Exploration of crystal chemical space using text-guided generative artificial intelligence

A research team of Hyunsoo Park, Anthony Onwuli and Aron Walsh from the Department of Materials, Imperial College London have developed Chemeleon, an AI model that generates crystal structures from text prompts. These models are trained with information on the chemistry and three-dimensional structure of materials. Generation, or sampling of novel configurations, is performed using a denoising diffusion model based on a graph neural network. This can be performed in several minutes on a laptop, rather than weeks on a supercomputer. The ultimate goal is to accelerate materials discovery by directly sampling compounds with properties suited to next-generation technologies, such as high-voltage batteries and efficient catalysts.

The open source code is available on Github: https://github.com/hspark1212/chemeleon

Authors: Hyunsoo Park, Anthony Onwuli and Aron Walsh

Nature Communications

DOI: https://doi.org/10.1038/s41467-025-59636-y