High-fidelity 3D object generation via retrieval-augmented Blender code synthesis
BlenderRAG is a Blender add-on that turns a natural-language description (e.g. “a modern wooden chair with armrests”) into an executable Blender Python script and runs it directly in the viewport. State-of-the-art LLMs frequently produce syntactically broken or geometrically inconsistent Blender code; BlenderRAG grounds generation in a curated multimodal dataset of 500 expert-validated examples (text, code, image) across 50 object categories.
Across four state-of-the-art LLMs, retrieval raises the compilation success rate from 40.8% to 70.0% and the CLIP semantic alignment from 0.41 to 0.77, without fine-tuning or specialized hardware.
The BlenderRAG pipeline: dataset construction with human correction, retrieval by embedding similarity, and LLM code synthesis.
How it works
Embed the query with a Nomic-AI sentence-embedding model.
Retrieve the top-_k_ most similar (description, code) pairs from a local Qdrant vector database.
Synthesize a Blender Python script with a user-selected LLM (open or closed source), prompted with the retrieved examples.
Execute the script in the active Blender session, leaving the mesh selected for further editing.
Automatic generation of executable Blender code from natural language remains challenging, with state-of-the-art LLMs producing frequent syntactic errors and geometrically inconsistent objects. We present BlenderRAG, a retrieval-augmented generation system that operates on a curated multimodal dataset of 500 expert-validated examples (text, code, image) across 50 object categories. By retrieving semantically similar examples during generation, BlenderRAG improves compilation success rates from 40.8% to 70.0% and semantic normalized alignment from 0.41 to 0.77 (CLIP similarity) across four state-of-the-art LLMs, without requiring fine-tuning or specialized hardware, making it immediately accessible for deployment. The dataset and code will be available at https://github.com/MaxRondelli/BlenderRAG.
@article{rondelli2026blenderrag,title={{BlenderRAG}: High-Fidelity {3D} Object Generation via Retrieval-Augmented Code Synthesis},author={Rondelli, Massimo and Pivi, Francesco and Gabbrielli, Maurizio},journal={arXiv preprint arXiv:2605.00632},year={2026},month=may,}
