June 26, 2024 update: A high-powered upgrade from my TryCycle series (descendants of this repo) is on its way if all goes well. It will use the multi-modal vision aspect of Claude-Sonnet 3.5 to replace much of the more cumbersome CLIP Interrogator/BLIP2 components and calls to ChatGPT. The superior performance and speed of Sonnet promises to be a big boon for this and other visual thinking projects.
Digital daydreaming with CLIP Interrogator and Diffusers
My eventual intention is to add a genetic algorithm into this mix but the simple feedback loop itself was so fascinating I couldn't pass up sharing. Please let me know if there are any problems getting it to work, it took me the better part of a week to find all the components that would work together (thanks Amin and Pharmapsycho)
(notice: this file may occasionaly be under edit. If it doesn't work please let me know and give it a 2nd chance later)
colab link on ClipLoop.ipynb page above.