AI-generated art check thread

Posted under General

ANON_TOKYO said in forum #421566:

You need to mention what makes it suspicious in the flag reason, just saying it is tells approvers nothing. The wiki also mentions that AI models cannot generate resolutions not divisible by 8, not that divisibility by 8 is any sort of evidence for anything (and even if the resolution matches the examples given in the wiki there's still cases where the artist knowingly or unknowingly uses an AI image as a reference, which might cause the resolution to match).

Also just check width:1448 height:2048. 2048x is a dimension limit of Twitter mobile, and it's a power of two, both of which contribute to it being extremely common.

Are the first two reasons insufficient? Or was I unclear in what I meant, namely that their style fits typical AI-generated works and the rate in which they post their works on Twitter seems unnaturally high.

I see the resolution is not suspicious now but that was more of an "adding to the list of points of concern". I wasn't sure how specific I had to be in the flag reason vs how much would be implicit.

Updated by luntoer

luntoer said in forum #421578:

Are the first two reasons insufficient? Or was I unclear in what I meant, namely that their style fits typical AI-generated works and the rate in which they post their works on Twitter seems unnaturally high.

I see the resolution is not suspicious now but that was more of an "adding to the list of points of concern".

The first reasons, in the context of this specific artists, were sufficient yeah. If an additional point is not suspicious, don't add it, since it's irrelevant if it's not suspicious.

ANON_TOKYO said in forum #411856:

It appears you have once again elected to contribute a series of declarations so exquisitely unburdened by evidence that one can only admire the consistency of your approach. Your steadfast dedication to presenting conjecture as though it were scholarship is, in its own way, almost artful.

However, this space is generally reserved for discussions supported by something more substantial than personal intuition and the philosophical school of “I just feel like it.” I would therefore gently, graciously, and with the utmost goodwill encourage you to continue cultivating your unique style of discourse in communities more hospitable to such… imaginative methodologies.

Unless you intend to provide information that rises above the level of vibes, impressions, or the absence of contradictory proof, I must politely request that you refrain from further derailing the conversation. We do try, after all, to maintain at least a passing acquaintance with evidence here.

You have a documented history of baseless accusations without any substance. That is not what this place is for. Feel free to continue that behavior on your circlejerk subreddit, but don't waste our time if you're not going to give any evidence beyond "vibes" and a lack of counter-evidence.

ANON_TOKYO said in forum #420128:

Post frequency isn't proof of anything, you have to show actual issues with the art.

Tankist447AV said in forum #421629:

Guys, about the artist "signallerc". Some of his works is tagged as 'AI' on pixiv, such as:

post #10159535
post #9822978
post #9782655
post #9638551

- while a few not, such as:

post #9343040 (and it's just a sketch)

Please pay notice to those tagged ones and do not flag and delete all of them. o7
edit: I reached my flag limit on the first 5 post, someone plz deal with it.

@Tankist447AV Anyone can add tags to Pixiv posts. Those tagged as AI are done by other users. Please do not flag for this reason. I have rejected your flags because they don't look AI to me. Someone is free to re-flag them with proper evidence.

@Tankist447AV When an artist actually marks their own work as AI-generated, the AI-generated will be at the beginning of the tags, in bold, and not have a # in front of it. That's what you have to look out for.
(And generally, when you want to check whether a tag was added by the artist or a user, you have to click the + at the end of the tags and see whether the tag has an X next to it or a lock. X are user-added and can be deleted by other users, lock means it was added by the artist and can't be removed.)

post #10356404

The character is lying on the futon on her back, and the pillow under the back of her head looks like it has differing sizes on the different sides of the character's head. A tray with a sake bottle seen at the image top border I would suppose should be square, and its projected form may be an attempt at perspective, same perspective could shorten the farther end of the pillow. Still, the portion of the pillow on the left side shows a generous part of its closer edge and if that straight line edge were to be continued, it makes actual portrayed right side of the pillow shorter than the left side by a quarter of the size.

Do I look at it wrong? Inconsistent object dimensions are a popular trait of generated images. Is a real artist mistake like that fathomable?

edit: to think more of it, the tray too, the corner behind the bottle doesn't connect that well.

KotokoIwanaga said in forum #421886:

chokopaii

Wiki: AI-generated content. Do not upload.

Why do people keep uploading this person's work?

that wiki is really misleading, it doesn't tell the whole story,

if i remember correctly, the reason this artist was under fire in the past was for using AI-based upscaling tools on the art which gives it the appearance of being ai-generated.

this is one of the examples the artist provided:

timelapse: asset #39969659 , original art (not upscaled): asset #32608595

whether that kind of treatment should be considered ai-generated or ai-assisted, i'm not sure.

this is all from last year, judging by newer posts it looks like they've stopped doing that.

yet we still see recent art like post #10783883 where the dress looks totally ai-generated while the head looks fine, post #10519866 where the headphones are nonsense but the rest looks alright, to name a couple of examples.

so ultimately we have an artist who seems to be fond of AI tools and can't be trusted, every post needs to be evaluated on a case-by-case basis.