Chemeleon Zoo is a directory for the family of Chemeleon generative models, developed for crystal structure and materials design tasks. It provides links and documentation to individual repositories, covering text-guided generation, de novo structure generation, and crystal structure prediction.

The lead developer is Dr Hyunsoo Park as part of the Materials Design Group at Imperial College London.

Models in the Zoo

🦎🦎 Chemeleon2: Reinforcement Learning for Guided Generation

Repo → chemeleon2

A reinforcement learning (RL) framework for latent diffusion models with:

• Three-stage pipeline: VAE encoding, LDM sampling, and RL fine-tuning.
• Allows the generative model to be guided by specific reward functions.
• Trained on mp-20 and alex_mp_20 datasets.

🦎 Chemeleon-DNG: De Novo Generation & Crystal Structure Prediction

Repo → chemeleon-dng

Implements models for:

• DNG (De Novo Generation): sample new crystals without conditioning.
• CSP (Crystal Structure Prediction): generate candidate structures consistent with known compositions.

Trained on mp-20 and alex_mp_20 datasets, using reduced diffusion steps for efficient inference.
Includes benchmarks with >10,000 generated structures. As simple as pip install chemeleon-dng.

🦎 Chemeleon: Text-Guided Diffusion Model

Repo → chemeleon

The original denoising diffusion model that generates crystal structures directly from text prompts.
It supports:

• Text-guided generation (e.g., “orthorhombic LiMnO₄”).
• Composition-based generation.
• Chemical system navigation.

Citation

If you use the models in this collection, please cite the following work:

@article{park2025chemeleon1,
  title={Exploration of crystal chemical space using text-guided generative artificial intelligence},
  author={Park, Hyunsoo and Onwuli, Anthony and Walsh, Aron},
  journal={Nature Communications},
  volume={16},
  pages={4379},
  year={2025},
  publisher={Nature Publishing Group}
}

@article{park2025chemeleon2,
  title={Guiding Generative Models to Uncover Diverse and Novel Crystals via Reinforcement Learning},
  author={Park, Hyunsoo and Walsh, Aron},
  journal={arXiv},
  year={2025},
  url={https://arxiv.org/abs/2511.07158}
}