[dragonwriter]

> You could probably do something similar with Stable Diffusion with just a little upfront effort tuning that system prompt.

And, indeed, someone has:

https://github.com/sayakpaul/caption-upsampling