Nowadays, LLMs are able to generate images, videos, music, and code. However, such abilities have not been well integrated. Given that video game assets involve all these kinds of resources, using LLMs as video game asset generators may be meaningful.
One potential use case is the generation of new scenes. When the user reaches the edge of the current scene, the LLM may take the assets in the current scene as the inputs and generate new assets for the next scene. In this way, the game world can be extended infinitely.
Another potential use case is the generation of customized items. When the user inputs a description of a skill/item/etc., the LLM may generate code, special effects, and other assets based on the description. In this way, the playability of the game will be much higher.
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{chen-llms-as-video-2026,
author = {Chen, Mike},
title = {LLMs as Video Game Asset Generators},
year = {2026},
url = {https://hypogenic.ai/ideahub/idea/HUBpe2Qqbe8rU3IK4URz}
}Please sign in to comment on this idea.
No comments yet. Be the first to share your thoughts!