Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ask HN: DALL-E was trained on watermarked stock images?

news.ycombinator.com

Post reply on HN
Print view View original

1–10 of 233 posts

Ask HN: DALL-E was trained on watermarked stock images?

#1

Post by whycombinetor »

I just got a Dall-E render with a very intact "gettyimages" watermark on it. I'm no legal expert on whether you have to own the license to something to use it as training input to your AI model, but surely you can't just... use stock photos without paying for the license? Maybe I'm just old fashioned.

Prompt: "king of belgium giving a speech to an audience, but the audience members are cucumbers"

All 4 results (all no good as far as the prompt is concerned): https://ibb.co/gz5RDkB

Fullsize of the one with the watermark https://ibb.co/DzGR063

Re: Ask HN: DALL-E was trained on watermarked stock images?

#2

Post by cercatrova »

Based on the new scraping ruling with LinkedIn [0], anything that is "open gate" (as in, accessible without logging in) can be scraped and (I assume) be used by neural networks. The onus, it appears, is to not use it to generate copyrighted works, like Iron Man from Marvel, just as one can use Photoshop as a tool but is still barred from making and selling an Iron Man digital painting.

[0] https://cdn.ca9.uscourts.gov/datastore/opinions/2022/04/18/1...

Re: Ask HN: DALL-E was trained on watermarked stock images?

#4

Post by resoluteteeth »

cercatrova wrote: post #2

Based on the new scraping ruling with LinkedIn [0], anything that is "open gate" (as in, accessible without logging in) can be scraped and (I assume) be used by neural networks. The onus, it appears, is to not use it to generate copyrighted works, like Iron Man from Marvel, just as one can use Photoshop as a tool but is still barred from making and selling an Iron Man digital painting. [0] https://cdn.ca9.uscourts.go…

> Based on the new scraping ruling with LinkedIn [0], anything that is "open gate" (as in, accessible without logging in) can be scraped and (I assume) be used by neural networks.

The ruling you are linking to is about whether scraping violates the Computer Fraud and Abuse Act.

This isn't really applicable here. First of all, that's a separate issue from copyright. Just because scraping publicly accessible data doesn't violate the CFAA doesn't mean that suddenly all images posted on the internet are public domain or that can use copyrighted images from websites for whatever you want, for example.

Furthermore, how copyright applies to training neural networks on copyrighted works is an open question right now.

Re: Ask HN: DALL-E was trained on watermarked stock images?

#5

Post by olliej »

cercatrova wrote: post #2

Based on the new scraping ruling with LinkedIn [0], anything that is "open gate" (as in, accessible without logging in) can be scraped and (I assume) be used by neural networks. The onus, it appears, is to not use it to generate copyrighted works, like Iron Man from Marvel, just as one can use Photoshop as a tool but is still barred from making and selling an Iron Man digital painting. [0] https://cdn.ca9.uscourts.go…

I would assume that for cases like this is is more a matter of whether you can redistribute copyrighted work that has not had any of the usual "creative use" things applied, rather than whether the original scanning was protected.

Re: Ask HN: DALL-E was trained on watermarked stock images?

#6

Post by im3w1l »

I remember when people used to say ianal. Innocent times when we thought there was an objective law and lawyers knew it. But that's not how these things work. The truth is that no one knows. Ultimately a bunch of people will decide how they feel about it. Well-read legal scholars trying really hard to be fair, but still just people. No one can predict with full certainty which way it will go.

Re: Ask HN: DALL-E was trained on watermarked stock images?

#7

Post by RcouF1uZ4gsC »

What is interesting is a human analogy.

Say you were an artist who went to every art show and museum and studied all the art there.

If you produced a work of art solely from memory that contained large portions of other people's copyrighted art, would that still fall under copyright/require licensing?

Re: Ask HN: DALL-E was trained on watermarked stock images?

#9

Post by whycombinetor »

RcouF1uZ4gsC wrote: post #7

What is interesting is a human analogy. Say you were an artist who went to every art show and museum and studied all the art there. If you produced a work of art solely from memory that contained large portions of other people's copyrighted art, would that still fall under copyright/require licensing?

Precedent in music says sometimes-yes. The "Blurred Lines" lawsuit found that Pharrell and Robin Thicke were liable in the tune of $7m for producing a work of art solely from memory that copied the "signature phrases, hooks, bass lines, keyboard chords, harmonic structures and vocal melodies" of a Marvin Gaye song. https://en.wikipedia.org/wiki/Pharrell_Williams_v._Bridgepor... https://www.npr.org/2015/03/11/392375390/-7-million-verdict-...

Re: Ask HN: DALL-E was trained on watermarked stock images?

#10

Post by cercatrova »

resoluteteeth wrote: post #4
cercatrova wrote: post #2

Based on the new scraping ruling with LinkedIn [0], anything that is "open gate" (as in, accessible without logging in) can be scraped and (I assume) be used by neural networks. The onus, it appears, is to not use it to generate copyrighted works, like Iron Man from Marvel, just as one can use Photoshop as a tool but is still barred from making and selling an Iron Man digital painting. [0] https://cdn.ca9.uscourts.go…

> Based on the new scraping ruling with LinkedIn [0], anything that is "open gate" (as in, accessible without logging in) can be scraped and (I assume) be used by neural networks. The ruling you are linking to is about whether scraping violates the Computer Fraud and Abuse Act. This isn't really applicable here. First of all, that's a separate issue from copyright. Just because scraping publicly accessible data doesn…

Perhaps a neural network's outputs can be deemed as transformative use and thus fair use. I don't know though, I'm not a lawyer.
Post reply on HN