Deep Dive: NVIDIA's Kimodo, a full open-source pipeline for text-to-3D-motion

maier_ak · x · 2026-08-31

This article provides an in-depth look at NVIDIA's Kimodo. Kimodo is the first open-source diffusion model capable of generating 3D human or humanoid-robot motion driven by natural language and kinematic constraints (e.g., keyframes, paths). The release includes a complete pipeline: CLI, web demo, 5 pre-trained checkpoints, a benchmark suite, and training scripts. It addresses the bottleneck of scarce high-quality motion data and runs on a single RTX 3090 or GPUs with <3GB VRAM, significantly lowering the barrier for robotics and interactive entertainment.

Related event: NVIDIA Open-Sources Kimodo for Text-to-3D Human Motion Generation(4 posts)→

Original post →

More from Embodied

Embodied channel →