ComfyUI Workflows
Browse real AI workflows from Flux, SDXL, Pony, ControlNet, video generation and commercial AI creators.
MonK ComfyUI Workflow
My workspace is for working in the ComfyUI environment and processing the finished image using Inpaint.
ZImage Turbo Essentials workflow
Note: If you have ANY issues with nodes not downloading, read the notes or reach out. There's nothing that special about any of them that aren't core modules.⛔⚠️🛑✋ Read the notes completly before using. Most common install and node problems are listed in the directions也有中文说明This model is set up for Z-Image Turbo. I have a seperate model for BaseA 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-breakdownThis is a modification to my advanced workflow designed for beginners:Checkpoint option for merged modelsSeed variance enhancerUltimate Upscaler for prescaling and hi res fixDetailing Suite for face, hands, & eyesEssentials Post Production SuiteSmart Noise scrubberFull tutorial available.也有中文说明Check out my other models: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 😉
Nvidia PiD 4K Image Upscaler Workflow (Z-Image, Flux.1, Flux.2, SD3 compatible) by Smittie
Smittie's PiD Upscaler WorkflowDescriptionWith this workflow you can upscale low-resolution images (512, 1K) to much higher resolutions (2K, 4K) using Nvidia's PiD (Pixel Diffusion). It is a simple image to image (I2I) workflow.Install Instructions1. Update ComfyUI to latest build with PiD support2. Install all missing custom nodes via the ComfyUI Manager3. Download all needed modelsDiffusion modelAll availableFor Z-Image or Flux.1 like images in 1K resolution use pid_flux1_1024_to_4096_4step_bf16.safetensorsText encodergemma_2_2b_it_elm_bf16.safetensorsVAEae.safetensorsVideo InstructionsIf you don't know what to do, you can watch the Nvidia PiD Tutorial by AI Search. He works with a different workflow, which I have built upon.AcknowledgementMost credits goes to the Nvidia, ComfyUI and CivitAI team, as well as kijai and AI Search. Thank you guys!
Zit Workflow Upscale Control Net and Ollama
This is my daily workflow for Zit. I think everything is pretty well organized with group bypassers so you only need to run what you need. The Lora Manager nodes and save nodes grab all relevant resources for posting to Civit saving the hassle of adding Metadata manually. There is a free Ram function at the end of the upscale because I noticed my pc holding it lol. Just delete it if you dont care. All of the nodes are pinned because I remote in from work on my phone and got sick of accidentally moving shit around. The couple of subgroups that are in the Workflow are for the save nodes so you will want to open them up and change your file names, or just expand them altogether, I like em, some people dont.Can't think of much more to say but if you have questions please ask and I'll try to help as much as I can.
Z-Image Base & Turbo Workflow I2I/T2I (Low or High VRAM)
This is an advanced workflow designed for those familiar with ComfyUI who want pro grade results. It is node heavy and has numerous options. I encourage you to take the time to learn it. If you feel overwhelmed, you can start with my simple model HERE也有中文说明Instagram: https://www.instagram.com/synth.studio.models/Buy me a☕ https://ko-fi.com/lonecatoneThis represents hundreds of hours of work. If you enjoy it, please 👍like, 💬 comment , and feel free to ⚡tip 😉Note: If you have ANY issues with nodes not downloading, read the notes or reach out. There's nothing that special about any of them that aren't core modules. Most issues can be resolved by updating everything⛔⚠️🛑✋ Read the notes completly before using. Most common install and node problems are listed in the directionsA 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-breakdownV14.0Because it is not in my nature to leave things alone....Gutted the stock face detailer nodes and replaced them with a SEGS suite for infinately better resultsNote: Yes, I will be going through all my workflows and replacing them, including teh NSFW onesControlnet Update.Added local bypass nodes so you don't have to zoom and scrollMajor update to notes, especially for detailersV13.0Overhauled the Controlnet:Now you can select the aspect ratio of the original image without having to "guess" it with the aspect ratio nodeLeft room for the resize node so you can choose which side to crop the image byRedid the dependencies so that every tie you shut it on and off, you don;t need to reset the switches so it will not give you an error 😡Massively improved the notes throughoutMinor layout tweaksV11.3Detailers are now able to handle float32 models, so I rewmoved the SDXL Based checkpoint.Please give me feedback with any issues on this, especailly RAM related.Added a crop section so you can...well... crop images easilyUpdated several areasAdded the ability to use a simple date with the metadata folderadded a "remove banned tags" nodeV8.0Removed Florence Prompt assist and replaced it with Qwen Advanced.Reorganized the controlnet section because it bugged me. Now it's easier to use.Added a LoRa controller for the detailer for better character facial and eye consistency as well as a localized promptGutted the backend of the post production suite and added 📷optical realism, giving a significantly better way to fine tune your images. Fixed some backend mathAdded a 📂 subfolder generator that seperates the metadata files according to the name you give the file for easier use.V7.0Added KSampler Configure Nodes so you can leave the setting in place for both ZIT and base without having to change themRemoved some problematic nodesRemoved Qwen Prompt enhancer. I refuse to put an API one in there.global scheduler and samplers for consistencyBetter mathImproved notes throughout.All in 1 workflow:ControlnetCheckpoint option for merged modelsI2IPrompt generationPrompt enhancementLora LoaderAdvanced KSamplerDraft mode for making corrections without having to rerun the workflow.Seed variance enhancerUltimate Upscaler for prescaling and hi res fixDetailing Suite for face, hands, eyes, and expressionsSeed VR2 Upscaler to get 4k on low VRAMMassive Post Production SuiteJPEG ScrubberSmart Noise scrubberFull tutorial available.也有中文说明Check out my other models:My NSFW version with Anantomy Correctors:https://civitai.com/models/2270894/zit-nsfw-anatomy-correctorFor ultimate Anime detail and control: https://civitai.com/models/2220766/zimage-ultra-anime-detail-workflow-get-the-most-out-of-your-generationsFor Realism and variety over the base model: https://civitai.com/models/2231181/z-image-ultra-real-workflow
Moody Minimal Zimage Workflow
I just made a new Telegram Group and Discord group drop in for chats, questions, feedbacks or just to share your work. 欢迎加入纸飞机和Discord群组。RH 体验地址：https://www.runninghub.ai/post/2062444041207373826/?inviteCode=cptpovgo喜欢我的作品吗？🎨 我会持续免费分享更多优质内容✨，欢迎支持一下，请我喝杯咖啡吧！💖 Like my work? 🎨 I share all my work for free ✨, so support me by buying me a coffee 💖!Ko-fi: ko-fi.com/catlover1937Fanvue: https://www.fanvue.com/catlover1937Please do not redistribute without consent. 如非作者同意禁止搬运。2026-06-04🚀 Moody Simple Workflow - Reborn! 🎉Guess what? 😆My Moody Simple Workflow has evolved so much over time that it's honestly not very "simple" anymore. 😂So I decided to create a brand new Minimal Workflow for everyone who:✨ Just wants to test ZImage quickly✨ Is completely new to ComfyUI✨ Wants an easy starting point without getting overwhelmed✨ Wants a small taste of what my full ZImage workflow can do 👀This workflow is intentionally kept as barebones and lightweight as possible 🪶✅ Minimal custom nodes✅ Beginner-friendly setup✅ Lots of notes and guides throughout the workflow✅ Easy to understand and modify✅ Great for learning the fundamentalsThink of it as the "starter pack" version of my full workflow — all the essentials, none of the chaos. 😎If you've ever opened my main workflow and thought:💀 "What am I even looking at?"This one is for you. 😂I hope this helps more people get started with ZImage and AI image generation without feeling overwhelmed.Have fun, and as always, happy generating! ❤️🔥🚀 Moody 简易工作流 - 重生啦！🎉猜猜怎么着？😆我的 Moody Simple Workflow 经过这么多次更新之后，已经进化到完全不「简单」了。😂所以我重新制作了一套全新的 极简版工作流，特别适合：✨ 想快速测试 ZImage 的朋友✨ 刚接触 ComfyUI 的新手✨ 不想一上来就被复杂节点吓到的人✨ 想先体验一下完整 ZImage 工作流能力的朋友 👀这套工作流刻意保持了 精简、轻量、易懂 的设计理念 🪶✅ 最少量的自定义节点✅ 新手友好✅ 内置大量注释和使用说明✅ 方便学习与修改✅ 快速掌握基础流程你可以把它理解成完整版工作流的「入门体验版」——保留核心功能，去掉各种花里胡哨的东西。😎如果你曾经打开我的完整工作流后：💀「这到底是什么天书？」那这套工作流就是为你准备的。😂希望它能帮助更多朋友轻松上手 ZImage 和 AI 绘图，不再被复杂的节点劝退。祝大家玩得开心！❤️🔥
Z-image turbo TXT2IMG, IMG2IMG, inpaint, controlnet, lora manager, caption, adetailer, SeedVR2.
A pretty simple workflow with a few common features. It is designed for 'legacy' mode. For now, I still think that Nodes 2.0 are mostly buggy and slower garbage, even though they have a few cool things. The metadata of a final image will not contain the diffusion model name; I still haven't found a good way to add it to the metadata. Fixed in v.1.2I didn't notice that Lora doesn't get into the metadata for automatic detection. Fixed in v2.0.
ZIT-GGUF-dAIver-v1.5
Optimized Low-VRAM Workflow for Z-Image-Turbo (GGUF) with CacheDiT Acceleration Roughly based on WikkedAI’s WikkedZITv4 – refined and enhanced by Experimental_dAIverThis workflow delivers the full power of Z-Image-Turbo in GGUF format, specially optimized for GPUs with less than 8 GB VRAM, like my RTX 4050 with only 6 GB. The integrated CacheDiT_Model_Optimizer and SageAttention2 (both optional!) provides a noticeable 1.4–1.6× speed boost with almost no quality loss. Two intelligent upscaling stages, automatic trigger-word integration via the Super LoRA Loader, and an extended save node complete this elegant setup.Version 1.5 brings significant improvements in speed, usability, and upscaling quality — while remaining extremely VRAM-efficient (tested on RTX 4050 with only 6 GB).What’s new in v1.5:PatchSageAttentionKJ integration for automatic Sage Attention optimization and faster sampling on supported hardwareselectLatentSizePlus — intuitive aspect-ratio and resolution selector with beautiful presets (including 7:12 Tall Vista and other golden-ratio-friendly options) plus easy orientation swapFull SEEDVR2 Video Upscaler Subgraph — powerful DiT-based (.safetensors or GGUF) high-end upscaler that delivers stunning 4K+ results with intelligent resolution handling, Lab color correction, and temporal settings. Works exceptionally well on still images too, producing superior detail and coherenceMain model updated to the higher-quality z-image-turbo-Q8_0.ggufCLIP switched to the abliterated Qwen3-4B-Instruct-2507.Q5_K_S.gguf (lumina2 type)Improved workflow organization, expanded notes, and more robust saving optionsRequired Custom Nodes (updated for v1.5):ComfyUI-GGUF - https://github.com/city96/ComfyUI-GGUF - UnetLoaderGGUF + CLIPLoaderGGUFComfyUI-CacheDiT - https://github.com/Jasonzzt/ComfyUI-CacheDiT - CacheDiT_Model_Optimizer – the turbo boost for DiT modelsnd-super-nodes - https://github.com/HenkDz/nd-super-nodes - NdSuperLoraLoader with tags, trigger words & beautiful UIsave-image-extended-comfyui - https://github.com/thedyze/save-image-extended-comfyui - Advanced saving with metadata & dynamic filenamesComfyUi-MzMaXaM - https://github.com/MzMaXaM/ComfyUi-MzMaXaM - selectLatentSizePlusComfyUI-SeedVR2_VideoUpscaler - https://github.com/numz/ComfyUI-SeedVR2_VideoUpscaler - SEEDVR2 Video Upscaler SubgraphModels & Downloads (exact paths)The following list explains the base models I am most frequently using with this workflow. The list as well explains where to put each file after you downloaded it.1. Main Model (Diffusion Model)File: z_image_turbo-Q8_0.gguf (higher quality Q8_0)Download: https://huggingface.co/jayn7/Z-Image-Turbo-GGUF/resolve/main/z_image_turbo-Q8_0.ggufTarget folder: ComfyUI/models/diffusion_models/2. Text Encoder (CLIP)File: Qwen3-4B-Instruct-2507-abliterated.Q5_K_S.ggufDownload: https://huggingface.co/mradermacher/Huihui-Qwen3-4B-Instruct-2507-abliterated-GGUF/resolve/main/Huihui-Qwen3-4B-Instruct-2507-abliterated.Q5_K_S.ggufTarget folder: ComfyUI/models/text_encoders/ (or clip/)3. VAEFile: ae.safetensors (~335 MB)Download: Usually included with Z-Image-Turbo setups or available here: https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/vae/ae.safetensorsTarget folder: ComfyUI/models/vae/4. Upscalers4× Upscaler: 4xLSDIRplusN.pth (variant of 4x-UltraSharp) → https://civitai.com/models/116225/4x-ultrasharp1× Skin-Contrast Upscaler: 1xSkinContrast-High-SuperUltraCompact.pth Download: https://huggingface.co/notkenski/upscalers/blob/main/1xSkinContrast-High-SuperUltraCompact.pthTarget folder: ComfyUI/models/upscale_models/5. SEEDVR2 Models (for the new high-end upscaler – optional but recommended):DiT Model: seedvr2_ema_3b-Q8_0.ggufVAE: ema_vae_fp16.safetensorsDownload from the official ComfyUI-SeedVR2_VideoUpscaler repository or Hugging Face and place in the folders required by the custom node.Key Nodes & Their FunctionsCacheDiT_Model_Optimizer + PathchSageAttentionKJ → next-generation turbo boostNdSuperLoraLoader with automatic trigger-word detection and clean tag interfaceselectLatentSizePlus → effortless aspect-ratio and resolution controlKSamplerAdvanced with proven settingsTwo-stage classic upscaler (4× LSDIR + 1× Skin-Contrast) in its own subgraph, orNew SEEDVR2 Video Upscaler Subgraph → for ultimate quality (optional, easily bypassed)SaveImageExtended with full metadata and dynamic filenamesRecommended Settings (already set in the workflow)Sampler: res_multistep or dpmpp_sdeScheduler: beta (or ddim_uniform)Steps: 8–11CFG Scale: 1.0–1.5Shift: 4–7Resolution: 864×1280 (portrait) – perfectly balanced for the golden ratio and typical ZIT outputs - the upscaler will automatically upscale by a factor of 4How to Use the WorkflowInstall all required custom nodesLoad the workflowEnter your positive prompt (NdSuperLoraLoader automatically adds trigger words)Adjust the negative promptChoose your desired aspect ratio and resolution in the Size SelectorKeep CacheDiT and Sage Attention enabled → GenerateOptionally run the SEEDVR2 upscaler for breathtaking 4K+ resultsDone — images saved with complete metadataSpecial thanks to @LumaRift who provided the SeedVR2 subworkflow and some good advise on optimizing my setup.
SoLordZ WF (upscale)
WF for upscale and adding more details for our SoLordZ and Lord Zolaris model series (ZIB and ZIT)we use our's TE:THM_UNC_Z-image_TE - v1.0 | ZImage Text Encoder | CivitaiZ_Image_TE_SoLordZ - Z_Image_TE_SoLordZ_v1 | ZImage Text Encoder | Civitaiand our models:SoLordZ - ZIT_v1.0 | ZImage Checkpoint | CivitaiLord Zolaris - ZIBv1.0 | ZImage Checkpoint | Civitaialso we use HDR VAE: Owen777/UltraFlux-v1 at mainI'm attaching two WF's, for ZIT and for ZIB because the settings are slightly different.There is also an additional upscale SeedVR2 stage in WF, I don't use it, but it might be useful to someone. To use this stage of the WF, you will need to download the official SeedVR2 models.If you like our work and want support us please support us directly, since I am in a country which cannot get money from Civitai https://www.buymeacoffee.com/tripleheadgPS: I used our unpublished models to generate the images.Stay tuned for updates :)Examples:
Z-Image Base & Turbo Pro Grade Realism Workflow (Low or High VRAM)
Realism...... I design the most professional workflows out there. I have better options, better samplers, better detailers, better prompt assist, better controlnet, and basiclly better everything, and I give them away for free. I'm not going to charge you $30 for a crappy workflow with a renamed Clownshark sampler, a horribly vibecoded detailer, and a bad Lut. I give you an almost infintite amount of adjustments and leave you notes on how to do it. If here is a feature you want, I'll add it. If you have questions, I answer them. I also have several other realism workflows. Just don't waste your money.All of the images I posted to V7 were zero LoRAs and basic settings. You can make any adjustment you want. I give you the tools to do it. Note: If you have ANY issues with nodes not downloading, read the notes or reach out. There's nothing that special about any of them that aren't core modules.⛔⚠️🛑✋ Read the notes completly before using. Most common install and node problems are listed in the directions也有中文说明This is an advanced workflow. I encourage you to take the time to learn it. If you feel overwhelmed, you can start here:https://civitai.com/models/2348914/zimage-base-and-turbo-essentials-workflowA 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-breakdownAll in 1 workflow:ControlnetCheckpoint option for merged modelsI2IPrompt generationPrompt enhancementLora LoaderFlowmatch Sampler (Rez4Lyf)Draft mode for making corrections without having to rerun the workflow.Seed variance enhancerUltimate Upscaler for prescaling and hi res fixDetailing Suite for face, hands, eyes, and expressionsNSFW detailersSeed VR2 Upscaler to get 4k on low VRAMMassive Post Production SuiteSmart Noise scrubberFull tutorial available.也有中文说明Check out my other models:My NSFW version with Anantomy Correctors:https://civitai.com/models/2270894/zit-nsfw-anatomy-correctorFor ultimate Anime detail and control: https://civitai.com/models/2220766/zimage-ultra-anime-detail-workflow-get-the-most-out-of-your-generationsFor Realism and variety over the base model: https://civitai.com/models/2231181/z-image-ultra-real-workflowInstagram: https://www.instagram.com/synth.studio.models/Buy me a☕ https://ko-fi.com/lonecatoneThis represents hundreds of hours of work. If you enjoy it, please 👍like, 💬 comment , and feel free to ⚡tip 😉
Moody i2i Image2Image ZIT (Zimage Turbo) Simple Workflow (LORA洗图流)
I just made a new Telegram Group and Discord group drop in for chats, questions, feedbacks or just to share your work. 欢迎加入纸飞机和Discord群组。喜欢我的作品吗？🎨 我会持续免费分享更多优质内容✨，欢迎支持一下，请我喝杯咖啡吧！💖 Like my work? 🎨 I share all my work for free ✨, so support me by buying me a coffee 💖!Ko-fi: ko-fi.com/catlover1937Fanvue: https://www.fanvue.com/catlover1937Please do not redistribute without consent. 如非作者同意禁止搬运。RH体验地址：https://www.runninghub.ai/post/2059781120304173057/?inviteCode=cptpovgo2025-05-27🚀 Hello great people!Image-to-Image Workflow V3.0 is finally here 🎉Before we dive into the updates, I wanna quickly explain what this workflow is actually made for 👀This workflow is for lazy people like me 😎You find a cute / beautiful image on Instagram, Xiaohongshu, Pinterest, or wherever you browse... (not judging if you're scrolling some XXX sites at work 💀) and want to recreate the same vibe with your own AI girl.That’s exactly what this workflow is built for ✨Just slot in your favorite character LoRA, hit generate, and boom — hands-free & prompt-free generation 🔥💖 A list of Moody Gals (character LoRAs) can be found through the collection link (more are coming).━━━ V3.0 Updates ━━━⚙️ Adjusted some preset stats✨ Minor optimizations for newer Moody Releases models🛠️ Added automatic detailer module🎭 Added manual masked detailer module🌊 Re-organized the workflow layout into a cleaner waterfall-style flow⚡ Added more QoL improvements inspired by my latest workflowsEnjoy & cheers 🍻🚀 大家好！Image-to-Image 工作流 V3.0 终于来了 🎉在介绍更新内容之前，我想先重新说明一下这个工作流到底是干嘛用的 👀这个工作流是专门为像我这种懒人准备的 😎当你在 Instagram、小红书、Pinterest，或者其他你平时逛的网站上（上班偷偷刷 XXX 网站我也不评价 💀）看到一张很好看的图，想用自己的 AI 女友 / AI 角色重新生成同样风格的时候——这个工作流就是为此而生的 ✨你只需要放入自己喜欢的角色 LoRA，然后点击生成，剩下全部自动完成 🔥无需写 Prompt，真正的 hands-free & prompt-free 体验。💖 Moody Gals（角色 LoRA）合集可以通过这里 collection link 找到，届时会有更多的人物Lora更新哦。━━━ V3.0 更新内容 ━━━⚙️ 调整了一些预设参数✨ 针对新版 Moody Releases 模型进行了小幅优化🛠️ 新增自动细节修复模块🎭 新增手动蒙版细节修复模块🌊 重新整理了工作流布局，让整体结构更加瀑布流化、更直观⚡ 加入了更多来自我最新工作流的 QoL 改进祝大家玩得开心，干杯 🍻2025-03-27Whats new in V2?New upscaler used for further attempt to eliminate the side effect of SD upscaleAdd pre-processor section to adjust image contrast and brightness, which will carry forward to the end resultImprove on auto prompt detectionsQoL updates as my other newer workflowsV2 有什么新内容？使用了新的放大器（SD upscaler），进一步尝试消除 SD 放大带来的副作用 （莫名的水渍，鸡皮疙瘩，油光）新增预处理（pre-processor）模块，可调整图像的对比度和亮度，并会影响最终输出结果优化了自动提示词检测功能加入了一些使用体验上的优化（QoL），整体流程更加顺手，并与我其他较新的工作流保持一致2025-03-05Tired of unreliable ControlNet options for Zimage or just tired of coming up with prompts?This workflow lets you re-generate your input image while preserving the original settings, pose, and likeness—without carrying over any real identity issues. It is compatible with all of my ZIT and ZID models. (Moody Porn V10 was used during testing and is recommended.) Prompts will be generated on the fly using Florence2. You can just input an image and it re-generates the image in hands-free (brains-free too) mode.厌倦了为 Zimage 寻找好用的 ControlNet 或者有洗图的烦恼吗？这个工作流可以重新生成你的输入图像，同时保留原始的结构、姿势和整体相似度，并且不会带来任何真实身份相关的问题。它兼容我所有的 ZIT 和 ZID 模型。（测试时使用的是 Moody Porn V10。）提示词会在工作流中通过Florence2 自动生成，做到全自动用脚都能喂图自我生成。Avaible ZIT Models / 可选 Turbo 模型：https://civitai.com/models/620406/moody-porn-mixhttps://civitai.com/models/621441/moody-real-mixhttps://civitai.com/models/2384856/moody-wild-mix?modelVersionId=2681752
ZImageTurboPiD Upscale Decode Workflow
ZImageTurboPiDv2+Color transfer from lora image as PiD tends to raise the gamma+Optional SageAttention+ModelSampling shift to PiD modelUpdate your ComfyUI for PixelDiT supportFor models/text encoder https://huggingface.co/Comfy-Org/PixelDiTINT8 Models https://civitai.red/models/2662932/pid-int8PiD reformulates the latent-to-pixel decoder as a conditional pixel-space diffusion model, unifying decoding and upsampling into a single generative module. It denoises directly in high-resolution pixel space and produces a super-resolved image in one pass. This repository hosts the released decoder checkpoints, plus the encoder/decoder ("VAE") weights they depend on.This workflow takes a ZImageTurbo/ZImage generation latent from 512x512>2048x2048/1024x1024>4096x4096 or 16:9, 4:3, 3:4 aspect ratios.RTX 3060 12gb + 32gb ram60 seconds upscale+decode for 1024 model 768x1024>3072x40968 seconds upscale+decode for 512 model 432x576>1728x2304
Z Image (Turbo & Base) Workflow
ComfyUI Workflow for Z ImageAny feedback would be appreciated.📂 Required ModelsDiffusion Model: z_image_turbo_bf16.safetensors .../ComfyUI/models/diffusion_models/https://huggingface.co/Comfy-Org/z_image_turbo/tree/main/split_files/diffusion_modelsGGUF Variants:https://huggingface.co/jayn7/Z-Image-Turbo-GGUF/tree/mainBase model:https://huggingface.co/Comfy-Org/z_image/tree/main/split_files/diffusion_modelsText Encoder: qwen_3_4b.safetensors .../ComfyUI/models/text_encoders/https://huggingface.co/Comfy-Org/z_image_turbo/tree/main/split_files/text_encodersGGUF Variants:https://huggingface.co/Qwen/Qwen3-4B-GGUF/tree/mainVAE: ae.safetensors .../ComfyUI/models/vae/https://huggingface.co/Comfy-Org/z_image_turbo/tree/main/split_files/vae📂 Optional ModelsUpscale Model: 4x_foolhardy_Remacri.pth .../ComfyUI/models/upscale_modelshttps://huggingface.co/FacehugmanIII/4x_foolhardy_RemacriDetection Model: face_yolov9c.pt .../ComfyUI/models/ultralytics/bboxhttps://huggingface.co/Bingsu/adetailer/blob/main/face_yolov9c.ptSAM Model: sam_vit_b_01ec64.pth .../ComfyUI/models/samshttps://github.com/facebookresearch/segment-anything#model-checkpointsFun-Controlnet-Union: Z-Image-Turbo-Fun-Controlnet-Union.safetensors .../ComfyUI/models/model_patcheshttps://huggingface.co/alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union/tree/mainZ Image Fun-Controlnet-Union requires the ComfyUI nightly version as of 3rd Dec. 2025🧩 Required Custom Nodes🟩 ComfyUI-Manager (by Comfy-Org)https://github.com/Comfy-Org/ComfyUI-Manager🟩 rgthree-comfy (by rgthree)https://github.com/rgthree/rgthree-comfy🟩 ComfyUI-Easy-Use (by yolain)https://github.com/yolain/ComfyUI-Easy-Use🟩 ComfyUI-Impact-Pack (by ltdrdata)https://github.com/ltdrdata/ComfyUI-Impact-Pack🟩 ComfyUI-Impact-Subpack (by ltdrdata)https://github.com/ltdrdata/ComfyUI-Impact-Subpack🟪 ComfyUI-Image-Saver (by alexopus)https://github.com/alexopus/ComfyUI-Image-Saver🟪 ComfyUI-Lora-Manager (by willmiao)https://github.com/willmiao/ComfyUI-Lora-Manager🟪 ComfyUI-KJNodes (by kijai)https://github.com/kijai/ComfyUI-KJNodes🟪 ComfyUI_essentials (by cubiq)https://github.com/cubiq/ComfyUI_essentials⚙️ Recommended SettingsTurbo Steps: 6 to 10 (Default set to 9).Turbo CFG: 1.0 (Default set to 1.0).Base Steps: 25 to 50Base CFG: 3.0 to 6.0Sampler: euler or res_multistepScheduler: simple[1:1] 1024x1024 [3:4] 896x1152 [5:8] 832x1216 [9:16] 768x1344 [9:21] 640x1536Experiment and enjoy!
Z-Image Base & Turbo Workflow I2I/T2I (Low or High VRAM)
Note: If you have ANY issues with nodes not downloading, read the notes or reach out. There's nothing that special about any of them that aren't core modules. Most issues can be resolved by updating everything⛔⚠️🛑✋ Read the notes completly before using. Most common install and node problems are listed in the directionsInstagram: https://www.instagram.com/synth.studio.models/Buy me a☕ https://ko-fi.com/lonecatoneThis represents hundreds of hours of work. If you enjoy it, please 👍like, 💬 comment , and feel free to ⚡tip 😉This is an advanced workflow designed for those who want pro grade results. It is node heavy and has numerous options. I encourage you to take the time to learn it. If you feel overwhelmed, you can start with my simple model HERE也有中文说明A complete user and troubleshooting guide inclusing a list of common error messages can be found here https://civitai.com/articles/24678/how-to-use-my-workflows-a-comprehensive-breakdownV13.0overhauled the Controlnet:Now you can select the aspect ratio of the original image without having to "guess" it with the aspect ratio nodeLeft room for the resize node so you can choose which side to crop the image byRedid the dependencies so that every tie you shut it on and off, you don;t need to reset the switches so it will not give you an error 😡Massively improved the notes throughoutMinor layout tweaksV11.3Detailers are now able to handle float32 models, so I rewmoved the SDXL Based checkpoint.Please give me feedback with any issues on this, especailly RAM related.Added a crop section so you can...well... crop images easilyUpdated several areasAdded the ability to use a simple date with the metadata folderadded a "remove banned tags" nodeV8.0Removed Florence Prompt assist and replaced it with Qwen Advanced.Reorganized the controlnet section because it bugged me. Now it's easier to use.Added a LoRa controller for the detailer for better character facial and eye consistency as well as a localized promptGutted the backend of the post production suite and added 📷optical realism, giving a significantly better way to fine tune your images. Fixed some backend mathAdded a 📂 subfolder generator that seperates the metadata files according to the name you give the file for easier use.V7.0Added KSampler Configure Nodes so you can leave the setting in place for both ZIT and base without having to change themRemoved some problematic nodesRemoved Qwen Prompt enhancer. I refuse to put an API one in there.global scheduler and samplers for consistencyBetter mathImproved notes throughout.All in 1 workflow:ControlnetCheckpoint option for merged modelsI2IPrompt generationPrompt enhancementLora LoaderAdvanced KSamplerDraft mode for making corrections without having to rerun the workflow.Seed variance enhancerUltimate Upscaler for prescaling and hi res fixDetailing Suite for face, hands, eyes, and expressionsSeed VR2 Upscaler to get 4k on low VRAMMassive Post Production SuiteJPEG ScrubberSmart Noise scrubberFull tutorial available.也有中文说明Check out my other models:My NSFW version with Anantomy Correctors:https://civitai.com/models/2270894/zit-nsfw-anatomy-correctorFor ultimate Anime detail and control: https://civitai.com/models/2220766/zimage-ultra-anime-detail-workflow-get-the-most-out-of-your-generationsFor Realism and variety over the base model: https://civitai.com/models/2231181/z-image-ultra-real-workflow
Z-Image-L2P/Z-Image-L2P-INT8
23/05/2026- For current ComfyUI support https://github.com/Comfy-Org/ComfyUI/pull/14055Go to ComfyUI_windows_portable/ComfyUIClick on the address bar at the top of the File Explorer window, type cmd, and press enter.In cmd type:git fetch origin pull/14055/head:pr-14055git checkout pr-14055Go back to ComfyUI_windows_portable folder and open a new cmd and type.\python_embeded\python.exe -m pip install -r .\ComfyUI\requirements.txtZ-Image-L2P https://huggingface.co/zhen-nan/L2P/tree/mainPlace the model in /models/checkpointsInt8Convrot quantization of the Z-Image-L2P modelLoad the model with https://github.com/BobJohnson24/ComfyUI-INT8-FastPlace the model in /models/diffusion_models
Multi Prompt Batch Generation
This ComfyUI workflow is designed to execute batch image generation efficiently through an automated prompt cycling system. It is highly optimized for resource-constrained hardware environments, ensuring stable performance on low-VRAM configurations.Key Features:- Automated multi-prompt generation using sequential text cycling.- Integrated face and hand detailing to enhance final output quality.- High-speed upscaling module for crisp, high-resolution rendering.
Moody Simple Zimage Turbo/Distilled Workflow
I just made a new Telegram Group and Discord group drop in for chats, questions, feedbacks or just to share your work. 欢迎加入纸飞机和Discord群组。RH 体验地址：https://www.runninghub.ai/post/2057506560754143234/?inviteCode=cptpovgo喜欢我的作品吗？🎨 我会持续免费分享更多优质内容✨，欢迎支持一下，请我喝杯咖啡吧！💖 Like my work? 🎨 I share all my work for free ✨, so support me by buying me a coffee 💖!Ko-fi: ko-fi.com/catlover1937Fanvue: https://www.fanvue.com/catlover1937Please do not redistribute without consent. 如非作者同意禁止搬运。2026-05-20Hey lads & gents 🍻V7.0 just dropped 🚀✨ Minor optimizations for the newer Moody Releases models🛠️ Added automatic detailer module🎭 Added manual masked detailer module🌊 Re-organized the workflow layout to feel more like a waterfall flow⚡ Plus a bunch of QoL improvements inspired by my latest workflowsEnjoy & cheers ❤️🇨🇳兄弟们来了 🍻V7.0 现已发布 🚀✨ 针对新版 Moody Releases 模型进行了小幅优化🛠️ 新增自动细节修复模块🎭 新增手动蒙版细节修复模块🌊 重新整理了工作流布局，让整体流程更像瀑布式结构⚡ 同时加入了不少我最新工作流里的 QoL 改进玩得开心，干杯 ❤️2026-03-31V6.0 ZIT Simple Dual Ksampler Enhanced WITH SDA workflow.DEMO images were generated by moody porn v10 model.预览图生成于 moody porn v10 model.With the help of SDA, the new workflow works almost as well as my ZIB+ZIT workflow just with minor drawbacks of bad poses and body anatomy. This is still a very good solution for the “Diversity Collapse” problem.Z-Image-Turbo-SDA is a highly efficient LoKr (Low-Rank Kronecker Product) adapter designed to rescue the “Diversity Collapse” problem in few-step distilled Flow Matching / Diffusion models. However, it also increases the chance of bad body anatomy.SDA Lora support added in 1ST samplerNew upscaler used for further attempt to eliminate the side effect of SD upscaleQoL updates as my other newer workflows在 SDA 的帮助下，这个新的工作流表现几乎和我的 ZIB + ZIT 工作流一样好，只是存在一些小缺点，比如姿势不自然和人体结构不准确。不过，这仍然是解决“多样性崩塌（Diversity Collapse）”问题的一个非常有效的方案。用于解决少步数蒸馏 Flow Matching / Diffusion 模型中的“多样性崩塌”问题的高效 LoKr 适配器，但也会在一定程度上增加人体结构异常的概率。增加了对SDA LORA 的支持使用了新的放大器（SD upscaler），进一步尝试消除 SD 放大带来的副作用 （莫名的水渍，鸡皮疙瘩，油光）加入了一些使用体验上的优化（QoL），整体流程更加顺手，并与我其他较新的工作流保持一致2026-03-20V5.0 ZIT Dual Ksampler Enhanced Simple workflowV5.0 is here. Some optimization and QoL update for the Moody ZIT and ZID models better support. Added an attempt to fix some anomolies from SD upscale node.V5.0 版本已发布。针对最新的moody模型 进行了一些优化和体验改进。尝试修复SD upscale奇怪纹路的问题。2026-02-25Dual K-Sampler setup with minor quality of life updates with support of my ZIT/ZID (Zimage distilled) models.Availble ZIT Models / 可选 Turbo 模型：https://civitai.com/models/620406/moody-porn-mixhttps://civitai.com/models/621441/moody-real-mixhttps://civitai.com/models/2384856/moody-wild-mix?modelVersionId=26817522026-01-20Add separate seeds for each Ksampler so you don't have to re-run the whole flow from beginning if you just want minor changesUse a different upscalers for better performance and qualityIncluded the facedetailer flow by defaultAdded optional SeedVR for upscaling to 4kAdded optional skin contrast filter2026-01-13Updated with 2 Ksamplers for better preview and optimization.2026-01-01Added a face detailer workflow as requested. Since some users have noted that character LoRAs are not fully compatible, my recommended workaround is to first describe the character using a text prompt, then apply the face detailer with your character LoRA to effectively perform a “face swap.”--As requested, here is a simplified, minimal workflow derived from my setup for the most basic Z-image AIGC tasks.
Wikked ZIT
Simple workflow with multistep upscale
diffusion_models+Upskale+facedetailer
easy to use
CyberRealistic Z-Image Turbo - ComfyUI Workflow
You can get this model, along with many others, by joining The Tinkerer on Whop. It’s one monthly membership, and you’ll also get early releases, private tools, and a bunch of extra stuff.👉 Join on WhopA clean and practical CyberRealistic Z-Image Turbo workflow for ComfyUI, built for strong results without making things unnecessarily complicated. The workflow is designed to stay user-friendly while still offering useful extras such as upscale options, cleaner organization, and improved saved-image metadata.CyberRealistic Z-Image Turbo - ComfyUI Workflow V5What's new in v5Cyberdelia Z-Engineer node — built-in LLM prompt engineering. Type your concept in plain language and a local LLM (LM Studio, Ollama, etc.) expands it into a definitive 200-250 word Z-Image-tuned prompt before it hits the sampler. CLIP encoding is built into the node, so no separate text-encode step is needed.Toggle the node between engineered (LLM) and passthrough (raw) on the fly. If you don't have an LLM endpoint running, set it to passthrough and the workflow runs as a normal text-to-image flow.Setup:1. Install [LM Studio](https://lmstudio.ai/) (or any OpenAI-compatible server)2. Load any 4B+ instruction-following model — Qwen3-4B is the recommended baseline3. Start the local server, defaults to http://localhost:1234/v14. Adjust the node's api_url and model fields if your setup differsGet the node via ComfyUI-Manager (search Cyberdelia) or from https://github.com/cyberdeliaAI/comfyui-cyberdelia-z-engineer.Also: the metadata save node has been rebranded — the workflow now uses https://github.com/cyberdeliaAI/comfyui-cyberdelia-metadata (renamed from revived_comfyui_image_metadata_extension).Functionally the same, just under the Cyberdelia AI Lab brand.Future updates and beta versions are available via the public Whop page:https://whop.com/cyberdelia-ai-lab/comfy-ui-workflows-zit-C80otuTsikeE8d/app/
Like a Boss workflow
My own workflow developed to generate Z-images (but it should work with any other type, as long as change the models and make minor adjustments).my github nodes: shirosaki33/shirotools: many utility tools made for different situationsI particularly hate that scattered clutter of nodes where it's extremely difficult to locate where everything goes and ughh.I ended up simplifying its use considerably with several functions along the way to the final result. You start by choosing the method: T2i, I2i, and hybrid with ControlNet.The workflow includes an InPaint, Adetailer, SeedVR2 Upscaler, and normal upscaler system.It's up to the user which steps will be used along the way, in the order I've listed them. The InPaint is divided into at least 3 areas, in case the user wants to change at least 3 different things in the image.If you want more areas, you just need to clone an additional section and connect the points correctly.It's not necessary to use all 3; you can use just 1 if you prefer.Inpaint contains an additional internal checkpoint system in this section, since the turbo checkpoint isn't very suitable for this. Loading z-base yielded a much more satisfactory result in this part.The seed has universal control. So if the user leaves it "fixed," all processes will remain until the user makes the desired adjustments in specific areas, without having to redo the entire process.Didn't like the result of the adetailer? Just disable what you didn't like and redo it; it will only redo the adetailer section, for example.3 KSampler systems to suit everyone's taste.A simple system that every workflow typically uses with 1 KSampler.The other 2 are in a 2 KSampler system, each already configured for turbo mode and base mode.It's up to the user to choose what they prefer.I hope it's satisfactory at least. this simplifies my personal use, without all those lines and boxes everywhere. A version with more features might come later :)
HiDream-O1 Dev 2604 + Z-Image-Turbo (Refiner)
HiDream Meets ZITA dual pass workflow for HiDream-O1 Image Dev 2604 with a refinement pass of Z-Image-TurboREQUIRES:REBELS HiDream-O1 Custom Node Sethttps://github.com/RealRebelAI/Rebels_HiDream-01_Image_Dev_NODESHiDream-O1 Image Dev 2604 (GGUF)https://civitai.com/models/2611889/rebels-hidream-01-image-dev-dev-2604Z-Image-Turbo (bf16)https://civitai.com/models/2169770/rebels-z-image-turbo-fp8bf16Runs 38 steps on hidream and then the second pass of zit runs 4 steps at a denoise of 0.35 to refine details missing in the dev model and clean up artifacting.
Zimage Turbo Workflow: Low-Step Preview, Fast Batch Picker & Fast Precision Resampling
### 🚀 OverviewAn optimized ComfyUI workflow for ZImage Turbo featuring a 2-stage generation process. Preview 5 variations in low steps, pick your favorite, and enhance it with high-quality resampling.### ✨ Key Features* Low-Step Preview: Generates 5 image variants simultaneously at ultra-low steps.* Fast Batch Picker: Custom routing module to select your favorite image index instantly.* Precision Resampling: Secondary pass to add rich details and sharp quality.* ControlNet Ready: Full composition/pose control without sacrificing Turbo speed.* Quick-Keys: Pre-configured keyboard shortcuts for a seamless UI experience.### ⚠️ Important Notes The *Quick-Key Module in the 'Quick start' node group** strictly relies on the rgthree-comfy custom nodes to function.* Due to a known bug in the node author's current source code, you must manually patch it. Please follow these steps to make the shortcuts work properly: 1. Locate the target folder: ..\custom_nodes\rgthree-comfy\web\comfyui\ 2. Replace the following two JS files in that folder with the ones provided: fast_actions_button.js and fast_groups_muter.js 3. Restart your ComfyUI server.
Donut-Z Image Turbo-workflow (upscale+face detailer)
download model and install tutorial: https://comfyanonymous.github.io/ComfyUI_examples/z_image/
basic Z Image Turbo Workflow + 3 Lora
A very simple Workflow for beginners. Simple, fast and easy to use.
ZImageCompact
Compact workflow, Zit+Zib+ControlNet+Upscale
Z-Image-i2L (Image to LoRA) Fast LoRA Training + ControlNet Testing + Upscale Workflow
This workflow is an expanded Z-Image-i2L production pipeline that combines fast Image-to-LoRA generation, ControlNet structure testing, and high-resolution tiled upscaling into one complete ComfyUI graph. It is designed for creators who do not only want to generate a quick LoRA from reference images, but also want to immediately test that LoRA under real production conditions and then push the result into a more polished final output.The first stage focuses on fast LoRA creation. Multiple reference images are loaded and combined into a training image batch, then passed into the RunningHub Z-Image-i2L system. This allows the workflow to generate a lightweight Z-Image LoRA from a small group of images without requiring a traditional local training setup, dataset folder preparation, caption files, or command-line configuration. It is especially useful for quickly capturing a character identity, fashion style, product look, object concept, creature design, or consistent visual aesthetic.After the LoRA is generated, the workflow immediately saves it and loads it back into Z-Image Base for testing. This is the key advantage of the pipeline: training and validation happen in the same graph. The user can quickly see whether the generated LoRA actually affects the output, whether it preserves the target identity, whether it introduces artifacts, and whether the strength needs to be adjusted. This makes the workflow much more practical than a training-only setup.The second stage adds ControlNet testing. A structure reference image is processed through DepthAnythingV2Preprocessor to create a depth map, then applied through Z-Image Fun ControlNet Union. This lets the newly generated LoRA be tested under controlled composition, depth, layout, and spatial guidance. A LoRA may look fine in a basic text-to-image test, but fail when the camera angle or scene structure becomes more demanding. This workflow helps reveal that immediately.The generation section uses Z-Image Base with qwen_3_4b text encoding, AE VAE, ControlNet guidance, SplitSigmas, DetailDaemonSamplerNode, CFGGuider, and SamplerCustomAdvanced. This gives the workflow a more controlled two-stage sampling structure, where the early phase builds the main layout and the later phase refines the image. It is useful for evaluating prompt compatibility, LoRA strength, structural stability, and final image coherence.The third stage is high-resolution enhancement. After the controlled LoRA test image is generated, the workflow sends the result into an upscale pipeline. It uses a traditional upscale model, then scales the image toward a target megapixel size. The image is split into tiles with TTP tile tools, each tile can be captioned with Florence2, refined through latent processing, decoded with tiled VAE decoding, and finally reconstructed into a complete high-resolution image. This makes the workflow suitable not only for LoRA testing, but also for producing sharper Civitai showcase images, RunningHub examples, thumbnails, and final publishing assets.In short, this is not just an Image-to-LoRA workflow. It is a full LoRA creation, ControlNet validation, and upscale finishing pipeline. If you want to see how the full node structure works, how the generated LoRA is tested with ControlNet, and how the final upscale stage improves the output, watch the full video tutorial from the YouTube link above.⚙️ Try the Workflow Online👉 Workflow: https://www.runninghub.ai/post/2023308180264067074/?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/BV1qXZMBwEC7/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/2023308180264067074/?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/BV1qXZMBwEC7/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。
Z-Image-i2L Image to LoRA Fast Training & Testing Workflow
This ComfyUI workflow is designed for Z-Image-i2L, also known as Image to LoRA. The main purpose of this workflow is to let creators quickly train a lightweight LoRA from a small group of reference images, save the generated LoRA, and immediately test it inside the same ComfyUI graph with Z-Image Base.Unlike a traditional LoRA training workflow that requires dataset preparation, caption files, training scripts, optimizer settings, command-line configuration, and manual model loading, this workflow is designed as a fast and practical image-to-LoRA pipeline. The user only needs to provide several reference images, run the i2L generation node, save the LoRA, and then test the newly generated LoRA through a normal Z-Image generation route.The workflow is built around the RunningHub Z-Image-i2L node system. It uses RunningHub_ZImageI2L_Loader to load the Image-to-LoRA pipeline, RunningHub_ZImageI2L_LoraGenerator to generate a LoRA from the uploaded training images, and RunningHub_ZImageI2L_Saver to save the generated LoRA file. This makes the workflow much more convenient for creators who want to quickly capture a character style, object style, visual identity, costume concept, creature design, or artistic direction from a few images.The training input section uses multiple LoadImage nodes and ImageBatchMulti. In the uploaded setup, the workflow accepts six image inputs and combines them into one image batch. These images become the training references for the Image-to-LoRA generator. This is useful because a single image may not be enough to define a stable visual concept. Multiple images help the i2L pipeline understand the repeated features across the references, such as face shape, clothing style, color theme, character identity, object design, or general aesthetic.The ImageBatchMulti node is important because it merges the reference images into one training batch. The workflow is configured with an input count of 6, which means users can provide a small set of images without preparing a full dataset folder manually. This is suitable for quick creator testing, lightweight character adaptation, concept extraction, and rapid LoRA prototyping.The i2L generation stage uses RunningHub_ZImageI2L_LoraGenerator. This node receives the loaded ZImageI2LPipeline and the batched training images, then generates a LoRA name and LoRA path. In the included setup, the seed is fixed, which helps make the LoRA generation process more reproducible. Users can change the seed if they want a different training result or want to test variation between generated LoRAs.After the LoRA is generated, the workflow uses RunningHub_ZImageI2L_Saver to save the result. The filename prefix is set to zimage_lora, making it easy to identify the generated file. The LoRA path is also previewed through PreviewAny, so users can confirm that the LoRA has been created and passed forward correctly.The workflow also includes easy cleanGpuUsed between the i2L pipeline and LoRA generation output. This is useful because training-like nodes and generation nodes can consume GPU memory. Cleaning GPU memory after the i2L stage helps make the later test generation stage more stable, especially in online or cloud environments.The second half of the workflow is the testing stage. After the LoRA is generated, it is loaded into the Z-Image generation route through LoraLoader. This means users do not need to manually move the LoRA file, restart ComfyUI, or create a separate test workflow. The LoRA can be trained and tested in one continuous graph.The testing route uses Z-Image Base with z_image_bf16.safetensors as the main diffusion model, qwen_3_4b.safetensors as the text encoder, and ae.safetensors as the VAE. The generated LoRA is applied through LoraLoader with model strength around 1.1 and clip strength around 1.0. These values are useful for testing whether the LoRA strongly affects the output without completely overwhelming the base model.The prompt test section uses CLIPTextEncode with a simple Chinese prompt: “一个武士鬼怪正在与恶鬼战斗,” meaning a samurai ghost or monster fighting an evil demon. This kind of prompt is useful for testing whether the generated LoRA can influence character identity, creature design, costume style, visual tone, or image composition in a new generation.The negative prompt suppresses common visual problems such as yellowed output, green tint, blur, low resolution, low quality, distorted limbs, eerie appearance, ugly results, AI-looking artifacts, noise, grid-like artifacts, JPEG compression artifacts, abnormal limbs, watermark, garbled text, and meaningless characters. This is practical for LoRA testing because newly generated LoRAs can sometimes introduce artifacts, overfitting, or unstable details if the reference images are inconsistent.The generation canvas is created with EmptySD3LatentImage at 1024 x 1536. This vertical format is suitable for character testing, portrait-style LoRA previews, concept art outputs, Civitai examples, RunningHub demos, and social media cover-style images. Users can adjust the size depending on whether they want portrait, square, or landscape results.The sampling stage uses KSampler with 50 steps, CFG 4, res_multistep sampler, simple scheduler, and full denoise. This is a relatively strong test route. Because the purpose is to evaluate a newly generated LoRA, using a higher step count can help reveal whether the LoRA is stable, expressive, and compatible with the base model. The output is decoded through VAEDecode and saved through SaveImage.This workflow is especially useful for rapid LoRA prototyping. Instead of spending a long time building a formal dataset, users can upload a small group of images and quickly generate a LoRA for testing. This is helpful for AI creators who want to test character concepts, style references, object concepts, mascot designs, fantasy creatures, costume sets, or visual branding ideas.It is also useful for RunningHub online workflow publishing. Many users want to experience LoRA creation without installing a full local training environment. This workflow lowers the barrier by packaging the training and testing process into one graph. Users can upload reference images, generate a LoRA, and immediately see whether it works.Main features:- Z-Image-i2L Image to LoRA workflow- Fast LoRA generation from reference images- Supports multiple training image inputs- Uses ImageBatchMulti to combine six reference images- RunningHub_ZImageI2L_Loader pipeline loading- RunningHub_ZImageI2L_LoraGenerator LoRA creation- RunningHub_ZImageI2L_Saver LoRA export- PreviewAny output for generated LoRA path checking- GPU cleanup support through easy cleanGpuUsed- Immediate LoRA testing inside the same workflow- Z-Image Base test generation route- Uses z_image_bf16.safetensors- Uses qwen_3_4b.safetensors text encoder- Uses ae.safetensors VAE- LoraLoader test route with adjustable model and clip strength- 1024 x 1536 vertical test canvas- KSampler test generation with res_multistepRecommended use cases:Fast LoRA training, Image to LoRA testing, character LoRA prototyping, style LoRA creation, object concept extraction, fantasy character adaptation, creature design testing, costume style capture, visual identity transfer, Civitai LoRA preview creation, RunningHub online LoRA workflow publishing, prompt testing with newly generated LoRA, and lightweight AI model customization.Suggested workflow:Start by preparing several reference images. Use images that share the same concept. If you are training a character LoRA, the images should show the same character or very similar identity. If you are training a style LoRA, the images should share the same visual style. If the images are too different, the generated LoRA may become unstable or unclear.Use clean images whenever possible. Avoid heavy watermarks, text overlays, excessive compression, very low resolution, strong motion blur, and unrelated background clutter. The i2L pipeline can work quickly, but the quality of the input images still matters. Better references usually create a more useful LoRA.Load the reference images into the six LoadImage nodes. The workflow batches them through ImageBatchMulti. You can use fewer or more references if the workflow is adjusted, but the uploaded version is structured around six input images. Six images are enough for a quick test while still giving the LoRA generator more information than a single reference.Run the RunningHub_ZImageI2L_LoraGenerator section to generate the LoRA. Keep the seed fixed if you want repeatable output. Change the seed if you want to test different LoRA generation results from the same reference images.Save the LoRA through RunningHub_ZImageI2L_Saver. The saved file uses the prefix zimage_lora. Check the PreviewAny output if you want to confirm the LoRA path. This helps verify that the generated LoRA is properly passed into the testing section.Use the testing section immediately after LoRA generation. The LoraLoader loads the generated LoRA into Z-Image Base. Start with moderate LoRA strength. If the LoRA effect is too weak, increase the model strength slightly. If the output becomes distorted or overfitted, reduce the strength.Write a prompt that tests the LoRA clearly. If the LoRA is meant to capture a character, write a prompt that asks for that character in a new scene. If it is meant to capture a style, use a subject that lets the style appear clearly. If it is meant to capture an object, describe that object directly and test whether the generated image keeps the reference features.Use the negative prompt to suppress LoRA artifacts. Newly generated LoRAs may create color cast, grid artifacts, strange limbs, low-quality details, or unwanted text if the input images contain those issues. Add targeted negative terms if specific problems appear.Use the 1024 x 1536 vertical canvas for character testing. This format is good for preview images and Civitai examples. For object or landscape testing, adjust the latent size to match the target composition.Run several test generations. A single image is not enough to judge a LoRA. Test different seeds and prompts. A good LoRA should influence the output consistently without destroying anatomy, composition, or general image quality.If the LoRA does not capture the concept well, improve the reference set. Use cleaner images, more consistent subject views, better lighting, and fewer unrelated background elements. The fastest way to improve an i2L result is usually to improve the input references.This workflow is designed for creators who want a fast, practical, and online-friendly Z-Image LoRA creation pipeline. It combines image batching, automatic LoRA generation, LoRA saving, GPU cleanup, LoRA loading, and Z-Image Base test generation into one graph. It is especially useful for quickly turning reference images into a usable LoRA prototype and testing the result without leaving ComfyUI.🎥 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/wT9ob7rFONMBefore 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/2023308170269036545/?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/BV1qXZMBwEC7/☕ 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/wT9ob7rFONM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/2023308170269036545/?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/BV1qXZMBwEC7/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。
Z-Image Base + Turbo Segmented Rendering Workflow
This ComfyUI workflow is designed for segmented rendering with Z-Image Base and Z-Image Turbo. Instead of using only one model for the entire generation process, this workflow splits the sampling process into different stages and lets Z-Image Base and Z-Image Turbo handle different parts of the render. The goal is to combine the stronger global structure and composition ability of Z-Image Base with the faster, sharper, and more detail-oriented finishing behavior of Z-Image Turbo.The core idea is simple: use Z-Image Base to build the main image foundation during the earlier high-noise stage, then hand the latent result to Z-Image Turbo for the later low-noise stage. This makes the workflow useful when a single-model workflow is not stable enough, or when Turbo alone is fast but not always strong enough for complex composition, and Base alone is more stable but slower or less efficient for final iteration. By separating the render into stages, the workflow gives creators more control over composition, detail, speed, and final polish.The workflow is built around two Z-Image models. The first model route uses z_image_bf16.safetensors as the Base model. This route is responsible for the main structure, subject placement, scene logic, atmosphere, and broad visual composition. The second model route uses z_image_turbo_bf16.safetensors as the Turbo model. This route is used for continuation, refinement, and detail strengthening after the Base model has already established the image direction.The workflow uses qwen_3_4b.safetensors as the text encoder and ae.safetensors as the VAE. The prompt is encoded through CLIPTextEncode, then passed into CFGGuider. The sampling process is handled through RandomNoise, BasicScheduler, SplitSigmas, DetailDaemonSamplerNode, SamplerEulerAncestral, and SamplerCustomAdvanced. This structure gives the workflow a more technical and controllable sampling chain than a normal KSampler-only setup.A key part of this workflow is SplitSigmas. The workflow generates a sigma schedule, then splits it into a high-sigma section and a low-sigma section. The high-sigma section represents the earlier generation stage, where the model is still deciding major image structure and composition. The low-sigma section represents the later refinement stage, where the image is already formed and the model mainly improves detail, texture, edge quality, lighting, and surface finish.In this workflow, Z-Image Base handles the earlier stage. This is useful because the Base model is better suited for establishing the image’s overall logic. It can help create more stable subject placement, stronger scene coherence, better spatial layout, and a more complete starting image. For complex prompts, surreal scenes, character-based illustrations, fashion photography concepts, product-like compositions, or cinematic visuals, this Base stage acts like the foundation pass.After the Base stage, the workflow passes the denoised latent into the Turbo stage. Z-Image Turbo then works on the later stage using the low-sigma portion. This stage can improve the final visual quality without fully rebuilding the image from zero. It can help sharpen local texture, strengthen lighting, improve object edges, add surface detail, and make the result cleaner and more finished. This is the part that makes the workflow feel like a segmented render rather than a normal one-pass generation.The workflow also uses DetailDaemonSamplerNode in the sampling chain. DetailDaemon is useful for controlling detail emphasis during generation. In the Base stage, it can help the model produce richer structure and stronger mid-frequency details. In the Turbo stage, it can help refine texture and micro-details without requiring a full redraw. This makes the workflow useful for images that need a more polished and high-density look.The first stage uses Z-Image Base with a stronger guidance value and a higher detail setting. This helps the image form with enough visual weight. The workflow also uses FlowMatchEulerDiscreteScheduler and Euler Ancestral-style sampling, which gives the render a more controlled progressive structure. The Base stage produces a denoised latent output that is then reused instead of being discarded.The second stage uses Z-Image Turbo with its own CFGGuider and DetailDaemon settings. This stage is more focused on speed and detail. Because the latent already contains the major image structure, the Turbo model does not need to solve the entire composition from noise. Instead, it can concentrate on making the result sharper, cleaner, and more visually complete.The workflow also includes an optional image scaling and second refinement section. After decoding the intermediate output, the workflow uses ImageScaleBy with a 1.5x scale setting. This allows the generated result to be enlarged and then refined again. This is useful when the image looks good but needs more resolution, cleaner detail, and stronger final output quality. The scaled image can be encoded back into latent space and passed through another refinement stage.This makes the workflow suitable not only for normal image generation, but also for high-quality render finishing. It can be used when creators want an image that feels more complete than a fast Turbo output, but faster and more flexible than relying only on Base for every stage. It is especially useful for testing how Base and Turbo behave when they are chained together in one controlled render pipeline.The workflow also includes preview and save nodes so users can inspect both intermediate and final outputs. This is important for segmented rendering because each stage may produce a different visual result. Users can compare the Base-stage result, the Turbo-refined result, and the upscaled or final refined result to decide which one is best for publishing.This workflow is useful for creators who want to explore model cooperation rather than simple model switching. Base and Turbo do not need to be treated as separate isolated workflows. In this graph, they are used as different rendering engines inside one pipeline. Base focuses on early structure. Turbo focuses on later refinement. The split-sigma design makes the handoff more intentional and more controllable.Main features:- Z-Image Base + Z-Image Turbo segmented rendering workflow- Uses z_image_bf16.safetensors for the Base stage- Uses z_image_turbo_bf16.safetensors for the Turbo stage- Qwen 3 4B text encoder support- AE VAE support- SplitSigmas for high-noise and low-noise stage separation- Base model used for global structure and composition- Turbo model used for continuation and final detail refinement- SamplerCustomAdvanced multi-stage sampling- CFGGuider control for each model stage- RandomNoise seed control- FlowMatchEulerDiscreteScheduler support- SamplerEulerAncestral route- DetailDaemonSamplerNode for detail control- Optional 1.5x upscale and second refinement pass- Preview and SaveImage output for stage comparisonRecommended use cases:Z-Image Base and Turbo comparison, segmented image rendering, high-quality text-to-image generation, complex composition testing, fantasy illustration, surreal concept art, cinematic scene generation, character image creation, fashion visual generation, product-style render testing, high-detail artwork finishing, prompt research, Base-to-Turbo pipeline testing, and Civitai workflow showcase examples.Suggested workflow:Start by writing a clear and detailed prompt. Because the Base stage is responsible for the main image structure, the prompt should describe the subject, scene, composition, lighting, color palette, atmosphere, and style clearly. Avoid writing only short vague prompts if you want to test the strength of the segmented pipeline.Use the Base stage to establish the image foundation. This stage is where the image decides the main structure, subject position, background layout, and visual direction. If the Base result already has weak composition, the Turbo stage may improve details but will not always fix the whole image logic. Therefore, tune the prompt and seed until the Base stage gives a good foundation.Use the SplitSigmas setting to control where the handoff happens. If the split happens too early, Turbo may take over before the image structure is stable. If the split happens too late, Base does most of the work and Turbo only has limited influence. The included split value gives a practical starting point, but users can adjust it depending on whether they want stronger Base structure or stronger Turbo finishing.Use the Turbo stage for final polish. This stage is useful for sharpening details, improving local texture, strengthening edges, and making the image feel more complete. If the Turbo output changes the image too much, lower the influence or adjust the low-sigma section. If the Turbo output is too weak, increase its detail settings or allow it more room in the later sampling stage.Use the DetailDaemon settings carefully. Higher detail values can create a richer image, but too much detail may introduce noise, harsh texture, or over-rendered surfaces. For clean fashion photography or product-style images, keep detail moderate. For fantasy, surreal, or illustrative images, stronger detail can create a more dramatic result.Use the optional upscale and refinement stage when the generated image is good but needs more resolution. The workflow includes a 1.5x image scaling section, followed by another latent refinement stage. This is useful for generating a stronger final image for Civitai examples, social media covers, thumbnails, posters, or RunningHub showcases.If you are testing speed, run only the Base + Turbo segmented generation first. If the result is good, enable or continue with the upscale/refinement section. This saves time during prompt testing and avoids wasting resources on high-resolution refinement before the core image is stable.When evaluating the results, do not only look at sharpness. Check whether the image follows the prompt, whether the composition is stable, whether the subject is coherent, whether textures are improved, and whether the Turbo stage preserved the Base structure. The best result should keep the Base model’s structural strength while gaining Turbo’s cleaner final detail.For character images, check facial stability, clothing detail, hands, body structure, and background consistency. For scene images, check depth, atmosphere, perspective, and object placement. For product or poster-style images, check whether the final output looks clean, controlled, and ready for publishing.This workflow is designed for creators who want a more advanced Z-Image rendering strategy inside ComfyUI. It is not just a simple Base workflow or a simple Turbo workflow. It uses staged sampling, sigma splitting, model handoff, detail control, and optional final refinement to create a more flexible image-generation pipeline. It is especially useful for users who want to study how Z-Image Base and Z-Image Turbo can cooperate in one workflow rather than being used separately.🎥 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/Y6L5qkA8ZYsBefore 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/2017271994244403202/?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/BV1DQ61B1Eix/☕ 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/Y6L5qkA8ZYs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/2017271994244403202/?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/BV1DQ61B1Eix/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。
Z-Image Base Upscale Workflow
This ComfyUI workflow is designed for Z-Image Base image upscaling, detail refinement, and high-resolution restoration. It combines a traditional 4x upscale model with Z-Image Base latent refinement, Florence2 automatic captioning, tiled processing, and final image reconstruction. The goal is to turn a lower-resolution or softer image into a cleaner, sharper, and more detailed high-resolution result while keeping the original composition and overall visual identity stable.This is not a simple one-click ESRGAN upscale workflow. It uses a multi-stage enhancement structure. First, the input image is enlarged with a classic upscale model. Then the image is scaled to a target megapixel size. After that, it is divided into tiles, automatically captioned with Florence2, refined through Z-Image Base, decoded with tiled VAE decoding, and finally stitched back into one complete image. This makes the workflow more useful for large images where direct full-frame processing may be unstable or too memory-heavy.The workflow uses z_image_bf16.safetensors as the main Z-Image model, qwen_3_4b.safetensors as the text encoder, and ae.safetensors as the VAE. It also uses 4x_NMKD-Siax_200k.pth as the first-stage upscale model. This gives the workflow a hybrid design: the traditional upscaler provides fast resolution expansion, while Z-Image Base adds AI-driven detail reconstruction, texture polishing, and local refinement.A key part of the workflow is the ImageUpscaleWithModel stage. This step uses the 4x_NMKD-Siax model to enlarge the input image before the Z-Image refinement stage. Traditional upscale models are useful because they preserve the original structure and provide a stable high-resolution base. However, pure traditional upscaling can sometimes look too smooth, too artificial, or lacking in new detail. That is why this workflow continues with a Z-Image refinement pass.After the first upscale, the workflow uses ImageScaleToTotalPixels to bring the image to a target output size. In the included setup, the image is scaled toward a high megapixel target using Lanczos scaling. This gives users a predictable way to control final resolution without manually calculating width and height. It is useful for social media covers, Civitai showcase images, posters, product visuals, portrait enhancement, and high-resolution AI artwork output.The workflow then uses TTP tile tools for large-image processing. TTP_Tile_image_size calculates tile size based on image dimensions, width factor, height factor, and overlap rate. TTP_Image_Tile_Batch splits the image into manageable tiles. This is important because high-resolution images can be too large to refine in one pass, especially on limited VRAM systems. Tiled processing allows the workflow to enhance large images while keeping memory usage more manageable.Florence2 is used to automatically generate captions for the tiled images. The workflow includes Florence2ModelLoader and Florence2Run, using a caption task to describe the content of each tile. The generated captions are then passed into CLIPTextEncode as the positive prompt for Z-Image refinement. This is useful because each tile may contain different visual content. Automatic captioning helps the model understand what is inside each tile, instead of using one vague global prompt for the entire image.This caption-guided refinement is one of the most practical parts of the workflow. For example, if one tile contains a face, another tile contains clothing, and another tile contains background lights, Florence2 can produce local descriptions that guide Z-Image to refine each region more appropriately. This helps improve details without forcing the whole image into one single prompt interpretation.The Z-Image refinement stage uses VAEEncode, KSampler, and VAEDecodeTiled. The enlarged tile image is encoded into latent space, then processed by Z-Image Base with a low denoise setting. In the included setup, the KSampler uses 20 steps, CFG 3, Euler sampler, simple scheduler, and a denoise value around 0.25. This is a conservative refinement setting. It is designed to improve texture and clarity without heavily changing the original image.Low denoise is important for upscale workflows. If denoise is too high, the model may redraw the image too aggressively and change faces, clothing, background details, or object structure. If denoise is too low, the result may not gain enough new detail. A value around 0.25 is a good starting point for faithful enhancement. Users can increase it slightly for stronger redraw or reduce it for more conservative preservation.The workflow uses ModelSamplingAuraFlow with a shift setting to match the model’s sampling behavior. This helps the Z-Image Base route work correctly in the refinement stage. The negative prompt includes simple artifact suppression terms such as blurry, ugly, and bad. This keeps the refinement direction clean without overloading the model with excessive negative tags.After each tile is processed, VAEDecodeTiled decodes the latent result. Tiled decoding helps reduce memory load and supports higher-resolution output. The tiles are then converted back into a batch and reconstructed with TTP_Image_Assy. The padding value helps reduce visible seams between tiles. This final assembly stage is important because tile-based workflows can produce edge artifacts if overlap and padding are not handled properly.The workflow also includes Image Comparer. This allows users to compare the original upscaled image and the final reconstructed image with a slide comparison view. This is useful for checking whether the Z-Image refinement actually improved the result. Good upscale evaluation should look at face detail, hair texture, fabric clarity, edge sharpness, background stability, seam visibility, and whether the image identity has changed too much.This workflow is suitable for creators who want more than basic resolution expansion. It is useful for AI-generated images that look slightly soft, screenshots that need enhancement, portraits that need more texture, fashion images that need sharper fabric detail, stage photos that need cleaner lights, product images that need better surface clarity, and Civitai examples that need high-resolution polish.Main features:- Z-Image Base upscale and refinement workflow- Uses z_image_bf16.safetensors- Qwen 3 4B text encoder support- AE VAE support- 4x_NMKD-Siax first-stage upscaling- ImageScaleToTotalPixels for target megapixel control- Florence2 automatic tile captioning- TTP tiled image splitting- Z-Image latent refinement per tile- Low-denoise detail enhancement- VAEDecodeTiled for high-resolution decoding- TTP_Image_Assy tile reconstruction- Padding and overlap support to reduce seams- Image Comparer for before/after checking- Suitable for high-resolution artwork, portraits, covers, and product imagesRecommended use cases:AI image upscaling, high-resolution restoration, portrait enhancement, anime artwork polishing, realistic photo refinement, product image cleanup, fashion image sharpening, social media cover enhancement, Civitai showcase image preparation, poster output, screenshot restoration, background detail recovery, texture improvement, and before/after comparison testing.Suggested workflow:Start by loading the image you want to upscale. Use an image with a clear subject and stable composition. The workflow can improve soft or lower-resolution images, but it cannot fully recover information from extremely damaged or heavily compressed images. A clean source image will always produce better results.Run the first-stage upscale with the 4x_NMKD-Siax model. This creates a larger base image while preserving the overall structure. Then use ImageScaleToTotalPixels to control the final target resolution. If your image becomes too large for your GPU, reduce the megapixel target before running the Z-Image refinement stage.Use the tiled processing section for large outputs. The workflow splits the image into tiles and processes them separately. If seams appear after reconstruction, increase overlap or padding. If memory usage is too high, use smaller tiles or reduce the final resolution. If details look inconsistent across tiles, try reducing denoise or using more stable caption settings.Let Florence2 generate captions for the tiles. These captions help guide the Z-Image refinement process. You can review the generated text in the ShowText node. If the captions are inaccurate, you can manually replace or adjust the prompt. For professional output, checking captions is useful because wrong tile descriptions can cause unwanted texture changes.Use low denoise for faithful enhancement. The included denoise setting around 0.25 is suitable for preserving the original image. If you want stronger detail reconstruction, increase denoise slightly. If faces or objects begin to change too much, reduce denoise and keep the prompt simpler.Check the final output with Image Comparer. Compare the original enlarged image with the refined image. Look for improvements in sharpness, texture, local detail, and clarity. Also check whether the model introduced artifacts, changed identity, damaged hands or faces, or created visible tile seams.For portrait images, focus on preserving facial identity and natural skin texture. For anime images, the workflow can be pushed slightly stronger because stylized linework often benefits from sharper redraw. For product images, keep denoise conservative so product shape and branding remain stable. For background-heavy images, inspect seams carefully after tile assembly.This workflow is designed as a practical Z-Image Base high-resolution enhancement pipeline for ComfyUI users. It combines classic upscaling, target-size control, automatic captioning, tile-based refinement, low-denoise Z-Image polishing, and final reconstruction into one usable graph. It is especially useful for creators who need cleaner, sharper, and more publishable images for Civitai, RunningHub, YouTube thumbnails, Bilibili posts, product visuals, and AIGC content production.🎥 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/JPA_qq5YusEBefore 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/2016768153547710466/?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/BV1D96wBxECM/☕ 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/JPA_qq5YusE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/2016768153547710466/?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/BV1D96wBxECM/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。
Title: LTX-2 Lip Sync Workflow
Description:LTX-2 Lip Sync Workflow is a ComfyUI workflow designed for audio-driven lip sync video generation, talking character animation, and image-to-video portrait performance using LTX-2. Instead of only creating a silent motion clip from an image, this workflow brings audio into the generation process and lets the video latent and audio latent work together, making it suitable for creating short speaking videos, AI presenters, dialogue clips, digital human previews, character voice performances, and social media talking-head content.The workflow is built around the LTX-2 19B Dev FP8 checkpoint, using both the main video model and the dedicated LTX audio VAE pipeline. The audio file is encoded into an audio latent, then combined with the video latent through an audio-video latent workflow. This design allows the model to use the input audio as part of the generation condition, instead of treating the audio as something added after the video is finished. The result is a more direct audio-to-mouth-motion relationship, which is important for lip sync, speech rhythm, facial timing, and natural talking performance.The core logic of the workflow is image + audio to lip-synced video. You provide a source image as the visual identity reference and an audio file as the speech or singing reference. The image is used to initialize the character appearance and video layout, while the audio latent guides the speaking rhythm. The workflow then generates a video where the character can appear to talk along with the provided audio.A key part of this workflow is the LTXVAudioVAEEncode stage. The input audio is processed by the LTX audio VAE and converted into an audio latent. This audio latent is then passed into the later video generation stage through LTXVConcatAVLatent, where it is combined with the video latent. After sampling, LTXVSeparateAVLatent is used to separate the final video latent from the audio-video latent structure. This gives the workflow a clear audio-video pipeline: load audio, encode audio, combine audio with video latent, sample, separate video latent, then decode or upscale the final result.The workflow also uses image-to-video logic through LTXVImgToVideoInplace. This helps preserve the source image identity and composition while allowing the generated frames to move. For portrait images, this is especially useful because the face, clothing, background, and general framing can remain close to the original image while the mouth, facial expression, and subtle head motion are animated according to the audio.The workflow includes an EmptyLTXVLatentVideo stage for setting the base video latent dimensions and frame length. In the included setup, the frame rate logic is based around 24 fps. This is important because lip sync quality depends heavily on matching the audio duration to the correct number of frames. For example, a 10-second clip at 24 fps usually needs 24 × 10 + 1 frames, which means 241 frames. If the frame count is wrong, the audio and mouth movement may drift or feel off-sync.The workflow also includes sampler control through SamplerCustomAdvanced, ManualSigmas, KSamplerSelect, CFGGuider, and RandomNoise. These nodes control how the latent video is generated, how strongly the prompt affects the result, and how the noise schedule behaves. The workflow is not just a basic video generation template; it is structured to support audio-conditioned motion, image identity preservation, and controlled sampling.Another important part is the LTXVLatentUpsampler stage. This allows the workflow to upscale the latent video after the first generation pass. The purpose is to improve output quality while keeping the initial motion and lip sync result. For faster previews, the upscale stage can be bypassed so users can test image, seed, prompt, and audio timing more quickly. After finding a good seed and prompt combination, the upscale stage can be enabled again for a cleaner final output.This workflow is suitable for AI creators who want to turn a still portrait into a speaking or singing video. It can be used for digital human demos, AI character narration, short-form video avatars, dialogue previews, virtual host content, product explanation clips, tutorial presenters, anime-style talking characters, realistic portrait animation, and creative voice-driven character tests.Main features:- LTX-2 audio-driven lip sync workflow- Built around LTX-2 19B Dev FP8- Image + audio to talking video generation- Audio VAE encoding for speech-driven motion- Audio latent and video latent combination- LTXVConcatAVLatent and LTXVSeparateAVLatent workflow- Image-to-video identity preservation- 24 fps video conditioning logic- Manual frame control based on audio duration- SamplerCustomAdvanced generation pipeline- ManualSigmas and CFG guidance control- Optional latent upscaling for final quality improvement- Suitable for portrait animation, digital humans, and character dialogue- Good for testing speech, singing, narration, and AI presenter workflowsRecommended use cases:AI talking head video, digital human demo, lip sync portrait animation, audio-driven character performance, virtual presenter video, product explanation avatar, short-form social media narration, anime character talking video, realistic portrait speech animation, singing character tests, voiceover-driven video creation, dialogue scene preview, ComfyUI audio-video workflow testing, and Civitai showcase examples.Suggested workflow:Start by preparing a clean source image. A front-facing or slightly angled portrait usually works best. The face should be visible, the mouth area should not be blocked, and the image should have enough resolution for facial detail. Avoid images with extreme face angles, heavy occlusion, tiny faces, very strong motion blur, or low-quality compression.Next, prepare the audio file. MP3 can work, but clean audio usually gives better results. Try to use a voice clip with clear speech, limited background noise, and stable volume. If the audio contains music, echo, overlapping voices, or heavy noise, the mouth movement may become less reliable. For a first test, use a short 5-second clip before trying longer videos.Set the audio duration and frame count carefully. Since the workflow uses 24 fps logic, the frame count should match the audio length. A 5-second clip should use about 121 frames. A 10-second clip should use about 241 frames. A 15-second clip should use about 361 frames. If the frame length is too short or too long, the final video may not match the audio timing.Write a prompt that describes the character and the intended performance. For lip sync videos, the prompt should not only describe the visual style, but also the behavior. You can describe a person speaking naturally, subtle head movement, realistic mouth movement, calm facial expression, stable camera, soft lighting, and shallow depth of field. If you want the video to stay close to the source image, keep the prompt simple and avoid changing identity details too aggressively.Use the negative prompt to suppress common lip sync problems. Useful negative terms include mismatched lip sync, distorted mouth, exaggerated expression, unnatural face movement, wrong gaze direction, robotic voice, audio delay, jittery motion, deformed face, flickering, duplicated mouth, missing microphone, incorrect expression, over-smiling, laughing, camera shake, and AI artifacts.For portrait videos, start with moderate resolution and short duration. A safe testing setup is 480 x 832 for portrait or 832 x 480 for widescreen, around 5 seconds. After confirming that the seed, prompt, and audio timing work well, you can increase duration and resolution. Longer videos require more VRAM and more generation time, so it is better to test in small steps.If the video has almost no motion, use a stronger motion prompt or camera-control guidance if available. If the face changes too much, simplify the prompt and reduce identity-changing descriptions. If the mouth does not match the audio, check the audio duration, frame count, and whether the audio has clear speech. If the video looks soft, enable the latent upscaler for the final pass. If you only want to preview quickly, bypass the upscale stage first.This workflow is designed for creators who need a practical LTX-2 lip sync pipeline inside ComfyUI. It combines source image preservation, audio latent conditioning, video latent generation, sampler control, and optional latent upscaling into one workflow. It is useful for testing LTX-2 audio-video generation, creating AI talking characters, preparing short digital human clips, and building publishable Civitai examples from image + voice input.🎥 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/LH1FquAz5O8Before 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/2011736436441092097/?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/BV1LLkFBhEgm/☕ 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/LH1FquAz5O8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/2011736436441092097/?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/BV1LLkFBhEgm/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。
Z-Image ControlNet 2.1-2601 Text-to-Image Workflow
Description:Z-Image ControlNet 2.1-2601 Text-to-Image Workflow is a ComfyUI generation workflow designed for high-quality text-to-image creation with stronger structural control, cleaner detail rendering, and more stable prompt interpretation. It is built around Z-Image Turbo and the Z-Image Turbo Fun ControlNet Union 2.1-2601 model patch, giving creators a practical way to generate polished images from text prompts while still keeping additional control options available for pose, composition, and detail refinement.Unlike a simple text-to-image workflow that only relies on a prompt and a sampler, this workflow uses a more structured generation design. It combines Z-Image Turbo, the Qwen 3 4B text encoder, the Z-Image VAE, ControlNet Union guidance, DetailDaemon sampling, high-noise prompting, low-noise prompting, and optional preprocessor support. The goal is to make text-to-image generation more controllable, especially when the user needs a specific visual direction such as cinematic lighting, cyberpunk characters, anime illustration, fantasy armor, product-style rendering, poster design, or social media cover images.The workflow is suitable for creators who want to generate images directly from text, but still need more control than a basic one-click setup. You can describe a character, environment, product, vehicle, scene atmosphere, lighting style, camera angle, color palette, and visual mood through prompts. The workflow then uses the Z-Image Turbo generation pipeline to create the base image, while ControlNet-related modules and DetailDaemon sampling help improve structure, detail density, and final sharpness.One important design point of this workflow is the high-noise and low-noise prompt logic. The high-noise prompt is used to define the main subject, scene, composition, and creative direction. This is where you write the core idea of the image: who or what appears in the frame, what the subject is doing, what the background looks like, what style you want, and what kind of atmosphere the image should have. For example, you can describe a cyberpunk female rider on a neon motorcycle, a fantasy warrior in glowing armor, a product hero shot, a futuristic city, or an anime character in a dramatic scene.The low-noise prompt is used for refinement. It helps polish the final appearance, including texture, edge quality, lighting consistency, color harmony, detail clarity, and surface finish. This two-layer prompt structure is useful because it separates the main creative idea from the final visual refinement. The high-noise prompt controls the large visual direction, while the low-noise prompt helps improve the generated result without completely changing the image concept.The workflow also includes ControlNet Union 2.1-2601 support. This gives the workflow more flexibility than a normal text-to-image pipeline. Depending on the setup, ControlNet guidance can help preserve structure, pose, silhouette, object direction, or spatial arrangement. This is useful when generating character-based images, action poses, fashion visuals, dynamic illustrations, concept art, or scenes where body structure and composition need to be more stable.The included preprocessor and pose-related nodes make this workflow more suitable for controlled image generation. When you want to create character images with a clearer body posture, or when you need stronger structure guidance, these modules can help guide the generation process. This makes the workflow useful for anime characters, realistic portraits, cyberpunk fashion, fantasy characters, game concept art, and cinematic poster-style outputs.DetailDaemonSamplerNode is used to enhance detail behavior during sampling. This is especially helpful for images that need richer local texture, sharper fabric edges, cleaner hair, more defined facial details, better armor surfaces, stronger neon highlights, metallic reflections, product materials, and polished illustration quality. Instead of producing a flat or soft image, the workflow can create a more refined final result with better visual density.The workflow also includes sampler and scheduler control. It uses structured sampling instead of a completely simplified default generation process. Users can adjust total steps, denoise behavior, CFG guidance, ControlNet strength, detail intensity, and seed settings depending on the desired result. For fast testing, you can keep the settings close to the default configuration. For more refined images, you can test different seeds, increase prompt specificity, and tune the detail settings carefully.This workflow is especially useful for AI creators who need a fast but controllable text-to-image pipeline inside ComfyUI. It is not only for casual generation, but also for production-style image creation, Civitai example image preparation, workflow testing, thumbnail design, social media cover creation, concept visualization, and polished AIGC content output.Main features:- Z-Image Turbo text-to-image generation workflow- Z-Image Turbo Fun ControlNet Union 2.1-2601 support- Qwen 3 4B text encoder support- Z-Image VAE generation pipeline- High-noise prompt for main subject and composition- Low-noise prompt for refinement and final visual polish- ControlNet-based structure guidance- Optional pose and preprocessor support- DetailDaemon sampling for improved texture and sharpness- Sampler and scheduler control for flexible output tuning- Suitable for anime, realistic, cyberpunk, fantasy, product, and poster-style generation- More controllable than a simple text-to-image workflow- Useful for Civitai showcase images and social media content productionRecommended use cases:Anime character generation, realistic portrait creation, cyberpunk scene design, fantasy warrior illustration, game concept art, product-style image generation, cinematic poster design, AI cover image production, social media thumbnails, character design drafts, fashion concept visuals, futuristic vehicle scenes, neon city compositions, stylized illustration, and high-quality text-to-image experiments.This workflow is also useful for creators who want to test Z-Image ControlNet 2.1-2601 behavior across different prompt styles. You can compare how it handles character prompts, environment prompts, product prompts, action scenes, lighting-heavy prompts, and illustration-style prompts. Because the workflow includes both high-noise and low-noise prompt areas, it is easier to separate creative control from final visual quality control.Suggested workflow:Start by writing a clear high-noise prompt. This should describe the main subject, action, scene, camera view, lighting, and style. For example, describe whether the image is a portrait, full-body shot, product shot, cinematic scene, anime illustration, or futuristic concept image. Include important subject details such as clothing, material, pose, environment, color tone, and mood.Then write the low-noise prompt for final polish. This prompt should focus on quality and consistency instead of rewriting the entire image. You can include terms like detailed texture, clean lighting, cinematic contrast, sharp edges, natural skin texture, refined fabric, polished metal, consistent color palette, realistic reflection, anime-style finish, or high-quality illustration.Use the seed setting to test different compositions. If the generated image does not match your idea, adjust the high-noise prompt first. If the composition is good but the details are weak, adjust the low-noise prompt or detail settings. If the structure looks unstable, use the ControlNet or pose-related options to guide the image more strongly.For character generation, keep the prompt focused on one subject when possible. Describe the character clearly, including hairstyle, clothing, pose, facial expression, lighting, and background. For product or object generation, avoid too many unrelated style words and focus on material, shape, lighting, and scene placement. For cinematic images, include camera language such as close-up, wide shot, low angle, shallow depth of field, rim light, volumetric lighting, or dramatic contrast.For anime and illustration generation, you can push the prompt more creatively. Use clear visual terms, strong color direction, and detailed subject descriptions. For realistic images, keep the prompt more controlled and avoid conflicting style tags. If the output becomes too chaotic, simplify the prompt and reduce unnecessary details.This workflow is designed as a practical Z-Image ControlNet text-to-image tool for ComfyUI users. With Z-Image Turbo, ControlNet Union 2.1-2601, Qwen text encoding, two-stage prompting, pose-aware preprocessing, and detail-enhanced sampling, it provides a flexible and efficient way to generate high-quality images from text while keeping more control over structure, detail, and final style.🎥 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/LH1FquAz5O8Before 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/2011731536432865281/?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/BV1LLkFBhEgm/☕ 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/LH1FquAz5O8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/2011731536432865281/?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/BV1LLkFBhEgm/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。
Z-Image ControlNet 2.1-2601 Local Inpainting Workflow
Z-Image ControlNet 2.1-2601 Local Inpainting Workflow is a ComfyUI workflow designed for precise local repainting, masked image editing, and structure-controlled visual correction. Instead of regenerating the entire image, this workflow focuses on editing only the selected mask area while keeping the original composition, pose, lighting, background, and main visual identity as stable as possible.This workflow is built around Z-Image Turbo, using the Qwen 3 4B text encoder, the Z-Image VAE, and the Z-Image Turbo Fun ControlNet Union 2.1-2601 model patch. The main goal is to give creators a more controllable way to repair or redesign specific parts of an image. You can use it to replace clothing, fix a face, repair hands, change an object, repaint a product area, modify a character, add new elements, or correct a broken AI-generated detail without destroying the rest of the picture.The workflow starts from an input image and a mask. The ImageLoader provides both the original image and the masked area, then the image is encoded into latent space through the VAE. The masked region is passed into the latent repainting stage, so the new generation is mainly applied to the area you selected. This makes the workflow suitable for real production correction, because you do not need to recreate the full image every time a small region fails.A key part of this workflow is the high-noise and low-noise prompt structure. The high-noise prompt controls the main semantic change. This is where you describe what the masked area should become: a new outfit, a new object, a different face detail, a new weapon, a corrected hand, a changed product, or a repaired background element. The low-noise prompt is used for refinement. It helps the generated area blend back into the original image by improving color consistency, lighting, texture, and edge transition.This two-stage prompt structure is useful because local repainting needs both creativity and stability. If the prompt is too strong, the edited area may break away from the original picture. If the prompt is too weak, the change may not be obvious enough. By separating the main concept from the final refinement, this workflow gives users more control over how much the masked region changes and how naturally it merges with the source image.The ControlNet module is another important part of this workflow. It uses the Z-Image Turbo Fun ControlNet Union 2.1-2601 model patch to provide structure guidance during generation. This helps the workflow preserve pose, silhouette, object direction, and scene layout when editing a selected region. For example, when editing a character, you can keep the original body posture while changing the outfit. When editing a product image, you can keep the product position and only modify surface details. When editing an illustration, you can preserve the composition while correcting a broken part.The workflow also includes pose and preprocessor-related nodes, such as DWPreprocessor and SDPose-related processing. These are useful when the source image contains a person or character and you need stronger body-structure preservation. This makes the workflow more suitable for character-based editing, anime illustration correction, fashion replacement, cyberpunk character repainting, fantasy armor modification, and pose-aware local redraw.DetailDaemonSamplerNode is included to enhance local detail during the sampling process. This helps the edited area avoid looking flat or blurry. It can improve fabric edges, hair strands, facial detail, metal reflections, product surfaces, armor texture, vehicle parts, neon materials, and other fine structures. This is especially useful when the original image is already high quality and the inpainted region must match the sharpness of the surrounding area.The workflow also includes sampler and scheduler control, using a structured sampling process rather than a simple one-click inpaint. The repaint strength, total steps, denoise level, CFG guidance, ControlNet influence, and mask size all affect the final result. For small corrections, a conservative denoise setting is recommended. For stronger replacement, use a larger mask and a clearer prompt. If the result drifts too much, reduce denoise or simplify the prompt. If the repaint is not strong enough, expand the mask area or increase the semantic strength of the high-noise prompt.Main features:- Z-Image Turbo local repainting workflow- Z-Image Turbo Fun ControlNet Union 2.1-2601 support- Mask-based local inpainting- High-noise prompt for main semantic change- Low-noise prompt for final texture and style refinement- Qwen 3 4B text encoder support- Z-Image VAE latent workflow- SetLatentNoiseMask for targeted region editing- ControlNet-guided structure preservation- Pose-aware preprocessing support- DetailDaemon sampling for sharper local details- Suitable for image repair, object replacement, and character editing- More stable than full image regeneration- Useful for AI artwork post-production and Civitai example preparationRecommended use cases:Local object replacement, face repair, hand correction, clothing replacement, product image cleanup, anime illustration repair, character redesign, cyberpunk outfit modification, fantasy armor repainting, background object removal, damaged image correction, AI-generated image fixing, masked area enhancement, social media cover correction, product visual adjustment, and Civitai showcase image refinement.Suggested workflow:Upload your source image first, then prepare the mask area. The mask should cover the region you want to edit, but it is usually better to make the mask slightly larger than the exact broken area so the model has enough space to blend edges naturally.Write the high-noise prompt to describe the main change. For example, if you want to replace clothing, describe the new clothing clearly. If you want to repair a face, describe the desired facial expression and quality. If you want to replace an object, describe the new object, material, angle, and lighting direction.Write the low-noise prompt to describe the final style. This can include terms such as natural lighting, consistent texture, clean edges, matching color tone, detailed fabric, cinematic lighting, realistic surface, or anime-style polish. Keep this prompt focused on refinement instead of rewriting the entire image.Use a lower denoise value when you only need small repairs. Use stronger denoise when the masked region needs to become something completely new. If the result looks too different from the source image, reduce denoise, reduce prompt complexity, or lower the detail intensity. If the result is too weak, increase the repaint strength, enlarge the mask, or make the high-noise prompt more direct.For character images, keep the pose reference active when you need body structure stability. For product images, keep the prompt clean and avoid describing unrelated background details. For anime and illustration work, the workflow can be pushed further creatively, but it is still recommended to control the mask carefully to avoid unwanted changes outside the target region.This workflow is designed for creators who need practical local editing inside ComfyUI. It is not only a demo workflow, but a useful production tool for image correction, publishable example generation, visual asset repair, character polishing, and before/after comparison creation. With Z-Image Turbo, ControlNet Union 2.1-2601, mask-based latent repainting, two-stage prompting, pose-aware preprocessing, and detail-enhanced sampling, it provides a flexible and efficient solution for controlled local inpainting.🎥 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/LH1FquAz5O8Before 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/2011731528195252226?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/BV1LLkFBhEgm/☕ 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/LH1FquAz5O8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/2011731528195252226?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/BV1LLkFBhEgm/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。
Z-Image - Three Upscaling Methods
Z-Image Three-Way Upscale Workflow is a ComfyUI image enhancement pipeline designed for high-resolution restoration, detail refinement, and controlled upscaling. This workflow provides three different upscale routes in one graph, so users can compare speed, texture quality, structure preservation, and final sharpness depending on the image type.The first route uses Z-Image Turbo with Z-Image Fun ControlNet Tile and UltimateSDUpscale. It is suitable for anime, illustration, realistic portraits, product images, and AI-generated artwork that needs stronger detail reconstruction. The Tile ControlNet helps preserve the original composition while allowing Z-Image to redraw micro-details, skin texture, hair, fabric, background objects, and edges with better clarity. The default denoise is low, so the image will not drift too far from the original.The second route combines a traditional 4x upscale model with Z-Image latent refinement. It first enlarges the image using 4x_NMKD-Siax, then scales the result to a high megapixel target, splits it into tiles, generates captions with Florence2, and sends the tiles back into Z-Image for controlled polishing. This path is useful when you want a balanced result: sharper than pure ESRGAN-style upscaling, but more stable than heavy redraw.The third route uses SeedVR2 as an AI restoration/upscale pass. Although SeedVR2 is often used for video enhancement, this workflow applies it to tiled image restoration. It is especially useful for soft images, compressed outputs, screenshots, old renders, and images that need cleaner reconstruction without excessive prompt influence. The workflow tiles the image, processes each section, and reassembles it with padding to reduce seams.Main features:- Three upscale methods in one workflow- Z-Image Turbo detail refinement- Z-Image Fun ControlNet Tile support- UltimateSDUpscale route for high-quality redraw- Florence2 automatic caption generation- 4x_NMKD-Siax classic upscale model route- SeedVR2 restoration/upscale route- Tiled processing for large images- Image comparison nodes for before/after checking- Low denoise settings for better identity and composition preservationRecommended use cases:Portrait enhancement, anime upscaling, AI artwork polishing, product image cleanup, social media cover restoration, high-resolution poster output, detail recovery, and comparing multiple upscale strategies from one source image.Suggested workflow:Load your source image, choose the upscale route you want to test, keep denoise low for faithful enhancement, increase tile overlap if seams appear, and compare the final outputs with the built-in image comparer. For realistic portraits, avoid pushing denoise too high. For anime or concept art, the Z-Image Tile route can create richer details. For compressed or blurry inputs, SeedVR2 may give the cleanest restoration.🎥 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/LH1FquAz5O8Before 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/2011736357118414850/?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/BV1LLkFBhEgm/☕ 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/LH1FquAz5O8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/2011736357118414850/?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/BV1LLkFBhEgm/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。
SAM 3 + Inpaint Crop and Stich
This uses ComfyUI native SAM 3 nodes to segment the image into masks, and then https://github.com/lquesada/ComfyUI-Inpaint-CropAndStitch nodes to upscale for inpaint editing and then stitch it back to the original image. It's an old idea from the SD 1.5 times, but it's still faster than using an editing model and can be useful for plenty of stuff, like improving faces in the background (or even the foreground) - as such, it works as an alternative to face detailer and segs detailer, with the more modern SAM 3 model.I've used Z-Image Turbo for this workflow, with the gguf clip, but you can change it for any other model you like. There are reroute nodes for the model, clip and vae if you need them.Besides lquesada's inpant crop and stitch custom nodes, it uses the https://github.com/city96/ComfyUI-GGUF nodes for the clip and https://github.com/pythongosssss/ComfyUI-Custom-Scripts for a sound notification at the end. If you use safetensors for the clip, you don't need the gguf loader and if you don't want the notification sound, you can simply delete that node
Anaglyph for ComfyUI
https://github.com/orion4d/Orion4D_anaglyph🎭 Orion4D Anaglyph for ComfyUIOrion4D Anaglyph is a high-performance custom node designed to transform 2D images into stereoscopic (3D) renders via a depth map. It offers total control over parallax, convergence, and depth processing to ensure optimal visual comfort.🚀 Key FeaturesTotal Compatibility: Works seamlessly with Depth Anything (V2/V3), Marigold, ZoeDepth, MiDaS, or even hand-edited maps.Various Rendering Modes: Supports Red/Cyan (Dubois, Color, Gray) and Red/Blue modes.Multiple Outputs: Generates the final anaglyph, Side-by-Side (SBS) format, isolated left/right views, and the processed depth map.Preset Manager: Integrated JS interface to save, load, and update your favorite settings on the fly.Native Processing: Optimized PyTorch implementation for maximum speed and cross-platform compatibility.🛠 InstallationCopy the Orion4D_anaglyph folder into your custom nodes directory:ComfyUI/custom_nodes/Orion4D_anaglyph RequirementsNo additional Python dependencies are required beyond a standard ComfyUI installation.This custom node uses only modules already available in ComfyUI:torchaiohttpComfyUI PromptServerPython standard library modulesFolder StructureOrion4D_anaglyph/ ├── orion4d_anaglyph.py # Core logic ├── preset_manager.py # JSON management ├── web/ # User Interface (JS) └── presets/ # Your saved configurations ⚙️ Configuration ParametersParameterDescriptionRecommended Valuestrength3D effect intensity (parallax).16 to 32convergenceFocal point where the image appears to sit on the screen plane.0.5shift_modeOffset method (symmetric, left_static, right_static).symmetricdepth_blurSmoothes edges to reduce visual artifacts.5 to 9depth_invertInverts depth if the relief appears backwards.Per source🔄 WorkflowDrag and drop orion4d_anaglyph_workflow.png onto the comfy canvasThe node is best placed immediately after a depth estimator:🎨 Available Anaglyph ModesOptimized Dubois (Red/Cyan): The gold standard for color fidelity and ghosting reduction.Color / Half-Color: Preserves more of the original vibrancy at the cost of potential color fringing.Gray: Ideal for maximum depth clarity without chromatic distraction.🆘 Troubleshooting & Tips👁️ Eye Strain?Reduce the strength value.Adjust convergence to place the main subject on the screen plane.Increase depth_blur.🔀 Inverted 3D Effect?Enable depth_invert.If the issue persists, use the swap_eyes option.🖼️ Edge Stretching (Warping)This is a standard limitation of 2D-to-3D reprojection. To minimize it:Use padding_mode: border.Reduce depth_contrast.Use a higher-precision depth map (e.g., Depth Anything V3 Large).🌟 Support the ProjectIf you find this tool useful, feel free to leave a ⭐ on GitHub!Developed with ❤️ for the ComfyUI community by Orion4D
Z-Image Turbo AIO Workflow for 4GBVRAM Laptops (TXT2IMG, IMG2IMG, Inpaint, Upscale, Face Detailer) LOW VRAM
Just a seriously basic workflow for those who have laptops with only 4GB video cards and who still wanna work with Z-Image turbo. Needless to say, with such low specs, you'll need to use extremely quantized models and text encoders. The ones I use:Text encoders- Qwen3-4B-Instruct-2507-Q4_K_M(https://huggingface.co/unsloth/Qwen3-4B-Instruct-2507-GGUF/blob/main/Qwen3-4B-Instruct-2507-Q4_K_M.gguf)GGUF Model- Z-image-turbo-q3_k_s(https://huggingface.co/gguf-org/z-image-gguf/resolve/main/z-image-turbo-q3_k_s.gguf?download=true)Only uses nodes you can find in the regular manager, nothing fancy. For image to image, make sure to activate the switch in the WF to "True". This is a frankensteined WF, so credits go to their respective authors.
Z-Image Turbo, fast simple t2i workflow + upscale
Z-Image Turbo workflow, fast and simple for t2i + upscale you can enable/disable. All instructions are in the workflow notes.
Orion4D-secret_agent
🕵️♂️ Orion4D Secret Agent - Steganography & Encryption ToolkitOrion4D Secret Agent is a comprehensive toolkit for ComfyUI dedicated to steganography, encryption, data encoding, and information transmission through audio and visual channels.This node suite allows you to hide, transform, and secure data (text, images, files) within multimedia streams. Whether you want to encode messages into images, convert audio to Base64, or inject workflows directly into image metadata, this is the ultimate spy toolkit for ComfyUI.🌟 Main Features & Node Categories1. Data Converters & EncodersBase64 / Binary / Hex / Decimal: Fully encode and decode text to various formats.Audio & Image to Base64: Convert audio streams or images into Base64 strings and vice versa.2. Steganography & CryptographyAudio Crypt & Modem Nodes: Encode data into audio signals (OFDM, Morse, Modem simulation).Military & Cipher Nodes: Advanced text and data encryption tools.Workflow Injector: Hide a specific ComfyUI JSON workflow directly inside an image's metadata without altering the visual.3. Multimedia & I/OAV Muxer: Create static videos (MP4/AVI) from a single image and an audio file, or replace audio in existing videos.Clean Save Image: Save images while stripping all metadata for total privacy.Advanced Audio Save: Manage WAV/FLAC/MP3 saving with timestamping and subfolders.Text Compression: Compress text (Zlib or LZMA) to Base64 to save space.4. Utility NodesSimple OCR: Read text contained within an image using Tesseract.PDF Report: Generate simple PDF reports combining images and text fields.Text Bus (Variables): Define global variables to transport text across your workflow without messy visible links!⚙️ InstallationOpen your terminal in the ComfyUI/custom_nodes/ directory and run:Bashgit clone https://github.com/orion4d/Orion4D-secret_agent Then, install the requirements (if any) and restart ComfyUI.🛠️ How to get started?Download the .json files or the demo workflow provided in this Civitai post and load it into your UI. It will guide you through the basic concepts of encoding and decoding data.🔗 Links & SupportGitHub Repository: https://github.com/orion4d/Orion4D-secret_agentIf you enjoy this project and find it useful, please consider giving it a ⭐ on GitHub and supporting the creator on Ko-fi!
MetaNode
👉 GitHub - Orion4D_MetaNode🚀 Orion4D_Metanode — Custom Nodes for ComfyUIInfo: 29+ Custom Nodes | Requires: Python 3.10+ | License: MIT ✔️ Fully Compatible with ComfyUI Nodes V2This project brings together all my work on ComfyUI.Orion4D_Metanode transforms ComfyUI into a true programmable environment.Thanks to the PyCode Max engine, you can execute Python directly within your workflows and build your own tools, logic, and interfaces. Featuring dynamic routing, variable buses, enriched UI, file management, and image processing... Everything is designed to push beyond the limits of traditional node-based systems.✨ Why Orion4D_Metanode?Create dynamic and scalable workflows tailored to your needs.Eliminate complex wiring thanks to an intelligent bus system.Develop your own tools directly inside the ComfyUI interface.Transform a standard workflow into a true programmable pipeline.🔗 Installation & DocumentationFor full installation instructions, detailed documentation, and the latest updates, please visit the official GitHub repository:👉 GitHub - Orion4D_MetaNode
MaskPro
https://github.com/orion4d/Orion4D_maskproOrion4D MaskProAdvanced mask editor for ComfyUI, non-destructive editing with professional selection tools, custom PNG brushes, embedded preview, and fullscreen editor.
ZIT simple inpaint workflow
ZImage Turbo Easy Inpaint WorkflowFast & Easy Inpainting workflow built around ZImage Turbo for high-quality, detailed results with minimal effort.Key Features:Optimized for ZImage Turbo (very fast generation)Easy inpainting setup with mask supportBuilt-in support for multiple LoRAsClean node layout focused on simplicity and speedGood anatomy and detail retention on inpaintsRecommended Settings:Steps: 6-9Sampler: heunpp2 or dpmpp_2m_sdeCFG: 1.0 – 1.5 (Turbo works best at low CFG)Denoise: 0.35 – 0.65 (depending on how much you want to change)Nodes Included in Workflow:Basic nodes for inpainting3 LoRa loadersPerfect for: Quick inpainting, outfit changes, body modifications, detail enhancement, and creative edits while keeping the speed advantage of Turbo models.After loading the workflow, just drag your image into the Load Image node, paint your mask, write your prompt, and hit Queue!Made this workflow because when i first started using ZIT there was a bunch of workflows on here that were overly complicated with custom nodes, this workflow uses only basic nodes included with ComfyUI.
IntoRealism ZIT Workflow
You can download the V2 Workflow with ControlNet on my Ko-Fi page for $5, or sign up for a Pro subscription for $25, which gives you access to all my models, early access, scripts, WF, and more...Find all my models, scripts, and beta versions on my Ko-Fi.IntoRealism ZIT ENZINO - Official WorkflowA complete and optimized workflow designed specifically for IntoRealism ZIT 4.0, focused on achieving ultra-realistic results with minimal effort and maximum consistency.FeaturesOptimized specifically for IntoRealism ZIT 4.0Clean and beginner-friendly layoutIntegrated FaceDetailerBuilt-in UpscalerFast generation setupHigh facial consistencyRealistic skin texture and lightingSupports LoRAs directly inside the workflowEasy prompt editingOptional style presets via Style SelectorTuned settings for cinematic realismIncluded ModulesZIT Text-to-ImageCore generation setup using :optimized sampler settingsbalanced CFGrealistic prompt structureclean negative promptsDesigned to produce:natural skinrealistic lightingnon-plastic facescinematic depthFaceDetailer IntegrationCarefully tuned FaceDetailer settings to:enhance realismpreserve facial identityavoid overprocessed "AI face" lookUnlike many workflows, this setup avoids:waxy skindistorted eyesaggressive sharpeningIntegrated UpscalerBuilt-in upscale pipeline for :sharper final rendersimproved detailscleaner texturesbetter social media/export qualityDesigned ForPerfect for:realistic portraitscinematic photographyinfluencer-style imagesfashion shotsamateur photography looksocial media contentNSFW realismcreative storytellingPhilosophyThis workflow was created with one goal :Make IntoRealism produce strong results immediately, without spending hours tweaking settings.Everything is already tuned to work together:samplerdenoiseFaceDetailerupscaleprompt structurerealism balanceRecommended UsageUse natural promptsAvoid overdescribing facesKeep CFG balancedLet the workflow handle realism naturallyThe workflow is intentionally designed to preserve :natural imperfectionsrealistic skin textureauthentic lightingbelievable facial structureOfficial WorkflowMade specifically for IntoRealism ZIT ENZINO by Enzino AI.
Workflow RAW Detailer
🔧 NEURO HARMONYPurpose: This workflow is designed to generate ultra-detailed RAW-style images using the RAW_Photography LoRA, ControlNet depth guidance, and multiple stages of high-resolution upscaling.🧠 Core Components:1. Model and LoRAUNET: flux_dev.safetensors (fp8)CLIP: clip_l.safetensors + t5xxl_fp16.safetensorsLoRA: RAW_Photography.safetensors (enabled at 100%) Model page here: https://civitai.com/models/1639720LoRA is integrated via the Power Lora Loader (rgthree) node.2. Prompting and DetailingBilboXPhotoPrompt is used to generate rich photo-style prompts.A separate prompt detailer enhances:high-frequency surface texture,edge and contour definition,natural gradients,depth and occlusion cues,realistic material response,and framing clarity.FluxGuidance is applied with guidance: 3.5 to boost semantic conditioning.3. Depth Guidance with ControlNetPreprocessor: MiDaS-DepthMapPreprocessor at 1280px resolutionControlNet: flux-depth-controlnet-v3.safetensorsUsed via ControlNetApplyAdvanced with settings:Strength: 0.6Start: 0.0End: 0.618This ensures stable structure and perspective.4. Detail Enhancement via DetailDaemonThree progressive sampling stages, each wrapped with a DetailDaemonSamplerNode using:detail_amount: 0.62start/end: 1bias: 0.62smooth: trueSampling is handled through SamplerCustomAdvanced and BasicScheduler nodes using karras/beta scheduling and controlled denoising.5. Hi-Res UpscalingAfter each sampling pass, image resolution is progressively increased with:002_lightweightSR_DIV2K_s64w8_SwinIR-S_x2.pth001_classicalSR_DIV2K_s48w8_SwinIR-M_x4.pth1xSkinContrast-SuperUltraCompact.pth — for final skin cleanupAll upscaling is performed with precision to preserve microtexture and avoid artifacts.📸 Finalization and OutputResults are compared using the Image Comparer (rgthree) in "Slide" mode.Images are saved with SaveImage for both intermediate and final outputs.✅ Summary:Flux_RAW_HIRes_V2 is a highly optimized multi-stage workflow tailored for photorealism and ultra-fine detail. It combines:a RAW photography LoRA,progressive DetailDaemon refinement,ControlNet depth stabilization,SwinIR upscaling layers,and enhanced semantic conditioning.Ideal for portraits where skin detail, structure, and texture fidelity matter most.
HARUKI MIX Simple ZimageTurbo T2I Workflow + Color Correct + Face Detailer
Z-image Turbo用のsimpleなワークフローです
Seed Explorer, Z Image Turbo, optional ControlNet
ComfyUI Workflow for Seed exploration with Z Image TurboStreamline your GenAI image creation process with this efficient workflow designed for Z Image Turbo. One of the primary challenges when starting with Z Image Turbo is finding a suitable seed for further processing, such as upscaling. This workflow addresses this issue by generating up to six previews with random seeds, providing a diverse set of options to choose from.Key Features:Seed Generation: Automatically generates up to six different seeds and corresponding previews, ensuring a variety of starting points for your image creation process.LoRA Integration: Easily add and adjust your preferred LoRAs on the fly to tailor the image generation to your specific needs.ControlNet Option: Optionally enable ControlNet to gain finer control over the image generation process, allowing for more precise and customized outputs.This workflow is ideal for artists, designers, and creators looking to enhance their GenAI image creation pipeline with flexibility and efficiency. Whether you are a seasoned professional or just starting out, this workflow will help you achieve high-quality results with minimal effort.Requirements - Custom NodesRGThreehttps://github.com/rgthree/rgthree-comfyEasy Usehttps://github.com/yolain/ComfyUI-Easy-UseWAS Node Suitehttps://github.com/ltdrdata/was-node-suite-comfyuiResolution Masterhttps://github.com/Azornes/Comfyui-Resolution-MasterKJNodeshttps://github.com/kijai/ComfyUI-KJNodesSageUtilshttps://github.com/arcum42/ComfyUI_SageUtilsControlNet AUX (optional)https://github.com/Fannovel16/comfyui_controlnet_aux
Z-Image Turbo + (High / Low VRAM) CivitAI metadata + 4k Upscaling + Detail Daemon
Standard Z-Image turbo text to image workflow.Some people have asked if I have a more simple workflow for ZIT with good results as the more complete one, well... here it is. This one is faster than the full version since it uses a standard LoRA pipe line instead of hooks.✅ You can delete the "CivitAI metadata" group if you are not planning on using it, the workflow can work perfectly fine without it, just be sure to enable the "Save image node" to enable standard image saving.✅ This workflow supports both standard and GGUF models. You can delete one model or the other if you won't use it, the workflow will work perfectly fine just with one.Main FeaturesSimple generationStandard Z-Image Turbo image generationDetail daemonStandard detail control using Detail Daemon.4K UpscalerUltra high-quality upscaling with SeedVR2.CivitAI metadataCustom metadata nodes allow you to seamlessly upload content to CivitAI without having to manually input generation data. It registers the model, sampling settings and LoRAs you use.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
BF95 ZIT Diffusion USDU Workflow for ComfyUI
This is a basic Z-Image Turbo workflow that I use. It works for any Diffusion_Model version of Z-Image Turbo, with the model suggested VAE and Text Encoder. I've kept the required nodes to a minimum to get you started.
BF95 ZIT AIO Checkpoint for Civitai. A ComfyUI Workflow
Looking for a simple workflow to get started with Z-Image Turbo in ComfyUI? This has an optional upscale pass group, and is based on an all in one Checkpoint style model. I've kept the required node pack down to a minimum.
ComfyUI - Z-Image Turbo - Easy to use, Various styles - Lora Manager + Triggers (By: Rafaelldestilo)
This is a workflow I developed entirely for my own use and have been improving for better experience and practicality.It includes the LoRa Loader, where you simply select the LoRa image using the LoRa Manager. The image already comes in the correct size and with the activation keys synchronized by Civitai; only the size needs to be configured separately. In my opinion, it's the best LoRa selector currently available.It includes the Style Selector for cat-shaped images, similar to Focus Styles, where you simply select the corresponding cat and the style is applied to the image with 275 styles.I've included two positive prompts; simply disable the Bypass of the second to manually apply a style to multiple prompts in the main prompt. When changing prompt 1, the style, camera angles, etc., of prompt 2 will be applied.Includes an image aspect selector (Select only 1 at a time)Sage Attention PatchSeedVarianceEnchancerIt is compatible with the Sage Attention Patch to disable Bypass, improving generation time for those who have the Sage Attention Patch.Includes SeedVarianceEnchancer. Simply disable Bypass to get more variation in the generated images.It's a practical workflow for any generation. Set up your LoRa files in the LoRa Loader, saving your favorites. Just hover over them and the cover image will appear synchronized with Civitate. Simply activate the LoRa file; the activation key is automatically activated.I decided to share this workflow because I've been improving it since the release of the Z Image Turbo model and I always use it. I hope you like it.
Simple Z Image Turbo Workflow
A relatively simple Z Image Turbo workflow with 2 stage KSampler, Detailers and Upscale. Nothing particularly special, just something I've been using to generate images and uploading it here as a backup.
Anchor Workflow ZImage Turbo
Since there was interest, I'm posting this workflow.What it does:This workflow is intended to create multiple scenes for the same character. The workflow generates a reference scene / character. Based on this scene more generations are possible.! Keep in mind: this workflow is experimental and its not guaranteed to work ! It works somehow, but it comes with a big speed penalty (around 4x).How to use:Select your model / clip / vae.The workflow has three positive prompt nodes. Example is in the workflow.1st one for the main description. Place your character description in there. This prompt is in all gens present.2nd one for the reference image. Describe the scene for the reference image.3rd one for the new scene. Describe the new scene here.Write the prompts idealy with names: "Samuel is a 25 year old men. Samuel is wearing a blue colored jacket." or "Samuel is standing in a crowded city. Background shows shops and signs."For new scenes, add to the new scene prompt (3rd one) a good and detailed background description. If not, the workflow will more likely drift into the scene of the reference image.Seeds are fixed, so you can create multiple new scenes, without changing the reference image.Reference image should be idealy prompted for close-ups. More face -> More likely character consistencyThere are three active preview windows: Reference image, New scene image and a new scene image without the anchors (for comparison). You can deactivate it with ctrl + b, if you dont want gens for this lane. The same goes for new scene image. Deactivate it, if you want to roll for a reference character, without starting the new scene image.What is happeningThis node builds side anchors and inpaints the center image in a multistage workflowFuture UpdatesThere are room for improvements. Making it faster, simpler and more reliable are the main goals. Workflows for SD15 are also possible.
Z-Image / Z-Image-Turbo + Lora T2I
Two workspaces for Comfy UI1. Custom clip2. Standard clip
high-resolution image . z image turbo
Workflow Title: Turbo-Photorealism Sovereign EngineOverviewThis is a high-end, professional ComfyUI workflow meticulously engineered for ultra-high-definition photorealism. Powered by the Z-Image Turbo architecture, this pipeline is designed to bypass the "synthetic" look of standard AI, delivering raw, lifelike imagery that is indistinguishable from professional cinematography or high-end photography.Technical ArchitectureCore Inference: Optimized for Z-Image Turbo models, achieving elite-level detail within minimal sampling steps, ensuring maximum efficiency without compromising texture density.Optics & Light Physics: Advanced node integration for simulating real-world light refraction and skin subsurface scattering. This eliminates the "plastic" effect found in filtered, commercial models.Pro-Level Image Degradation: Instead of artificial smoothing, this workflow introduces "Analog Imperfections" (natural film grain, lens chromatic aberration, and sensor noise) to replicate the output of a physical DSLR camera.High-Resolution Scaling: A custom upscaling chain designed for 4K+ output, maintaining structural integrity and micro-details such as skin pores and fabric weaves.Philosophy & DesignThis workflow is built for creators who demand Absolute Creative Sovereignty. It operates on a technical logic that respects the user's expertise, free from the restrictive filters and "moral" handicaps imposed by corporate AI platforms. It is a tool for High-Fidelity Visual Documentation, prioritizing raw truth and artistic freedom.Key FeaturesZero-Filter Freedom: Full control over every pixel without corporate interference.Hardware Optimized: Specifically tuned to maximize the performance of high-RAM/VRAM local setups for rapid, high-res iteration.Cinematic Textures: Focuses on natural shadows and authentic highlights to ensure the results never look "AI-generated."
ZImage Turbo Essentials workflow
Note: If you have ANY issues with nodes not downloading, read the notes or reach out. There's nothing that special about any of them that aren't core modules.⛔⚠️🛑✋ Read the notes completly before using. Most common install and node problems are listed in the directions也有中文说明This model is set up for Z-Image Turbo. I have a seperate model for BaseA 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-breakdownThis is a modification to my advanced workflow designed for beginners:Checkpoint option for merged modelsSeed variance enhancerUltimate Upscaler for prescaling and hi res fixDetailing Suite for face, hands, & eyesEssentials Post Production SuiteSmart Noise scrubberFull tutorial available.也有中文说明Check out my other models: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 😉
Z Image Extreme Aesthetics Text to Image (NSFW)
A workflow uses the latest Z Image loras (FDPO, Skin Texture...) to Generate High Quality ImagesOther model details are within the workflow.Try the workflow online for free: https://www.runninghub.ai/post/2028999747696070657https://www.runninghub.ai/post/2042835007500197890Please check the video tutorial before you use this workflow: https://rumble.com/v76v5li-comfyui-z-image-extreme-asthetics.html
Z-Image Power Nodes
Z-Image Power Nodes is a collection of nodes designed specifically for the Z-Image / Z-Image Turbo model. They are based on some ideas and discoveries I made while developing the Amazing Z-Image Workflow.This page includes example workflows that utilize these nodes. The nodes can be installed via ComfyUI-Manager or manually from the GitHub repository. For more information, please visit:https://github.com/martin-rizzo/ComfyUI-ZImagePowerNodesThese are the nodes currently included in the project:⚡Z-Sampler Turbo ^G2A specialized sampler designed for Z-Image Turbo that achieves sufficient quality to eliminate the need for further post-processing.⚡Style Prompt EncoderApplies a selected visual styles to your prompt and encodes both of them using a text-encoder model (clip).⚡Style String InjectorSeamlessly integrates a chosen style into your prompt text. It accepts a string as input and modifies it based on the selected style.⚡My Top-10 StylesAllows you to create a list of favorite styles for quick selection of your most used ones.⚡VAE Encode (for Soft Inpainting)Encodes images into a latent representation, embedding the mask that indicates where inpainting will be applied.⚡Save ImageSaves generated images with the option to embed CivitAI-compatible metadata, making it easy to share generation parameters through that platform.⚡Empty Z-Image Latent ImageCreates an empty latent image of the appropriate size for Z-Image, selecting aspect ratio, scale, and orientation.
DBZit2Klein with Swap
ComfyUI WorkflowRequires the latest Version of: ComfyUI_Eclipse v3.3+This workflow is a heavily modified version of the original workflow by AIMetatron (DarkBeast). Since only the design remains from the original workflow, I’m just uploading it here because, in essence, it’s not even a merge anymore. All Nodes have been placed 1 by 1 into a new Workflow.The workflow is straightforward: select model/clip/vae (Zit/F2Klein), activate the groups you want to use in the fast muter, and get started. Everything you need to see is on the left side. Flux2 tend to be too bright and lack contrast/saturation, for that reason i've added a Color Match node in this group. Apart from this, no post happens to the image in this workflow.I've reduced the colors of the Nodes by a lot so you can see the important one's easily.In the tests a face crop worked better most of the times. if that doesnt work for you bypass the crop node. Also, the BFS1 Upscale Factor shouldn't be too high, it works better around 1.5 (it uses the size that you have set in the Smart Folder node * factor)
Z-image Turbo (AIO or Split) - Detailer + SeedVR2.5.24
The workflow can be used for both versions of z-image (base and turbo). V1 has Detail Daemon and SeedVR2.5.24 are used for better detail and upscaling.The reworked version is split-only and detail daemon has been replaced by FaceDetailer. Added prompt enhancer (QwenVL) and FaceSwap.
Z-Image Turbo | Detailer(s) | Dual Sampling | Upscaling
Z-Image (Turbo) with Ollama prompt enhancerSimple prompt is fed to an Ollama model for feature rich enhancing.Starts with standard Z-Image Turbo image generation, passed to a secondary multi-pass sampler.Selectable detailers (face, hands, skin etc).Allows selection of a normal or advanced (SeedVR2) upscaler (or disable upscaling)