Hugging Face's logo

akatz-ai
/
MiniMax-H3-Character-Swap-LoRA

Video-to-Video
Diffusion Single File
MiniMax H3
English
lora
character-swap
video-editing
ref2va
comfyui
ai-toolkit
Model card Files
xet
 Community
13

How can I preserve some face details from the source video?

#10
by ywwynm - opened
Discussion
ywwynm

First, thanks for this wonderful LoRA!
I'm using the very latest weight and workflow. I want to preserve some face details from the source video, like decorations, stickers, liquid on face, etc. How can I do that?
akatz-ai
 Owner

dale-walking-dead

Working on it chief 🫡
ywwynm

You are absolutely a master!
BTW, I also find that for some videos, the face swap simply doesn't work even when SAM detects the person accurately and the video is not complicated. Probably because of the face decorations, liquid or stickers on face like I mentioned?
Tamilboy1

99% success rate for me with this LORA, this is fantastic work, only thing that is a hit of miss for me is the teeth :) keep up the good work.
ywwynm

@Tamiboy1 How does your workflow look like?

99% success rate for me with this LORA, this is fantastic work, only thing that is a hit of miss for me is the teeth :) keep up the good work.
Tamilboy1

I use the workflow that is on this repo, posted by the creator.
ywwynm

@akatz-ai And is it possible to only swap head? I also tried to change the prompt to not preserve the reference character's clothing or body details, but failed.

Sign up or log in to comment