AI-generated art check thread

TenaciousTurtleDucks said:

post #6234437

I'm positive it's generated. Firstly, the image is unlike all the other posts under this artist name. You can't really make out the lines around/of the character which is a common sign. The right eye under the hair has strange artifacting & no blue at all even though her left eye has full color. Position of the legs at the bottom, along with her mouth/chin. Her left armpit/shoulder don't look natural. The overall color palette/tone/contrast of the image is characteristic of models like NovelAI/Anything. The lighting looks like there should be light on her torso but there isn't any.

The heterochromia is a design trait, seen more clearly on post #4990946. However, now that you mention it, the eye looks suspiciously like it's artifacted rather than containing a normal pattern.

edit: nevermind

Edit: Actually it's a tough one to make a final decision, the more you zoom in.

It could just be an anime problem where so many styles start looking the same after a while.

https://danbooru.donmai.us/posts/6241000

TenaciousTurtleDucks said:

post #6234437

I thought that this might be ai-assisted (coloring/rendering) when posting it, but I don't believe it is purely AI generated because everything is too clear and coherent and the details of the design are accurate. Plus, this outfit was only revealed on April 1st and as far as I'm aware it's not possible to train AI generation unless a decent amount of material is available (I'm probably wrong on this, my technical knowledge is rather limited)

Hereinafter said:

I thought that this might be ai-assisted (coloring/rendering) when posting it, but I don't believe it is purely AI generated because everything is too clear and coherent and the details of the design are accurate. Plus, this outfit was only revealed on April 1st and as far as I'm aware it's not possible to train AI generation unless a decent amount of material is available (I'm probably wrong on this, my technical knowledge is rather limited)

You can train A.I on images it has never seen before.

Watch the timestamp at 2:28.

https://youtu.be/W4Mcuh38wyM?t=148

feline_lump said:

AI filter over post #6112079. Her hand is behind her head.

Wolf_of_Gubbio said:

post #6233132 is probably AI-generated. the left hand, under the head, seems non-existent

Came here to talk about these images and respond to this "no hand" accusation. post #6233120 and post #6233132 are redraws of post #5688075 and post #6112079. And there is no doubt the latter two are drawn by humans. The artist just intends the left hand to support the head. That is not a valid reason to discount these pictures at all.

The artist declared using AI assistance, but look at the resolutions. 4950x7875 is miles above what any AI is currently capable of producing. Zooming in, her face and right hand are very sharply defined with overlapping semi-transparent brushstrokes that look way too human. No color leaks, no artefacts, object-behind-object continuity is flawless, and we can find the artist's original work as the first draft.

To me, this is an example of AI assistance that does not take away the majority of human work. In fact, I cannot point out a single place where the artist clearly has not painted over AI's output. The most I can say is that maybe the artist learned how to color and shade the picture by generating an AI reference, but all those brushstrokes suggest that the artist then hand-painted the colors and shading themselves.

That's why I'm feeling a bit unfair that these pictures are not getting approved.

ice_cream said:

I'm positive it's generated. Firstly, the image is unlike all the other posts under this artist name. You can't really make out the lines around/of the character which is a common sign. The right eye under the hair has strange artifacting & no blue at all even though her left eye has full color. Position of the legs at the bottom, along with her mouth/chin. Her left armpit/shoulder don't look natural. The overall color palette/tone/contrast of the image is characteristic of models like NovelAI/Anything. The lighting looks like there should be light on her torso but there isn't any.

I'm not convinced.

You can't really make out the lines around/of the character

Yet each component of the image has clear boundaries. An AI-generated image usually has color leaks giving bright objects unwanted glows or color patches that exceed contours.

The right eye under the hair has strange artifacting & no blue at all

I don't see problems with the shape of the eye.

Isekai Joucho, the character, has heterochromia. You can see it from her official artworks like this one and this one. The eye under hair in post #6234437 has the correct color.

Position of the legs at the bottom, along with her mouth/chin.

Look strange indeed.

Her left armpit/shoulder don't look natural.

Foreshortening is a really challenging perspective technique and this isn't perfect but OK to me. AI would only mess this up even more. (In fact have we ever seen AI do foreshortening this well? I'm not aware but please enlighten me)

The overall color palette/tone/contrast of the image is characteristic of models like NovelAI/Anything. The lighting looks like there should be light on her torso but there isn't any.

The artist might have used AI to generate a reference for coloring and shading.

The whole image is pretty consistently lit by a no-diffusion strobe light from the viewer's left. I see no problems.

---

There are other things in this image that I find hard to believe can be done so well by AI.

The very well-defined frills on her dress, which are shaded accurately from light to dark layer by layer.

The window on the top layer of her dress, showing the black inner layer. The window casts a subtle but accurate shadow on the inner layer.

The shading of creases on her waist.

The complex intersections between her hair ornaments and hair.

Hair in distant background is blurred not in the AI "gaussian blur" way but in a pretty human way.

Etc.

8253803 said:

Came here to talk about these images and respond to this "no hand" accusation. post #6233120 and post #6233132 are redraws of post #5688075 and post #6112079. And there is no doubt the latter two are drawn by humans. The artist just intends the left hand to support the head. That is not a valid reason to discount these pictures at all.

The artist declared using AI assistance, but look at the resolutions. 4950x7875 is miles above what any AI is currently capable of producing. Zooming in, her face and right hand are very sharply defined with overlapping semi-transparent brushstrokes that look way too human. No color leaks, no artefacts, object-behind-object continuity is flawless, and we can find the artist's original work as the first draft.

To me, this is an example of AI assistance that does not take away the majority of human work. In fact, I cannot point out a single place where the artist clearly has not painted over AI's output. The most I can say is that maybe the artist learned how to color and shade the picture by generating an AI reference, but all those brushstrokes suggest that the artist then hand-painted the colors and shading themselves.

Check the background. There are still resizing artifacts all over the place from upscaling from a much lower resolution. These two images were clearly painted directly over the output from an AI generator. I wouldn’t call that “learn[ing] how to color and shade the picture by generating an AI reference.” Unlike the background, the character was 100% painted over with no missed spots, as far as I can tell.

kittey said:

Check the background. There are still resizing artifacts all over the place from upscaling from a much lower resolution. These two images were clearly painted directly over the output from an AI generator. I wouldn’t call that “learn[ing] how to color and shade the picture by generating an AI reference.” Unlike the background, the character was 100% painted over with no missed spots, as far as I can tell.

I wouldn’t call that “...”

I respect your different opinion but I consider painting over AI's output a learning process that helps with understanding color choices and depiction of lights and shadows.

the character was 100% painted over with no missed spots

That's what counts, isn't it? The artist did all the contouring and detailing, and as far as I can tell most of the coloring and shading, too. The dominant majority of the image comes from human labor.

Is post #6193801 AI? The background and shading look kind of generic, but the fingers look fine and it also resembles some of Oyaman's older works, so I'm not sure.

8253803 said:

The artist declared using AI assistance, but look at the resolutions. 4950x7875 is miles above what any AI is currently capable of producing. Zooming in, her face and right hand are very sharply defined with overlapping semi-transparent brushstrokes that look way too human. No color leaks, no artefacts, object-behind-object continuity is flawless, and we can find the artist's original work as the first draft.

To me, this is an example of AI assistance that does not take away the majority of human work. In fact, I cannot point out a single place where the artist clearly has not painted over AI's output.

I won't go over the rest because i can't argue against it, but:

• the resolution being big doesnt mean much, a typical AI resolution is a good indicator of ai-generated but the inverse is not true, upscaling is a thing, and some scammers do it better than others, so not a valid point
• I don't get what you mean by "I cannot point out a single place where the artist clearly has not painted over AI's output", i can't myself be sure about the degree of human intervention in the picture, but there are no real visible brush strokes, the closest to "visible brush strokes" i see are on the frills and the hair, and this definitely falls in the realm of "could easily be emulated by AI", i've seen many other ai generated stuff have this kind of look, and seen even more "convincing" brush strokes from AI

Is post #6193801 AI? The background and shading look kind of generic, but the fingers look fine and it also resembles some of Oyaman's older works, so I'm not sure.

I'm sure there is at least a significant degree of AI intervention in this, the hands, forearms and thighs have a distinct look that seems human-made, and then it kinda clashes with the style of the rest of the picture, there is a change in rendering style where those connect to the rest of the body, i guess the hair is at least retouched by a humen too, but those specific parts clash too much with the rest of the pic
The face definitely looks 100% ai-gen, even has artifacts around the eyes and ears, and has the same rendering style as the whole torso and abdomen
And speaking of this artist's older works, stuff that is just months old clashes a lot with the recent pics that do have an AI look, i can't spend hours checking everything to differenciate between ai-assisted and ai-generated but a quick look at some of the other recent posts of this artist lets me see some blatant AI shenanigans

gotb4096 said:

Is post #6193801 AI? The background and shading look kind of generic, but the fingers look fine and it also resembles some of Oyaman's older works, so I'm not sure.

Scrolling through his twitter timeline and he posts sketches/WIP's very frequently.

He also posted this piece which has a lot of visible brush strokes, especially in the hair.

https://twitter.com/oyaman_comic/status/1645384673194029057

I think "generic" in this case is a by-product of the Artist being around for a decade. A bit like Sakimichan who has made enough works that over time they become similar to each other. But they still bear proof it was made by a Human.

post #6245259
post #6245269

Previous post was tagged with AI-generated.

Edit:
Interestingly, artist removed the "disclaimer" from his Twitter. (forum #239147).

VR-Man said:

post #6245259
post #6245269

Previous post was tagged with AI-generated.

Edit:
Interestingly, artist removed the "disclaimer" from his Twitter. (forum #239147).

I hate that shit so much, it makes people distrust artists.
People lying about this does harm to honest artists...

The whole '' ai-assisted '' tag is honestly a cancer imo, it just allows people to get away with it doing minimum work.

Mayhem-Chan said:

• the resolution being big doesnt mean much, a typical AI resolution is a good indicator of ai-generated but the inverse is not true, upscaling is a thing, and some scammers do it better than others, so not a valid point

Without exhaustively listing everything, I'll just point out two details that rule out upscaling.

Look at her eyes (especially eyebrows) and how her swimsuit is outlined. Only when you set your canvas size to that big, zoom in, and use a brush hardness of >50%, can you produce lines as sharp as that. Upscaling, with anti-aliasing either on or off, will make lines' boundaries blurred.

Side note: you will also find hints of layers here and there, due to the coloring of the layer on top overlapping the outline of the layer below.

The background might be upscaled, fair enough, but not the character.

there are no real visible brush strokes

All the outlines with anti-aliasing off are brushstrokes. AI does not produce lines that have zero anti-aliasing. Pick a curve on her body, hair, or clothes, and follow it. You'll see in parts of it the jagged edges indicating human brushstrokes.

Some of her hair is shaded with the hatching technique. I don't think AIs do that.

The shadow cast by her swimsuit on her body is clearly shaded with a thick transparent brush.

Chilly93 said:

I hate that shit so much, it makes people distrust artists.
People lying about this does harm to honest artists...

True. For example, people questioning post #6234437 are just seeing things due to their paranoia and I don't blame them for having it.

Edit to add:

Throwaway9999 said:

You can train A.I on images it has never seen before.

AI can do well to understand what a dog looks like, or remember a human's face, but costume designs as complex as the one we're talking about, reproduced with no mistake? Still beyond its ability as of now.

Hereinafter said:

I thought that this might be ai-assisted (coloring/rendering) when posting it, but I don't believe it is purely AI generated because everything is too clear and coherent and the details of the design are accurate. Plus, this outfit was only revealed on April 1st and as far as I'm aware it's not possible to train AI generation unless a decent amount of material is available (I'm probably wrong on this, my technical knowledge is rather limited)

I'm worried about setting this precedence of "tagging pictures that resemble AI's shading ai-assisted to be safe". AI "learned" its artstyle from humans that had been drawing like that for years. It's probably not healthy to build an implication of "drawing like this = suspected AI" in a community.

Not going to challenge the tagging of that picture, but I just want to express my concern.

8253803 said:

kittey said:
the character was 100% painted over with no missed spots

That's what counts, isn't it? The artist did all the contouring and detailing, and as far as I can tell most of the coloring and shading, too. The dominant majority of the image comes from human labor.

Yes, this definitely counts as AI-assisted and not AI-generated.

8253803 said:
The background might be upscaled, fair enough, but not the character.

Presses X to doubt. The artist obviously put his older work through an AI generator to improve it and then upscaled it. Why would anyone then bother to cut out the rather simple background to copy it and not the much more detailed character, and instead leave a hole where the character was, just to look at the generated image side by side? No way. I’d definitely place my bets on the character being upscaled and painted over as well. The artist was simply very thorough and there are no parts of the character that were not painted over.