Is there a copyright protection in terms of consuming a copyright-protected image? I thought it was only for the purpose of displaying that image. If you're reading the file and reading the data, but not displaying it, is that also protected?
Ask HN: DALL-E was trained on watermarked stock images?
11–20 of 233 posts
Re: Ask HN: DALL-E was trained on watermarked stock images?
#12Wondered the same thing recently … https://news.ycombinator.com/item?id=31159231
Re: Ask HN: DALL-E was trained on watermarked stock images?
#13> but surely you can't just... use stock photos without paying for the license?
They aren't hosting the infringing content. Training on the data is probably covered under fair use. Generations are of _learned_ representations of the dataset, not the dataset itself. This makes it closer to outputting original works (probably owned by the person who used the model).
The players involved here are known for being litigious, however. I wouldn't be surprised if OpenAI did in fact pay some hefty fee upfront to get full permission to use these images.
Re: Ask HN: DALL-E was trained on watermarked stock images?
#14What 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/2015/03/11/39237539…
Yes, but that's an outlier ruling that was widely criticized.
Re: Ask HN: DALL-E was trained on watermarked stock images?
#15What 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?
Another human analogy could be: you take a photo from every art show and museum, and use those for reference as you paint.
Re: Ask HN: DALL-E was trained on watermarked stock images?
#16It doesn't matter. I could put a Getty watermark on anything. Getty would have to show that a generated image was at least in part the same as their image.
Re: Ask HN: DALL-E was trained on watermarked stock images?
#17Relevant earlier discussion about this issue:
https://news.ycombinator.com/item?id=32436203
Re: Ask HN: DALL-E was trained on watermarked stock images?
#18> but surely you can't just... use stock photos without paying for the license? They aren't hosting the infringing content. Training on the data is probably covered under fair use. Generations are of _learned_ representations of the dataset, not the dataset itself. This makes it closer to outputting original works (probably owned by the person who used the model). The players involved here are known for being litigio…
If they had been paying for the images upfront, wouldn't you expect them to train the model on the non-watermarked versions?
Re: Ask HN: DALL-E was trained on watermarked stock images?
#19> but surely you can't just... use stock photos without paying for the license? They aren't hosting the infringing content. Training on the data is probably covered under fair use. Generations are of _learned_ representations of the dataset, not the dataset itself. This makes it closer to outputting original works (probably owned by the person who used the model). The players involved here are known for being litigio…
if they paid for access, or permission, why train on the watermark versions?
I’m guessing they assumed fair use and there will be lawsuits.
Re: Ask HN: DALL-E was trained on watermarked stock images?
#20> but surely you can't just... use stock photos without paying for the license? They aren't hosting the infringing content. Training on the data is probably covered under fair use. Generations are of _learned_ representations of the dataset, not the dataset itself. This makes it closer to outputting original works (probably owned by the person who used the model). The players involved here are known for being litigio…
If they had been paying for the images upfront, wouldn't you expect them to train the model on the non-watermarked versions?
Good point! They certainly had no obligation to pay, either. Perhaps they just scraped it all.