showlab/MotionDirector
A text-to-video diffusion model research project that adapts pre-trained models to generate videos with desired motions from example video clips.
Velocity · 7d
+1.1
★ / day
Trend
→steady
star history
MotionDirector customizes text-to-video diffusion models to produce videos with specific motions given reference video clips. The method uses lightweight fine-tuning to decouple appearance and motion, enabling customizable motion generation while preserving visual quality. Published at ECCV 2024 as an Oral presentation.