prompt2rig: type a motion, get a rigged Mixamo animation by way of a video model

prompt2rig was created on Oct 19 and has about 360 GitHub stars. The idea is clever: instead of generating motion data directly, it generates a video of a performer doing the motion, then runs motion capture on that video. Your prompt goes to a video model on fal (the README lists FLUX 3 Max at 720p, about 8–9 s, or H3 Max Turbo at 480p, about 2–3 s). MediaPipe gives an instant preview in the browser. The HQ pass runs GVHMR on your own GPU for 3D body motion (the README says ~3.5–4 s per 5-second clip on an RTX 4090) and WiLoR for fingers. The result is retargeted onto X Bot, Michelle, Soldier or any Mixamo .fbx/.glb and exported as .glb for Blender or Unity. A stream mode chains clips together from each one's last frame.

Requirements: Node 18+ and a FAL_KEY. The HQ pass needs Windows with an NVIDIA GPU, uv and ffmpeg, about 5 GB of checkpoints, and SMPL-X registration (MANO too if you want hands). The README's prompting advice: keep the performer facing the camera with limbs clear. Flips, spins and floor work confuse it.

Why it's on ReelStack: a wave, bow or kick for a 3D character used to mean a mocap suit or an afternoon of keyframing. This turns it into a prompt plus a few seconds of GPU time. That's useful for Blender blockouts that feed Seedance or LTX layout-to-render workflows, and for stylized 3D explainers.

Our take: the output can only be as clean as the generated clip, so expect foot sliding and jitter on fast moves. Check the SMPL-X licence terms before using it for client work. Spotted, not tested.