Disco AI

↝      Description

Exploring and experimenting with OpenAI unconditional diffusion models, mainly using Disco Diffusion. Thank you to the DElab of the Hasso Plattner Institute for letting me use the power of a few A100 GPUs.
It was incredibly interesting to learn more and more of how the AI interprets the input, both the prompt as well as all the other parameters, and getting better at formulating of what I want the AI to do. In the midst of exploring, I followed the development of StabilityAI's Stable Diffusion, which blew these models out of the water in terms of coherency and speed. Feel free to see my experiments with Stable Diffusion here →

You may notice the number of iterations for each image; to get the results I had in mind, I had to fiddle around a lot with the parameters of the generation as well as the prompt formulation. Nth iteration means that this was the nth image generated from the same conceptual idea. If you are interested in the nitty gritty details, an example prompt can be found at the bottom of this page.

↝      Tools used

Disco Diffusion, Python, Photoshop

flamingo • 39th iteration
tucan • 21st iteration
robo battle of colors • 211th iteration
temple of petra • 182nd iteration
nepal • 218th iteration
royal portrait • 14th iteration
bratty portrait • 9th iteration
samurai portrait • 3rd iteration
sunflower portrait • 8th iteration
overgrown battlefield • 137th iteration
vision of ecstasy • 6th iteration
alien plant still life • 46th iteration
pastel rock waterfall • 11th iteration
holographic ocean cliff • 34th iteration
Example prompt with parameters

←      Go Back