People Twitter

I’ve commented about this before, but I do actually miss those first few generations of image generators. The first DeepDream style stuff was interesting but didn’t go very far, but this image in particular is a milestone for what came directly after.

I would love to run VQGAN+CLIP locally, for example, in some efficient way. It was fun to play with and to see how the model interpreted the input. And it wasn’t as scary as the tools we have now (especially when those are paired with the deep fake face swap stuff, for example)

I genuinely think slop is the perfect word for the current iteration of these image generators, both the image outputs themselves and the role they’re playing in the already-bleak digital media landscape.