Uthana centres on three capabilities that reinforce one another. Under Models, it converts 2D video into editable 3D motion and generates motion from text descriptions that can specify action, sequence, pacing, posture, and style. Under Embodiment, it retargets that motion across different bipedal character skeletons and adapts human movement to humanoid robot joint representations, plus auto-rigging to add a skeleton to an unrigged character. Under Data, it offers studio-captured motion — over 1,000 hours across collections from a 120-camera AAA production setup — alongside services to turn real-world video collections into structured 3D pose and motion datasets with labels and segmentation for AI training.
The target audience spans game studios that need production-ready character animation, simulation and robotics teams training physical AI, and AI labs building world models. Motion can be previewed in the browser, applied to a compatible character, and exported in editable FBX, GLB, or BVH — or in tabular formats for training pipelines. The platform quotes metrics including sub-50ms time to first motion token, 45,000+ characters animated, and 1.7 million+ seconds of motion generated.
For developers, Uthana exposes a single GraphQL API covering generation, motion extraction from video, retargeting, and auto-rigging, plus plugins for Blender and Maya and an Uthana MCP server for Claude, Codex, and Cursor. It notes honestly that robot retargeting outputs motion in the target representation but does not by itself provide a control policy, dynamics model, or guarantee of hardware executability.
