Text to motion · Video to mocap
Character motion from text or a video.
Describe an action or upload a clip, and Loopcut returns humanoid body animation as a BVH file for Blender. Motion from text costs $0.02 a second, made with NVIDIA Kimodo. Motion from video costs $0.03 a second of video, captured with NVIDIA GEM-X. Both are priced before they run.
Two ways in
From words, or from a performance.
Paid from Loopcut credits (1 credit = $0.01), and a failed job is refunded.
Motion from text
$0.02 per secondDescribe one simple physical action and get humanoid body animation. Good at grounded motion: walks, dances, kicks, gestures and fighting in place. Flips, spins in the air and big jumps come out small or missing, and the fingers stay still.
2 credits a second of motion, 2–10 seconds a clip, one free redo. NVIDIA Kimodo. Built with Meta Llama 3.
Motion from video
$0.03 per secondUpload an MP4, MOV or WebM clip and Loopcut captures the person's full-body motion, hands included. Works best with one person, the whole body in view, a mostly still camera and good light.
3 credits a second of video, rounded up, 2-second minimum, up to 60 seconds a clip, captured with NVIDIA GEM-X. Putting the motion on your rig happens in Blender, free.
Where each one works
Text works everywhere; video needs Loopcut on your computer.
| Feature | The Loopcut app | Local connector (Claude Code, Codex, Cursor) | Claude and ChatGPT apps |
|---|---|---|---|
| Motion from text | Yes | Yes | Yes |
| Motion from video | Yes | Yes | No. The video is on your computer |
| Put the motion on your own character | In Blender, free | On a saved .blend, with your Blender | No. You get SOMA's own skeleton |
| What comes back | The motion on your character, or on its own skeleton | motion.bvh and motion.json files | Links to motion.bvh and motion.json, each good for 1 hour |
Into Blender
From a BVH to your character.
The BVH from text uses a SOMA 77-joint skeleton in centimetres, Y up, at 30 fps. In the app, the motion goes onto the humanoid armature you select, or comes in on its own skeleton with a simple body.
- 01
Make the motion
Describe the action, or capture a clip on your computer with the app or the local connector. From a shell:
shelluvx loopcut run motion_from_video --file video=clip.mp4 - 02
Import the BVH
File › Import › Motion Capture (.bvh) at Scale 0.01, or from Python. It brings in SOMA's own skeleton with the motion, in a T-pose rest.
pythonbpy.ops.import_anim.bvh(filepath=path, global_scale=0.01, axis_forward='-Z', axis_up='Y', update_scene_fps=True, update_scene_duration=True) - 03
Put it on your character
For a Mixamo character, the Unreal mannequin or another humanoid rig, the Loopcut app applies the motion in Blender, free. The local connector does the same on a saved file:
shelluvx loopcut motion-apply --blend scene.blend --bvh motion.bvh
Questions
- What is text to motion?
- Character animation from a description. Loopcut's motion from text uses NVIDIA Kimodo, built with Meta Llama 3, and returns 2–10 seconds of humanoid body motion as a BVH file, for $0.02 a second.
- Can I turn a video into motion capture?
- Yes. Motion from video captures a person's full-body motion from an MP4, MOV or WebM clip with NVIDIA GEM-X, for $0.03 a second of video, rounded up, with a 2-second minimum and up to 60 seconds a clip. It runs in the Loopcut app or the local connector, because the video is on your computer.
- What file do I get?
- motion.bvh and motion.json. The BVH from text uses a SOMA 77-joint skeleton in centimetres, Y up, at 30 fps. In Blender, import it at scale 0.01.
- Does it animate fingers and faces?
- Motion from text animates the body only: the fingers stay still, and it isn't for faces, props or finger work. Motion from video captures the hands too.
- What if the motion isn't right?
- Motion from text includes one free redo: the same request again within 24 hours costs nothing. A new prompt is a new charge. Say what the body does in one short sentence per action, since each sentence becomes its own segment.
- Is my video used for training?
- No. Both tools run on our own workers on RunPod, with the models inside them, so no model developer receives your video or prompt, and nothing is used for training.