If you take a model trained on Getty and ask it for Indiana Jones or Harry Potter, what does it give you? These things are popular enough that it's likely to be present in any large set of training data, either erroneously or because some specific works incorporated them in a way that was licensed or fair use for those particular works even if it isn't in general.
And then when it conjures something like that by description rather than by name, how are you any better off than something trained from random social media? It's not like you get to make unlicensed AI India Jones derivatives just because Getty has a photo of Harrison Ford.
I work in this space. In traditional diffusion-based regimes (paired image and text), one can absolutely check the text to remove all occurrences of Indiana Jones. Likewise, Adobe Stock has content moderation that ensures (up to human moderation limit) no dirty content. It is a world without Indiana Jones to the model
I don't know the data distribution, but are you sure that's generated by an Adobe model? I can only see that it is in Stock + it is tagged as AI generated (that is, was that image generated by some other model?)
Disclaimer: I used to work at Adobe GenAI. Opinions are of my own ofc.
Yeah, there's no way Indiana Jones was not in the training data that created that image. To even say it's not in there is James Clapper in front of Congress level lying.
> one can absolutely check the text to remove all occurrences of Indiana Jones
How do you handle this kind of prompt:
“Generate an image of a daring, whip-wielding archaeologist and adventurer, wearing a fedora hat and leather jacket. Here's some back-story about him: With a sharp wit and a knack for languages, he travels the globe in search of ancient artifacts, often racing against rival treasure hunters and battling supernatural forces. His adventures are filled with narrow escapes, booby traps, and encounters with historical and mythical relics. He’s equally at home in a university lecture hall as he is in a jungle temple or a desert ruin, blending academic expertise with fearless action. His journey is as much about uncovering history’s secrets as it is about confronting his own fears and personal demons.”
Try copy-pasting it in any image generation model. It looks awfully like Indiana Jones for all my attempts, yet I've not referenced Indiana Jones even once!
It comes down to who is liable for the edge cases, I suspect. Adobe will compensate the end user if they get sued for using a Firefly-generated image (probably up to some limit).
Getting sued occasionally is a cost of doing business in some industries. It’s about risk mitigation rather than risk elimination.
Feels like "paying extra for the extended warranty" vibes. What it covers isn't much (do you expect someone to come after you in small claims court and if they do, was that your main concern?) meanwhile the big claim you're actually worried about is what it doesn't cover.
And if you really wanted insurance then why not get it from an actual insurance company?
Because almost everything is risk mitigation or reduction, not elimination.
In particular, in the US, the legal apparatus has been gamified to the point that the expectation becomes people will sue if their expected value out of it is positive even if the case is insane on its merits, because it's much more likely someone with enough risk and cost will settle as the cheaper option.
And in that world, there is nothing that completely eliminates the risk of being sued in bad faith - but the more things you put in your mitigation basket, the narrower the error bars are on the risk even if the 99.999th percentile is still the same.
Risk mitigation is about a) lowering the risk and b) lowering the cost in the event you get unlucky.
This doesn't seem to do the first one. It doesn't actually stop you from doing the things you could get in trouble for doing, even though that was ostensibly the original point.
And the second one is where you want to shave off the high side, not the low side. This is why normal insurance is typically a policy large enough to cover whatever plausible amount of damages there could be and then a deductible so you don't have tons of people filing claims for amounts small enough they could reasonably cover themselves instead of paying the insurance company a margin to do it. But this is the opposite of that, they'd cover a little claim you'd have been able to shrug off regardless but not the one where you'd most want coverage.
I believe the specific purpose of that is to appeal to the parts of companies that engage in professional ass covering for the hot potato.
That is, if a well-established company offers you a service explicitly promising no human bones in the soup, when all the other services don't promise that, you can make a reasonable argument that, in the event you were required by circumstances to pick one of the services to use, you have done the best you can in that circumstance, even if human bones ended up in the soup, as long as the company doesn't have a known history of not honoring its commitments.
And in that case, it becomes easier to make an argument to your management chain that you didn't deliberately put in human bones to save on cost, you picked the one that explicitly promised no human bones.
If you take a model trained on Getty and ask it for Indiana Jones or Harry Potter, what does it give you? These things are popular enough that it's likely to be present in any large set of training data, either erroneously or because some specific works incorporated them in a way that was licensed or fair use for those particular works even if it isn't in general.
And then when it conjures something like that by description rather than by name, how are you any better off than something trained from random social media? It's not like you get to make unlicensed AI India Jones derivatives just because Getty has a photo of Harrison Ford.