Consistent Characters in ComfyUI! [IP Adapter]

In this video we go over using IP Adapter to create a consistent character in ComfyUI. We are able to use this system, specifically IP Adapter PlusV2, to create and embedding that is passed into the model, and used to reconstruct the character in the new image in such a way that it looks similar to the character in the original image.

We download the models from github at:
https://github.com/cubiq/ComfyUI_IPAdapter_plus
^^^Recomended method

or from huggingface at:
http://huggingface.co/h94/IP-Adapter-FaceID/tree/main
^For model model and safetensors
https://huggingface.co/h94/IP-Adapter/tree/main/models/image_encoder
^For Clip Vision, which we will need to rename

The workflow may be found here:
https://github.com/VictoryCATiii3/ComfyUI-Workflows/blob/main/IP_AdapterFace.json

The models that need to be downloaded are:
CLIP-ViT-H-14-laion2B-s32B-b79K.safetensors
-Note, Clip will be downloaded as model.safetensors, so you are going to need to rename it.

Then if you have sdxl you will want to download:
ip-adapter-faceid-plusv2_sdxl.bin
^- Put this in your /ComfyUI/models/ipadapter directory. You might need to make a new one.
ip-adapter-faceid-plusv2_sdxl_lora.safetensors
^- Goes in /ComfyUI/models/loras

If you are using SD 1.5 you will download:
ip-adapter-faceid-plusv2_sd15.bin
^- Goes in /ComfyUI/models/ipadapter
ip-adapter-faceid-plusv2_sd15_lora.safetensors
^- Goes in /ComfyUI/models/loras

If you download both the SD 1.5 models and the SDXL models, the workflow is smart enough to pick the right one for the checkpoint you are using, however if you only intend to use SDXL or SD 1.5 you only need the .bin and lora for the one you are using.

Then we will be using ComfyUI-Manager to download the custom nodes for:
ComfyUI_IPAdapter_plus
If you haven't installed ComfyUI-Manager yet here is a video on how we set it up:
https://youtu.be/5FsA6-OET5c

Once you install the manager you will need to run two pip install commands making sure that you are in the same venv that you use to run ComfyUI:
pip install insightface
pip install onnxruntime

Then you will be able to follow along with the workflow in the video, there are two nodes we are using from IPAdapter_plus, in case you struggle to fine them these are:
IP Adapter FaceID, found at:
Right Click/Add Node/ipadapter/faceid/IPAdapter FaceID

IPAdapter Unified Loader FaceID, found at:
Right Click/Add Node/ipadapter/faceid/IPAdapter Unified Loader FaceID

These will of course only show up if you have downloaded the correct custom nodes, as we go over in the video.

The sdxl checkpoint I am using is juggernaut-xl which can be found on civit.com here:
https://civitai.com/models/133005/juggernaut-xl
Or for the Australian viewers who have informed me that civit is not available in their region, it can be found on huggingface here:
https://huggingface.co/RunDiffusion/Juggernaut-XL-v9

#tutorial #comfyui #aiimagegeneration #consistentcharacters #ai #facetransfer #educational #technology

Видео Consistent Characters in ComfyUI! [IP Adapter] канала VictoryCat