Which is AI-generated? It looks to me like real pixel art (except the 360) very lazily resized in a non-nearest-neighbor fractional scale and anti-aliased to mush.
What is “ChatGPT font”? ChatGPT and its image tool are distillation models that do not have fonts. They produce images based on per-pixel relational distillation, they are guessing what pixels should be next to each other and do not use fonts. Current models do produce text that can be indistinguishable from fonts, but there is no single “ChatGPT font.” If there is a generic font appearing here, that doesn’t tell us anything new.
For the PS1, I don’t understand what you are referring to. The blurriness and uneven lines happen from compression artifacting and/or resizing to a non-divisible fractional resolution. You can get the same effect now if you go into Photoshop, create a 32x32 pixel image, resize to nearest-neighbor 10x, then set an arbitrary similar but non-divisible resolution with a different resampler (e.g., 56x56 bicubic), and save as JPG at <40 quality. That’s extreme, but you get aliased artifacting, interpolated stepping, and so on.
If you’re taking some other features as evidence of AI, let me know.
I don’t understand you defending ai art, but here, the extremely obvious odd coloring texturing the surface of the ps1 would be enough for anyone to notice, but the generation error in the top is proof. This smudge line is where the ai failed on its final pass, likely to do with clipskip or whatever crap open ai uses as this is clearly open ai’s image model.
Why are you defending ai art so hard, what do you get from defending a massive cooperation?
Buddy, I’m not defending AI, and you making some conspiratorial allegation about my motivation is just weirdly aggressive. You and other people don’t seem to understand what happens with typical generational lossy compression and resizing. Randomly resize and save any image to jpeg 12 times, and see if you don’t see similar artifact noise patterns. That’s a technical literacy thing and not your fault, but the overconfidence here is. The exact thing you’ve marked above is very typical artifacting that occurs for non-AI reasons.
I also know enough to say that I can’t be 100% positive it was or wasn’t AI at some point in the chain. But I can confidently say nobody has identified credible evidence it is AI compared to a multi-generational lossy resize by a lazy designer (and no, posting a screenshot with a vague circle and “that’s obviously AI” is not great evidence - these are not twelve fingers or mush pseudo text, this is pixel level inconsistency).
The things you and others are pointing out here are very explainable without AI, and AI likely would not be reliable enough to create some of the details you see which survived the lossy compression.
I for the life of me cannot believe that a nsfw account is arguing with me about confidence while not giving an inkling of evidence themselves. You know nothing of AI “art” and it shows.
Here is a ps3 i generated from chatgpt just now, it details every single thing you stated. The prompt was: “Generate me a Playstation 3 pixel art image”
If you aren’t willing to give even an inkling of proof to your claims, or even try to show proof when talking about jpeg errors then you have no grounds to stand on.
You are defending ai art, weather you realize it or not.
buttons and their colors make no sense and aren’t correlated to a real PS3 controller.
there’s nine status lights and literally status lights on the disk drive.
the text is spelled “PLAYSTA.TION”.
Yes, I’m aware AI can do “pixel art.” No, this doesn’t invalidate the specific examples and logic from my prior posts. I’ve been discussing this is good faith, but you are not, you’re just reiterating and increasing the volume and insults. Have a nice day.
Sorry, none of this is a clear indicator of AI. The “latent noise” you refer to is perfectly consistent with compression and resizing artifacting and noise. Proportions are often off when making “chibi” icon-sized consoles, but notably, they are consistently or coherently off. Other features are strongly suggestive it isn’t AI. For example:
All of the controllers have consistent layouts, including the correct number and orientation of buttons, player indicators, etc (e.g., the Wii controllers).
Consistent diagonal step effects, even if blurred from poor resizing (see the PS4).
Consistent text for all system indicators that is legible without AI artifacting, even if blurred from poor resizing.
The fact that the 360 and PS3 (didn’t notice initially) are not even pixel art suggests they just grabbed random icons from the web, not ran them through AI generators.
Ok - Yes, Adobe does have insidiously integrated AI tools. But again, nothing you point to here is strongly indicative of AI, and again, just consistent with sloppy & lazy resizing (which you could just as likely see pre-2020, before AI). Adobe also has a very extensive stock library which may be where these came from.
There are some really hard to spot AI generated materials possible now, but the sloppy inconsistency here is - conversely - an indicator that they don’t care much what we do or don’t notice so wouldn’t be spending the time to generate something with all of the consistent details (see list above). Instead, the consistent details suggest human-created versions based on the real systems.
Which is AI-generated? It looks to me like real pixel art (except the 360) very lazily resized in a non-nearest-neighbor fractional scale and anti-aliased to mush.
Ps3 in chatgpt font,
Its very obvious for the ps1.
Sorry…Again, what should I be taking from this?
What is “ChatGPT font”? ChatGPT and its image tool are distillation models that do not have fonts. They produce images based on per-pixel relational distillation, they are guessing what pixels should be next to each other and do not use fonts. Current models do produce text that can be indistinguishable from fonts, but there is no single “ChatGPT font.” If there is a generic font appearing here, that doesn’t tell us anything new.
For the PS1, I don’t understand what you are referring to. The blurriness and uneven lines happen from compression artifacting and/or resizing to a non-divisible fractional resolution. You can get the same effect now if you go into Photoshop, create a 32x32 pixel image, resize to nearest-neighbor 10x, then set an arbitrary similar but non-divisible resolution with a different resampler (e.g., 56x56 bicubic), and save as JPG at <40 quality. That’s extreme, but you get aliased artifacting, interpolated stepping, and so on.
If you’re taking some other features as evidence of AI, let me know.
I don’t understand you defending ai art, but here, the extremely obvious odd coloring texturing the surface of the ps1 would be enough for anyone to notice, but the generation error in the top is proof. This smudge line is where the ai failed on its final pass, likely to do with clipskip or whatever crap open ai uses as this is clearly open ai’s image model.
Why are you defending ai art so hard, what do you get from defending a massive cooperation?
Buddy, I’m not defending AI, and you making some conspiratorial allegation about my motivation is just weirdly aggressive. You and other people don’t seem to understand what happens with typical generational lossy compression and resizing. Randomly resize and save any image to jpeg 12 times, and see if you don’t see similar artifact noise patterns. That’s a technical literacy thing and not your fault, but the overconfidence here is. The exact thing you’ve marked above is very typical artifacting that occurs for non-AI reasons.
I also know enough to say that I can’t be 100% positive it was or wasn’t AI at some point in the chain. But I can confidently say nobody has identified credible evidence it is AI compared to a multi-generational lossy resize by a lazy designer (and no, posting a screenshot with a vague circle and “that’s obviously AI” is not great evidence - these are not twelve fingers or mush pseudo text, this is pixel level inconsistency).
The things you and others are pointing out here are very explainable without AI, and AI likely would not be reliable enough to create some of the details you see which survived the lossy compression.
I for the life of me cannot believe that a nsfw account is arguing with me about confidence while not giving an inkling of evidence themselves. You know nothing of AI “art” and it shows.
Here is a ps3 i generated from chatgpt just now, it details every single thing you stated. The prompt was: “Generate me a Playstation 3 pixel art image”
If you aren’t willing to give even an inkling of proof to your claims, or even try to show proof when talking about jpeg errors then you have no grounds to stand on.
You are defending ai art, weather you realize it or not.
Super obvious AI signals:
Yes, I’m aware AI can do “pixel art.” No, this doesn’t invalidate the specific examples and logic from my prior posts. I’ve been discussing this is good faith, but you are not, you’re just reiterating and increasing the volume and insults. Have a nice day.
the proportions are off and the lines are crooked. there’s also the typical latent noise on places like the controller buttons.
Sorry, none of this is a clear indicator of AI. The “latent noise” you refer to is perfectly consistent with compression and resizing artifacting and noise. Proportions are often off when making “chibi” icon-sized consoles, but notably, they are consistently or coherently off. Other features are strongly suggestive it isn’t AI. For example:
Visual Capitalist says they make all their own graphics (so not just grabbed from the Web) from a “veriaty of tools from Adobe Creative Cloud”.
Adobe has been making a huge push for so-called “AI tools” in their Creative Cloud service.
Ok - Yes, Adobe does have insidiously integrated AI tools. But again, nothing you point to here is strongly indicative of AI, and again, just consistent with sloppy & lazy resizing (which you could just as likely see pre-2020, before AI). Adobe also has a very extensive stock library which may be where these came from.
There are some really hard to spot AI generated materials possible now, but the sloppy inconsistency here is - conversely - an indicator that they don’t care much what we do or don’t notice so wouldn’t be spending the time to generate something with all of the consistent details (see list above). Instead, the consistent details suggest human-created versions based on the real systems.
And a couple of them (PS3, 360) aren’t even pixelated.
And the ps2 controller is unsymmetrical.