ArtiMo: agent-driven framework for animating articulated 3D meshes from text

On August 21, 2026, the paper ArtiMo: Agent-Driven Articulated Mesh Animation was published on arXiv. The authors propose a framework that animates articulated 3D meshes from a text description without fine-tuning models. It is based on an agent pipeline using an LLM and a VLM: the language model plans the motion of object parts considering URDF kinematic constraints, while the visual model iteratively checks the result against rendered keyframes and corrects errors. The authors have not released the code.

What it means

For our section on video generators and 3D animation, this is more of a related work than a direct competitor. ArtiMo does not generate video or create meshes — it animates ready-made articulated 3D models based on their kinematic scheme. There are no direct analogs among the tools in the reference. The closest video generators work differently. LTX-Video is an image-to-video model from Lightricks with an open source license (Apache-2.0), the largest weights file is 26.6 GB, and as described by the author it requires only 1 GB of VRAM and supports MPS on macOS. CogVideoX from Zhipu AI tackles the text-to-video task, weighs 9.2 GB in the largest file, and runs on CUDA. Pyramid Flow is also text-to-video, with an MIT license for the code, 7.8 GB in the largest file, and requires from 8 GB of VRAM. ArtiMo differs fundamentally from all three: it is not a generative model but an orchestrator that uses ready-made LLMs and VLMs to plan and verify motion. For a user looking for a “text-to-video” tool, ArtiMo is useless — it does not render the final video sequence but only controls the animation of an articulated mesh. For robotics tasks or character animation with a rigid skeleton, this approach may be interesting, but without published code it cannot yet be tested in practice.