https://github.com/comfyanonymous/ComfyUI
---python3.11
https://abc.htmltoo.com/thread-46907.htm
---Conda
https://abc.htmltoo.com/thread-46888.htm
git clone https://github.com/comfyanonymous/ComfyU
---pip3
https://abc.htmltoo.com/thread-46890.htm
-注意python版本
https://abc.htmltoo.com/thread-46907.htm
apt-get --reinstall install python3.11
cd /data/site/htmltoo.opt/ComfyUI
chmod -R 777 /data/site/htmltoo.opt/ComfyUI
export PATH=~/anaconda3/bin:$PATH
source ~/.bashrc
conda create -n comfyenv python=3.11
conda activate comfyenv
-Nvidia
conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia
apt-get install -y opus-tools ffmpeg libjpeg-dev libtiff5* libavcodec-dev libavformat-dev libswscale-dev libv4l-dev libxvidcore-dev libx264-dev libgtk2.0-dev libgtk-3-dev libatlas-base-dev gfortran python3-dev
pip3 install opencv-python
pip3 install --upgrade opencv-python
pip3 install --upgrade torch
#
cd /data/site/htmltoo.opt/ComfyUI/custom_nodes/ComfyUI-Stable-Video-Diffusion
-
cd /data/site/htmltoo.opt/ComfyUI/custom_nodes/ComfyUI-Frame-Interpolation
cd /data/site/htmltoo.opt/ComfyUI/custom_nodes/ComfyUI-Impact-Pack
-
python3 install.py
-
cd /data/site/htmltoo.opt/ComfyUI/custom_nodes/was-node-suite-comfyui/cstr
cd /data/site/htmltoo.opt/ComfyUI/custom_nodes/was-node-suite-comfyui/ffmpy
cd /data/site/htmltoo.opt/ComfyUI/custom_nodes/was-node-suite-comfyui/img2texture
python3 setup.py install
cd /data/site/htmltoo.opt/ComfyUI/
pip3 install -r requirements.txt
pip3 install -r requirements-two.txt
python3 main.py
nohup python3 main.py --listen 0.0.0.0 > main.log 2>&1 &
-清理缓存
pip3 cache purge
-更新要求文件：如果你更新了软件包的版本或要求文件，确保要求文件中的哈希值与新版本匹配
pip3 install --generate-hashes -r requirements.txt
-禁用哈希校验：在一些情况下，你可能希望临时禁用哈希校验来解决问题
pip3 install --no-cache-dir --no-deps -r requirements.txt
-设置一个 5 到 6 的软连接
ln -s /usr/lib/x86_64-linux-gnu/libtiff.so.6.0.0 /lib/x86_64-linux-gnu/libtinfo.so.5
-No module named 'huggingface_hub'
pip3 install -U huggingface_hub
pip3 install -U aiohttp
pip3 install -U multidict
pip3 install -U deepdiff
pip3 install -U insightface
pip3 install -U piexif
pip3 install -U onnxruntime
pip3 install -U accelerate
pip3 install -U py-cpuinfo
pip3 install -U pynvml
pip3 install -U segment_anything
pip3 install -U ultralytics
pip3 install -U ollama
pip3 install -U pydantic
pip3 install -U decord
pip3 install -U cv2
pip3 install -U translate
pip3 install -U gguf
pip3 install -U instructor
pip3 install -U openai
pip3 install -U transformers
pip3 install -U torchsde
pip3 install -U kornia
pip3 install -U regex
pip3 install -U trampoline
pip3 install -U kornia-rs
pip3 install -U matplotlib
pip3 install -U numba
pip3 install -U numpy==1.26
pip3 install -U llama-cpp-python==0.2.89
---版本---
transformers 4.44.0 requires regex!=2019.12.17, which is not installed.
torchsde 0.2.6 requires trampoline>=0.1.2, which is not installed.
kornia 0.7.3 requires kornia-rs>=0.1.0, which is not installed.
transformers 4.44.0 requires tokenizers<0.20,>=0.19, but you have tokenizers 0.20.0 which is incompatible
-
pip3 install -U numpy==1.26
---异常---
pip3 install -U llama_cpp
#存放位置
-大模型
ComfyUI/models/checkpoints
# 导入工作流
左上角， browse，导入工作流。
右上角， Manager，跳过更新检查，数据库本地，
节点管理，安装缺失管理， 点标题，新打开链接，git，
上传到，custom_nodes，pip3 install requirements.txt， 重启。
#插件
cd /data/site/htmltoo.opt/ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Manager.git
git clone https://github.com/AIGODLIKE/AIGODLIKE-ComfyUI-Translation.git
git clone https://github.com/11cafe/comfyui-workspace-manager
git clone https://github.com/crystian/ComfyUI-Crystools
git clone https://github.com/pythongosssss/ComfyUI-WD14-Tagger
git clone https://github.com/ssitu/ComfyUI_UltimateSDUpscale
git clone https://github.com/ArtVentureX/comfyui-animatediff
git clone https://github.com/yolain/ComfyUI-Easy-Use
-异常调试
git clone https://github.com/Gourieff/comfyui-reactor-node
git clone https://github.com/biegert/ComfyUI-CLIPSeg
--
-ComfyUI_Custom_Nodes_AlekPet
提供了提示词的翻译功能。有了这个插件，再也不用使用第三方翻译软件翻译提示词后，再粘贴到CLIP文本编码器中了。
同时提供的PoseNode、PainterNode两个节点可以编辑和绘画姿态，实现所画即所得。
-ComfyUI-Custom-Scripts
给模型分类，更加便捷的调用模型
针对输入自动补齐，信息的索引
更丰富的功能设定页面
-ComfyUI-Advanced-ControlNet
ControlNet插件，使用线稿、姿态、深度、IP等控制数据和响应的ControlNet 模型控制图片的生成，Stable Diffusion 必备。
-comfyui_controlnet_aux
ControlNet的预处理器，用于从参考图片中提取线稿、姿态、深度、IP等控制数据。
-ComfyUI_IPAdapter_plus
用于图像到图像调节的非常强大的模型。图像的主题甚至风格都可以迁移到所生成的图像中。
-ComfyUI Workspace Manager
ComfyUI 工作流程和模型管理器扩展，可在一个地方组织和管理所有工作流程、模型和生成的图像。在工作流程之间无缝切换，跟踪版本历史记录和图像生成历史记录，一键安装 Civit ai 的模型，浏览/更新已安装的模型
-ComfyUI-Crystools
资源监视器、进度条和经过的时间、元数据和两个图像之间的比较、两个JSON之间的比较、向控制台/显示器显示任何值、管道等等！ 这提供了更好的节点来加载/保存图像、预览等，并在不加载新工作流程的情况下查看“隐藏”数据。
-ComfyUI-WD14-Tagger
允许从图像中查询 booru 标签，支持标记和输出多个批处理输入。可搭配ipadapter一起食用，实现一键复刻目标图像
-ReActor Node for ComfyUI 换脸插件
用于自动检测图片中的人脸并置换指定人脸，但无法改变脸型。还原度还行，仁者见仁智者见智
-ComfyUI-CLIPSeg
通过语义文本识别图片中的元素，并输出遮罩，支持输出热成像图；例如输入face，检测人物面部，自动对面部区域进行遮罩
-SD放大 SD分块放大
将图像拆分成N个小块，并结合放大模型对每一个块进行放大。放大效果很不错，但是速度非常慢，使用时要注意
-AnimateDiff 动画节点
动画生成节点，支持通过文生图方式生成动画，并且可以搭配运镜lora实现不同的运镜效果。还可以搭配cn实现动画控制，玩法非常多
-ComfyUI_IPAdapter_plus 风格参考组件
风格参考插件，超强控图节点，通过参考图风格生成类似风格的图片。玩法灵活，可搭配组合多张图片，或者多个风格embbding，另外还支持ipadapter faceid，通过该技术可以提取人物面部特征并生成全新的图片
-ComfyUI Easy Use 简单易用的comfyui节点封装，并支持美化节点
一个化繁为简的节点整合包， 在 tinyterraNodes 的基础上进行延展，并针对了诸多主流的节点包做了整合与优化，以达到更快更方便使用ComfyUI的目的，在保证自由度的同时还原了本属于Stable Diffusion的极致畅快出图体验
# 下载comfyUI_IPAdapter_plus
https://github.com/cubiq/ComfyUI_IPAdapter_plus
手功下载后，解压，放置 ComfyUI/custom_nodes目录下面
预先训练的模型可以在huggingface上获得，下载并将它们放置在ComfyUI/custom_nodes/ComfyUI_IPAdapter_plus/models目录中。
https://huggingface.co/h94/IP-Adapter/resolve/main/models/ip-adapter_sd15.bin
https://huggingface.co/h94/IP-Adapter/blob/main/models/ip-adapter_sd15_light.bin
https://huggingface.co/h94/IP-Adapter/resolve/main/models/ip-adapter-plus_sd15.bin
https://huggingface.co/h94/IP-Adapter/resolve/main/models/ip-adapter-plus-face_sd15.bin
https://huggingface.co/h94/IP-Adapter/resolve/main/sdxl_models/ip-adapter_sdxl.bin
https://huggingface.co/h94/IP-Adapter/resolve/main/sdxl_models/ip-adapter_sdxl_vit-h.bin
https://huggingface.co/h94/IP-Adapter/resolve/main/sdxl_models/ip-adapter-plus_sdxl_vit-h.bin
https://huggingface.co/h94/IP-Adapter/resolve/main/sdxl_models/ip-adapter-plus-face_sdxl_vit-h.bin
此外，您需要将图像编码器放置在ComfyUI/models/clip_vision/目录中:
https://huggingface.co/h94/IP-Adapter/resolve/main/models/image_encoder/model.safetensors
https://huggingface.co/h94/IP-Adapter/resolve/main/sdxl_models/image_encoder/model.safetensors
下载好，放置在正确的目录，重启，就可以了
# ComfyUI-CLIPSeg， clipseg.py
vim /data/site/htmltoo.opt/ComfyUI/custom_nodes/ComfyUI-CLIPSeg/custom_nodes/clipseg.py
# Predict the segemntation mask with torch.no_grad(): outputs = model(**input_prc) tensor = torch.sigmoid(outputs[0]) # get the mask
改成
# Predict the segemntation mask with torch.no_grad(): outputs = model(**input_prc) preds = outputs.logits.unsqueeze(1) tensor = torch.sigmoid(preds[0][0]) # get the mask
# civitai
API: htmltoo / ce4adcc781c53ac7013641a3f9f03bfc
#
-ComfyUI\models\checkpoints\SVD
wget https://hdongquane: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@huggingface.co/stabilityai/stable-video-diffusion-img2vid-xt/resolve/main/svd_xt.safetensors
-ComfyUI\models\checkpoints\hunyuan_dit_comfyui
wget https://hdongquane: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@huggingface.co/comfyanonymous/hunyuan_dit_comfyui/resolve/main/hunyuan_dit_1.2.safetensors
wget https://hdongquane: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@huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned-emaonly.ckpt
-ComfyUI/models/clip_vision/
wget https://hdongquane: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@huggingface.co/stabilityai/control-lora/resolve/main/revision/clip_vision_g.safetensors
# 工作流
-文件夹名称 作用
checkpoints 出图大模型
controlnet controlnet
embeddings 打包好的提示词，比如 embedding:EasyNegativeV2
loras Lora 模型
upscale_models 放大模型
vae vae
#
extra_model_paths.yaml.example -> extra_model_paths.yaml
# 下载插件
https://github.com/talesofai/comfyui-browser.git
->
https://kkgithub.com/talesofai/comfyui-browser.git
# MiniCPM-V-2_6-int4
https://github.com/IuvenisSapiens/ComfyUI_MiniCPM-V-2_6-int4
实现已经无缝集成到ComfyUI平台中，支持基于文本的查询、视频查询、单幅图像查询和多幅图像查询，以生成标题或响应
cd custom_nodes
git clone https://github.com/IuvenisSapiens/ComfyUI_MiniCPM-V-2_6-int4
pip3 install -U -r requirements.txt
#
https://www.fszx-ai.cn/portal.php
https://github.com/kijai/ComfyUI-Florence2
Florence-2是一个高级视觉基础模型，它使用基于提示的方法来处理各种视觉和视觉语言任务。Florence-2可以解释简单的文本提示，以执行字幕、对象检测和分段等任务。它利用我们的FLD-5B数据集，包含1.26亿张图像的54亿条注释，来掌握多任务学习。该模型的序列到序列架构使其能够在零射击和微调设置中表现出色，证明是一个有竞争力的视觉基础模型。
https://github.com/kijai/ComfyUI-segment-anything-2
使用segment-any-2的ComfyUI节点
# requirements.txt
torch torchsde torchvision torchaudio einops transformers tokenizers sentencepiece safetensors aiohttp pyyaml Pillow scipy tqdm psutil #non essential dependencies: kornia spandrel soundfile
# requirements-two.txt
translate opencv-python timm transformers fairscale pycocoevalcap opencv-python qrcode[pil] pytorch_lightning kornia pydantic segment_anything omegaconf boto3 tqdm pandas requests deepdiff torch Pillow pynvml py-cpuinfo piexif pilgram openai diffusers accelerate clip_interrogator lark-parser onnxruntime opencv-python #sentencepiece spandrel torch einops opencv-contrib-python kornia scipy Pillow torchvision tqdm cupy-wheel piexif gguf segment-anything scikit-image piexif transformers opencv-python-headless GitPython scipy ollama instructor pydantic openai pillow scipy color-matcher matplotlib huggingface_hub mss opencv-python GitPython PyGithub matrix-client transformers huggingface-hub typer rich typing-extensions py-cpuinfo gitpython moviepy opencv-python scikit-build typing diskcache ollama pyyaml opencv-python imageio-ffmpeg onnxruntime send2trash torch importlib_metadata huggingface_hub scipy opencv-python filelock Pillow einops torchvision pyyaml scikit-image python-dateutil mediapipe svglib fvcore yapf omegaconf ftfy addict yacs trimesh[easy] albumentations scikit-learn numba colour-science rembg pixeloe transparent-background insightface onnxruntime onnxruntime-gpu torch torchvision torchaudio pillow huggingface_hub transformers decord bitsandbytes clip-interrogator simpleeval cmake fairscale gitpython imageio joblib matplotlib numba pilgram rembg scikit-image scikit-learn scipy timm tqdm transformers tqdm aiohttp-retry amqp billiard blinker celery certifi cfgv charset-normalizer click click-didyoumean click-plugins click-repl distlib filelock Flask fsspec identify idna itsdangerous Jinja2 kombu MarkupSafe mpmath networkx nodeenv packaging platformdirs pre-commit prompt-toolkit psutil PyYAML requests show-in-file-manager six sympy tqdm typing_extensions tzdata vine virtualenv wcwidth Werkzeug fsspec matplotlib requests omegaconf transformers pytorch-lightning packaging regex ftfy tqdm sentencepiece protobuf pyyaml huggingface_hub blake3 tqdm aiohttp-retry aiofiles ultralytics matplotlib==3.8 python-dateutil==2.9.0.post0 numpy==1.26.4 antlr4-python3-runtime==4.9.3 opencv-python-headless[ffmpeg]
