Type | Workflows |
Stats | 237 |
Reviews | (11) |
Published | Mar 28, 2024 |
Base Model | |
Hash | AutoV2 8B8A894296 |
(check version pages for more images)
This is a face detailer that takes an image of actual face instead of prompt! Allows to transfer face into an image.
SDXL only.
Installation and dependencies
Install Impact pack custom nodes;
Download Photomaker model and place it in "\ComfyUI\ComfyUI\models\photomaker\";
Download ViT-B SAM model and place it in "\ComfyUI\ComfyUI\models\sams\";
Download and open the workflow.
How to use
Load an image to inpaint into (toImage version) or write prompts to generate it (toGen version). Hair is not transferred! So it's good to prompt for hair. Prompting for similar face features also helps a lot.
Load an image with the face you want.
Run it.
FAQ
Q: Face not detected.
A: Increase "threshold" in "BBOX Detector (SEGS)" node.
Q: Face remains the same, no transfer, only color change.
A: Make sure you use SDXL model, not SD1.5 model. If problem persists, try basic SDXL 1.0 model.
Q: Face is still to close to general image, not recognizable face image.
A: Increase denoise
Q: Increasing denoise leads to garbage instead of the face.
A: Not any transfer is possible. You can't inpaint a human face over a Shrek face, unless using specific models or LoRAs.