Motion2MixamoRig Turns a Phone Video Into a Rigged 3D Animation
A new open-source pipeline takes a single-person video and spits out a Mixamo-rigged GLB you can drop straight into Blender or Unity.
Takeaways
- Motion2MixamoRig retargets human motion from a single-person video onto an Adobe Mixamo-rigged 3D character.
- Built on GVHMR for world-grounded motion recovery plus SMPL-X body model and Mixamo retargeting.
- Outputs skinned
.glb, per-bone rotations, skeleton.npz, and four preview videos per run. - Apache-2.0 code; around 1,000 GitHub stars already on the repo.
- Runs on CPU; first run pulls ~5 GB of weights and takes 8 to 15 minutes per 30-second clip.
- Limited to single-person clips; hand detail limited; SMPL-X assets are gated for non-commercial use by default.
Motion2MixamoRig is an Apache-2.0 command-line pipeline that accepts a monocular video of one person and a Mixamo FBX character, recovers the person’s body motion, retargets it to the character, and exports a skinned, animated .glb. “Monocular” means the source can come from one ordinary camera, including a phone, with no depth sensor, marker suit, or capture studio.
The repository has 2 GitHub stars. Its practical contribution is packaging several research components behind a small CLI, complete with preview renders and intermediate motion files for inspection. The Apache license covers the repository’s code; SMPL-X, Mixamo assets, and downloaded model weights carry separate terms.
Gravity anchors the recovered motion
The pipeline uses GVHMR to estimate human motion from the source video. GVHMR defines a frame-by-frame coordinate system from estimated gravity and the camera’s viewing direction, giving the recovered body a consistent reference for orientation and movement through the scene. Its per-frame regression also limits the accumulated drift associated with autoregressive methods that feed each prediction into the next.
Motion2MixamoRig represents the recovered person with an SMPL-X body model, then maps those joints and rotations onto the target Mixamo skeleton. The final GLB contains a mesh bound to the rig plus its animation, while the intermediate files preserve the motion data for reuse or debugging.