Synergy Diffusion studio: A code for evolutionary latent diffusion models.

This code can connect to several latent diffusion models, for text to image. Using evolutionary computation, we can combine or modify previously generated images.

The associated paper is nominated for best paper award!

Paper: https://dl.acm.org/doi/abs/10.1145/3583131.3590471 Nomination: https://gecco-2023.sigevo.org/Best-Paper-Nominations

Requirements

Google colab.

Building

Just run and it should work, ping us if it does not :-)

Installing / running (possibly without installation)

You can open our notebook and run it in google colab notebook

You will be able to download dissusion-based model (some presets are also provided), generate images, select parts which you want to change, or combine several images. For example if you want to change mouse paw, click on it to select area.

and then all mouse will be kept, but got new pow

We can also combine different pictures, generate animation and apply different neural network to imporve the result. All instruction is in the notebook.

Authors

Github loginss of authors: irumata (main author), teytaud, mzameshina, mathuvu, artemru

License

MIT License