3 ms·

Biggest signs for me were the edges you mentioned, compositions that werent technically bad but were visually uninteresting (and had minor logical flaws like mi

Biggest signs for me were the edges you mentioned, compositions that werent technically bad but were visually uninteresting (and had minor logical flaws like missing details in a reflection), but also the subject matter is the sort of thing that could be described in a succint text prompt designed to challenge a machine (I actually guessed a couple of the prompts, and "tessellated Hello Kitty" and "Origami car" are less likely to be human art concepts than challenges for machines). I'm browsing on a phone so some of the fine detail is lost. The fact that I'm using the subject matter as much as the image itself to discern AI art is a sign of how good the tech is though. Especially since the subject matter in question is typically "stuff that would be really challenging for an AI to conceptualise" and the image itself (not necessarily the AI's only attempt of course) is a sound attempt at rendering it. I see most of the utility as the opposite: the artist still needs to come up with the original concept, the overall compositions are unlikely to be better than an artist would conceive, but even in the hands of a non-artist a bit of text can produce "good enough" images (that can be cropped or touched up to remove rough edges) instantaneously which would take a lot of time for a skilled artist to produce.

linux2647 4y ago
> are less likely to be human art concepts than challenges for machines Counter-example: there was one image I got that was a Gameboy as a concrete brick. That seems like a prompt you’d give DALL-E 2, but it was actually of human creation
- cowtools 4y ago
  Yeah, but I guessed that one correctly because that sort of art (taking symbolism like a game-boy and putting it in a "cool" medium) is pretty common on the internet already. It makes sense that someone would think to make that because a gameboy is basically already a grey brick. So it's really just a change of texture- it's a comparison of two things that look similar that a person would probably make but an AI might not. it has some artistic intent. The other DALL-E prompts are just "a <random subject> doing <random thing> in <random medium>", like "a dog building a fence in watercolor". Would it make sense for a person to paint a dog building a fence in watercolor? Probably not unless they're writing a children's book.

linux2647 4y ago

> are less likely to be human art concepts than challenges for machines Counter-example: there was one image I got that was a Gameboy as a concrete brick. That seems like a prompt you’d give DALL-E 2, but it was actually of human creation

cowtools 4y ago
Yeah, but I guessed that one correctly because that sort of art (taking symbolism like a game-boy and putting it in a "cool" medium) is pretty common on the internet already. It makes sense that someone would think to make that because a gameboy is basically already a grey brick. So it's really just a change of texture- it's a comparison of two things that look similar that a person would probably make but an AI might not. it has some artistic intent. The other DALL-E prompts are just "a <random subject> doing <random thing> in <random medium>", like "a dog building a fence in watercolor". Would it make sense for a person to paint a dog building a fence in watercolor? Probably not unless they're writing a children's book.