MimicMotion
Tencent/MimicMotionReviewed by AI
Checked by RepoPicks Radar on
What is MimicMotion?
MimicMotion is the research code from Tencent for a model that animates a person. You give it one photo and a video of someone moving, and it produces a video of the person in the photo repeating those movements, up to 72 frames at 576x1024 per run. There is no interface: you build a conda environment, download several model files by hand and run inference.py with a configuration file. An NVIDIA card with 16 GB of memory is needed for the full quality.
MimicMotion is free and open source (Apache-2.0) and can replace Runway and Viggle.
How easy is it to install MimicMotion?
D · terminal, several tools or a specific graphics card · Windows
- Windows D
- Linux D
On Linux and Windows the path is the same and it is all terminal: install conda, create the environment from environment.yaml with CUDA PyTorch, download the DWPose and MimicMotion weights with wget, then run "python inference.py --inference_config configs/test.yaml". It also needs an NVIDIA card, 16 GB of video memory for the 72-frame model.
Coming soon: install in 1 click with the RepoPicks app.
MimicMotion on GitHub
- Stars
- 2,690on Nov 10, 2026
- Downloads
- —all releases
- Growth
- +2stars in 6 days
Stars over time
Facts
- Runs on
- Windows, Linux
- You need to install
- Python, NVIDIA graphics card
- To build from source
- Python, NVIDIA graphics card
- Disk space
- 150 MB–450 MB · estimate
- Accounts or keys
- None
- Cost
- Free
- License
- Apache-2.0 (Permissive)
- Can replace
- Runway, DeepMotion, Viggle
- Last change on GitHub
- Nov 18, 2025
MimicMotion risk signals (5)
- Attentioncan animate a real person
The tool produces "High-Quality Human Motion Video Generation" from a photo plus a motion video, so it can show a real person making movements they never made. There is no watermark or consent step in the repository.
AI review
- Attentiondepends on third party model terms
LICENSE: "MimicMotion is licensed under the Apache License Version 2.0 except for the third-party components listed below." The README says "The SVD model stabilityai/stable-video-diffusion-img2vid-xt-1-1 will be automatically downloaded", and that Stability AI model carries its own usage terms.
AI review
- Infono updates for about a year
The last commit is from 2025-11-18 and there are no releases; the newest entry in the News section is the ICML acceptance on 2025-05-03, while 93 issues are open. The pinned dependencies are also old (pytorch 2.0.1, diffusers 0.27.0, transformers 4.32.1).
AI review
- Infoheavy hardware
README.md: "the 72-frame model requires 16GB VRAM (4060ti) and finishes in 20 minutes on a 4090 GPU" for a 35 second video, and "The minimum VRAM requirement for the 16-frame U-Net model is 8GB; however, the VAE decoder demands 16GB".
AI review
- Infoweights downloaded from outside
Setup fetches files from Hugging Face by hand (yolox_l.onnx, dw-ll_ucoco_384.onnx, MimicMotion_1-1.pth) and suggests the mirror "export HF_ENDPOINT=https://hf-mirror.com". The models are not in the repository.
AI review
Evidence is shown in its original language.
What Radar could not check about MimicMotion
- The quality of the result with your own photo and motion video
- The total disk space after all the model files are downloaded
- Whether the old pinned versions still install on a current graphics driver
- Whether it runs on a card with less memory using the CPU for the decoder
- What the third-party component list in the truncated part of the LICENSE allows