Comment on Useful things at the mall
nialv7@lemmy.world 17 hours agoLook at the barcode on the price label. To me that says AI slop
snooggums@piefed.world 17 hours ago
nialv7@lemmy.world 16 hours ago
Lol it even has OpenAI’s watermark…
I know gloating is bad but like, I told you so, feel dumb now?
snooggums@piefed.world 16 hours ago
The plug image portion is literally from the page I linked. Does AI merge images into other images?
What watermark? What are you using to verify that it is AI, an unreliable AI checking tool?
nialv7@lemmy.world 16 hours ago
Yes, AI has been able to take reference images for quite a while now.
And OpenAI generated images have invisible watermarks embedded into them so you can check if it’s AI generated definitively. It’s not an AI detector. Here: openai.com/research/verify/
It’s sad how tech illiterate and yet still confidently wrong people can be.
snooggums@piefed.world 16 hours ago
MrQuallzin@pie.eyeofthestorm.place 16 hours ago
Imagine using the hallucination tool to check for hallucinations
nialv7@lemmy.world 16 hours ago
Honestly sad people can be so wrong and arrogant. Educate yourself a bit: youtu.be/kVXp6UNVPTo
Dyskolos@lemmy.zip 16 hours ago
If you say things like “educate yourself” followed by youtube-links…I mean, on a us-american schoolyard this might be valued as “education”, but…well…
DarthFreyr@lemmy.world 17 hours ago
That just looks like typical camera and compression artifacts on a fine detail pattern, not like a generative AI that doesn’t actually know what it’s drawing so it starts making inaccurate extrapolations? What about that makes you think it was done by AI?
nialv7@lemmy.world 17 hours ago
It’s the typical artifact created by AI when drawing multiple parallel lines, it’s immediately noticeable to me. This doesn’t look at all like image compression artifacts to me if that’s what you are saying.
It’s fine I guess. Ordinary people (not a nerd like me) don’t notice this kind of stuff.
DarthFreyr@lemmy.world 15 hours ago
Can you specifically describe what you are seeing then that stands out as an AI artifact? I am genuinely interested, not just try to challenge your statement. Here’s my observations and thoughts on it:
What I see is the JPEG compression blocks around the lines and characters where the chunk changes from a smooth color to a much noisier square; what looks to have once been much blockier aliased pixels forming the barcode lines that got scaled up and blurred, so now there are distinct vertical segments a within each line, as well as a noticeable gridding on the small characters (some of which might also be my viewer); and then possibly some sort of moire effect from the slightly angled barcode lines interacting with the grid of the pixels and/or JPEG compression chunk boundaries. As with many images on the internet, it looks like it has changed size and format in the past, and it’s possible that someone composting an image from different sources would end up with different artifacting in different places.
Given how small the details of the barcode are, I think they would have to come pretty much entirely from the VAE decoding itself, not from the typical generative AI denoising that we would typically associate with hallucinations, extra fingers, and all the other “classic” tells of older image generation models as well as the less-specific “wrongness” of some more recent ones. So it would AFAIK end up having to be a pretty specific artifact to the actual base model being used, and even then just doing a VAE encoding into the compressed, many-dimensional latent-space and immediately decoding would produce the same artifact, and that’s hardly an “AI” process on its own.
Redjard@reddthat.com 5 hours ago
First, @nialv7@lemmy.world was roght, the background by itself without the box has a synthid, I commented some screenshots below.
I also see those “ai artifacts” on the barcode, though I wasn’t as sure about it before seeing the synthid.
I can make out that the lines don’t run quite parallel but form these V and Λ shapes, closer together on one end and merging together.
Counting the left most blurry line as one, the 5th is leaning into the 4th line at the top for example. The compression is strong and the image doesn’t have enough resolution to make it out very clearly, but it still looks like an actual detail not a compression artifact to me. The lines are all a bit wonky, more than the noise could do.
Here’s a way more obvious example:
Image
This has some other artifacts too that I don’t see in the post, but you can see the general wonkiness. And also how some lines have “bled” together, though this isn’t a great example of that.