that's not always desirable (not always the best upscaler). I'm looking into it but haven't found another solution yet - if you happen to know a way to fix it, please let me know!Note on resolutionsI've left some good starting resolutions in there, but the model is capable of going higher. Just start with a lower res, then resize it and feed it into the inpainting workflow to do img-to-img at around 40 denoise or so. I've tested 1440p and it works pretty well.
LTX2.3 Voice-over Video Production Workflow
SongGeneration 2 (ChatMusician) Workflow: AI Composer[Best for: Original Music Videos, Background Tracks, Musical Theory Logic]🚀 FREE Cloud Computing (No VRAM Limits): Run this music generation model in the cloud without worrying about complex environment setups. Register on RunningHub for 1,000 FREE credits: 👉 https://www.runninghub.ai/post/2037055296702521345/?inviteCode=rh-v1497📺 Full Video Tutorial: Learn how to use LTX2.3 to make Conversation and singing videos: 👉 💡 Key Features: This workflow integrates Tencent AI Lab's ChatMusician project. It doesn't just "patch" sounds together; it treats music as a language.Compose via Chat: Simply type a description (e.g., "A melancholic jazz piano piece") and let the AI generate the full score.Strong Logical Coherence: Perfect for structured music segments. It can even help you complete unfinished melodies.High-Fidelity Output: Specifically tuned for LTX 2.3 singing videos, producing high-sample-rate audio that syncs beautifully with visual movements.