Description
Motion capture is the expensive part of character animation. Pro rigs are out of reach, and a phone clip usually gives you jittery data with feet sliding on the floor that you then fix frame by frame in Blender. RGB MoCap is a native Windows app that turns an ordinary video into 3D human motion, with the AI running locally on your own GPU so the footage never leaves your machine.
It doesn't stop at raw data. After reconstruction you can see foot and hand contact spans on the timeline, adjust them, apply a correction to cut the sliding, then retarget the motion onto your own character and export.
Several AI models: ExMR, FlowHMR and GVHMR come with the base install, and you can add MAMMA for synchronized multi-camera video and PromptHMR for steadier relative placement between people in one video. In the author's own subjective tests ExMR had better overall motion quality than PromptHMR, while PromptHMR helped with multi-person interaction.
Contact correction: Foot and palm contact correction with editable timings for every lock on the timeline, aimed at foot sliding and floor penetration.
Character retargeting: Move the motion onto Mixamo FBX, VRM and PMX humanoid characters.
Export formats: Source animation to FBX, characters to FBX with or without meshes, and PMX character motion to VMD for MMD.
Multiple people and projects: Track one or several people in a clip, and projects keep a separate result per AI model so you can compare models.
Tunable VRAM: Resource Mode defaults to Automatic and falls back to a low-memory path when needed. In the author's short-clip tests peak VRAM dropped from 5.68 to 3.87 GiB for ExMR and from 6.10 to 2.77 GiB for FlowHMR, which are short-clip numbers, not a full-job guarantee.
It needs 64-bit Windows, a compatible NVIDIA GPU and OpenGL 4.3. Install by cloning the repo and running install.cmd, then start with RGBMoCap.cmd. Model files are downloaded separately in Model Manager, and some need registration and a manual download. All five models take about 63 GiB, so check your disk first.
It doesn't stop at raw data. After reconstruction you can see foot and hand contact spans on the timeline, adjust them, apply a correction to cut the sliding, then retarget the motion onto your own character and export.
Features
Several AI models: ExMR, FlowHMR and GVHMR come with the base install, and you can add MAMMA for synchronized multi-camera video and PromptHMR for steadier relative placement between people in one video. In the author's own subjective tests ExMR had better overall motion quality than PromptHMR, while PromptHMR helped with multi-person interaction.
Contact correction: Foot and palm contact correction with editable timings for every lock on the timeline, aimed at foot sliding and floor penetration.
Character retargeting: Move the motion onto Mixamo FBX, VRM and PMX humanoid characters.
Export formats: Source animation to FBX, characters to FBX with or without meshes, and PMX character motion to VMD for MMD.
Multiple people and projects: Track one or several people in a clip, and projects keep a separate result per AI model so you can compare models.
Tunable VRAM: Resource Mode defaults to Automatic and falls back to a low-memory path when needed. In the author's short-clip tests peak VRAM dropped from 5.68 to 3.87 GiB for ExMR and from 6.10 to 2.77 GiB for FlowHMR, which are short-clip numbers, not a full-job guarantee.
It needs 64-bit Windows, a compatible NVIDIA GPU and OpenGL 4.3. Install by cloning the repo and running install.cmd, then start with RGBMoCap.cmd. Model files are downloaded separately in Model Manager, and some need registration and a manual download. All five models take about 63 GiB, so check your disk first.
