MovementModeler

MovementModeler turns everyday video into clean, accurate human motion—fully on device.
Visualize movement using stick figures, capsule rigs, or sprite-based rigs.
Export motion as MP4 videos or BODY-25 JSON for analysis, research, and ML pipelines.

MovementModeler converts ordinary video into precise motion visualizations using fast, fully on-device processing. Built for creators, athletes, coaches, researchers, and computer-vision practitioners, it provides a straightforward way to extract, review, and export human motion—without accounts, cloud uploads, or privacy trade-offs.

Load a video from your library and MovementModeler automatically generates a real-time motion preview for every detected person. Multi-person scenes are handled through persistent track identification, keeping individuals consistently labeled even during motion, occlusion, or crossing paths. Any combination of people can be isolated for preview or export, making it easy to focus on a single subject or compare movements side by side.

Choose between multiple visualization styles—including classic stick figures, capsule rigs, and sprite-based rigs—to match your use case or aesthetic preference.

Built-in refinement controls help clean up real-world motion artifacts in real time. Smooth jittery joints, stabilize motion, suppress low-confidence frames, and fine-tune joint continuity for both global playback and individual frames—all without modifying the underlying data. Playback matches the original video speed, with frame-accurate scrubbing for detailed inspection.

For export, generate lightweight MP4 motion videos using your selected visualization style, or BODY-25 JSON keypoint data suitable for research, robotics, sports analysis, rehabilitation, animation, and machine-learning workflows. JSON exports include per-frame keypoints, confidence values, and stable multi-person track IDs—ready for downstream analysis or model training.