LLM-Based 3D Scene Generation Automating Asset Retrieval

The patent replaces the manual mechanical process of designers creating 3D scenes with an automated system using large language models and rendering engines. The LLM automatically generates scene descriptions, asset lists, and configuration parameters from user input, eliminating the need for manual designer intervention in routine scene creation tasks while maintaining quality through automated rendering.

The patent introduces large language models as an intermediary between user input and the rendering engine. The LLM processes user descriptions, generates detailed scene configurations, and translates them into parameters the rendering engine can execute. This intermediary layer enables automated generation while maintaining accuracy by ensuring proper interpretation and translation of creative intent into technical specifications.

The patent merges multiple discrete steps of the scene creation process (conceptualization, asset selection, parameter configuration, and rendering) into a single automated workflow driven by the large language model. Instead of requiring separate manual operations for each step, the system combines them into an integrated process that takes user input and automatically produces the final scene, significantly simplifying the overall process.

The patent employs large language models that have been pre-trained on extensive datasets of scene descriptions, asset libraries, and rendering parameters. This preliminary training and preparation of the LLM enables it to efficiently process user inputs and generate accurate scene configurations without requiring excessive computational resources during actual scene generation, as the heavy lifting of learning patterns and relationships has already been completed in advance.