fudan-generative-vision/champ

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

4261 stars · 485 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Champ is a research framework for controllable and consistent human image animation using 3D parametric guidance (SMPL-based depth, normal, and semantic maps) with diffusion models. It animates a single reference human image according to motion from a driving video, and includes training code and SMPL preprocessing scripts.

Use cases

• animate a photo of a person using motion from a dance video
• generate consistent human motion videos from a single image
• preprocess videos into SMPL pose data for animation
• train a custom human animation model
• retarget human motion between characters

When to choose

• you need pose-controllable human image-to-video animation with 3D guidance
• you want an open research codebase for human animation with training support
• you have GPU resources and want to animate character images with driving motion

When to avoid

• you need general-purpose text-to-video generation without human pose control
• you lack a GPU or cannot run diffusion model inference locally
• you need a polished end-user application rather than a research codebase

Facets

library · maturity maintenance

video-processing machine-learning deep-learning image-processing computer-vision deep-learning artificial-intelligence python human-animation video-generation diffusion-models smpl pose-guided-animation eccv-2024 video gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="fudan-generative-vision/champ")