Kimodo Models

Several variations of Kimodo-v1 are available trained on various skeletons and datasets. All models support text-to-motion and kinematic controls.

ModelSkeletonTraining DataRelease DateLicenseDownload
Kimodo-SOMA-RP-v1SOMARigplay 1 (700hrs)March 16, 2026NVIDIA Open Model HF
Kimodo-G1-RP-v1G1Rigplay 1 (700hrs)March 16, 2026NVIDIA Open Model HF
Kimodo-SOMA-SEED-v1SOMABONES-SEED (288hrs)March 16, 2026NVIDIA Open Model HF
Kimodo-G1-SEED-v1G1BONES-SEED (288hrs)March 16, 2026NVIDIA Open Model HF
Kimodo-SMPLX-RP-v1SMPL-XRigplay 1 (700hrs)March 16, 2026NVIDIA R&D Model HF

Which Model Should I Use?

For best quality motion generation, use models trained on the full Rigplay 1 dataset (700 hours).

For benchmarking and comparison, use BONES-SEED models (288 hours of publicly available data).

For humanoid robotics, use G1 skeleton models with MuJoCo-compatible output.

For compatibility with existing motion pipelines, use the SMPL-X model with AMASS format output.