Krea2-H3-Style-Lora / README.md
TenStrip's picture
Update README.md
6e80e18 verified
|
Raw
History Blame Contribute Delete
867 Bytes
metadata
license: other
license_name: minimax-h3-community-license-agreement
license_link: https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/LICENSE
pipeline_tag: image-text-to-video
base_model:
  - MiniMaxAI/MiniMax-H3
tags:
  - text-to-video
  - image-text-to-video
  - image-to-video

This is a lora for the video h3 model, using krea2 attn k/v. Only effective for pure T2V generations as a lora.

This is extraction result from Krea2 branch of my attn graft experiments. At rank 512 the lora only captures 52% of the SLERP orthagonal delta inside the generated Krea2 full model version for test1 and rank 128 is only 24% (too negligible), those degree shifts are hard extraction targets, but it produces a large shift in T2V usage and visual style. All experimental. Refer to Eros Max page for exact methodology of how cross-model attn grafting works inside H3 model.