InstantID+Style+Composition. Or use InstantID without style/compositon and use the included One Button Prompts to come up with new ideas. Combine multiple faces into the ultimate visage of your own design.
Includes various post-processing filters to enhance your final output.
This is an advanced workflow. It wil require all sorts of fiddling to get what you want. Don't be afraid to change all the settings, especially in the Style/Composition node and the PerturbedAttentionGuidance node. Don't be afraid to change things up.
InstantID Settings Example
Only InstantID + Prompting was used for this, no extra style/composition.
ControlNet weight is very important for likeness, but going too high can apply too much of the style of the face image (e.g. a photo, in this case, produces a more photographic image at a high weight).
IPAdapter weight only counts for approximately 20~30% of the face, so it can be kept lower normally.
Noise can be used to make small variations in the final image.
Recommended Upscale Models
I recommended several upscale models in the workflow. I didn't think to link them in the note, so here they are:
Description
Original version
FAQ
Comments (5)
Is there any other known name for "instantID-controlnet_model.safetensors"?
You can find links to all models in the link below. The model in question is under the "ControlNet" link, and is named "diffusion_pytorch_model.safetensors". I renamed mine for organizational purposes.
https://github.com/cubiq/ComfyUI_InstantID?tab=readme-ov-file#installation
@Monster Thanks! I'll verify if I have already it under its original name or another one. I think ComfyUI should have "model hash" system like A1111. It would be way more easier for workflow sharing.
@XX007 Absolutely.
Hello! Can you send final workflow with images and other settings? I'm truing to use downloaded workflow, but i have really ugly result((