-
just read about a project to use an AI to generate prompts for image-generation AIs and i can't tell if it is causing me to sink into further depths of despair or cause me to find joy in the irrepressibility of human nature
-
the way reality is always more interesting than the salespitch... "you just have to type a quick description of an image, and the computer will generate it for you in high quality"
-
anyway, the whole prompt engineering focus is kind of weird anyway, the words are used to specify a latent vector that is "actually" what the machine generates images based on. there are richer & more interesting ways to manipulate a latent vector than words!
-
like a power steering device that bolts onto a steering wheel to help you turn it, rather than something that is built into the steering column.
-
anyway, previously on: @v21/1538618706586124290
-
also the whole thing where OpenAI was bolting on the words "black" or "female" to prompts to try to disguise the lack of diversity in the output newscientist.com/article/2329690-ai-art-tool-dall-e-2-adds-black-or-female-to-some-image-prompts/
-
creepypasta thread of a particular cluster in Stable Diffusion's latent space... that was found by using tools to explore the latent space that are more interesting than just prompting @supercomposite/1567180909044023297
-
strong point here: @proc_gen/1567228819530194944 it's notable what interesting stuff is being done in terms of new non-prompt based ways to explore the latent space now that there's a model people can run themselves & write their own tooling for...