Blender Python — generates the mesh
Loading…BlenderRAG turns natural language descriptions into high-fidelity 3D objects in Blender by retrieving semantically similar (description, code) pairs from a curated dataset and conditioning an LLM on them to synthesize executable Python.
An overview of the work, the authors, and how to access it.
Generating 3D content from text remains challenging: end-to-end mesh generators often produce low-fidelity results, while code-driven approaches struggle with the breadth of Blender's API. BlenderRAG bridges this gap by retrieving semantically similar (description, code) pairs from a curated dataset of 500 objects and conditioning a large language model on them to produce executable Blender Python code. The result: cleaner geometry, more controllable outputs, and a workflow that lives directly inside Blender as a native add-on.
A short walkthrough of BlenderRAG generating meshes from natural language inside Blender.
docs/assets/demo.mp4 (or set data-video on the card to a YouTube URL) and it will appear here automatically.500 objects · 50 categories · 25 indoor + 25 outdoor · 10 variants each. Each variant ships with a rendered preview, a natural-language description, and the Blender Python script that generates the mesh. Below is a fresh random sample — click any mesh for the full image, description, and code.
If BlenderRAG is useful in your research, please cite:
@misc{rondelli2026blenderraghighfidelity3dobject,
title={BlenderRAG: High-Fidelity 3D Object Generation via Retrieval-Augmented Code Synthesis},
author={Massimo Rondelli and Francesco Pivi and Maurizio Gabbrielli},
year={2026},
eprint={2605.00632},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2605.00632},
}
Loading…