We give each essay to GPT-4, and ask it to generate 3 visual metaphors for the essay (with some scaffolding so that it doesn’t just ask for things like “someone reading a book on rationality”). Those are then put inside a prompt for Midjourney that was built using a bunch of /describe and trial-and-error (it also uses an image prompt to help lock-in the style). Then mods go through and pick their favourite of the ~12 images that end up generated.
We give each essay to GPT-4, and ask it to generate 3 visual metaphors for the essay (with some scaffolding so that it doesn’t just ask for things like “someone reading a book on rationality”). Those are then put inside a prompt for Midjourney that was built using a bunch of
/describeand trial-and-error (it also uses an image prompt to help lock-in the style). Then mods go through and pick their favourite of the ~12 images that end up generated.