Kandinsky v3.0

This is an implementation of a Cog model. Cog packages machine learning models as standard containers.

First, download the pre-trained weights:

cog run script/download-weights

Then, you can run

text2img predictions:

cog predict -i prompt="a moss covered astronaut with a black background"

img2img predictions:

cog predict -i image=@astro.png -i prompt="photo of a lone astronaut standing on a barren planet"