BlenderRAG: High-Fidelity 3D Object Generation via Retrieval-Augmented Code Synthesis

arXiv cs.AI / 5/4/2026

📰 NewsDeveloper Stack & InfrastructureTools & Practical UsageModels & Research

共有:

Key Points

BlenderRAG addresses the difficulty of converting natural language into executable Blender code, which current leading LLMs often fail due to syntax errors and inconsistent 3D geometry.
The system uses retrieval-augmented code synthesis with a curated multimodal dataset of 500 expert-validated examples (text, code, images) across 50 object categories.
By retrieving semantically similar examples during generation, BlenderRAG raises Blender code compilation success from 40.8% to 70.0% and improves semantic alignment (CLIP similarity) from 0.41 to 0.77 across four SOTA LLMs.
The approach does not require fine-tuning or specialized hardware, aiming to make high-fidelity 3D object generation easier to deploy.
The authors plan to release the dataset and code publicly via the provided GitHub repository.

Abstract

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.