According to a recent tweet shared by AI enthusiast Nick St. Pierre, the alleged theft occurred last Saturday. It is claimed that employees from Stability AI infiltrated Midjourney’s database and stole all prompt and image pairs, an action that also caused a 24-hour outage. In response, MJ reportedly banned all Stable Diffusion developers from its services, a move supposedly disclosed internally within the company on Wednesday.
Lol, after both steal every image on the internet.
No wonder the images look similar.
Have you seen those images of a bunch of people overlaid on each other so that shared features become clear and outliers become fuzzy. The result is an average human but it doesn’t actually look like anyone in particular because it’s a human with no striving feature or bold choice. Thats why AI look the same, they all have the same dull parts of real art but none of the interesting bits.
Thats such a good point i hadnt thought about before. Training data helps the ai know what is most common, so its products tend to be tropy, predictable and a bit bland (which is great for some things). They are often lacking that ooomph that makes great works truely unique and fascinating.
Live by the sword, die by the sword.
AI generated images are not, and should not be considered copyright able, and they don’t own the right to the image they generated, as I understand it.
Otherwise, Midjourney are certainly very welcome to start paying royalties to certain popular celebrities whose images they are profiting off of. You can’t have it both ways.
Iirc the only precedent we have is that the AI algorithm itself cannot hold the copyright to what it creates, as one artist wanted it to be. Basically the same thing as the famous monkey selfie.
If the copyright of a generated image can be claimed by the creator of the algorithm, or the user who wrote the prompt, and how much human effort is required for it is still unknown.
The image output themselves might not be protected by copyrights. However, that does not mean that there are no rights over the code (or prompts) used to generate those images or over the database compilations themselves (https://www.copyright.gov/reports/appendix.pdf).
The code is obviously protected by copyright. Not sure why anyone would question that?
If the prompt is protected, then the output image will be too (and by the same owner). I suspect it depends on how detailed the prompt is (just like a tweet might not be eligible for copyright, unless it’s a particularly creative joke/haiku/etc).
I agree with you insofar as no image should be copyrightable.
Don’t steal my stolen data.
No honor among thieves.
Infiltrating databases is too fancy a description for what was actually happening. Someone was just scraping the gallery website a little too fast, and that caused a mild DoS attack.
As happens to everyone scraping data. Sometimes you forget the delay and then you wonder why the interface is burning.
I don’t think they care about the images being used, just the disruption of service. It’s pretty clear that this wasn’t a coordinated thing from Stability and was at most a lone individual acting in bad faith.
It’s pretty ironic though that the company that practices mass scraping has no rate limits to prevent outages due to mass scraping.
They care about Ownership of data?
Only when it’s theirs.
No they care about dollar bills
Quite possibly the most hypocritical thing I’ve ever seen.
Lol its absurd to claim ownership of training data. You didnt create or license it to begin with!
Begun, the AI Wars have
Ironic
While also being completely predictable.
Oh no…anyway.
JFC this time line is stupid