Recently I've been experimenting with DALL-E 2, one of the models that uses CLIP to generate images from my text descriptions. It was trained on internet text and images, so there's a lot it can do, and a lot of ways it can remix the stuff

One thing I've noticed with image-generating algorithms is that the more of something they have to put in an image, the worse it is.

I first noticed this with the kitten-generating variant of StyleGAN, which often does okay on one cat:

but is