Content creators need to realize that this is an unsolvable problem. They were the ones who forced this tagging in place, and now the outcome is exactly as they were told it would be.
It seems like the tagging standards accommodate a wide range of approaches to communicating provenance information. To address this author’s grumps, it seems like more specific tagging, rather than no tagging at all, might serve the interests of these photographers who really did make photos and really did use genAI just to create harmless elements of polish rather than to fabricate subject matter.
In fact, the CAI and the C2PA’s efforts seem to be aimed as much toward vouching and communicating that an image was captured by a camera’s sensor as they are toward flagging up generative AI (e.g. [0], [1]).
Which makes sense: it’s probably always going to be pretty easy to strip out metadata or digital watermarks. So as the flood of AI trash becomes the norm, the long term value would seem to lie in being able to prove that your image is not that.
Would love to see that tagged as well. Dislike how much so much "AI beauty enhancement" has been normalized, to the point where people feel tremendous shame over a raw photo of themselves.
Which would seem a perfectly accurate description. If an AI tool was used to remove something, thereby replacing the removed pixels with something AI-generated, then the image has indeed been "Made with AI".
I played quite a bit with AI photo detection tools, and they definitely trigger on high-quality portraits with bokeh and specific color mood-setting combinations that are overused by generative models (e.g., orange-teal or orange-purple).
AI text detectors are not perfect, but they are right most of the time. These photo tools are absolutely atrocious in comparison. I'm sure it looks good in Meta's metrics, but they almost certainly just have a detector that tells middling photos from good ones, and confuses that for AI. As it happens, for user uploads, it overlaps pretty strongly, but it's still wrong.
Of course in this case, the author appears to be reacting to tech that literally just reads the content credential metadata that Photoshop embeds when you use its generative AI features.
The complaint seems to be more that “made with AI” implies (to him) “created from whole cloth with generative AI,” not “created or edited with AI” as the tag’s tooltip explains.
> Souza told TechCrunch in an email that Adobe changed how its cropping tool works and you have to “flatten the image” before saving it as a JPEG image. He suspects that this action has triggered Meta’s algorithm to attach this label.
You have to flatten the image when you use something that makes a new layer. Probably they used generative fill, not a simple crop.
That said "Made with AI" is a little misleading for this as it implies the whole image is AI created.
Interesting that it says "Made with AI" right below the account name and above the title of the post. Not on the actual photo. To me it looks as if they're saying the entire account / post is made with AI, which is completely plausible.
The slop itself was always destined to be just one part of the problem. But there was money to be made with it, so no one cared about the unintended consequences.
Title is a little misleading: looks like some people see “real” as “no AI components at all” and others as “not generated by AI”. Some countries (like Norway) have required disclaimers whenever Photoshop was used - seems like similar thing.
> As PetaPixel reported last week, Meta seems to be applying the “Made with AI” label when photographers use tools such as Adobe’s Generative AI Fill to remove objects.
While Meta hasn’t clarified when it automatically applies the label, some photographers have sided with Meta’s approach, arguing that any use of AI tools should be disclosed.
26 comments
[ 4.5 ms ] story [ 60.3 ms ] threadIn fact, the CAI and the C2PA’s efforts seem to be aimed as much toward vouching and communicating that an image was captured by a camera’s sensor as they are toward flagging up generative AI (e.g. [0], [1]).
Which makes sense: it’s probably always going to be pretty easy to strip out metadata or digital watermarks. So as the flood of AI trash becomes the norm, the long term value would seem to lie in being able to prove that your image is not that.
[0] https://helpx.adobe.com/creative-cloud/help/content-credenti...
[1] https://c2pa.org/
AI text detectors are not perfect, but they are right most of the time. These photo tools are absolutely atrocious in comparison. I'm sure it looks good in Meta's metrics, but they almost certainly just have a detector that tells middling photos from good ones, and confuses that for AI. As it happens, for user uploads, it overlaps pretty strongly, but it's still wrong.
The complaint seems to be more that “made with AI” implies (to him) “created from whole cloth with generative AI,” not “created or edited with AI” as the tag’s tooltip explains.
You have to flatten the image when you use something that makes a new layer. Probably they used generative fill, not a simple crop.
That said "Made with AI" is a little misleading for this as it implies the whole image is AI created.
Title is a little misleading: looks like some people see “real” as “no AI components at all” and others as “not generated by AI”. Some countries (like Norway) have required disclaimers whenever Photoshop was used - seems like similar thing.
> As PetaPixel reported last week, Meta seems to be applying the “Made with AI” label when photographers use tools such as Adobe’s Generative AI Fill to remove objects. While Meta hasn’t clarified when it automatically applies the label, some photographers have sided with Meta’s approach, arguing that any use of AI tools should be disclosed.
A weak, universally applicable label allows Slop everywhere.