A Framework for Procedural Generation and Evaluation of 3D Assets Driven by Large Language Models

Key Points

• The aim is to develop a method for automated 3D content generation driven by natural language instructions, enhancing quality and control.
• Translating natural language descriptions into script commands for the InfiniGen library to create 3D models.
• Employing the CLIP model for filtering and optimizing the generated 3D assets.
• Automatically selecting 3D models based on evaluation scores.
• The proposed approach significantly improves the quality of generated 3D models compared to baseline methods without CLIP filtering.
• Higher CLIP matching scores achieved across multiple categories of 3D object generation tasks.
• Demonstrates the feasibility of combining large models with procedural 3D generation techniques.