InstantID is one of methods extending Stable Diffusion, and can fix face of person output by inputting face image. In this article I explain method to use this InstantID in ComfyUI.
Work Flow
1. Introduction of ComfyUI and update to latest version
Introduce ComfyUI. Those already introduced please update ComfyUI to latest version too.
>-
2. Installation of Extension
Install following extension.
Contribute to cubiq/ComfyUI_InstantID development by creating an account on GitHub.
3. Manual Installation of Library
Since python library (insightface) necessary for InstantID is partially not installed by error, install manually.
Download wheel of insightface from link below, then hit install command with pip.
Contribute to Gourieff/Assets development by creating an account on GitHub.
python -m pip install (path to downloaded whl) onnxruntime onnxruntime-gpuIn case of standalone version replace python command with path to execution binary of python. python.exe should be placed around below.
- In
python-embeddeddirectory in case of Comfyui-Standalone - In
Data/Packages/{Package Name}/venv/Scriptsdirectory in case of StabilityMatrix
4. Download and Placement of Model
First download + unzip antelopev2 and place in ComfyUI/models/insightface/models/antelopev2. Can be obtained from either of ↓.
-
antelopev2.zipdrive.google.com
-
MonsterMMORPG/tools at main
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
Next place main model of InstantID in ComfyUI/models/instantid.
Finally add model of controlnet to directory of controlnet.
5. Import of Workflow
Since there is workflow of example, use it.
Contribute to cubiq/ComfyUI_InstantID development by creating an account on GitHub.
6. Generate
Change model to one placed earlier, and finally generate specifying face image and prompt.
Generation Result
Used Animagine XL for model.
Conclusion
From generation result, I think it was felt that image is generated along input face image. By using InstantID, controlling face to generate becomes possible, expanding breadth of image generation. (Please be careful about portrait right)
This time I used only InstantID, but by combining with ControlNet, it seems posture/composition etc. also become controllable. Since around here becomes story of application, I don’t touch in this article, but those interested should try combining while referring to following article and workflow.
>-
Contribute to cubiq/ComfyUI_InstantID development by creating an account on GitHub.
