ComfyUI Extension: ComfyUI_MS_Diffusion

Youu can using MS_Diffusion in ComfyUI

-- MS-Diffusion origin From: MS-Diffusion

NEW Update

• 2024/09/06:fix runway error

Previous updates

• del clip repo，Add comfyUI clip_vision loader/加入comfyUI的clip vision节点，不再使用 clip repo。

1.Installation

In the ./ComfyUI /custom_node directory, run the following:

git clone https://github.com/smthemex/ComfyUI_MS_Diffusion.git

2.requirements

pip install -r requirements.txt

缺啥装啥。
If the module is missing, please pip install

3 Need model

3.1 base model:
底模使用社区单体模型，首次使用需要下载一些必备的config文件，没连外网或者没有做镜像站映射的，肯定是用不了的。
Choose one of the three options above the model loading node, and change the rest to "none" or leave it blank. The first option is to select a standalone SDXL community model, the second option is to select a pre downloaded SDXL diffuser model, and the third option is to use the "repo_id" method
The default repo_id models are stabilityai/stable-diffusion-xl-base-1.0 (for example,you can using :G161222/RealVisXL_V4.0,sd-community/sdxl-flash...)

3.2 adapter and encoder model
必须的模型，位置如图：
Need download "ms_adapter.bin" : link

├── ComfyUI/models/
|      ├──photomaker/
|             ├── ms_adapter.bin

clip vision model （ any base from CLIP-ViT-bigG-14-laion2B-39B-b160k）

├── ComfyUI/models/
|      ├──clip_vision/
|             ├── clip_vision_g.safetensors  or CLIP-ViT-bigG-14-laion2B-39B-b160k.safetensors

3.3 only supports SDXL controllnet
以下是controlnet的模型和config文件，ComfyUI的用的单体controlnet模型用不了。用了太占显存了，改回pipe基本加载模式。

├── ComfyUI/models/diffusers/   
|     ├──xinsir/controlnet-openpose-sdxl-1.0    
|         ├── config.json   
|         ├── diffusion_pytorch_model.fp16.safetensors   
|     ├──xinsir/controlnet-scribble-sdxl-1.0   
|         ├── config.json   
|         ├── diffusion_pytorch_model.fp16.safetensors   
|     ├──diffusers/controlnet-canny-sdxl-1.0   
|         ├── config.json   
|         ├── diffusion_pytorch_model.fp16.safetensors   
|     ├──diffusers/controlnet-depth-sdxl-1.0   
|         ├── config.json   
|         ├── diffusion_pytorch_model.fp16.safetensors
|     ├──/controlnet-zoe-depth-sdxl-1.0  
|         ├── config.json   
|         ├── diffusion_pytorch_model.fp16.safetensors
|     ├──TheMistoAI/MistoLine 
|         ├── config.json   
|         ├── diffusion_pytorch_model.fp16.safetensors
|     ├──xinsir/controlnet-tile-sdxl-1.0
|         ├── config.json   
|         ├── diffusion_pytorch_model.fp16.safetensors
   

Control_img image preprocessing, please use other nodes

4 using tips

--生成物体名称需要用[ ]括起来，有多少个物体，就要有多少张图片输入;
--To generate object names, they need to be enclosed in [ ]. As many objects as there are, there must be as many images to input;

5 Example

normal 2 boject 常规双主体图生图 最新示例。

normal 4 boject 4主体图生图。更多的主体没有测试，旧示例，仅供参考。

2 object zero shot and controlnet img2img 双物体加controlnet引导 图生图,旧示例，仅供参考

6 Citation

MS-Diffusion

@misc{wang2024msdiffusion,
  title={MS-Diffusion: Multi-subject Zero-shot Image Personalization with Layout Guidance}, 
  author={X. Wang and Siming Fu and Qihan Huang and Wanggui He and Hao Jiang},
  year={2024},
  eprint={2406.07209},
  archivePrefix={arXiv},
  primaryClass={cs.CV}
}

IP-Adapter

@article{ye2023ip-adapter,
  title={IP-Adapter: Text Compatible Image Prompt Adapter for Text-to-Image Diffusion Models},
  author={Ye, Hu and Zhang, Jun and Liu, Sibo and Han, Xiao and Yang, Wei},
  booktitle={arXiv preprint arxiv:2308.06721},
  year={2023}
}