This LoRA contextually fills in the green area in image1 with head from image2.
Basic prompt:
fill in the green area in image1 with head from image2
Sometimes it helps to expand the basic prompt a bit:
fill in the green area in image1 with head from image2, maintain correct anatomy (maintain identity, from back)
Mix with other face swap or consistence LoRA sometimes helps sometimes not. The green area should be specifically RGB(0,255,00). You guys can do it with Mask Editor in ComfyUI (latest), but personally I convert mask to area with RGB(0,255,0).
Best resolution for image2 (source head) is 1024 x 1024. There are no rules about background, zoom nor pose of the source head.
Workflow i'm using:
https://civarchive.com/models/2307457/aint-another-head-swap-workflow?modelVersionId=2596183
Description
Rank 64, more trening, more images, another prompt
FAQ
Comments (13)
What's new in v3.2?
Has the issue of head proportions been resolved?
Yes, but head size match is best when image2 (head reference) contains not only head but also shoulders.
New trening session in rank64, new prompt, more pictures in dataset.
Aaaa... and better face consistence... i hope ;)
Flux.2 Klein 9B will be the next model for our concept.
I give up :( That model is untrainable.
Back to QIE2511.
Haha, I also realized that Klein isn’t really suitable for tasks like adding makeup or head swapping—it’s much better suited for natural-language–driven editing.
hi wondering if i give you a safetensors file you can know how to retrain the lora for another model? like for ex a lora from edit 2509 to adapt it and maybe improve it for 2511 just with the .safetensors?
Sorry mate, I don't know how to do this. The only way i believe is to train the same imageset on the new base model 2511.
What do you mean by "First Pass" and "Second Pass" in the examples?
There is a workflow dedictaed the LoRA. There are 2 passes. The second is for head area upscaled.