stable_diffusion_study3

This image is brought to you by a new Python script that allows you analyze an image with CLIP and then have it describe the images using weighted percentages per word.

I created an image, modified it with JAX, and then scanned it with the CLIP interrogator and ran that prompt as unmodified text in Stable Diffusion. This is something of a feedback loop in the cybernetic sense, an AI critiquing another instance of it's own program's output.