LLM-Evolved Digital Art

Published:

A digital art project exploring what happens when you let a language model incrementally evolve a generative simulation. Starting from a small set of rules and an initial p5.js sketch, a screenshot is taken 30 seconds into the simulation, and shared with the LLM together with the current code, asking the LLM to add a single feature based on a thematic prompt. The new code, together with a new output screenshot, becomes the input for the next iteration, and the cycle repeats — so each generation builds on the previous version’s behaviour and look.

This is similar to autoresearch, but instead of a hard objective, we aim to incrementally increase beauty and complexity.

Note that all these simulations run locally in your browser.

Agent-interactions

Browse all iterations of this type. ·

//: # ( title=”LLM-evolved simulation, ‘Multi-agent’”)

Chaos vs Order

Browse all iterations of this type. ·

About