ARTifical Intelligence

Eric Nord (@dirtcastle) on Threads

When it comes to highly stylized, unique, and experimental images, there are limits to what prompt text can do. Image inputs offer vastly more information, can express subtle/ineffable aesthetics, and they avoid many of the ambiguities, simplifications, and biases of a word-based approach.
Ideally, the last step of a series of prompts is often better off being image-only. Output images visually encode the previous input text(s). Removing text allows for broader interpretation by the model.

This Midjourney prompt represents a breakthrough for me, a culmination of techniques: experimental "hacking", mixing media (paint, photo, vector), fluke c/s/w variables (+ style raw), and most relevant here... geometric griddings in 3D space. This prompt is like a *meta* rabbit hole: I've already got crazy variations way beyond this. Stay tuned!... PROMPT: <imgA><imgB><imgC> --c 27 --ar 4:5 --style raw --s 80 --w 33 --niji 6 Input images are in a comment, 2nd image is the first gen