showlab/MotionDirector

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1050 stars · 62 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

MotionDirector is a research library for customizing text-to-video diffusion models to generate videos with desired motions from a small set of reference clips. It uses a dual-path LoRA architecture and appearance-debiased temporal loss to decouple motion learning from appearance.

Use cases

• generate videos with a specific motion from reference clips
• customize text-to-video diffusion models with LoRA
• mix appearance from one video with motion from another
• animate a single image with a customized motion
• create videos with prescribed camera movements
• research on motion customization for video diffusion

When to choose

• you need to teach a video diffusion model a specific motion concept
• you want decoupled control over appearance and motion in generated videos
• you are reproducing or building on ECCV 2024 motion customization research

When to avoid

• you need production-grade video generation without training
• you lack a GPU or cannot run diffusion model fine-tuning
• you only need simple text-to-video generation without motion customization

Facets

library · maturity active

machine-learning deep-learning video-processing image-processing deep-learning artificial-intelligence image-processing python diffusion-models text-to-video motion-customization lora eccv-2024 video-generation research video gpu linux docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="showlab/MotionDirector")