IP-Adapter-FaceID

[C0untFloyd]

I stumbled over https://huggingface.co/h94/IP-Adapter-FaceID which is an ingenious use of the insight swapper model. I know this is quite different from plain swapping but it might prove useful for inpainting problems.

[Jynkz41]

I dont know if it will help inpaint problems, but using ip adapter full face + reactor/roop/whatever works too good in stable diffusion. Ip adapter gets the hair/expresion/glasses/even skin texture and color.. ect .. you get 80% there with that then the last 20% with the face swap...
it works so well becuase face swap mapping itself to a face with the right geometry as your face your going to swap. This also works combining with other source vids and pics using multiple controlnets. Ive done clips in a1111 trying this.. its very slow and there wont be any lip sync.. i hope u can figure out how to make this work, it would be the next big step... also "IP adapter 1.5 full face" works much better than the new face id based swap for doing deepfakes imho

[Jynkz41]

just add same face to ip adaper and reactor/roop and simple prompt, heres the workflow and then results with ip adapter face full 1.5 and then ip adapter with reactor/roop.. not cherry picked first run...

[C0untFloyd]

https://github.com/TencentARC/PhotoMaker

Developments are so quick currently...