Blog

Introducing Video-to-Motion 3.0

Video-to-Motion 3.0 improves hand and finger detail and reduces foot sliding.

A signer making the ASL sign for blinds, beside the same motion from Video-to-Motion 2.1 and 3.0 on two 3D characters and a 3.0 skeleton.
The ASL sign for “blinds” at half speed: Video-to-Motion 2.1 (red), Video-to-Motion 3.0 (blue) and the 3.0 skeleton. Same input video, shown as generated, with the body held in place so the hands are easy to follow.

Today we're releasing Video-to-Motion 3.0. It improves two details that decide whether captured motion looks right: what the hands are doing, and whether the feet stay planted.

The workflow hasn't changed. Upload a video of one person, get back editable skeletal motion, apply it to your character and export FBX, GLB or BVH. What changes is how much of the original performance survives the trip.

Hands and fingers that keep their shape

3.0 picks up the wrist and the pointing finger, and more of the finger movement through each gesture, where 2.1 holds a mostly flat hand. Same input video, shown as generated. Left to right: source video, Video-to-Motion 2.1, Video-to-Motion 3.0.

Video-to-Motion has captured finger motion since 2.0. Version 3.0 adds a dedicated hand refinement step, so finger poses come through more clearly: an open hand stays open, a pointed finger stays pointed, and hand shapes that change mid-gesture are easier to read on the character.

That matters most when the hands carry the meaning: gestures, signing, counting, grips and anything a character does with its hands close to the camera.

Feet that stay planted

3.0 keeps better contact with the floor: less foot sliding, and the standing foot floats above the floor far less often. Same input video, shown as generated. Left to right: source video, Video-to-Motion 2.1, Video-to-Motion 3.0.

Sliding feet are one of the quickest ways to make captured motion look wrong. Video-to-Motion 3.0 adds foot IK, which adjusts the legs so a foot that should be planted stays put. You'll see less sliding during weight shifts, pauses and steps.

Choose 3.0 when you need it

Video-to-Motion 3.0 sits alongside the existing versions. Pick it in the model selector in the app, or pass the model value video-to-motion-3.0 in the API. You can also run two versions on the same video and compare them side by side before you decide.

Versions 2.0 and 2.1 stay available, and the API default doesn't change.

Pricing

Video-to-Motion 3.0 costs $0.06 per second of output motion on pay-as-you-go credits. Versions 2.0 and 2.1 stay at $0.05 per second. There's no subscription or minimum commitment. See pricing.

Get started

Video-to-Motion 3.0 is available now in the Uthana app and the API. The API documentation has examples, and the Video-to-Motion page shows more examples.