Pixray is an image generation system. It combines previous ideas including:

• Perception Engines which uses image augmentation and iteratively optimises images against an ensemble of classifiers
• CLIP guided GAN imagery from Ryan Murdoch and Katherine Crowson as well as modifictions such as CLIPDraw from Kevin Frans
• Useful ways of navigating latent space from Sampling Generative Networks
• (more to come)

pixray it itself a python library and command line utility, but is also friendly to running on line in Google Colab notebooks.

There is currently some documentation on options. Also checkout THE DEMO NOTEBOOKS or join in the discussion on discord.

Be sure to git clone --recursive to also get submodules.

You can install pip install -r requirements.txt and then pip install basicsr manually in a fresh python 3.8 environment (eg: using conda). After that you can use the included pixray.py command line utility:

python pixray.py --drawer=pixel --prompt=sunrise --outdir sunrise01

pixray can also be run from within your own python code, like this

import pixray
pixray.run("an extremely hairy panda bear", "vdiff", custom_loss="aesthetic", outdir="outputs/hairout")

Examples of pixray colab notebooks can be found in this separate repo.

running in a Docker using Cog is also possible. First, install Docker and Cog, then you can use cog run to run Pixray inside Docker. For example:

cog run python pixray.py --drawer=pixel --prompt=sunrise --outdir sunrise01