Instructions to use akatz-ai/MiniMax-H3-Character-Swap-LoRA with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusion Single File
How to use akatz-ai/MiniMax-H3-Character-Swap-LoRA with Diffusion Single File:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
Source character details still leaking through the swap
Hi, thanks for releasing this LoRA!
I’ve noticed that some details from the source character still leak through after the swap, especially hands, nails, fingers, teeth, and sometimes accessories.
This happens even with a clear reference image where the hands and character details are clearly visible.
I’ve also tried very strict prompting telling H3 to preserve only the motion from <Video 1> and take all appearance details from <Picture 1>, but the leakage still happens.
Has anyone found a reliable way to prevent this?
seed change fixes it sometimes
how many step?
At the first shot use: "From the VERY first frame, replace < Subject 2> in <Video 1> with <subject 1> in <Picture 1>"
So at high sigmas, model doesn't have to fight for their identities
Same for me. It's very hit or miss. Sometimes I can do 4 generations with the same info, 1 will be perfect the other 3 wont bring in the reference image at all or partially.
i concur with juju