Tasks and inputs
Kimodo-SOMA-RP-v1.2 is listed for Data & control, Text-to-3D. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Kimodo-SOMA-RP-v1.2 is NVIDIA's open 282M motion diffusion model that turns text and pose constraints into 3D human animation, now on clean splits.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Open weights for self-hosting; no hosted API price. Hardware cost is yours.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Kimodo-SOMA-RP-v1.2 is a correction release, and it says so. In version 1.1, some test sequences from the BONES-SEED dataset had leaked into training, which flattered its scores on the Kimodo motion-generation benchmark. Version 1.2, released on 15 October 2026, retrains on corrected splits. Foot skating drops, constraint accuracy holds, and text-following scores (R-precision and FID) slip slightly, partly because the leakage is gone. If you benchmark against Kimodo, this is the checkpoint to use.
Kimodo stands for kinematic motion diffusion. It generates 3D skeletal animation from a text prompt, from constraints, or from both. Constraints can be full-body poses at given frames, end-effector joint positions, paths and waypoints. Up to ten seconds come out per call: 300 frames at 30 fps, as root translation and per-joint rotation matrices.
The family spans five variants. SOMA-RP, this one, uses the SOMA skeleton and the proprietary Bones Rigplay data. Others cover SOMA with the open BONES-SEED data, the Unitree G1 robot skeleton and SMPL-X.
| Spec | Value |
|---|---|
| Architecture | Two-stage transformer, diffusion |
| Parameters | 282M |
| Max duration | 10 s (300 frames at 30 fps) |
| Inputs | Text, duration, pose constraints |
| Outputs | Root translation, joint rotations |
| Licence | NVIDIA Open Model License, commercial use allowed |
The model description refers to a 30-joint SOMA skeleton. The project's changelog, however, says model inputs and outputs moved to a 77-joint SOMA layout in March 2026. Before you wire outputs into a retargeting pipeline, print the joint dimension the installed code actually returns.
The wider project was trained on roughly 700 hours of commercially friendly optical motion capture and ships a CLI, a timeline demo for mixing prompts with constraints, and the benchmark itself.
Write twenty prompts from your own shot list. For each, add one hard constraint, a hand position or a waypoint. Measure constraint error and count foot-sliding frames, then have an animator rate naturalness. Compare v1.2 with whatever you use today, not with v1.1's numbers.
Text-driven 3D generators are grouped in the text-to-3D directory, and NVIDIA's other motion and robotics models are on the provider page.
Compare the model on the work you need to complete.
Kimodo-SOMA-RP-v1.2 is listed for Data & control, Text-to-3D. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
The weights are downloadable under NVIDIA Open Model License. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
NVIDIA
MotionBricks is NVIDIA's open real-time motion generator for SOMA human characters and the Unitree G1 robot, filling motion between context and target poses.
alexgusevski
alexgusevski/LLaMA-Mesh-mlx is a source-linked Hugging Face repository assigned to Text-to-3D. Its recorded task interface accepts text and produces 3d.
alexgusevski
Hugging Face repository for Text-to-3D.
alexgusevski
Hugging Face repository for Text-to-3D.
alexgusevski
alexgusevski/LLaMA-Mesh-q8-mlx is a source-linked Hugging Face repository assigned to Text-to-3D. Its recorded task interface accepts text and produces 3d.
AvaLovelace
AvaLovelace/BrickGPT is a source-linked Hugging Face repository assigned to Text-to-3D. Its recorded task interface accepts text and produces 3d.