[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-Comfy-Org--ComfyUI":3,"tool-Comfy-Org--ComfyUI":64},[4,17,27,35,48,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",138956,2,"2026-04-05T11:33:21",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,43,44,45,15,46,26,13,47],"数据工具","视频","插件","其他","音频",{"id":49,"name":50,"github_repo":51,"description_zh":52,"stars":53,"difficulty_score":10,"last_commit_at":54,"category_tags":55,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,46],{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},519,"PaddleOCR","PaddlePaddle\u002FPaddleOCR","PaddleOCR 是一款基于百度飞桨框架开发的高性能开源光学字符识别工具包。它的核心能力是将图片、PDF 等文档中的文字提取出来，转换成计算机可读取的结构化数据，让机器真正“看懂”图文内容。\n\n面对海量纸质或电子文档，PaddleOCR 解决了人工录入效率低、数字化成本高的问题。尤其在人工智能领域，它扮演着连接图像与大型语言模型（LLM）的桥梁角色，能将视觉信息直接转化为文本输入，助力智能问答、文档分析等应用场景落地。\n\nPaddleOCR 适合开发者、算法研究人员以及有文档自动化需求的普通用户。其技术优势十分明显：不仅支持全球 100 多种语言的识别，还能在 Windows、Linux、macOS 等多个系统上运行，并灵活适配 CPU、GPU、NPU 等各类硬件。作为一个轻量级且社区活跃的开源项目，PaddleOCR 既能满足快速集成的需求，也能支撑前沿的视觉语言研究，是处理文字识别任务的理想选择。",74913,"2026-04-05T10:44:17",[26,14,13,46],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":79,"owner_email":80,"owner_twitter":81,"owner_website":82,"owner_url":83,"languages":84,"stars":100,"forks":101,"last_commit_at":102,"license":103,"difficulty_score":23,"env_os":104,"env_gpu":105,"env_ram":106,"env_deps":107,"category_tags":121,"github_topics":122,"view_count":128,"oss_zip_url":79,"oss_zip_packed_at":79,"status":16,"created_at":129,"updated_at":130,"faqs":131,"releases":160},2271,"Comfy-Org\u002FComfyUI","ComfyUI","The most powerful and modular diffusion model GUI, api and backend with a graph\u002Fnodes interface.","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。","\u003Cdiv align=\"center\">\n\n# ComfyUI\n**The most powerful and modular visual AI engine and application.**\n\n\n[![Website][website-shield]][website-url]\n[![Dynamic JSON Badge][discord-shield]][discord-url]\n[![Twitter][twitter-shield]][twitter-url]\n[![Matrix][matrix-shield]][matrix-url]\n\u003Cbr>\n[![][github-release-shield]][github-release-link]\n[![][github-release-date-shield]][github-release-link]\n[![][github-downloads-shield]][github-downloads-link]\n[![][github-downloads-latest-shield]][github-downloads-link]\n\n[matrix-shield]: https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FMatrix-000000?style=flat&logo=matrix&logoColor=white\n[matrix-url]: https:\u002F\u002Fapp.element.io\u002F#\u002Froom\u002F%23comfyui_space%3Amatrix.org\n[website-shield]: https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FComfyOrg-4285F4?style=flat\n[website-url]: https:\u002F\u002Fwww.comfy.org\u002F\n\u003C!-- Workaround to display total user from https:\u002F\u002Fgithub.com\u002Fbadges\u002Fshields\u002Fissues\u002F4500#issuecomment-2060079995 -->\n[discord-shield]: https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdynamic\u002Fjson?url=https%3A%2F%2Fdiscord.com%2Fapi%2Finvites%2Fcomfyorg%3Fwith_counts%3Dtrue&query=%24.approximate_member_count&logo=discord&logoColor=white&label=Discord&color=green&suffix=%20total\n[discord-url]: https:\u002F\u002Fwww.comfy.org\u002Fdiscord\n[twitter-shield]: https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Ffollow\u002FComfyUI\n[twitter-url]: https:\u002F\u002Fx.com\u002FComfyUI\n\n[github-release-shield]: https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fv\u002Frelease\u002Fcomfyanonymous\u002FComfyUI?style=flat&sort=semver\n[github-release-link]: https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Freleases\n[github-release-date-shield]: https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Frelease-date\u002Fcomfyanonymous\u002FComfyUI?style=flat\n[github-downloads-shield]: https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fdownloads\u002Fcomfyanonymous\u002FComfyUI\u002Ftotal?style=flat\n[github-downloads-latest-shield]: https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fdownloads\u002Fcomfyanonymous\u002FComfyUI\u002Flatest\u002Ftotal?style=flat&label=downloads%40latest\n[github-downloads-link]: https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Freleases\n\n![ComfyUI Screenshot](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FComfy-Org_ComfyUI_readme_dfe3a8cb4502.png)\n\u003C\u002Fdiv>\n\nComfyUI lets you design and execute advanced stable diffusion pipelines using a graph\u002Fnodes\u002Fflowchart based interface. Available on Windows, Linux, and macOS.\n\n## Get Started\n\n### Local\n\n#### [Desktop Application](https:\u002F\u002Fwww.comfy.org\u002Fdownload)\n- The easiest way to get started.\n- Available on Windows & macOS.\n\n#### [Windows Portable Package](#installing)\n- Get the latest commits and completely portable.\n- Available on Windows.\n\n#### [Manual Install](#manual-install-windows-linux)\nSupports all operating systems and GPU types (NVIDIA, AMD, Intel, Apple Silicon, Ascend).\n\n### Cloud\n\n#### [Comfy Cloud](https:\u002F\u002Fwww.comfy.org\u002Fcloud)\n- Our official paid cloud version for those who can't afford local hardware.\n\n## Examples\nSee what ComfyUI can do with the [newer template workflows](https:\u002F\u002Fcomfy.org\u002Fworkflows) or old [example workflows](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002F).\n\n## Features\n- Nodes\u002Fgraph\u002Fflowchart interface to experiment and create complex Stable Diffusion workflows without needing to code anything.\n- NOTE: There are many more models supported than the list below, if you want to see what is supported see our templates list inside ComfyUI.\n- Image Models\n   - SD1.x, SD2.x ([unCLIP](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Funclip\u002F))\n   - [SDXL](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fsdxl\u002F), [SDXL Turbo](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fsdturbo\u002F)\n   - [Stable Cascade](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fstable_cascade\u002F)\n   - [SD3 and SD3.5](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fsd3\u002F)\n   - Pixart Alpha and Sigma\n   - [AuraFlow](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Faura_flow\u002F)\n   - [HunyuanDiT](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fhunyuan_dit\u002F)\n   - [Flux](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fflux\u002F)\n   - [Lumina Image 2.0](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Flumina2\u002F)\n   - [HiDream](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fhidream\u002F)\n   - [Qwen Image](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fqwen_image\u002F)\n   - [Hunyuan Image 2.1](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fhunyuan_image\u002F)\n   - [Flux 2](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fflux2\u002F)\n   - [Z Image](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fz_image\u002F)\n- Image Editing Models\n   - [Omnigen 2](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fomnigen\u002F)\n   - [Flux Kontext](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fflux\u002F#flux-kontext-image-editing-model)\n   - [HiDream E1.1](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fhidream\u002F#hidream-e11)\n   - [Qwen Image Edit](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fqwen_image\u002F#edit-model)\n- Video Models\n   - [Stable Video Diffusion](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fvideo\u002F)\n   - [Mochi](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fmochi\u002F)\n   - [LTX-Video](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fltxv\u002F)\n   - [Hunyuan Video](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fhunyuan_video\u002F)\n   - [Wan 2.1](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fwan\u002F)\n   - [Wan 2.2](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fwan22\u002F)\n   - [Hunyuan Video 1.5](https:\u002F\u002Fdocs.comfy.org\u002Ftutorials\u002Fvideo\u002Fhunyuan\u002Fhunyuan-video-1-5)\n- Audio Models\n   - [Stable Audio](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Faudio\u002F)\n   - [ACE Step](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Faudio\u002F)\n- 3D Models\n   - [Hunyuan3D 2.0](https:\u002F\u002Fdocs.comfy.org\u002Ftutorials\u002F3d\u002Fhunyuan3D-2)\n- Asynchronous Queue system\n- Many optimizations: Only re-executes the parts of the workflow that changes between executions.\n- Smart memory management: can automatically run large models on GPUs with as low as 1GB vram with smart offloading.\n- Works even if you don't have a GPU with: ```--cpu``` (slow)\n- Can load ckpt and safetensors: All in one checkpoints or standalone diffusion models, VAEs and CLIP models.\n- Safe loading of ckpt, pt, pth, etc.. files.\n- Embeddings\u002FTextual inversion\n- [Loras (regular, locon and loha)](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Flora\u002F)\n- [Hypernetworks](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fhypernetworks\u002F)\n- Loading full workflows (with seeds) from generated PNG, WebP and FLAC files.\n- Saving\u002FLoading workflows as Json files.\n- Nodes interface can be used to create complex workflows like one for [Hires fix](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002F2_pass_txt2img\u002F) or much more advanced ones.\n- [Area Composition](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Farea_composition\u002F)\n- [Inpainting](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Finpaint\u002F) with both regular and inpainting models.\n- [ControlNet and T2I-Adapter](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fcontrolnet\u002F)\n- [Upscale Models (ESRGAN, ESRGAN variants, SwinIR, Swin2SR, etc...)](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fupscale_models\u002F)\n- [GLIGEN](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fgligen\u002F)\n- [Model Merging](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fmodel_merging\u002F)\n- [LCM models and Loras](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Flcm\u002F)\n- Latent previews with [TAESD](#how-to-show-high-quality-previews)\n- Works fully offline: core will never download anything unless you want to.\n- Optional API nodes to use paid models from external providers through the online [Comfy API](https:\u002F\u002Fdocs.comfy.org\u002Ftutorials\u002Fapi-nodes\u002Foverview) disable with: `--disable-api-nodes`\n- [Config file](extra_model_paths.yaml.example) to set the search paths for models.\n\nWorkflow examples can be found on the [Examples page](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002F)\n\n## Release Process\n\nComfyUI follows a weekly release cycle targeting Monday but this regularly changes because of model releases or large changes to the codebase. There are three interconnected repositories:\n\n1. **[ComfyUI Core](https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI)**\n   - Releases a new stable version (e.g., v0.7.0) roughly every week.\n   - Starting from v0.4.0 patch versions will be used for fixes backported onto the current stable release.\n   - Minor versions will be used for releases off the master branch.\n   - Patch versions may still be used for releases on the master branch in cases where a backport would not make sense.\n   - Commits outside of the stable release tags may be very unstable and break many custom nodes.\n   - Serves as the foundation for the desktop release\n\n2. **[ComfyUI Desktop](https:\u002F\u002Fgithub.com\u002FComfy-Org\u002Fdesktop)**\n   - Builds a new release using the latest stable core version\n\n3. **[ComfyUI Frontend](https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI_frontend)**\n   - Weekly frontend updates are merged into the core repository\n   - Features are frozen for the upcoming core release\n   - Development continues for the next release cycle\n\n## Shortcuts\n\n| Keybind                            | Explanation                                                                                                        |\n|------------------------------------|--------------------------------------------------------------------------------------------------------------------|\n| `Ctrl` + `Enter`                      | Queue up current graph for generation                                                                              |\n| `Ctrl` + `Shift` + `Enter`              | Queue up current graph as first for generation                                                                     |\n| `Ctrl` + `Alt` + `Enter`                | Cancel current generation                                                                                          |\n| `Ctrl` + `Z`\u002F`Ctrl` + `Y`                 | Undo\u002FRedo                                                                                                          |\n| `Ctrl` + `S`                          | Save workflow                                                                                                      |\n| `Ctrl` + `O`                          | Load workflow                                                                                                      |\n| `Ctrl` + `A`                          | Select all nodes                                                                                                   |\n| `Alt `+ `C`                           | Collapse\u002Funcollapse selected nodes                                                                                 |\n| `Ctrl` + `M`                          | Mute\u002Funmute selected nodes                                                                                         |\n| `Ctrl` + `B`                           | Bypass selected nodes (acts like the node was removed from the graph and the wires reconnected through)            |\n| `Delete`\u002F`Backspace`                   | Delete selected nodes                                                                                              |\n| `Ctrl` + `Backspace`                   | Delete the current graph                                                                                           |\n| `Space`                              | Move the canvas around when held and moving the cursor                                                             |\n| `Ctrl`\u002F`Shift` + `Click`                 | Add clicked node to selection                                                                                      |\n| `Ctrl` + `C`\u002F`Ctrl` + `V`                  | Copy and paste selected nodes (without maintaining connections to outputs of unselected nodes)                     |\n| `Ctrl` + `C`\u002F`Ctrl` + `Shift` + `V`          | Copy and paste selected nodes (maintaining connections from outputs of unselected nodes to inputs of pasted nodes) |\n| `Shift` + `Drag`                       | Move multiple selected nodes at the same time                                                                      |\n| `Ctrl` + `D`                           | Load default graph                                                                                                 |\n| `Alt` + `+`                          | Canvas Zoom in                                                                                                     |\n| `Alt` + `-`                          | Canvas Zoom out                                                                                                    |\n| `Ctrl` + `Shift` + LMB + Vertical drag | Canvas Zoom in\u002Fout                                                                                                 |\n| `P`                                  | Pin\u002FUnpin selected nodes                                                                                           |\n| `Ctrl` + `G`                           | Group selected nodes                                                                                               |\n| `Q`                                 | Toggle visibility of the queue                                                                                     |\n| `H`                                  | Toggle visibility of history                                                                                       |\n| `R`                                  | Refresh graph                                                                                                      |\n| `F`                                  | Show\u002FHide menu                                                                                                      |\n| `.`                                  | Fit view to selection (Whole graph when nothing is selected)                                                        |\n| Double-Click LMB                   | Open node quick search palette                                                                                     |\n| `Shift` + Drag                       | Move multiple wires at once                                                                                        |\n| `Ctrl` + `Alt` + LMB                   | Disconnect all wires from clicked slot                                                                             |\n\n`Ctrl` can also be replaced with `Cmd` instead for macOS users\n\n# Installing\n\n## Windows Portable\n\nThere is a portable standalone build for Windows that should work for running on Nvidia GPUs or for running on your CPU only on the [releases page](https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Freleases).\n\n### [Direct link to download](https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Freleases\u002Flatest\u002Fdownload\u002FComfyUI_windows_portable_nvidia.7z)\n\nSimply download, extract with [7-Zip](https:\u002F\u002F7-zip.org) or with the windows explorer on recent windows versions and run. For smaller models you normally only need to put the checkpoints (the huge ckpt\u002Fsafetensors files) in: ComfyUI\\models\\checkpoints but many of the larger models have multiple files. Make sure to follow the instructions to know which subfolder to put them in ComfyUI\\models\\\n\nIf you have trouble extracting it, right click the file -> properties -> unblock\n\nThe portable above currently comes with python 3.13 and pytorch cuda 13.0. Update your Nvidia drivers if it doesn't start.\n\n#### Alternative Downloads:\n\n[Experimental portable for AMD GPUs](https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Freleases\u002Flatest\u002Fdownload\u002FComfyUI_windows_portable_amd.7z)\n\n[Portable with pytorch cuda 12.6 and python 3.12](https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Freleases\u002Flatest\u002Fdownload\u002FComfyUI_windows_portable_nvidia_cu126.7z) (Supports Nvidia 10 series and older GPUs).\n\n#### How do I share models between another UI and ComfyUI?\n\nSee the [Config file](extra_model_paths.yaml.example) to set the search paths for models. In the standalone windows build you can find this file in the ComfyUI directory. Rename this file to extra_model_paths.yaml and edit it with your favorite text editor.\n\n\n## [comfy-cli](https:\u002F\u002Fdocs.comfy.org\u002Fcomfy-cli\u002Fgetting-started)\n\nYou can install and start ComfyUI using comfy-cli:\n```bash\npip install comfy-cli\ncomfy install\n```\n\n## Manual Install (Windows, Linux)\n\nPython 3.14 works but some custom nodes may have issues. The free threaded variant works but some dependencies will enable the GIL so it's not fully supported.\n\nPython 3.13 is very well supported. If you have trouble with some custom node dependencies on 3.13 you can try 3.12\n\ntorch 2.4 and above is supported but some features and optimizations might only work on newer versions. We generally recommend using the latest major version of pytorch with the latest cuda version unless it is less than 2 weeks old.\n\n### Instructions:\n\nGit clone this repo.\n\nPut your SD checkpoints (the huge ckpt\u002Fsafetensors files) in: models\u002Fcheckpoints\n\nPut your VAE in: models\u002Fvae\n\n\n### AMD GPUs (Linux)\n\nAMD users can install rocm and pytorch with pip if you don't have it already installed, this is the command to install the stable version:\n\n```pip install torch torchvision torchaudio --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Frocm7.2```\n\nThis is the command to install the nightly with ROCm 7.2 which might have some performance improvements:\n\n```pip install --pre torch torchvision torchaudio --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fnightly\u002Frocm7.2```\n\n\n### AMD GPUs (Experimental: Windows and Linux), RDNA 3, 3.5 and 4 only.\n\nThese have less hardware support than the builds above but they work on windows. You also need to install the pytorch version specific to your hardware.\n\nRDNA 3 (RX 7000 series):\n\n```pip install --pre torch torchvision torchaudio --index-url https:\u002F\u002Frocm.nightlies.amd.com\u002Fv2\u002Fgfx110X-all\u002F```\n\nRDNA 3.5 (Strix halo\u002FRyzen AI Max+ 365):\n\n```pip install --pre torch torchvision torchaudio --index-url https:\u002F\u002Frocm.nightlies.amd.com\u002Fv2\u002Fgfx1151\u002F```\n\nRDNA 4 (RX 9000 series):\n\n```pip install --pre torch torchvision torchaudio --index-url https:\u002F\u002Frocm.nightlies.amd.com\u002Fv2\u002Fgfx120X-all\u002F```\n\n### Intel GPUs (Windows and Linux)\n\nIntel Arc GPU users can install native PyTorch with torch.xpu support using pip. More information can be found [here](https:\u002F\u002Fpytorch.org\u002Fdocs\u002Fmain\u002Fnotes\u002Fget_start_xpu.html)\n\n1. To install PyTorch xpu, use the following command:\n\n```pip install torch torchvision torchaudio --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fxpu```\n\nThis is the command to install the Pytorch xpu nightly which might have some performance improvements:\n\n```pip install --pre torch torchvision torchaudio --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fnightly\u002Fxpu```\n\n### NVIDIA\n\nNvidia users should install stable pytorch using this command:\n\n```pip install torch torchvision torchaudio --extra-index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu130```\n\nThis is the command to install pytorch nightly instead which might have performance improvements.\n\n```pip install --pre torch torchvision torchaudio --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fnightly\u002Fcu130```\n\n#### Troubleshooting\n\nIf you get the \"Torch not compiled with CUDA enabled\" error, uninstall torch with:\n\n```pip uninstall torch```\n\nAnd install it again with the command above.\n\n### Dependencies\n\nInstall the dependencies by opening your terminal inside the ComfyUI folder and:\n\n```pip install -r requirements.txt```\n\nAfter this you should have everything installed and can proceed to running ComfyUI.\n\n### Others:\n\n#### Apple Mac silicon\n\nYou can install ComfyUI in Apple Mac silicon (M1 or M2) with any recent macOS version.\n\n1. Install pytorch nightly. For instructions, read the [Accelerated PyTorch training on Mac](https:\u002F\u002Fdeveloper.apple.com\u002Fmetal\u002Fpytorch\u002F) Apple Developer guide (make sure to install the latest pytorch nightly).\n1. Follow the [ComfyUI manual installation](#manual-install-windows-linux) instructions for Windows and Linux.\n1. Install the ComfyUI [dependencies](#dependencies). If you have another Stable Diffusion UI [you might be able to reuse the dependencies](#i-already-have-another-ui-for-stable-diffusion-installed-do-i-really-have-to-install-all-of-these-dependencies).\n1. Launch ComfyUI by running `python main.py`\n\n> **Note**: Remember to add your models, VAE, LoRAs etc. to the corresponding Comfy folders, as discussed in [ComfyUI manual installation](#manual-install-windows-linux).\n\n#### Ascend NPUs\n\nFor models compatible with Ascend Extension for PyTorch (torch_npu). To get started, ensure your environment meets the prerequisites outlined on the [installation](https:\u002F\u002Fascend.github.io\u002Fdocs\u002Fsources\u002Fascend\u002Fquick_install.html) page. Here's a step-by-step guide tailored to your platform and installation method:\n\n1. Begin by installing the recommended or newer kernel version for Linux as specified in the Installation page of torch-npu, if necessary.\n2. Proceed with the installation of Ascend Basekit, which includes the driver, firmware, and CANN, following the instructions provided for your specific platform.\n3. Next, install the necessary packages for torch-npu by adhering to the platform-specific instructions on the [Installation](https:\u002F\u002Fascend.github.io\u002Fdocs\u002Fsources\u002Fpytorch\u002Finstall.html#pytorch) page.\n4. Finally, adhere to the [ComfyUI manual installation](#manual-install-windows-linux) guide for Linux. Once all components are installed, you can run ComfyUI as described earlier.\n\n#### Cambricon MLUs\n\nFor models compatible with Cambricon Extension for PyTorch (torch_mlu). Here's a step-by-step guide tailored to your platform and installation method:\n\n1. Install the Cambricon CNToolkit by adhering to the platform-specific instructions on the [Installation](https:\u002F\u002Fwww.cambricon.com\u002Fdocs\u002Fsdk_1.15.0\u002Fcntoolkit_3.7.2\u002Fcntoolkit_install_3.7.2\u002Findex.html)\n2. Next, install the PyTorch(torch_mlu) following the instructions on the [Installation](https:\u002F\u002Fwww.cambricon.com\u002Fdocs\u002Fsdk_1.15.0\u002Fcambricon_pytorch_1.17.0\u002Fuser_guide_1.9\u002Findex.html)\n3. Launch ComfyUI by running `python main.py`\n\n#### Iluvatar Corex\n\nFor models compatible with Iluvatar Extension for PyTorch. Here's a step-by-step guide tailored to your platform and installation method:\n\n1. Install the Iluvatar Corex Toolkit by adhering to the platform-specific instructions on the [Installation](https:\u002F\u002Fsupport.iluvatar.com\u002F#\u002FDocumentCentre?id=1&nameCenter=2&productId=520117912052801536)\n2. Launch ComfyUI by running `python main.py`\n\n\n## [ComfyUI-Manager](https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI-Manager\u002Ftree\u002Fmanager-v4)\n\n**ComfyUI-Manager** is an extension that allows you to easily install, update, and manage custom nodes for ComfyUI.\n\n### Setup\n\n1. Install the manager dependencies:\n   ```bash\n   pip install -r manager_requirements.txt\n   ```\n\n2. Enable the manager with the `--enable-manager` flag when running ComfyUI:\n   ```bash\n   python main.py --enable-manager\n   ```\n\n### Command Line Options\n\n| Flag | Description |\n|------|-------------|\n| `--enable-manager` | Enable ComfyUI-Manager |\n| `--enable-manager-legacy-ui` | Use the legacy manager UI instead of the new UI (requires `--enable-manager`) |\n| `--disable-manager-ui` | Disable the manager UI and endpoints while keeping background features like security checks and scheduled installation completion (requires `--enable-manager`) |\n\n\n# Running\n\n```python main.py```\n\n### For AMD cards not officially supported by ROCm\n\nTry running it with this command if you have issues:\n\nFor 6700, 6600 and maybe other RDNA2 or older: ```HSA_OVERRIDE_GFX_VERSION=10.3.0 python main.py```\n\nFor AMD 7600 and maybe other RDNA3 cards: ```HSA_OVERRIDE_GFX_VERSION=11.0.0 python main.py```\n\n### AMD ROCm Tips\n\nYou can enable experimental memory efficient attention on recent pytorch in ComfyUI on some AMD GPUs using this command, it should already be enabled by default on RDNA3. If this improves speed for you on latest pytorch on your GPU please report it so that I can enable it by default.\n\n```TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1 python main.py --use-pytorch-cross-attention```\n\nYou can also try setting this env variable `PYTORCH_TUNABLEOP_ENABLED=1` which might speed things up at the cost of a very slow initial run.\n\n# Notes\n\nOnly parts of the graph that have an output with all the correct inputs will be executed.\n\nOnly parts of the graph that change from each execution to the next will be executed, if you submit the same graph twice only the first will be executed. If you change the last part of the graph only the part you changed and the part that depends on it will be executed.\n\nDragging a generated png on the webpage or loading one will give you the full workflow including seeds that were used to create it.\n\nYou can use () to change emphasis of a word or phrase like: (good code:1.2) or (bad code:0.8). The default emphasis for () is 1.1. To use () characters in your actual prompt escape them like \\\\( or \\\\).\n\nYou can use {day|night}, for wildcard\u002Fdynamic prompts. With this syntax \"{wild|card|test}\" will be randomly replaced by either \"wild\", \"card\" or \"test\" by the frontend every time you queue the prompt. To use {} characters in your actual prompt escape them like: \\\\{ or \\\\}.\n\nDynamic prompts also support C-style comments, like `\u002F\u002F comment` or `\u002F* comment *\u002F`.\n\nTo use a textual inversion concepts\u002Fembeddings in a text prompt put them in the models\u002Fembeddings directory and use them in the CLIPTextEncode node like this (you can omit the .pt extension):\n\n```embedding:embedding_filename.pt```\n\n\n## How to show high-quality previews?\n\nUse ```--preview-method auto``` to enable previews.\n\nThe default installation includes a fast latent preview method that's low-resolution. To enable higher-quality previews with [TAESD](https:\u002F\u002Fgithub.com\u002Fmadebyollin\u002Ftaesd), download the [taesd_decoder.pth, taesdxl_decoder.pth, taesd3_decoder.pth and taef1_decoder.pth](https:\u002F\u002Fgithub.com\u002Fmadebyollin\u002Ftaesd\u002F) and place them in the `models\u002Fvae_approx` folder. Once they're installed, restart ComfyUI and launch it with `--preview-method taesd` to enable high-quality previews.\n\n## How to use TLS\u002FSSL?\nGenerate a self-signed certificate (not appropriate for shared\u002Fproduction use) and key by running the command: `openssl req -x509 -newkey rsa:4096 -keyout key.pem -out cert.pem -sha256 -days 3650 -nodes -subj \"\u002FC=XX\u002FST=StateName\u002FL=CityName\u002FO=CompanyName\u002FOU=CompanySectionName\u002FCN=CommonNameOrHostname\"`\n\nUse `--tls-keyfile key.pem --tls-certfile cert.pem` to enable TLS\u002FSSL, the app will now be accessible with `https:\u002F\u002F...` instead of `http:\u002F\u002F...`.\n\n> Note: Windows users can use [alexisrolland\u002Fdocker-openssl](https:\u002F\u002Fgithub.com\u002Falexisrolland\u002Fdocker-openssl) or one of the [3rd party binary distributions](https:\u002F\u002Fwiki.openssl.org\u002Findex.php\u002FBinaries) to run the command example above.\n\u003Cbr\u002F>\u003Cbr\u002F>If you use a container, note that the volume mount `-v` can be a relative path so `... -v \".\\:\u002Fopenssl-certs\" ...` would create the key & cert files in the current directory of your command prompt or powershell terminal.\n\n## Support and dev channel\n\n[Discord](https:\u002F\u002Fcomfy.org\u002Fdiscord): Try the #help or #feedback channels.\n\n[Matrix space: #comfyui_space:matrix.org](https:\u002F\u002Fapp.element.io\u002F#\u002Froom\u002F%23comfyui_space%3Amatrix.org) (it's like discord but open source).\n\nSee also: [https:\u002F\u002Fwww.comfy.org\u002F](https:\u002F\u002Fwww.comfy.org\u002F)\n\n## Frontend Development\n\nAs of August 15, 2024, we have transitioned to a new frontend, which is now hosted in a separate repository: [ComfyUI Frontend](https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI_frontend). This repository now hosts the compiled JS (from TS\u002FVue) under the `web\u002F` directory.\n\n### Reporting Issues and Requesting Features\n\nFor any bugs, issues, or feature requests related to the frontend, please use the [ComfyUI Frontend repository](https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI_frontend). This will help us manage and address frontend-specific concerns more efficiently.\n\n### Using the Latest Frontend\n\nThe new frontend is now the default for ComfyUI. However, please note:\n\n1. The frontend in the main ComfyUI repository is updated fortnightly.\n2. Daily releases are available in the separate frontend repository.\n\nTo use the most up-to-date frontend version:\n\n1. For the latest daily release, launch ComfyUI with this command line argument:\n\n   ```\n   --front-end-version Comfy-Org\u002FComfyUI_frontend@latest\n   ```\n\n2. For a specific version, replace `latest` with the desired version number:\n\n   ```\n   --front-end-version Comfy-Org\u002FComfyUI_frontend@1.2.2\n   ```\n\nThis approach allows you to easily switch between the stable fortnightly release and the cutting-edge daily updates, or even specific versions for testing purposes.\n\n### Accessing the Legacy Frontend\n\nIf you need to use the legacy frontend for any reason, you can access it using the following command line argument:\n\n```\n--front-end-version Comfy-Org\u002FComfyUI_legacy_frontend@latest\n```\n\nThis will use a snapshot of the legacy frontend preserved in the [ComfyUI Legacy Frontend repository](https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI_legacy_frontend).\n\n# QA\n\n### Which GPU should I buy for this?\n\n[See this page for some recommendations](https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Fwiki\u002FWhich-GPU-should-I-buy-for-ComfyUI)\n","\u003Cdiv align=\"center\">\n\n# ComfyUI\n**功能最强大、模块化程度最高的视觉AI引擎与应用。**\n\n\n[![官网][website-shield]][website-url]\n[![动态JSON徽章][discord-shield]][discord-url]\n[![Twitter][twitter-shield]][twitter-url]\n[![Matrix][matrix-shield]][matrix-url]\n\u003Cbr>\n[![][github-release-shield]][github-release-link]\n[![][github-release-date-shield]][github-release-link]\n[![][github-downloads-shield]][github-downloads-link]\n[![][github-downloads-latest-shield]][github-downloads-link]\n\n[matrix-shield]: https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FMatrix-000000?style=flat&logo=matrix&logoColor=white\n[matrix-url]: https:\u002F\u002Fapp.element.io\u002F#\u002Froom\u002F%23comfyui_space%3Amatrix.org\n[website-shield]: https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FComfyOrg-4285F4?style=flat\n[website-url]: https:\u002F\u002Fwww.comfy.org\u002F\n\u003C!-- 用于显示总用户数的变通方法，源自 https:\u002F\u002Fgithub.com\u002Fbadges\u002Fshields\u002Fissues\u002F4500#issuecomment-2060079995 -->\n[discord-shield]: https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdynamic\u002Fjson?url=https%3A%2F%2Fdiscord.com%2Fapi%2Finvites%2Fcomfyorg%3Fwith_counts%3Dtrue&query=%24.approximate_member_count&logo=discord&logoColor=white&label=Discord&color=green&suffix=%20total\n[discord-url]: https:\u002F\u002Fwww.comfy.org\u002Fdiscord\n[twitter-shield]: https:\u002F\u002Fimg.shields.io\u002Ftwitter\u002Ffollow\u002FComfyUI\n[twitter-url]: https:\u002F\u002Fx.com\u002FComfyUI\n\n[github-release-shield]: https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fv\u002Frelease\u002Fcomfyanonymous\u002FComfyUI?style=flat&sort=semver\n[github-release-link]: https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Freleases\n[github-release-date-shield]: https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Frelease-date\u002Fcomfyanonymous\u002FComfyUI?style=flat\n[github-downloads-shield]: https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fdownloads\u002Fcomfyanonymous\u002FComfyUI\u002Ftotal?style=flat\n[github-downloads-latest-shield]: https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fdownloads\u002Fcomfyanonymous\u002FComfyUI\u002Flatest\u002Ftotal?style=flat&label=downloads%40latest\n[github-downloads-link]: https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Freleases\n\n![ComfyUI截图](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FComfy-Org_ComfyUI_readme_dfe3a8cb4502.png)\n\u003C\u002Fdiv>\n\nComfyUI允许您通过基于图、节点和流程图的界面来设计并执行先进的稳定扩散工作流。支持Windows、Linux和macOS系统。\n\n## 开始使用\n\n### 本地\n\n#### [桌面应用程序](https:\u002F\u002Fwww.comfy.org\u002Fdownload)\n- 最简单的入门方式。\n- 支持Windows和macOS。\n\n#### [Windows便携版](#installing)\n- 获取最新提交，完全便携。\n- 仅适用于Windows。\n\n#### [手动安装](#manual-install-windows-linux)\n支持所有操作系统及各类GPU（NVIDIA、AMD、Intel、Apple Silicon、Ascend）。\n\n### 云端\n\n#### [Comfy Cloud](https:\u002F\u002Fwww.comfy.org\u002Fcloud)\n- 我们的官方付费云版本，专为无法负担本地硬件的用户设计。\n\n## 示例\n请查看[新版模板工作流](https:\u002F\u002Fcomfy.org\u002Fworkflows)或旧版[示例工作流](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002F)，了解ComfyUI的强大功能。\n\n## 功能\n- 节点\u002F图\u002F流程图界面，无需编写任何代码即可实验和创建复杂的 Stable Diffusion 工作流。\n- 注意：支持的模型远不止下方列表所示，如需查看完整支持列表，请参阅 ComfyUI 内部的模板列表。\n- 图像模型\n   - SD1.x、SD2.x（[unCLIP](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Funclip\u002F)）\n   - [SDXL](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fsdxl\u002F)、[SDXL Turbo](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fsdturbo\u002F)\n   - [Stable Cascade](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fstable_cascade\u002F)\n   - [SD3 和 SD3.5](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fsd3\u002F)\n   - Pixart Alpha 和 Sigma\n   - [AuraFlow](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Faura_flow\u002F)\n   - [HunyuanDiT](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fhunyuan_dit\u002F)\n   - [Flux](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fflux\u002F)\n   - [Lumina Image 2.0](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Flumina2\u002F)\n   - [HiDream](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fhidream\u002F)\n   - [Qwen Image](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fqwen_image\u002F)\n   - [Hunyuan Image 2.1](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fhunyuan_image\u002F)\n   - [Flux 2](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fflux2\u002F)\n   - [Z Image](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fz_image\u002F)\n- 图像编辑模型\n   - [Omnigen 2](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fomnigen\u002F)\n   - [Flux Kontext](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fflux\u002F#flux-kontext-image-editing-model)\n   - [HiDream E1.1](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fhidream\u002F#hidream-e11)\n   - [Qwen Image Edit](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fqwen_image\u002F#edit-model)\n- 视频模型\n   - [Stable Video Diffusion](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fvideo\u002F)\n   - [Mochi](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fmochi\u002F)\n   - [LTX-Video](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fltxv\u002F)\n   - [Hunyuan Video](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fhunyuan_video\u002F)\n   - [Wan 2.1](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fwan\u002F)\n   - [Wan 2.2](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fwan22\u002F)\n   - [Hunyuan Video 1.5](https:\u002F\u002Fdocs.comfy.org\u002Ftutorials\u002Fvideo\u002Fhunyuan\u002Fhunyuan-video-1-5)\n- 音频模型\n   - [Stable Audio](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Faudio\u002F)\n   - [ACE Step](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Faudio\u002F)\n- 3D 模型\n   - [Hunyuan3D 2.0](https:\u002F\u002Fdocs.comfy.org\u002Ftutorials\u002F3d\u002Fhunyuan3D-2)\n- 异步队列系统\n- 多种优化：仅重新执行工作流中每次执行之间发生变化的部分。\n- 智能内存管理：可通过智能卸载技术，在显存低至 1GB 的 GPU 上自动运行大型模型。\n- 即使没有 GPU 也能运行：使用 ```--cpu``` 参数（速度较慢）。\n- 支持加载 ckpt 和 safetensors 文件：包括一体化检查点以及独立的扩散模型、VAE 和 CLIP 模型。\n- 安全加载 ckpt、pt、pth 等文件。\n- 嵌入\u002F文本反演\n- [LoRA（常规、locon 和 loha）](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Flora\u002F)\n- [超网络](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fhypernetworks\u002F)\n- 可从生成的 PNG、WebP 和 FLAC 文件中加载包含种子的完整工作流。\n- 支持将工作流保存为 Json 文件并加载。\n- 节点界面可用于创建复杂的工作流，例如用于 [高分辨率修复](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002F2_pass_txt2img\u002F) 或更高级别的工作流。\n- [区域合成](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Farea_composition\u002F)\n- [图像修复](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Finpaint\u002F)，支持常规修复模型和 inpainting 模型。\n- [ControlNet 和 T2I-Adapter](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fcontrolnet\u002F)\n- [超分辨率模型（ESRGAN、ESRGAN 变体、SwinIR、Swin2SR 等）](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fupscale_models\u002F)\n- [GLIGEN](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fgligen\u002F)\n- [模型合并](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Fmodel_merging\u002F)\n- [LCM 模型和 LoRA](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002Flcm\u002F)\n- 使用 [TAESD](#how-to-show-high-quality-previews) 进行潜在空间预览。\n- 完全离线工作：核心组件不会在您未主动要求的情况下下载任何内容。\n- 可选 API 节点，可通过在线 [Comfy API](https:\u002F\u002Fdocs.comfy.org\u002Ftutorials\u002Fapi-nodes\u002Foverview) 使用外部提供商的付费模型；可通过 `--disable-api-nodes` 禁用。\n- [配置文件](extra_model_paths.yaml.example)，用于设置模型搜索路径。\n\n工作流示例可在 [示例页面](https:\u002F\u002Fcomfyanonymous.github.io\u002FComfyUI_examples\u002F) 中找到。\n\n## 发布流程\n\nComfyUI 采用每周一为目标的发布周期，但由于模型发布或代码库的重大变更，发布时间可能会经常调整。该项目由三个相互关联的仓库组成：\n\n1. **[ComfyUI Core](https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI)**\n   - 大约每周发布一个稳定版本（例如 v0.7.0）。\n   - 自 v0.4.0 起，补丁版本将用于向当前稳定版回迁修复。\n   - 小版本将用于从主分支发布的版本。\n   - 在不适合回迁修复的情况下，主分支也可能使用补丁版本进行发布。\n   - 稳定版标签之外的提交可能非常不稳定，并可能导致许多自定义节点失效。\n   - 是桌面版的基础。\n\n2. **[ComfyUI Desktop](https:\u002F\u002Fgithub.com\u002FComfy-Org\u002Fdesktop)**\n   - 使用最新的稳定核心版本构建新版本。\n\n3. **[ComfyUI Frontend](https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI_frontend)**\n   - 每周将前端更新合并到核心仓库中。\n   - 针对即将发布的核心版本冻结功能。\n   - 开发工作将继续进行下一版本的迭代。\n\n## 快捷键\n\n| 快捷键                            | 说明                                                                                                        |\n|------------------------------------|--------------------------------------------------------------------------------------------------------------------|\n| `Ctrl` + `Enter`                      | 将当前图谱加入生成队列                                                                              |\n| `Ctrl` + `Shift` + `Enter`              | 将当前图谱设为首个待生成任务                                                                     |\n| `Ctrl` + `Alt` + `Enter`                | 取消当前的生成任务                                                                                          |\n| `Ctrl` + `Z`\u002F`Ctrl` + `Y`                 | 撤销\u002F重做                                                                                                          |\n| `Ctrl` + `S`                          | 保存工作流                                                                                                      |\n| `Ctrl` + `O`                          | 加载工作流                                                                                                      |\n| `Ctrl` + `A`                          | 全选所有节点                                                                                                   |\n| `Alt `+ `C`                           | 折叠\u002F展开选中的节点                                                                                 |\n| `Ctrl` + `M`                          | 静音\u002F取消静音选中的节点                                                                                         |\n| `Ctrl` + `B`                           | 跳过选中的节点（效果相当于将该节点从图谱中移除，并重新连接相关连线）            |\n| `Delete`\u002F`Backspace`                   | 删除选中的节点                                                                                              |\n| `Ctrl` + `Backspace`                   | 删除当前图谱                                                                                           |\n| `Space`                              | 按住并移动鼠标时，可在画布上平移画面                                                             |\n| `Ctrl`\u002F`Shift` + `点击`                 | 将点击的节点添加到选区                                                                                      |\n| `Ctrl` + `C`\u002F`Ctrl` + `V`                  | 复制并粘贴选中的节点（不保留未选节点的输出连接）                                                     |\n| `Ctrl` + `C`\u002F`Ctrl` + `Shift` + `V`          | 复制并粘贴选中的节点（保留未选节点的输出连接至已粘贴节点的输入） |\n| `Shift` + `拖动`                       | 同时移动多个选中的节点                                                                      |\n| `Ctrl` + `D`                           | 加载默认图谱                                                                                                 |\n| `Alt` + `+`                          | 画布放大                                                                                                     |\n| `Alt` + `-`                          | 画布缩小                                                                                                    |\n| `Ctrl` + `Shift` + 左键 + 垂直拖动 | 画布放大\u002F缩小                                                                                                 |\n| `P`                                  | 固定\u002F取消固定选中的节点                                                                                           |\n| `Ctrl` + `G`                           | 将选中的节点分组                                                                                               |\n| `Q`                                 | 切换任务队列的可见性                                                                                     |\n| `H`                                  | 切换历史记录的可见性                                                                                       |\n| `R`                                  | 刷新图谱                                                                                                      |\n| `F`                                  | 显示\u002F隐藏菜单                                                                                                      |\n| `.`                                  | 自动调整视图以适应选区（若无选中内容，则显示整个图谱）                                                        |\n| 双击左键                   | 打开节点快速搜索面板                                                                                     |\n| `Shift` + 拖动                       | 一次性移动多条连线                                                                                        |\n| `Ctrl` + `Alt` + 左键                   | 断开点击插槽的所有连线                                                                             |\n\n对于 macOS 用户，`Ctrl` 键也可替换为 `Cmd` 键。\n\n# 安装\n\n## Windows 便携版\n\n在 [发布页面](https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Freleases) 上提供了适用于 Windows 的便携式独立版本，既可在 NVIDIA GPU 上运行，也可仅使用 CPU 运行。\n\n### [直接下载链接](https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Freleases\u002Flatest\u002Fdownload\u002FComfyUI_windows_portable_nvidia.7z)\n\n只需下载，使用[7-Zip](https:\u002F\u002F7-zip.org)或在较新版本的Windows中用文件资源管理器解压，然后运行即可。对于较小的模型，通常只需将检查点文件（那些巨大的ckpt\u002Fsafetensors文件）放入：ComfyUI\\models\\checkpoints 文件夹中。但许多较大的模型包含多个文件，请务必按照说明将它们放置到正确的子文件夹中：ComfyUI\\models\\。\n\n如果解压时遇到问题，可以右键点击文件 -> 属性 -> 取消“解除锁定”选项。\n\n当前提供的便携版自带Python 3.13和PyTorch CUDA 13.0。如果无法启动，请更新您的NVIDIA显卡驱动程序。\n\n#### 其他下载选项：\n\n[适用于AMD显卡的实验性便携版](https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Freleases\u002Flatest\u002Fdownload\u002FComfyUI_windows_portable_amd.7z)\n\n[配备PyTorch CUDA 12.6和Python 3.12的便携版](https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Freleases\u002Flatest\u002Fdownload\u002FComfyUI_windows_portable_nvidia_cu126.7z)（支持NVIDIA 10系列及更早型号的GPU）。\n\n#### 如何在其他UI与ComfyUI之间共享模型？\n\n请参阅[配置文件](extra_model_paths.yaml.example)，以设置模型的搜索路径。在独立的Windows版本中，该文件位于ComfyUI目录下。将此文件重命名为extra_model_paths.yaml，并使用您喜欢的文本编辑器进行编辑。\n\n\n## [comfy-cli](https:\u002F\u002Fdocs.comfy.org\u002Fcomfy-cli\u002Fgetting-started)\n\n您也可以使用comfy-cli来安装并启动ComfyUI：\n```bash\npip install comfy-cli\ncomfy install\n```\n\n## 手动安装（Windows、Linux）\n\nPython 3.14可以正常工作，但某些自定义节点可能会出现问题。免费线程版本虽然可用，但由于部分依赖会启用GIL，因此并不完全支持。\n\nPython 3.13则得到了很好的支持。如果您在3.13环境下遇到某些自定义节点依赖的问题，可以尝试使用Python 3.12。\n\nPyTorch 2.4及以上版本均受支持，但部分功能和优化可能仅在较新版本中才能生效。我们通常建议使用最新稳定版的PyTorch，并搭配最新的CUDA版本，除非该版本发布不足两周。\n\n### 安装步骤：\n\n使用Git克隆本仓库。\n\n将您的SD检查点文件（那些巨大的ckpt\u002Fsafetensors文件）放入：models\u002Fcheckpoints 文件夹。\n\n将VAE文件放入：models\u002Fvae 文件夹。\n\n\n### AMD显卡（Linux）\n\nAMD用户可以通过pip安装ROCm和PyTorch（如果尚未安装）。以下是安装稳定版的命令：\n```pip install torch torchvision torchaudio --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Frocm7.2```\n\n若要安装带有ROCm 7.2的夜间版，可能带来一些性能提升，可使用以下命令：\n```pip install --pre torch torchvision torchaudio --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fnightly\u002Frocm7.2```\n\n### AMD显卡（实验性：Windows和Linux），仅支持RDNA 3、3.5和4。\n\n这些版本对硬件的支持不如上述构建，但在Windows上仍可运行。此外，您还需要根据自己的硬件安装特定版本的PyTorch。\n\nRDNA 3（RX 7000系列）：\n```pip install --pre torch torchvision torchaudio --index-url https:\u002F\u002Frocm.nightlies.amd.com\u002Fv2\u002Fgfx110X-all\u002F```\n\nRDNA 3.5（Strix Halo\u002FRyzen AI Max+ 365）：\n```pip install --pre torch torchvision torchaudio --index-url https:\u002F\u002Frocm.nightlies.amd.com\u002Fv2\u002Fgfx1151\u002F```\n\nRDNA 4（RX 9000系列）：\n```pip install --pre torch torchvision torchaudio --index-url https:\u002F\u002Frocm.nightlies.amd.com\u002Fv2\u002Fgfx120X-all\u002F```\n\n### Intel显卡（Windows和Linux）\n\nIntel Arc GPU用户可以通过pip安装原生支持torch.xpu的PyTorch。更多信息请参见[此处](https:\u002F\u002Fpytorch.org\u002Fdocs\u002Fmain\u002Fnotes\u002Fget_start_xpu.html)。\n\n1. 要安装torch.xpu版本的PyTorch，可使用以下命令：\n```pip install torch torchvision torchaudio --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fxpu```\n\n若要安装torch.xpu夜间版，可能带来性能提升，可使用：\n```pip install --pre torch torchvision torchaudio --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fnightly\u002Fxpu```\n\n### NVIDIA\n\nNVIDIA用户应使用以下命令安装稳定版PyTorch：\n```pip install torch torchvision torchaudio --extra-index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu130```\n\n若要安装夜间版PyTorch，也可能带来性能提升，可使用：\n```pip install --pre torch torchvision torchaudio --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fnightly\u002Fcu130```\n\n#### 故障排除\n\n如果出现“Torch未编译为启用CUDA”的错误，请先卸载PyTorch：\n```pip uninstall torch```\n\n然后再使用上述命令重新安装。\n\n### 依赖项安装\n\n打开终端，进入ComfyUI文件夹，执行以下命令以安装所有依赖项：\n```pip install -r requirements.txt```\n\n完成之后，您应该已经安装好所有必要的组件，可以继续运行ComfyUI了。\n\n### 其他：\n\n#### Apple Mac silicon\n\n您可以在搭载 M1 或 M2 芯片的 Apple Mac 上，使用任何最新版本的 macOS 安装 ComfyUI。\n\n1. 安装 PyTorch Nightly。有关说明，请参阅 Apple 开发者指南中的 [在 Mac 上加速 PyTorch 训练](https:\u002F\u002Fdeveloper.apple.com\u002Fmetal\u002Fpytorch\u002F)（请确保安装最新的 PyTorch Nightly）。\n2. 按照适用于 Windows 和 Linux 的 [ComfyUI 手动安装](#manual-install-windows-linux)说明进行操作。\n3. 安装 ComfyUI 的 [依赖项](#dependencies)。如果您已经安装了其他 Stable Diffusion UI，[或许可以复用这些依赖项](#i-already-have-another-ui-for-stable-diffusion-installed-do-i-really-have-to-install-all-of-these-dependencies)。\n4. 通过运行 `python main.py` 启动 ComfyUI。\n\n> **注意**：请务必将您的模型、VAE、LoRA 等文件添加到相应的 Comfy 文件夹中，如 [ComfyUI 手动安装](#manual-install-windows-linux)中所述。\n\n#### Ascend NPUs\n\n适用于与 PyTorch Ascend 扩展（torch_npu）兼容的模型。要开始使用，请确保您的环境满足 [安装](https:\u002F\u002Fascend.github.io\u002Fdocs\u002Fsources\u002Fascend\u002Fquick_install.html) 页面中列出的先决条件。以下是针对您的平台和安装方法的分步指南：\n\n1. 如果需要，首先按照 torch-npu 安装页面的说明，为 Linux 安装推荐或更新的内核版本。\n2. 接着，按照适用于您特定平台的说明，安装包含驱动程序、固件和 CANN 的 Ascend Basekit。\n3. 然后，根据 [安装](https:\u002F\u002Fascend.github.io\u002Fdocs\u002Fsources\u002Fpytorch\u002Finstall.html#pytorch) 页面上的平台特定说明，安装 torch-npu 所需的软件包。\n4. 最后，遵循适用于 Linux 的 [ComfyUI 手动安装](#manual-install-windows-linux)指南。所有组件安装完成后，您可以按照之前描述的方式运行 ComfyUI。\n\n#### Cambricon MLUs\n\n适用于与 PyTorch Cambricon 扩展（torch_mlu）兼容的模型。以下是针对您的平台和安装方法的分步指南：\n\n1. 按照 [安装](https:\u002F\u002Fwww.cambricon.com\u002Fdocs\u002Fsdk_1.15.0\u002Fcntoolkit_3.7.2\u002Fcntoolkit_install_3.7.2\u002Findex.html) 页面上的平台特定说明，安装 Cambricon CNToolkit。\n2. 接着，按照 [安装](https:\u002F\u002Fwww.cambricon.com\u002Fdocs\u002Fsdk_1.15.0\u002Fcambricon_pytorch_1.17.0\u002Fuser_guide_1.9\u002Findex.html) 页面上的说明，安装 PyTorch（torch_mlu）。\n3. 通过运行 `python main.py` 启动 ComfyUI。\n\n#### Iluvatar Corex\n\n适用于与 Iluvatar PyTorch 扩展兼容的模型。以下是针对您的平台和安装方法的分步指南：\n\n1. 按照 [安装](https:\u002F\u002Fsupport.iluvatar.com\u002F#\u002FDocumentCentre?id=1&nameCenter=2&productId=520117912052801536) 页面上的平台特定说明，安装 Iluvatar Corex Toolkit。\n2. 通过运行 `python main.py` 启动 ComfyUI。\n\n\n## [ComfyUI-Manager](https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI-Manager\u002Ftree\u002Fmanager-v4)\n\n**ComfyUI-Manager** 是一个扩展，可让您轻松安装、更新和管理 ComfyUI 的自定义节点。\n\n### 设置\n\n1. 安装管理器依赖项：\n   ```bash\n   pip install -r manager_requirements.txt\n   ```\n2. 在运行 ComfyUI 时，使用 `--enable-manager` 标志启用管理器：\n   ```bash\n   python main.py --enable-manager\n   ```\n\n### 命令行选项\n\n| 标志 | 描述 |\n|------|-------------|\n| `--enable-manager` | 启用 ComfyUI-Manager |\n| `--enable-manager-legacy-ui` | 使用旧版管理器界面而非新版界面（需配合 `--enable-manager` 使用） |\n| `--disable-manager-ui` | 禁用管理器界面和端点，同时保留后台功能，例如安全检查和计划性安装完成（需配合 `--enable-manager` 使用） |\n\n\n# 运行\n\n```python main.py```\n\n### 对于 ROCm 官方不支持的 AMD 显卡\n\n如果您遇到问题，可以尝试使用以下命令运行：\n\n对于 6700、6600 及可能的其他 RDNA2 或更早型号显卡：```HSA_OVERRIDE_GFX_VERSION=10.3.0 python main.py```\n\n对于 AMD 7600 及可能的其他 RDNA3 显卡：```HSA_OVERRIDE_GFX_VERSION=11.0.0 python main.py```\n\n### AMD ROCm 小贴士\n\n您可以在 ComfyUI 中，通过此命令在部分 AMD GPU 上启用实验性的内存高效注意力机制；RDNA3 显卡上默认已启用。如果该设置在您使用的最新 PyTorch 版本下提升了 GPU 性能，请告知我们，以便我们将其设为默认值。\n\n```TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1 python main.py --use-pytorch-cross-attention```\n\n您还可以尝试设置环境变量 `PYTORCH_TUNABLEOP_ENABLED=1`，这可能会加快速度，但会导致初始运行非常缓慢。\n\n# 注意事项\n\n只有那些拥有完整正确输入并产生输出的部分图节点才会被执行。\n\n每次执行时，只有发生变化的部分会被执行。如果您两次提交相同的图，仅第一次会被执行。如果您只修改了图的最后一部分，则只会执行您修改的部分及其依赖部分。\n\n将生成的 PNG 文件拖放到网页上或直接加载，即可查看完整的流程，包括用于生成该图像的种子。\n\n您可以使用括号来调整某个词或短语的权重，例如：(good code:1.2) 或 (bad code:0.8)。括号的默认权重是 1.1。如果要在提示中实际使用括号，请对其进行转义，例如 \\\\( 或 \\\\)。\n\n您还可以使用 {day|night} 来创建通配符或动态提示。例如，使用 `{wild|card|test}` 语法，前端会在每次排队提示时随机替换为“wild”、“card”或“test”。如果要在提示中实际使用花括号，请对其进行转义，例如 \\\\{ 或 \\\\}。\n\n动态提示还支持 C 语言风格的注释，例如 `\u002F\u002F comment` 或 `\u002F* comment *\u002F`。\n\n要在文本提示中使用文本反演概念或嵌入，请将其放入 models\u002Fembeddings 目录，并在 CLIPTextEncode 节点中按如下方式引用（可省略 .pt 扩展名）：\n\n```embedding:embedding_filename.pt```\n\n\n## 如何显示高质量预览？\n\n使用 `--preview-method auto` 启用预览。\n\n默认安装包含一种快速但低分辨率的潜在空间预览方法。要启用更高品质的预览，请使用 [TAESD](https:\u002F\u002Fgithub.com\u002Fmadebyollin\u002Ftaesd)，下载 [taesd_decoder.pth、taesdxl_decoder.pth、taesd3_decoder.pth 和 taef1_decoder.pth](https:\u002F\u002Fgithub.com\u002Fmadebyollin\u002Ftaesd\u002F)，并将它们放置在 `models\u002Fvae_approx` 文件夹中。安装完成后，重启 ComfyUI，并以 `--preview-method taesd` 参数启动，即可启用高质量预览。\n\n## 如何使用 TLS\u002FSSL？\n通过运行以下命令生成自签名证书（不适用于共享或生产环境）和密钥：`openssl req -x509 -newkey rsa:4096 -keyout key.pem -out cert.pem -sha256 -days 3650 -nodes -subj \"\u002FC=XX\u002FST=StateName\u002FL=CityName\u002FO=CompanyName\u002FOU=CompanySectionName\u002FCN=CommonNameOrHostname\"`\n\n使用 `--tls-keyfile key.pem --tls-certfile cert.pem` 来启用 TLS\u002FSSL，应用程序现在将可以通过 `https:\u002F\u002F...` 而不是 `http:\u002F\u002F...` 访问。\n\n> 注意：Windows 用户可以使用 [alexisrolland\u002Fdocker-openssl](https:\u002F\u002Fgithub.com\u002Falexisrolland\u002Fdocker-openssl) 或者其中一种 [第三方二进制分发版](https:\u002F\u002Fwiki.openssl.org\u002Findex.php\u002FBinaries) 来运行上述示例命令。\n\u003Cbr\u002F>\u003Cbr\u002F>如果您使用容器，请注意 `-v` 挂载的卷路径可以是相对路径，例如 `... -v \".\\:\u002Fopenssl-certs\" ...` 将会在您命令提示符或 PowerShell 终端的当前目录下创建密钥和证书文件。\n\n## 支持与开发通道\n\n[Discord](https:\u002F\u002Fcomfy.org\u002Fdiscord)：尝试 #help 或 #feedback 频道。\n\n[Matrix 社区：#comfyui_space:matrix.org](https:\u002F\u002Fapp.element.io\u002F#\u002Froom\u002F%23comfyui_space%3Amatrix.org)（类似于 Discord，但为开源平台）。\n\n更多信息请参阅：[https:\u002F\u002Fwww.comfy.org\u002F](https:\u002F\u002Fwww.comfy.org\u002F)\n\n## 前端开发\n\n截至 2024 年 8 月 15 日，我们已切换到全新的前端，该前端现托管在独立的仓库中：[ComfyUI Frontend](https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI_frontend)。此仓库现在在 `web\u002F` 目录下托管由 TS\u002FVue 编译生成的 JS 文件。\n\n### 报告问题与请求功能\n\n如有关于前端的任何错误、问题或功能请求，请使用 [ComfyUI Frontend 仓库](https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI_frontend)。这将有助于我们更高效地管理和解决前端相关的问题。\n\n### 使用最新前端\n\n新的前端现已成为 ComfyUI 的默认设置。但请注意：\n\n1. 主 ComfyUI 仓库中的前端每两周更新一次。\n2. 独立的前端仓库则提供每日发布的版本。\n\n要使用最新的前端版本：\n\n1. 若要获取最新的每日发布版，请使用以下命令行参数启动 ComfyUI：\n\n   ```\n   --front-end-version Comfy-Org\u002FComfyUI_frontend@latest\n   ```\n\n2. 若需特定版本，可将 `latest` 替换为所需版本号：\n\n   ```\n   --front-end-version Comfy-Org\u002FComfyUI_frontend@1.2.2\n   ```\n\n这种方式使您可以轻松地在稳定的双周发布版与前沿的每日更新版之间切换，甚至可以根据测试需求选择特定版本。\n\n### 访问旧版前端\n\n若您因任何原因需要使用旧版前端，可通过以下命令行参数访问：\n\n```\n--front-end-version Comfy-Org\u002FComfyUI_legacy_frontend@latest\n```\n\n这将使用保存在 [ComfyUI Legacy Frontend 仓库](https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI_legacy_frontend) 中的旧版前端快照。\n\n# QA\n\n### 我应该购买哪款显卡来运行这个？\n\n[请参阅此页面以获取一些建议](https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Fwiki\u002FWhich-GPU-should-I-buy-for-ComfyUI)","# ComfyUI 快速上手指南\n\nComfyUI 是一款功能强大且模块化的可视化 AI 引擎，允许用户通过节点\u002F流程图界面设计和执行复杂的 Stable Diffusion 工作流，无需编写代码。支持 Windows、Linux 和 macOS。\n\n## 1. 环境准备\n\n### 系统要求\n- **操作系统**: Windows 10\u002F11, Linux (Ubuntu\u002FDebian 等), macOS (Intel 或 Apple Silicon)。\n- **GPU**: \n  - NVIDIA (推荐，显存建议 4GB+，智能优化后可低至 1GB)。\n  - AMD (ROCm)、Intel (Arc)、Apple Silicon (M1\u002FM2\u002FM3) 或华为 Ascend。\n  - 若无独立显卡，可使用 CPU 模式（速度较慢）。\n- **磁盘空间**: 建议预留 20GB+ 用于存放模型和依赖。\n\n### 前置依赖\n- **Python**: 3.8 - 3.12 (官方安装包通常自带，手动安装需自行配置)。\n- **Git**: 用于克隆仓库和管理自定义节点。\n- **CUDA\u002FcuDNN**: NVIDIA 用户通常由 PyTorch 自动处理，特殊环境需手动安装。\n\n> **国内加速提示**: 如遇网络问题，建议配置 pip 国内镜像源（如清华源、阿里源）加速 Python 依赖下载。\n\n## 2. 安装步骤\n\n根据您的需求选择以下任一方式：\n\n### 方案 A：桌面应用程序（推荐新手）\n最简单的方式，内置所有依赖，开箱即用。\n1. 访问 [ComfyUI 下载页](https:\u002F\u002Fwww.comfy.org\u002Fdownload)。\n2. 下载对应系统（Windows 或 macOS）的安装包。\n3. 运行安装程序并按提示完成安装。\n\n### 方案 B：Windows 便携版 (Portable)\n无需安装，解压即用，适合希望保持最新版本的 Windows 用户。\n1. 下载 `ComfyUI_windows_portable_nvidia_cu121_or_cpu.7z` (NVIDIA) 或 CPU 版本。\n   *下载地址*: [GitHub Releases](https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI\u002Freleases)\n2. 解压文件到任意目录（路径不要包含中文或空格）。\n3. 双击运行 `run_nvidia.bat` (或 `run_cpu.bat`)。\n\n### 方案 C：手动安装 (通用\u002FLinux\u002FmacOS)\n适合高级用户或需要特定环境配置的场景。\n\n**1. 克隆仓库**\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fcomfyanonymous\u002FComfyUI.git\ncd ComfyUI\n```\n> **国内加速**: 若 GitHub 连接缓慢，可使用镜像源：\n> `git clone https:\u002F\u002Fgitee.com\u002Fmirrors\u002FComfyUI.git` (如有可用镜像) 或通过代理克隆。\n\n**2. 创建虚拟环境并安装依赖**\n```bash\n# 创建虚拟环境\npython -m venv venv\n\n# 激活环境 (Windows)\nvenv\\Scripts\\activate\n# 激活环境 (Linux\u002FmacOS)\nsource venv\u002Fbin\u002Factivate\n\n# 安装依赖 (建议使用国内 pip 源)\npip install -r requirements.txt -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n```\n\n**3. 启动 ComfyUI**\n```bash\n# NVIDIA GPU\npython main.py\n\n# CPU 模式 (无显卡或调试用)\npython main.py --cpu\n\n# 指定监听地址 (允许局域网访问)\npython main.py --listen 0.0.0.0\n```\n\n## 3. 基本使用\n\n### 启动与界面\n启动成功后，终端会显示类似 `Starting server` 的日志，并在浏览器自动打开 `http:\u002F\u002F127.0.0.1:8188`。若未自动打开，请手动在浏览器访问该地址。\n\n### 生成第一张图片\nComfyUI 默认加载时会带入一个基础的“文生图”工作流。\n\n1. **检查节点**: 界面上应已有默认连接的节点组（包括 Load Checkpoint, CLIP Text Encode, KSampler, VAE Decode, Save Image 等）。\n2. **输入提示词**: \n   - 找到 `CLIP Text Encode (positive)` 节点，在 `text` 框中输入正向提示词，例如：`a beautiful cat, high quality, 8k`。\n   - (可选) 在 `CLIP Text Encode (negative)` 节点输入负向提示词，例如：`blurry, low quality`。\n3. **选择模型**: \n   - 点击 `Load Checkpoint` 节点。\n   - 确保您的模型文件 (`*.safetensors` 或 `*.ckpt`) 已放入 `ComfyUI\u002Fmodels\u002Fcheckpoints` 目录。\n   - 在下拉菜单中选择对应的模型。\n4. **执行生成**:\n   - 点击右侧菜单栏的 **\"Queue Prompt\"** 按钮。\n   - 或者使用快捷键：`Ctrl` + `Enter`。\n5. **查看结果**: \n   - 等待进度条完成，生成的图片将显示在 `Save Image` 节点中。\n   - 右键点击图片可选择保存或预览大图。\n\n### 常用快捷键\n| 快捷键 | 功能 |\n| :--- | :--- |\n| `Ctrl` + `Enter` | 队列当前工作流并生成 |\n| `Ctrl` + `Z` \u002F `Ctrl` + `Y` | 撤销 \u002F 重做 |\n| `Ctrl` + `S` | 保存工作流 (JSON) |\n| `Ctrl` + `O` | 加载工作流 |\n| `Delete` | 删除选中的节点 |\n| `Space` + 拖拽 | 移动画布视图 |\n| `Ctrl` + `D` | 加载默认工作流 |\n\n### 管理工作流\n- **保存**: 点击 `Save` 按钮可将当前节点连线布局保存为 `.json` 文件。\n- **加载**: 直接将包含工作流信息的图片 (PNG\u002FWebP) 拖入浏览器窗口，或点击 `Load` 加载 JSON 文件，即可还原完整工作流（含参数和种子）。","一位独立游戏开发者需要为角色生成一套风格统一但表情各异的立绘素材，同时要求精确控制构图和细节。\n\n### 没有 ComfyUI 时\n- **流程黑盒难调整**：使用传统一键生成工具时，无法干预中间步骤，一旦手部崩坏或光影错误，只能重新随机生成，效率极低。\n- **多模型串联繁琐**：想要结合 SDXL 底模、特定 LoRA 风格和控制网（ControlNet）姿态，需编写复杂脚本或在多个软件间手动倒腾图片。\n- **复现一致性差**：调整某个参数后，难以保留之前的成功设置，导致生成的系列角色在画风或色调上出现细微偏差，破坏整体感。\n- **硬件资源浪费**：缺乏对显存的精细化调度，尝试高分辨率修复时经常直接爆显存崩溃，被迫降低画质妥协。\n\n### 使用 ComfyUI 后\n- **可视化节点编排**：通过拖拽节点构建清晰的工作流，可单独微调“提示词编码”、“潜空间降噪”或“面部修复”环节，精准修正瑕疵而不影响其他部分。\n- **灵活组装高级管线**：像搭积木一样将 SD3 底模、Flux 架构与多个 ControlNet 串联，轻松实现“保持角色脸不变仅换表情”的复杂需求。\n- **工作流即时复用**：保存好的节点图可随时加载，只需替换参考图或种子数，即可批量产出风格完全统一的数百张素材。\n- **高效显存管理**：得益于其模块化后端，ComfyUI 能智能释放闲置显存，让中端显卡也能流畅运行高分辨率的大模型生图任务。\n\nComfyUI 将原本依赖运气的“抽卡式”生图，转变为可控、可复用且高度定制化的工业化生产流程。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FComfy-Org_ComfyUI_dfe3a8cb.png","Comfy-Org","Comfy Org","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FComfy-Org_bd077766.png","",null,"hello@comfy.org","comfyui","https:\u002F\u002Fcomfy.org","https:\u002F\u002Fgithub.com\u002FComfy-Org",[85,89,93,97],{"name":86,"color":87,"percentage":88},"Python","#3572A5",99.5,{"name":90,"color":91,"percentage":92},"GLSL","#5686a5",0.5,{"name":94,"color":95,"percentage":96},"Batchfile","#C1F12E",0,{"name":98,"color":99,"percentage":96},"Mako","#7e858d",107662,12434,"2026-04-03T11:11:01","GPL-3.0","Windows, Linux, macOS","支持 NVIDIA, AMD, Intel, Apple Silicon (M 系列), Ascend (华为昇腾)。非必需，可通过 --cpu 参数在无 GPU 环境下运行（速度较慢）。智能内存管理支持低至 1GB 显存的 GPU。","未说明（取决于运行的模型大小，建议充足以加载大模型）",{"notes":108,"python":109,"dependencies":110},"1. 提供多种安装方式：桌面应用（Windows\u002FmacOS）、Windows 便携包、手动安装（全平台）。2. 支持多种硬件后端，包括苹果芯片和华为昇腾。3. 具有智能显存卸载功能，可在低显存设备上运行大模型。4. 核心完全离线运行，除非用户主动使用在线 API 节点。5. 支持加载 ckpt, safetensors, pt, pth 等多种模型格式。6. 异步队列系统优化工作流执行，仅重新执行变更部分。","未说明",[111,112,113,114,115,116,117,118,119,120],"torch","torchaudio","torchvision","numpy","Pillow","safetensors","aiohttp","pyyaml","scipy","transformers",[13,14,15],[123,124,125,126,127,81],"stable-diffusion","pytorch","ai","python","comfy",91,"2026-03-27T02:49:30.150509","2026-04-06T01:14:46.308664",[132,137,142,147,151,156],{"id":133,"question_zh":134,"answer_zh":135,"source_url":136},10425,"遇到 TypeError: forward_orig() got an unexpected keyword argument 'attn_mask' 错误如何解决？","该错误通常由旧版 PuLID 节点与新版 ComfyUI 不兼容引起。解决方案是：卸载所有现有的 PuLID 相关插件，然后安装最新的 'ComfyUI-PuLID-Flux-Enhanced'（作者 sipie800）或切换到 'Flux Pulid II' 版本及其对应的新模型。安装后问题即可解决。","https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fissues\u002F6077",{"id":138,"question_zh":139,"answer_zh":140,"source_url":141},10426,"Stable Cascade 报错 'expected input to have 16 channels, but got 64 channels' 怎么办？","此错误通常是因为 Stage C 模型未正确加载导致的。请检查工作流中模型加载节点，确保 'stage_c' 模型已正确选择并加载。重新加载该模型后，通道数不匹配的错误通常会消失。","https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fissues\u002F2785",{"id":143,"question_zh":144,"answer_zh":145,"source_url":146},10427,"在 Mac (M2 Pro\u002FSonoma) 上运行几步后图像变黑如何解决？","这是 Mac 上的精度计算问题。解决方法是在启动命令中添加 '--force-upcast-attention' 参数以提高计算精度。如果使用 Pinokio 启动，需修改启动脚本中的命令行为：'python main.py --force-fp32'（针对 Mac），或者在终端直接运行带有该参数的启动命令。此外，避免使用 'Karras' 调度器也可能缓解此问题。","https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fissues\u002F3521",{"id":148,"question_zh":149,"answer_zh":150,"source_url":141},10428,"Stable Cascade 生成特定分辨率（如 1080x1080）图像时出现伪影或模糊怎么办？","Stable Cascade 对生成分辨率敏感，非标准分辨率（如 1080x1080 或 1024x1080）容易产生伪影或模糊。建议优先使用模型训练的原生分辨率（如 1024x1024）以获得最清晰的效果。如果必须使用其他分辨率，请尝试接近 1024 的倍数或进行后期放大，避免直接使用导致伪影的中间分辨率。",{"id":152,"question_zh":153,"answer_zh":154,"source_url":155},10429,"启用 Pinned Memory 时使用 GGUF 模型报 CUDA error: invalid argument 错误？","该问题出现在启用 Pinned Memory 且使用 GGUF 量化模型（如 Wan 2.2）进行高视频分辨率生成时，通常在模型卸载阶段触发。目前的临时解决方案是：在 ComfyUI 启动参数中禁用 Pinned Memory，或者降低视频生成的分辨率。同时确保 'ComfyUI-GGUF' 节点和 ComfyUI 主程序均已更新到最新版本。","https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fissues\u002F10662",{"id":157,"question_zh":158,"answer_zh":159,"source_url":141},10430,"Stable Cascade 的 ControlNet 模型无法使用，报 'NoneType' object has no attribute 'copy' 错误？","这是因为 Stable Cascade 的 ControlNet 支持尚未完全集成到默认节点中。目前内置的 ControlNetApply 节点可能无法直接识别这些新模型。用户需要等待官方更新支持，或者寻找社区开发的专门适配 Stable Cascade ControlNet 的自定义节点插件。",[161,166,171,176,181,186,191,196,201,206,211,216,221,226,231,236,241,246,251,256],{"id":162,"version":163,"summary_zh":164,"released_at":165},71007,"v0.18.2","**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.18.1...v0.18.2","2026-03-25T04:22:37",{"id":167,"version":168,"summary_zh":169,"released_at":170},71008,"v0.18.1","## What's Changed\r\n* Fix canny node not working with fp16. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13085\r\n* Fix sampling issue with fp16 intermediates. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13099\r\n* Fix fp16 intermediates giving different results. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13100\r\n* wan: vae: Fix light\u002Fcolor issue by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13101\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.18.0...v0.18.1","2026-03-23T16:58:47",{"id":172,"version":173,"summary_zh":174,"released_at":175},71009,"v0.18.0","## What's Changed\r\n* Revert \"Revert \"feat: Add CacheProvider API for external distributed caching\"\" by @deepme987 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12915\r\n* fix(api-nodes): Tencent TextToModel and ImageToModel nodes by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12680\r\n* Bump comfyui-frontend-package to 1.41.19 by @comfy-pr-bot in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12923\r\n* bump manager version to 4.1b4 by @ltdrdata in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12930\r\n* comfy aimdo 0.2.11 + Improved RAM Pressure release strategies - Windows speedups by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12925\r\n* Update README. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12931\r\n* comfy-aimdo 0.2.12 by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12941\r\n* fix: use no-store cache headers to prevent stale frontend chunks by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12911\r\n* feat: Support mxfp8 by @kijai in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12907\r\n* Add --fp16-intermediates to use fp16 for intermediate values between … by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12953\r\n* Update comfyui-frontend-package to version 1.41.20 by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12954\r\n* LTXV: Accumulate VAE decode results on intermediate_device by @kijai in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12955\r\n* bump manager version to 4.1b5 by @ltdrdata in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12957\r\n* ops: opt out of deferred weight init if subclassed by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12967\r\n* Make EmptyLatentImage follow intermediate dtype. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12974\r\n* Enable Pytorch Attention for AMD gfx1150 (Strix Point) by @lostdisc in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12973\r\n* feat: add essentials_category to nodes and blueprints for Essentials tab by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12573\r\n* feat(assets): align local API with cloud spec by @luke-mino-altherr in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12863\r\n* ci: add check to block AI agent Co-authored-by trailers in PRs by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12799\r\n* Skip running model finalizers at exit by @blepping in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12994\r\n* Add --enable-dynamic-vram options to force enable it. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13002\r\n* [Trainer] FP4, 8, 16 training by native dtype support and quant linear autograd function by @KohakuBlueleaf in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12681\r\n* Fix potential issue. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13009\r\n* fix: atomic writes for userdata to prevent data loss on crash by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12987\r\n* Disable SageAttention for Hunyuan3D v2.1 DiT by @paulomuggler in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12772\r\n* Update workflow templates to v0.9.26 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13012\r\n* Mark weight_dtype as advanced input in Load Diffusion Model node by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12769\r\n* Reduce LTX VAE VRAM usage and save use cases from OOMs\u002FTiler by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13013\r\n* Reduce WAN VAE VRAM, Save use cases for OOM\u002FTiler by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13014\r\n* bump manager version to 4.1b6 by @ltdrdata in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13022\r\n* Inplace VAE output processing to reduce peak RAM consumption. by @kijai in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13028\r\n* Fix case where pixel space VAE could cause issues. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13030\r\n* cascade: remove dead weight init code by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13026\r\n* fix: run text encoders on MPS GPU instead of CPU for Apple Silicon by @k06a in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12809\r\n* fix(api-nodes): add support for \"thought_image\" in Nano Banana 2 by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13038\r\n* Update comfyui-frontend-package version to 1.41.21 by @DrJKL in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13035\r\n* Make more intermediate values follow the intermediate dtype. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13051\r\n* Further Reduce LTX VAE decode peak RAM usage by @kijai in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13052\r\n* Fix regression. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13053\r\n* fp16 intermediates doen't work for some text enc models. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13056\r\n* ltx: vae: implement chunked encoder + CPU IO chunking (Big VRAM reductions) by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F13062\r\n* memory: Add more exclus","2026-03-21T01:16:27",{"id":177,"version":178,"summary_zh":179,"released_at":180},71010,"v0.17.2","**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.17.1...v0.17.2","2026-03-15T02:28:33",{"id":182,"version":183,"summary_zh":184,"released_at":185},71011,"v0.17.1","**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.17.0...v0.17.1","2026-03-13T19:58:49",{"id":187,"version":188,"summary_zh":189,"released_at":190},71012,"v0.17.0","## What's Changed\r\n* refactor(assets): modular architecture + async two-phase scanner & background seeder by @luke-mino-altherr in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12621\r\n* comfy-aimdo 0.2.9 by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12840\r\n* Import assets seeder later, print some package versions. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12841\r\n* Fix text encoder lora loading for wrapped models by @kijai in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12852\r\n* Update workflow templates to v0.9.18 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12857\r\n* fix(manager): improve install guidance when comfyui-manager is not installed and bump manager version to v4.1b2 by @ltdrdata in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12810\r\n* Add a way for nodes to add pre attn patches to flux model. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12861\r\n* model_detection: deep clone pre edited weights by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12862\r\n* Widen OOM_EXCEPTION to AcceleratorError form by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12835\r\n* main: switch on python faulthandler by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12868\r\n* Allow model patches to have a cleanup function. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12878\r\n* Add pre attention and post input patches to qwen image model. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12879\r\n* comfy-aimdo 0.2.10 by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12890\r\n* feat(api-nodes): add Reve Image nodes by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12848\r\n* fix: guard torch.AcceleratorError for compatibility with torch \u003C 2.8.0 by @adiborochov in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12874\r\n* Fix issue with batch_size > 1 on some models. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12892\r\n* Bump comfyui-frontend-package to 1.41.15 by @comfy-pr-bot in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12891\r\n* Bump comfyui-frontend-package to 1.41.16 by @comfy-pr-bot in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12894\r\n* Bump comfy-kitchen version to 0.2.8 by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12895\r\n* Support flux 2 klein kv cache model: Use the FluxKVCache node. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12905\r\n* add painter node by @jtydhr88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12294\r\n* fix: use frontend-compatible format for Float gradient_stops by @jtydhr88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12789\r\n* Update workflow templates to v0.9.21 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12908\r\n* Lower kv cache memory usage. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12909\r\n* Patch frontend to 1.41.18 (from 1.41.16) by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12910\r\n* feat: Add CacheProvider API for external distributed caching by @deepme987 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12056\r\n* Revert \"feat: Add CacheProvider API for external distributed caching\" by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12912\r\n* Fix audio extraction and truncation bugs by @PxTicks in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12652\r\n\r\n## New Contributors\r\n* @luke-mino-altherr made their first contribution in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12621\r\n* @adiborochov made their first contribution in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12874\r\n* @deepme987 made their first contribution in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12056\r\n* @PxTicks made their first contribution in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12652\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.16.4...v0.17.0","2026-03-13T01:33:15",{"id":192,"version":193,"summary_zh":194,"released_at":195},71013,"v0.16.4","## What's Changed\r\n* feat: add Math Expression node with simpleeval evaluation by @dante01yoon in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12687\r\n* feat(api-nodes): add TencentSmartTopology node by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12741\r\n* feat(api-nodes): add Gemini 3.1 Flash-Lite model to LLM node by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12803\r\n* Fix fp16 audio encoder models by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12811\r\n* Don't run TE on cpu when dynamic vram enabled. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12815\r\n* Fix requirements version. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12817\r\n* Update workflow templates to v0.9.11 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12821\r\n* mm: reset_cast_buffers: sync compute stream before free by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12822\r\n\r\n## New Contributors\r\n* @dante01yoon made their first contribution in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12687\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.16.3...v0.16.4","2026-03-07T22:37:49",{"id":197,"version":198,"summary_zh":199,"released_at":200},71014,"v0.16.3","## What's Changed\r\n* Add manual cast to LTX2 vocoder conv_transpose1d by @kijai in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12795\r\n* LTX audio vae novram fixes. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12796\r\n* feat: add causal_fix parameter to add_keyframe_index and append_keyframe by @tavihalperin in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12797\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.16.2...v0.16.3","2026-03-05T23:18:25",{"id":202,"version":203,"summary_zh":204,"released_at":205},71015,"v0.16.2","## What's Changed\r\n* comfy-aimdo 0.2.7 by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12791\r\n* ops: Handle CPU weight in VBAR caster by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12792\r\n* Update workflow templates to v0.9.10 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12793\r\n* Refactor. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12794\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.16.1...v0.16.2","2026-03-05T20:07:22",{"id":207,"version":208,"summary_zh":209,"released_at":210},71016,"v0.16.1","## What's Changed\r\n* feat(api-nodes-xAI): updated models and pricing by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12756\r\n* feat(api-nodes): enable Kling 3.0 Motion Control by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12785\r\n* Update workflow templates to v0.9.8 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12788\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.16.0...v0.16.1","2026-03-05T17:07:27",{"id":212,"version":213,"summary_zh":214,"released_at":215},71017,"v0.16.0","## What's Changed\r\n* Portable with cu128 isn't useful anymore. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12666\r\n* feat(ace-step): add ACE-Step 1.5 lycoris key alias mapping for LoKR #… by @fappaz in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12665\r\n* bug_report template: Push harder for logs by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12657\r\n* feat: Support SDPose-OOD by @kijai in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12661\r\n* Enable Pytorch Attention for gfx950 by @vickytsang in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12641\r\n* Fix GLSL node requiring all input images to have same dimensions by @pythongosssss in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12673\r\n* feat: add ResolutionSelector node for aspect ratio presets by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12199\r\n* Class WanVAE, def encode, feat_map is using self.decoder instead of self.encoder by @ReinerBforartists in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12682\r\n* ops: implement lora requanting for non QuantizedTensor fp8 by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12668\r\n* Limit overlap in image tile and combine nodes to prevent issues. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12688\r\n* Native LongCat-Image implementation by @Talmaj in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12597\r\n* PyOpenGL-accelerate is not necessary. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12692\r\n* refactor: use enum members as ASPECT_RATIOS dict keys by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12689\r\n* fix(ace15): handle missing lm_metadata in memory estimation during ch… by @fappaz in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12686\r\n* feat: Support SCAIL WanVideo model by @kijai in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12614\r\n* Disable dynamic_vram when weight hooks applied by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12653\r\n* cli_args: Default comfy to DynamicVram mode by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12658\r\n* Move parsing of requirements logic to function. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12701\r\n* Disable dynamic vram on wsl. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12706\r\n* refactor: rename Mahiro CFG to Similarity-Adaptive Guidance by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12172\r\n* feat: add text preview support to jobs API by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12169\r\n* fix: handle substep sigmas in context window set_step by @drozbay in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12719\r\n* comfy aimdo 0.2.3 by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12720\r\n* Make --disable-smart-memory disable dynamic vram. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12722\r\n* model_mangament: make dynamic --disable-smart-memory work by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12724\r\n* comfy aimdo 0.2.4 - Fix unaligned allocator sizes by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12727\r\n* feat(api-nodes): allow to use \"IMAGE+TEXT\" in NanoBanana2 by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12729\r\n* Update workflow templates to v0.9.5 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12732\r\n* Support loading zeta chroma weights properly. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12734\r\n* Feat: z-image pixel space (model still training atm) by @lodestone-rock in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12709\r\n* Fix VAEDecodeAudioTiled ignoring tile_size input (#12735) by @xeinherjer-dev in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12738\r\n* ModelPatcherDynamic: Force load all non-comfy weights by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12739\r\n* main: Load aimdo after logger is setup by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12743\r\n* Adjust memory usage factor of zeta model. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12746\r\n* CURVE type by @jtydhr88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12581\r\n* DynamicVram: Add casting \u002F fix torch Buffer weights by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12749\r\n* Fix VideoFromComponents.save_to crash when writing to BytesIO by @artokun in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12683\r\n* model_patcher: Improve dynamic offload heuristic by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12759\r\n* Comfy Aimdo 0.2.5 + Fix offload performance in DynamicVram by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12754\r\n* Pass tokens when loading text gen model for text generation by @kijai in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12755\r\n* comfy-aimdo 0.2.6 by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12764\r\n* Initial load device should be cpu when using dynamic vram. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12766\r\n* Support the LTXAV 2","2026-03-05T10:35:07",{"id":217,"version":218,"summary_zh":219,"released_at":220},71018,"v0.15.1","## What's Changed\r\n* Fix KeyError when prompt entries lack class_type key by @Kosinkadink in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12595\r\n* comfy aimdo 0.2.1 by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12620\r\n* Disable dynamic_vram when using torch compiler by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12612\r\n* Fix Aimdo fallback on probe to not use zero-copy sft by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12634\r\n* comfy aimdo 0.2.2 by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12635\r\n* Patch frontend to 1.39.19 (from 1.39.16) (fix progress text serialization in API nodes) by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12642\r\n* Fix LTXAV text enc min length. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12640\r\n* Fix ltxav te mem estimation. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12643\r\n* feat: Support WanVideo based segmentation model FlowRVS by @kijai in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12637\r\n* feat: per-guide attention strength control in self-attention by @tavihalperin in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12518\r\n* Cleanups to the last PR. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12646\r\n* fix: move essentials_category to correct replacement nodes by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12568\r\n* Fix GLSL node with pyopengl accelerate numpy textures by @pythongosssss in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12650\r\n* main: load aimdo earlier by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12655\r\n* feat(api-nodes): add NanoBanana2 by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12660\r\n* Update workflow templates to v0.9.4 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12664\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.15.0...v0.15.1","2026-02-26T22:01:35",{"id":222,"version":223,"summary_zh":224,"released_at":225},71019,"v0.15.0","## What's Changed\r\n* ops: limit return of requants (Fix perf of some fp8 dynamic_vram workflows) by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12506\r\n* Bump comfyui-frontend-package to 1.39.14 by @comfy-pr-bot in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12494\r\n* BBox widget by @jtydhr88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F11594\r\n* fix: use glob matching for Gemini image MIME types by @huntcsg in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12511\r\n* fix(api-nodes): add price badge for Rodin Gen-2 node by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12512\r\n* Add simple 3 band equalizer node for audio. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12519\r\n* feat: Add basic text generation support with native models, initially supporting Gemma3 by @kijai in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12392\r\n* fix(api-nodes): force Gemini to return uncompressed images by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12516\r\n* feat: add essentials_category by @Yourz in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12357\r\n* feat: mark 429 widgets as advanced for collapsible UI by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12197\r\n* Small cleanup and try to get qwen 3 work with the text gen. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12537\r\n* Force min length 1 when tokenizing for text generation. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12538\r\n* Add GLSL shader node using PyOpenGL by @pythongosssss in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12148\r\n* [API Nodes] add ElevenLabs nodes by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12207\r\n* chore: add CodeRabbit configuration for automated code review by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12539\r\n* Update nightly installation command for ROCm by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12547\r\n* Fix non-contiguous audio waveform crash in video save by @artokun in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12550\r\n* feat: add gradient-slider display mode for FLOAT inputs by @jtydhr88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12536\r\n* update glsl blueprint with gradient by @jtydhr88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12548\r\n* feat: add essential subgraph blueprints by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12552\r\n* fix: swap essentials_category from CLIPTextEncode to PrimitiveStringM… by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12553\r\n* Add support for pyopengl \u003C 3.1.4 where the size parameter does not exist by @pythongosssss in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12555\r\n* comfy-aimdo 0.2 - Improved pytorch allocator integration by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12557\r\n* fix: specify UTF-8 encoding when reading subgraph files by @DrJKL in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12563\r\n* Add category to Normalized Attention Guidance node by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12565\r\n* chore: tune CodeRabbit config to limit review scope and disable for drafts by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12567\r\n* Move LTXAV av embedding connectors to diffusion model. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12569\r\n* Fix dtype issue in embeddings connector. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12570\r\n* Fix issue loading fp8 ltxav checkpoints. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12582\r\n* feat(api-nodes): add KlingAvatar node by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12591\r\n* Update workflow templates to v0.9.2 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12596\r\n* Add SplitImageToTileList and ImageMergeTileList nodes. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12599\r\n* Temporality unbreak some LTXAV (LTX 2.0) workflows to give people time to migrate. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12605\r\n* Patch frontend to 1.39.16 (from 1.39.14) by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12604\r\n* feat(api-nodes): add ByteDance Seedream-5 model by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12609\r\n* Update workflow templates to v0.9.3 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12610\r\n* Don't crash right away if op is uninitialized. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12615\r\n* Update embedded docs to v0.4.3 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12601\r\n\r\n## New Contributors\r\n* @huntcsg made their first contribution in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12511\r\n* @Yourz made their first contribution in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12357\r\n* @artokun made their first contribution in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12550\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.14.1...v0.15.0","2026-02-24T20:56:09",{"id":227,"version":228,"summary_zh":229,"released_at":230},71020,"v0.14.2","* Fix for Gemini\u002FNano banana api nodes that sometimes returned a blank image.\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.14.1...v0.14.2","2026-02-18T06:12:02",{"id":232,"version":233,"summary_zh":234,"released_at":235},71021,"v0.14.1","## What's Changed\r\n* Fix anima LLM adapter forward when manual cast by @chaObserv in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12504\r\n* feat(api-nodes): add \"viduq3-turbo\" model support by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12482\r\n* feat(api-nodes): add Recraft V4 nodes by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12502\r\n* Update workflow templates to v0.8.43 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12507\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.14.0...v0.14.1","2026-02-17T20:31:49",{"id":237,"version":238,"summary_zh":239,"released_at":240},71022,"v0.14.0","## What's Changed\r\n* mp: dont deep-clone objects from model_options by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12382\r\n* sd: delay VAE dtype archive until after override by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12388\r\n* dynamic_vram: ops: Fix vanilla-fp8 loaded lora quality by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12390\r\n* Fix 3D models not showing in outputs window by @guill in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12381\r\n* Add a VideoSlice node by @AustinMroz in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12107\r\n* [Trainer] training with proper offloading by @KohakuBlueleaf in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12189\r\n* Make built in lora training work on anima. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12402\r\n* fix(api-nodes): retry on connection errors during polling instead of aborting by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12393\r\n* Dispatch desktop auto-bump when a ComfyUI release is published by @benceruleanlu in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12398\r\n* [API Nodes] enable Magnific Upscalers by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12179\r\n* dynamic_vram: Fix windows Aimdo crash + Fix LLM performance by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12408\r\n* ace15: Use dynamic_vram friendly trange by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12409\r\n* model_patcher: guard against none model_dtype by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12410\r\n* Update ace15.py to allow min_p sampling by @askmyteapot in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12373\r\n* Add a tip for common error. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12414\r\n* fix(api-nodes): add separate retry budget for 429 rate limit responses by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12421\r\n* ModelPatcherDynamic: force load non leaf weights by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12433\r\n* llama: use a more efficient rope implementation by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12434\r\n* Update command to install AMD stable linux pytorch. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12437\r\n* dynamic_vram: Training fixes by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12442\r\n* Use torch RMSNorm for flux models and refactor hunyuan video code. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12432\r\n* Support generating attention masks for left padded text encoders. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12454\r\n* Fix some custom nodes. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12455\r\n* Add left padding to LTXAV text encoder. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12456\r\n* Add working Qwen 2512 ControlNet (Fun ControlNet) support by @krigeta in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12359\r\n* Patch frontend to 1.38.14 (from 1.38.13) by @christian-byrne in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12469\r\n* Remove unsafe pickle loading code that was used on pytorch older than 2.4 by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12473\r\n* chore(api-nodes): remove \"gpt-4o\" model by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12467\r\n* Node Replacement API by @Kosinkadink in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12014\r\n* feat(api-nodes-Tencent): add ModelTo3DUV, 3DTextureEdit, 3DParts nodes by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12428\r\n* feat(api-nodes): add Bria RMBG nodes by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12465\r\n* Fix lora Extraction in offload conditions (+ dynamic_vram mode) by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12479\r\n* MPDynamic: force load flux img_in weight (Fixes flux1 canny+depth lora crash) by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12446\r\n* Remove workaround for old pytorch. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12480\r\n* Update workflow templates to v0.8.42 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12491\r\n* Remove code to support RMSNorm on old pytorch. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12499\r\n* gitignore venv* by @alexheretic in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12431\r\n* Allow control_after_generate to be type ControlAfterGenerate in v3 schema by @Kosinkadink in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12187\r\n* Implement NAG on all the models based on the Flux code. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12500\r\n* Fix anima preprocess text embeds not using right inference dtype. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12501\r\n\r\n## New Contributors\r\n* @askmyteapot made their first contribution in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12373\r\n* @krigeta made their first contribution in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12359\r\n\r\n**Full Changelog","2026-02-17T06:29:10",{"id":242,"version":243,"summary_zh":244,"released_at":245},71023,"v0.13.0","## What's Changed\r\n* Bump comfyui-frontend-package to 1.38.13 by @comfy-pr-bot in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12238\r\n* Add a Create List node by @AustinMroz in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12173\r\n* Make ace step 1.5 work without the llm. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12311\r\n* Fix some lowvram stuff with ace step 1.5 by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12312\r\n* Fix issue when using disable_unet_model_creation by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12315\r\n* EasyCache: Support LTX2 by @kijai in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12231\r\n* Make ace step 1.5 base model work properly with default workflow. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12337\r\n* Fix return_word_ids=True with Anima tokenizer by @asagi4 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12328\r\n* Fix bug with last pr by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12338\r\n* Support fp16 for Cosmos-Predict2 and Anima by @tdrussell in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12249\r\n* Some fixes to previous pr. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12339\r\n* Pad ace step 1.5 ref audio is not long enough. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12341\r\n* Fix LazyCache by @kijai in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12344\r\n* Add search_aliases to sa-solver and seeds-2 node by @chaObserv in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12327\r\n* Update embedded docs to v0.4.1 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12346\r\n* Disable prompt weights for ltxv2. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12354\r\n* Make tonemap latent work on any dim latents. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12363\r\n* Dynamic VRAM fixes - Ace 1.5 performance + a VRAM leak by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12368\r\n* Iimprovements to ACE-Steps 1.5 text encoding (part 2) by @blepping in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12350\r\n* Ace step prompts match now. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12376\r\n* Update workflow templates to v0.8.37 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12377\r\n* fix(Moonvalley-API-Nodes): adjust \"steps\" parameter to not raise exception by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12370\r\n* feat(api-nodes-Kling): add new models (V3, O3) by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12389\r\n* Update workflow templates to v0.8.38 by @comfyui-wiki in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12394\r\n\r\n## New Contributors\r\n* @tdrussell made their first contribution in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12249\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.12.3...v0.13.0","2026-02-10T20:27:38",{"id":247,"version":248,"summary_zh":249,"released_at":250},71024,"v0.12.3","## What's Changed\r\n* Fix crash on safetensors load for some 3rd party Models by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12266\r\n* mp: Fix checkpoint saving by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12268\r\n* add File3DAny output to Load3D node; extend SaveGLB to accept File3DAny as input by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12276\r\n* Fix ace step nan issue on some hardware\u002Fpytorch configs. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12289\r\n* Try to fix ace text encoder slowness on some configs. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12290\r\n* Add llm sampling options and make reference audio work on ace step 1.5 by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12295\r\n* Disable sage attention on ace step 1.5 by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12297\r\n* Improvements to ACE-Steps 1.5 text encoding by @blepping in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12283\r\n* Add VAE tiled decode node for audio. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12299\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.12.2...v0.12.3","2026-02-05T07:04:07",{"id":252,"version":253,"summary_zh":254,"released_at":255},71025,"v0.12.2","## What's Changed\r\n* Support the 4B ace step 1.5 lm model. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12257\r\n* Dynamic_vram: Remove Aimdo exemption for empty_cache (fixes VRAM leak) by @rattus128 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12260\r\n* Fix crash with ace step 1.5 by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12264\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.12.1...v0.12.2","2026-02-04T06:09:31",{"id":257,"version":258,"summary_zh":259,"released_at":260},71026,"v0.12.1","## What's Changed\r\n* Add progress bar to ace step. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12242\r\n* Fix some issues with mac. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12247\r\n* llama: cast logits as a comfy-weight by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12248\r\n* Fix mac issue. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12250\r\n* feat(comfy_api): add basic 3D Model file types by @bigcat88 in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12129\r\n* Support ace step 1.5 base model loras. by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12252\r\n* Fix tiled vae for ace step 1.5 by @comfyanonymous in https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fpull\u002F12253\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI\u002Fcompare\u002Fv0.12.0...v0.12.1","2026-02-03T21:27:07"]