GitHub radar
NVIDIA Kimodo in C++/GGML: Local Text-to-Motion
The LocalAI community has ported the NVIDIA Kimodo text-to-motion model to C++ / GGML so you can use text to animate 3d models on your computer!
The LocalAI community has ported the NVIDIA Kimodo text-to-motion model to C++ / GGML so you can use text to animate 3d models on your computer! The Kimodo model uses a text prompt (or LLM2Vec embedding) to generate 3d human motion data (in SMPL-X format: root translations + joint rotations). Under the hood, it’s powered by a Llama 3 text encoder and the Kimodo-SMPLX-RP-v1 diffusion model, both available as out-of-the-box GGUF weights from the LocalAI-io organization on HuggingFace (download with a single line of code). Works with CPU or Vulkan compatible GPU. Compile with CMake. Run a browser demo server locally at localhost:8094 and test out prompts. Licensed under Apache-2.0. Model weights for non-commercial research use only.
Why a vibe-coder should care
Kimodo makes it possible to develop applications using AI agents and characters that generate motion on your local computer instead of requiring API calls to the cloud or subscriptions. There was a clear need for a local text-to-motion solution: Kimodo went from 0 to 460 stars in 3 days! It previously required special setup to run.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up kimodo.cpp from https://github.com/localai-org/kimodo.cpp — follow the README, download the GGUF weights with the provided script, build and start the demo at http://localhost:8094; show me the first generated animation. Ask for my HuggingFace credentials if needed.
Requires a Mac or Linux machine with 16+ GB of RAM — builds from source (C++ and CMake); no GPU required.
Open on GitHub▌ More finds