Viggle/Viggle-Animate
Viggle-Animate offers character replacement in video using a single repainted frame as input. This full finetune of MiniMax-H3 avoids pose estimators and segmentation masks, relying instead on direct propagation. The model runs three forward passes, achieving 26 seconds per shot on a B200 GPU. It claims faster performance than Wan2.2-Animate-14B while maintaining motion accuracy for fast actions.
README
Viggle/Viggle-Animate View on Hugging Face
Loading the README from Hugging Face…