Art-directing AI

I want to draw your eye to the images in this recent post from Amelia Wattenberger, which seem to me an example of someone trying hard to art-direct AI image gen­er­a­tion in a rec­og­niz­ably edi­to­rial way.

Clearly, Amelia was going for a par­tic­ular look. There is a clear idea at work here, exactly the kind you’d specify to an artist (or pitch to an art director). However, the fun­da­mental fuzzi­ness of the AI approach is apparent; while the images do all have the same “texture”, they don’t seem to have come from the same source, or indeed to have been made by the same “person”.

Anyone who has worked with AI tools will rec­og­nize the feeling of “close enough”-ness. If you squint, you can see all the images Amelia rejected — a pile of crumpled-up draw­ings just beyond the frame of the browser. The images pub­lished with the post were, for sure, the best options, even if together they don’t quite form a coherent package.

Anyway, it’s inter­esting and useful to encounter this strategy for illus­tra­tion “fully expressed”, rather than just imagined. I don’t think it succeeds, but/and I’m glad to have the example to consider.