But you can already see the idea of a “prompt” evolving into...

But you can already see the idea of a “prompt” evolving into something more complex in the open source AI world. The big new thing over the last few months on the Stable Diffusion subreddit are “LoRas,” or “Low-Rank Adaptations”. They’re essentially aesthetic packages that you can load into an image generator. Here’s a good example of one for a children’s book. It was trained on 100 or so images (I don’t know from where, but that’s a whole other can of worms) and if you download it and load it into your instance of Stable Diffusion it should spit out images that look consistent in that particular style. It basically moves the prompt work behind the scenes.

Comments
www.joshbeckman.org/notes/526627235