Prompt
walk forward, confidently
Locomotion
The product · ArenaLabs Studio
A working motion-generation surface built on AlphaMotion. Five prompts, five behaviors, one model. No per-task fine-tune.
The system runs
Each clip below is a single inference call against the same model, with the prompt shown beneath the video. The skeleton overlay is the live pair-coordination state, not a post-hoc visualization.
How it works
01
Prompt
Natural language
Describe the motion. Studio sends the prompt to AlphaMotion as a single text input — no skeleton authoring, no keyframes.
02
Preview
Real-time skeleton
AlphaMotion returns a full-body trajectory in seconds. Inspect joint-level motion, pair-coordination heatmap, and per-joint confidence directly in Studio.
03
Export
Drop into your stack
Export to BVH, SMPL-H, NumPy npz, or the ArenaLabs SOMA-77 skeleton format. Drop straight into your renderer, simulator, or controller.
Supported targets
AlphaMotion produces motion on the SOMA-77 reference skeleton with bidirectional converters to SMPL-H 52, SMPL 22, Unitree G1 23-DOF, and Boston Dynamics Atlas (28-DOF). Custom retargeting for hardware partners on request.
Early access
Studio is in closed beta through Q3 2027. Tell us who you are and what you'd build — we onboard integrators in waves and read every submission.
We read every submission. Founder typically replies within 72 hours.