Daily Generations: Can’t Bear it

Created in: Dall-E

Just like generative text can “sound like AI” I often feel like generative imaging can “look like AI” no matter what style was requested.

If we ignore for a moment things like too many limbs, not enough fingers. AI art is often too perfect. Like the artist had zero limitations on how they could place ink on paper. Like the the paint was applied one particle at a time rather than by a brush held by a shaky imperfect human hand.

This was my experiment to get away from the “AI look”. I think it was somewhat successful.

Overall, at first glance I don’t think I would guess that it was generated. Though there are still some odd bits; the shadows on the side of the headstone are a little digital looking as well as some weirdness under the plant. The biggest tell is obviously the reflection of his boots in the puddle.

An AI can’t reason that the puddle should be ink outlined like all the rest, or make the artistic choice maybe the red reflection detracts from the contrast of the bear being the only thing in the artwork that has color. The ability to apply artistic choice in this way is why we will likely always need a human in the loop.

In this final iteration it forgot to color his hat red so I just did that myself in Photoshop. This will probably get better, but for now small edits like this can be identified and made faster and easier by a human.