Wonder Dynamics can now track actors even when hidden

If you have ever tried turning regular video into 3D animation, you know the absolute nightmare of occlusions: an actor walks behind a lamppost, crosses paths with another person, or dips out of frame, and the 3D skeleton instantly twitches, breaks, or collapses.

Wonder Dynamics just dropped a new model called Motion Prediction inside Wonder Animation to fix exactly that.

Integrated directly into the platform's Video to 3D Scene tool, the AI model predicts an actor's movements even when they are partially or completely blocked from view, turning messy footage into usable motion capture.

Filling in the blind spots

Traditional markerless mocap relies entirely on clear lines of sight. The moment an object gets between the camera and the actor, the tracking usually falls apart.

Motion Prediction introduces pose prediction for partial obstructions, estimating where hidden limbs should be so joints do not suddenly flip or freeze.

It also handles short interval occlusion. If a character walks completely behind a pillar or another actor for a brief second, the model calculates the expected trajectory and bridges the gap smoothly rather than popping back into place on the other side.

On top of that, Wonder Dynamics tackled out-of-frame limbs in close-up shots. When the camera only frames an actor from the waist up, the model now estimates lower body positioning, resulting in significantly better foot-to-ground contact and way less sliding around the virtual floor. A built-in noise filter also trims down the jittery twitching typical of vision-based mocap.

Why it matters

For solo creators, indie visual effects artists, and small game studios, this update cuts out some of the most soul-crushing cleanup work in 3D production.

Normally, when mocap breaks due to an obstacle or a camera cutoff, an artist has to dive into software like Maya or Blender to manually keyframe the missing poses. Eliminating even half of that rotoscoping and manual retargeting changes video mocap from a messy experiment into an actual production shortcut.

Across the industry, this looks like markerless video capture chipping away at the necessity of dedicated motion capture suits. The biggest weakness of single-camera optical capture has always been occlusion; software that can reliably guess what it cannot see levels the playing field against hardware-heavy setups.

That said, we think there are real limits to how far pure prediction can go. Estimating a continuous walking pace behind a pillar is straightforward math, but if an actor suddenly pivots, ducks, or makes an unpredictable movement while hidden, the model will inevitably guess wrong. It dramatically raises the baseline quality of casual footage, but high-end hero shots will still demand human oversight.