Motion reconstruction guide
Reconstruct a body performance from video
JST-2 video-to-motion uses a recorded performance as the reference for full-body 3D movement. The task is to recover motion for animation, not to replace the person in a finished video.
Inspect real JST-2 examplesChoose footage that makes the movement readable
Use footage you have permission to process. As an evaluation starting point, choose one clearly visible performer with the whole body in frame and minimal camera movement. This is a practical recommendation, not a statement of the model’s maximum supported complexity.
Avoid introducing several sources of ambiguity in your first test. Fast cuts, motion blur and long periods behind obstacles make it harder to distinguish an input limitation from a reconstruction problem. Check the official interface for its current file, duration and upload requirements.
Visible motion is evidence; hidden motion is inferred
Viggle says JST-2 reconstructs visible movement and uses learned dynamics to infer how unseen body parts move. This allows a complete body performance even when footage does not show every limb throughout the action.
An inferred pose is not a measurement of the hidden limb. Compare reconstructed timing and body orientation with the source, particularly during turns or occlusion. Review hands and feet where they become visible again to see whether the transitions remain plausible.
Compare the source and reconstructed performance
Judge rhythm and changes of speed before fine detail. Mark the moments when feet land, the torso turns or an arm changes direction. Compare those moments against the animation rather than relying on a single attractive still frame.
Then apply the result to your target character and inspect joint behavior. Viggle documents direct use on JST-2 characters and compatibility with standard rigs. Check the actual export and destination software; compatibility does not eliminate the need to review scale and orientation.
Keep body motion separate from a scene trajectory
A person can walk convincingly while their placement in a reconstructed environment remains unresolved. The JST-2 release distinguishes full-body motion from reconstructing the person’s path through a scene, describing the latter as an upcoming capability.
Do not plan on recovering a complete camera-calibrated world trajectory from this announcement alone. If your project needs exact distances, prop contact or alignment with a filmed set, evaluate those requirements separately. Viggle also identifies further motion-accuracy improvements as future work.
Questions & answers
Does video-to-motion copy the person’s appearance?
The task described here reconstructs body movement. Character appearance is a separate asset or generation workflow. Applying the motion to another character should not be confused with producing a finished character-replacement video.
What happens when a hand or leg is hidden?
Viggle describes using learned dynamics to infer unseen body movement. Review those intervals carefully because the output is an estimate. Clearer source footage gives you more visible evidence against which to judge the reconstruction.
Can I recover the performer’s exact route through a room?
The release labels scene-path reconstruction from video as upcoming. Current full-body motion claims should not be read as a guarantee of exact world-space trajectory, measured distance or alignment with a reconstructed room.
Should I use video or a text prompt?
Use video when an existing performance’s rhythm and poses matter. Use text when you want to specify an action without reference footage. Either result still needs review on the character and in the application where you plan to use it.
Check the original sources.
Reviewed . Model claims are attributed to Viggle; practical suggestions are editorial guidance.