Donmai

AI-generated art check thread

Posted under General

nonamethanks said:

FYI last week photoshop added generative fill to the beta version (example, news article), so in the coming months there's a good chance many if not most uploads will have some degree of AI in them by default.

Looks very efficient at generating landscape and background/foreground, and at learning the artstyle from the image itself. It's going to require much keener eyes that what it took for AI-generated characters to find traces of using these features.

8253803 said:

Who keeps flagging these long-established artists that publish their PSD/CLIP files for all their works? Seen it happen to misaka_12003-gou for post #6112583 and now waterring. It's hard to believe they're not trolls. I don't know if Danbooru has a system to filter out bad flaggers.

I doubt it's intentional, the program is trained on long-established artists. People see a style they've seen the machine spit out and assume something is AI without bothering to check the artist's body of work to see how long they've been drawing.

More carelessness than witch-hunting.

I'm very suspicious of this guy personal160400. I'm not saying that he's faking it for sure, but... He was working exclusively with 3D in Blender until may and then suddenly he's posting art at the art god level that looks very close to a popular ai art style. Especially the hair. Very sus.
But here's where it gets messy. He posted a lot, and I mean, A LOT of images of the process, but none of them completely prove this legitimacy. All those process images can be faked by just tracing over an ai generated image and then combining some parts of the final image with the trace and some flat colors. By the way, his flat colors never match the lines, but always go over the lines. As someone who actually paints this to me seems like the most pain in the ass way of painting, cause you can't actually rely on any of your edges and will have to fix all of them as you render. And here if you zoom in on the final image and look in the bottom left corner you'll see the lineart sitting on top of the rest of the image, including the leaves. Which is very perplexing, cause lineart doesn't exactly match the image behind it and also why is it going over the leaves? He never posted a version where he would fix this. In fact he posted "lineart over leaves" version twice. He also posted a process video, but it only shows a rough sketch that doesn't actually match the final image. Then he posted that he did a second pass of the lineart here and it still doesn't match the final image. So there either has to be a third pass or he is such an art god that he improved his lineart as he was rendering, but then he wouldn't be needing two passes of it, would he...
I'm not saying he's definitely faking it, and if he is, he's putting a lot of effort, but until he posts a timelapse of the entire rendering process I'll very suspicious, and I think you should be too.

Updated

Yoksven said:

art at the art god level

Rather a default AI level...

That dude trying so hard to make an illusion of working process... and it works! Someone's really buying it for that show.

Updated

kittey said:

Unrelated to the AI question, is that a paid reward or 0 yen-tier?

I think it's a 0-yen tier one (but it has exclusive versions.)

On a related note, I look through the exclusive content for the post that image came from and there's a lot of images available (maybe variations or whatever) and it seems like quite a high output, so I'm thinking AI is involved here (assisted or not), unless it's a usual thing for artists on there. I wouldn't really know, admittedly..

狐霊 said:

I think it's a 0-yen tier one (but it has exclusive versions.)

On a related note, I look through the exclusive content for the post that image came from and there's a lot of images available (maybe variations or whatever) and it seems like quite a high output, so I'm thinking AI is involved here (assisted or not), unless it's a usual thing for artists on there. I wouldn't really know, admittedly..

Variations. It would be harder to produce so many variations from an AI image than from a PSD because the latter only comes down to turning on and off layers like clothes, fluids, and facial expressions, and can be done in minutes. Clearly, the high volume does not indicate AI.

8253803 said:

Variations. It would be harder to produce so many variations from an AI image than from a PSD because the latter only comes down to turning on and off layers like clothes, fluids, and facial expressions, and can be done in minutes. Clearly, the high volume does not indicate AI.

Yeah, that does make sense.

1 31 32 33 34 35 36 37 38 39 118