Mapping the limitations of Artificial Intelligence is a fascinating hobby. There’s a function in midjourney to “describe” a given image. You can feed it anything and it’ll attempt to produce a prompt that might make something similar.
The above image produced the following prompts, it does them four at a time:
The above image produced the following prompts, it does them four at a time:
A woman with glasses in black and white, in the style of fujifilmneopan, impressionistic color harmonies, canon ae-1, steinheil quinon 55mm f/1.9, northwest school, bibliopunk, cinestill 50d
black and white photo of a woman, in the style of fujifilm fp-100c,
michael sowa, muted tonalities, mikhail larionov, close up, jeffrey t. larson, surprisingly absurd
an image of a woman wearing glasses, in the style of fujifilm neopan, mikhail larionov, myroslava sviridova, jeffrey t. larson, kodak vision3 250d 5207, grainy, blurred
a woman in glasses looks away from a window, in the style of fujifilm neopan, canon ae-1
black and white photo of a woman, in the style of fujifilm fp-100c,
michael sowa, muted tonalities, mikhail larionov, close up, jeffrey t. larson, surprisingly absurd
an image of a woman wearing glasses, in the style of fujifilm neopan, mikhail larionov, myroslava sviridova, jeffrey t. larson, kodak vision3 250d 5207, grainy, blurred
a woman in glasses looks away from a window, in the style of fujifilm neopan, canon ae-1
The prompts themselves aren’t so bad I guess. Suggesting the camera, lens and f-stop is curious and even clever. I don’t know what bibliopunk means and I’ve no intention of looking it up. I’ve not heard of any of those photographers either, and I think it’s a little bit rude that they’ve not capitalized their names. It’s also rude that they assume I’m shooting with something as pedestrian as a Canon ae-1, which also not capitalized.
You’d think it’d go for something incredibly esoteric and expensive, the camera doesn’t actually exist, why not shoot for the moon?
As interesting and even informative as those reverse-prompts are, the images you get when you run them are not. They’re miserable.
You’d think it’d go for something incredibly esoteric and expensive, the camera doesn’t actually exist, why not shoot for the moon?
As interesting and even informative as those reverse-prompts are, the images you get when you run them are not. They’re miserable.
I’m not sure I’ve seen less interesting photographs. Except maybe for the left most one because of that fold or drip mark or some sort of artifact across her face. That’s pretty strange.
Another strange thing: AI can’t make ugly people. You can tell it to, but it’ll just ignore you and turn whatever prompt you give it into model fodder, it’s excruciating and boring. Maybe in the future facial scars will make a come back.
...we can only hope.
Another strange thing: AI can’t make ugly people. You can tell it to, but it’ll just ignore you and turn whatever prompt you give it into model fodder, it’s excruciating and boring. Maybe in the future facial scars will make a come back.
...we can only hope.