[{"data":1,"prerenderedAt":456},["ShallowReactive",2],{"page-ComfyUI-1":3,"page-count-ComfyUI":212},[4,237],{"id":5,"title":6,"body":7,"date":221,"description":13,"extension":222,"meta":223,"navigation":229,"path":230,"seo":231,"stem":232,"tags":233,"__hash__":236},"blogs\u002F_legacy\u002F2026\u002F2026-03-22-wan2.2-dasiwa-comfyui-note.md","ComfyUI中安装DaSiWa Wan2.2记录",{"type":8,"value":9,"toc":211},"minimark",[10,14,17,36,39,43,46,53,56,62,67,72,75,78,89,98,102,105,128,131,134,142,145,148,151,168,171,174,183,186,189,195,198,204,207],[11,12,13],"p",{},"Wan2.2 Dasiwa这个之前看到有人说8G显存也能跑，所以试着安装了下。",[11,15,16],{},"使用工具版本：",[18,19,20,24,27,30,33],"ul",{},[21,22,23],"li",{},"ComfyUI 0.12.0",[21,25,26],{},"ComfyUI_frontend v1.37.11",[21,28,29],{},"rgthree-comfy v1.0.2512112053",[21,31,32],{},"EasyUse v1.3.6",[21,34,35],{},"ComfyUI-Manager V3.39.2",[11,37,38],{},"与当前的版本可能有点差距了，因为这篇文章其实是之前在本地的草稿，后面我忘记更新了……",[40,41,42],"h2",{"id":42},"使用效果",[11,44,45],{},"8G内存确实能跑GGUF的版本，但是效果差强人意，如果已经习惯了商用模型的话可能无法适应。其实HuggingFace上能够查到GGUF模型的兼容性",[11,47,48],{},[49,50],"img",{"alt":51,"src":52},"Wan2.2在4060Ti8G上的可用性","\u002Fimages\u002F2026\u002Fwan2.2-gguf-40608g-comp.png",[11,54,55],{},"使用Q2模型的时间消耗",[57,58,59],"blockquote",{},[11,60,61],{},"I2V, 5s, 768x768\nPrompt executed in 479.62 seconds",[57,63,64],{},[11,65,66],{},"I2V, 5s, 432x576\nPrompt executed in 152.66 seconds",[57,68,69],{},[11,70,71],{},"FLF2V, 1s, 672x384\nPrompt executed in 32.20 seconds",[11,73,74],{},"很慢了，而且由于是Q2的模型，其实质量也不是很好。",[40,76,77],{"id":77},"安装过程",[11,79,80,81,88],{},"其实官方有使用说明的，可能一下子找不到，基本按照",[82,83,87],"a",{"href":84,"rel":85},"https:\u002F\u002Fcivitai.com\u002Farticles\u002F26508",[86],"nofollow","官方说明","安装就可以了。",[11,90,91,92,97],{},"官方还提供了",[82,93,96],{"href":94,"rel":95},"https:\u002F\u002Fcivitai.com\u002Farticles\u002F20293\u002Fdarksidewalkers-wan-22-14b-i2v-usage-guide-definitive-edition",[86],"概念解释","。",[99,100,101],"h3",{"id":101},"模型下载",[11,103,104],{},"如果显存不高可以用GGUF版本，",[18,106,107,114,121],{},[21,108,109],{},[82,110,113],{"href":111,"rel":112},"https:\u002F\u002Fcivitai.com\u002Fmodels\u002F1981116?modelVersionId=2555652",[86],"Safetensor版本",[21,115,116],{},[82,117,120],{"href":118,"rel":119},"https:\u002F\u002Fcivitai.com\u002Fmodels\u002F2269796\u002Fdasiwa-wan-22-i2v-14b-synthseduction-v9-or-lightspeed-or-gguf",[86],"GGUF版本-civitai",[21,122,123],{},[82,124,127],{"href":125,"rel":126},"https:\u002F\u002Fhuggingface.co\u002FBedovyy\u002FdasiwaWAN22I2V14B-GGUF",[86],"GGUF版本-huggingface",[11,129,130],{},"Huggingface上的版本不确定是不是作者本人发的，看说明应该版本没有civitai新，有条件的话还是用civitai比较好。",[99,132,133],{"id":133},"快速安装",[11,135,136,137,97],{},"节点相关最快的是，使用官方提供的",[82,138,141],{"href":139,"rel":140},"https:\u002F\u002Fcivitai.com\u002Fmodels\u002F1823089?modelVersionId=2712329",[86],"WorkFlow",[11,143,144],{},"然后ComfyUI在有安装管理器的情况下，会提示缺失的节点并安装。",[11,146,147],{},"以下是遇到的问题",[11,149,150],{},"问题一：No module named 'sageattention'",[152,153,158],"pre",{"className":154,"code":155,"language":156,"meta":157,"style":157},"language-shell shiki shiki-themes github-light-high-contrast github-dark monokai","python -m pip install -U sageattention\n","shell","",[159,160,161],"code",{"__ignoreMap":157},[162,163,166],"span",{"class":164,"line":165},"line",1,[162,167,155],{},[11,169,170],{},"问题二：No module named 'triton'",[11,172,173],{},"似乎triton对windows的兼容性有问题，如果安装失败，可以用社区的windows版本",[152,175,177],{"className":154,"code":176,"language":156,"meta":157,"style":157},"python -m pip install triton-windows\n",[159,178,179],{"__ignoreMap":157},[162,180,181],{"class":164,"line":165},[162,182,176],{},[99,184,185],{"id":185},"使用",[11,187,188],{},"GGUF模型需要将左上角的配置改为使用GGUF",[11,190,191],{},[49,192],{"alt":193,"src":194},"开启GGUF开关","\u002Fimages\u002F2026\u002Fwan2.2-dasiwa-workflow-gguf.png",[11,196,197],{},"以及针对GGUF模型进行配置",[11,199,200],{},[49,201],{"alt":202,"src":203},"配置GGUF模型","\u002Fimages\u002F2026\u002Fwan2.2-gguf-config.png",[11,205,206],{},"测试的时候可以把时间长度和分辨率先降下来，不然可能会卡半天。",[208,209,210],"style",{},"html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html .sepia .shiki span {color: var(--shiki-sepia);background: var(--shiki-sepia-bg);font-style: var(--shiki-sepia-font-style);font-weight: var(--shiki-sepia-font-weight);text-decoration: var(--shiki-sepia-text-decoration);}html.sepia .shiki span {color: var(--shiki-sepia);background: var(--shiki-sepia-bg);font-style: var(--shiki-sepia-font-style);font-weight: var(--shiki-sepia-font-weight);text-decoration: var(--shiki-sepia-text-decoration);}",{"title":157,"searchDepth":212,"depth":213,"links":214},2,3,[215,216],{"id":42,"depth":212,"text":42},{"id":77,"depth":212,"text":77,"children":217},[218,219,220],{"id":101,"depth":213,"text":101},{"id":133,"depth":213,"text":133},{"id":185,"depth":213,"text":185},"2026-03-22","md",{"layout":224,"excerpt":225},"post",{"type":8,"value":226},[227],[11,228,13],{},true,"\u002F2026-03-22-wan2.2-dasiwa-comfyui-note",{"title":6,"description":13},"_legacy\u002F2026\u002F2026-03-22-wan2.2-dasiwa-comfyui-note",[234,235],"ComfyUI","Wan2.2","ji2rvwwNsiBu0XwUih7V2AiaB-W-pZBvRJaZmeDvG9c",{"id":238,"title":239,"body":240,"date":444,"description":244,"extension":222,"meta":445,"navigation":229,"path":450,"seo":451,"stem":452,"tags":453,"__hash__":455},"blogs\u002F_legacy\u002F2026\u002F2026-02-04-qwen-tts-comfyui-note.md","ComfyUI中使用Qwen3-tts记录",{"type":8,"value":241,"toc":434},[242,245,248,251,263,266,275,279,286,295,298,307,310,315,318,327,330,337,341,348,352,355,364,367,370,375,378,381,390,393,404,407,416,419,421,429,432],[11,243,244],{},"Qwen3-tts这个模型，据说效果不错，于是试着安装了下。",[11,246,247],{},"使用设备4060Ti，8GB显存。",[11,249,250],{},"版本信息:",[18,252,253,255],{},[21,254,23],{},[21,256,257,258],{},"flybirdxx\u002FComfyUI-Qwen-TTS ",[82,259,262],{"href":260,"rel":261},"https:\u002F\u002Fgithub.com\u002Fflybirdxx\u002FComfyUI-Qwen-TTS\u002Fcommit\u002Fbebdd5bdd8345aaafc35c96dab47a6db22514fe3",[86],"Commits on Jan 29, 2026",[40,264,234],{"id":265},"comfyui",[11,267,268,269,274],{},"ComfyUI使用的是官方的ComfyUI_windows_portable版本，在",[82,270,273],{"href":271,"rel":272},"https:\u002F\u002Fgithub.com\u002FComfy-Org\u002FComfyUI",[86],"ComfyUI官方库","的Release中可以找到。这个版本的依赖库是装好的，可以减少很多麻烦。用Desktop Application版本应该也是可以的，但是之前不知道点了什么功能，不小心把环境搞坏了，就换了portable的版本手动操作了。",[99,276,278],{"id":277},"custom-node安装","Custom node安装",[11,280,281,282,285],{},"Portable版本可以很方便自己手动安装，在安装前在powershell中先调整python环境，让python.exe指向portable中嵌入的版本即可。例如，我是解包在",[159,283,284],{},"F:\\ComfyUI_windows_portable\\","，那么先执行以下指令，临时修改env便可：",[152,287,289],{"className":154,"code":288,"language":156,"meta":157,"style":157},"$env:PATH = \"F:\\ComfyUI_windows_portable\\python_embeded;\" + $env:PATH\n",[159,290,291],{"__ignoreMap":157},[162,292,293],{"class":164,"line":165},[162,294,288],{},[11,296,297],{},"修改后执行",[152,299,301],{"className":154,"code":300,"language":156,"meta":157,"style":157},"python -m pip --version\n",[159,302,303],{"__ignoreMap":157},[162,304,305],{"class":164,"line":165},[162,306,300],{},[11,308,309],{},"查看输出以确保当前powershell环境正确设置：",[57,311,312],{},[11,313,314],{},"pip 26.0 from F:\\ComfyUI_windows_portable\\python_embeded\\Lib\\site-packages\\pip (python 3.12)",[11,316,317],{},"不过需要留意，这样执行之后pip不会直接修改，需要使用",[152,319,321],{"className":154,"code":320,"language":156,"meta":157,"style":157},"python.exe -m pip install -r .\\requirements.txt\n",[159,322,323],{"__ignoreMap":157},[162,324,325],{"class":164,"line":165},[162,326,320],{},[11,328,329],{},"来安装python包，而不是直接执行pip。",[11,331,332,333,336],{},"之后",[159,334,335],{},"cd .\\ComfyUI\\custom_nodes\\","切换到目录执行各个自定义节点提供的安装步骤即可。",[40,338,340],{"id":339},"comfyui-qwen-tts","ComfyUI-Qwen-TTS",[11,342,343,344],{},"这个在节点管理器里面有好几个，目前用的是这个版本：",[82,345,340],{"href":346,"rel":347},"https:\u002F\u002Fgithub.com\u002Fflybirdxx\u002FComfyUI-Qwen-TTS",[86],[99,349,351],{"id":350},"huggingface镜像设置","Huggingface镜像设置",[11,353,354],{},"使用这个插件的节点会自动下载模型，由于是huggingface的地址，如果有访问问题可以在执行ComfyUI前设置镜像地址",[152,356,358],{"className":154,"code":357,"language":156,"meta":157,"style":157},"$env:HF_ENDPOINT = \"https:\u002F\u002Fhf-mirror.com\"\n",[159,359,360],{"__ignoreMap":157},[162,361,362],{"class":164,"line":165},[162,363,357],{},[99,365,366],{"id":366},"pad_token_id问题",[11,368,369],{},"执行节点时报错",[57,371,372],{},[11,373,374],{},"'Qwen3TTSTalkerConfig' object has no attribute 'pad_token_id'",[11,376,377],{},"这个问题在Github上有，最终解决方案是降级transformers",[11,379,380],{},"可以先检查当前版本",[152,382,384],{"className":154,"code":383,"language":156,"meta":157,"style":157},"python -m pip show transformers\n",[159,385,386],{"__ignoreMap":157},[162,387,388],{"class":164,"line":165},[162,389,383],{},[11,391,392],{},"我这边修改前的版本是",[57,394,395,398,401],{},[11,396,397],{},"Name: transformers",[11,399,400],{},"Version: 5.0.0",[11,402,403],{},"Summary: Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, > audio, and multimodal models, for both inference and training.",[11,405,406],{},"执行以下指令安装目标版本",[152,408,410],{"className":154,"code":409,"language":156,"meta":157,"style":157},"python -m pip install transformers==4.57.3\n",[159,411,412],{"__ignoreMap":157},[162,413,414],{"class":164,"line":165},[162,415,409],{},[11,417,418],{},"即可正常执行节点",[99,420,42],{"id":42},[11,422,423,424,97],{},"文档中有提供各个节点的功能，对于Qwen3-tts本身的能力范畴，可以参考",[82,425,428],{"href":426,"rel":427},"https:\u002F\u002Fqwen.ai\u002Fblog?id=qwen3tts-0115",[86],"Qwen3-tts官方说明",[11,430,431],{},"这边测试的Qwen3-TTS CustomVoice，使用预设的几个声音进行生成，1.7B和0.6B模型都能执行，且生成速度在可接受范围内。",[208,433,210],{},{"title":157,"searchDepth":212,"depth":213,"links":435},[436,439],{"id":265,"depth":212,"text":234,"children":437},[438],{"id":277,"depth":213,"text":278},{"id":339,"depth":212,"text":340,"children":440},[441,442,443],{"id":350,"depth":213,"text":351},{"id":366,"depth":213,"text":366},{"id":42,"depth":213,"text":42},"2026-02-04",{"layout":224,"excerpt":446},{"type":8,"value":447},[448],[11,449,244],{},"\u002F2026-02-04-qwen-tts-comfyui-note",{"title":239,"description":244},"_legacy\u002F2026\u002F2026-02-04-qwen-tts-comfyui-note",[234,454],"Qwen3-tts","trzWv2bGeWame7j8FrpOztXJah97fASk7S8NZV1csHs",1788763176491]