This thread died back in September 2021, I revived it in November 2022, it's fun to see how 12 new pages appeared since then.
towforce wrote: ↑Sun Jan 01, 2023 8:11 pm
Here's a simple, unmistakable expression of what I'm saying:
"you can't just copy other people's products and sell them as your own".
Apparently that's all what people care about. Recently Getty Images sued Stability AI. In case you don't know, Stability AI is responsible for the cutting edge technology of AI powered text to image generation that we have, called Stable Diffusion. Even its competition (Midjourney, NovelAI, AI Roguelite, et al) are powered by Stable Diffusion.
Getty Images is a company that makes and sells stock photos, they made a large amount of them, and those images were scrapped and Stability AI used them to train their model (among many other places they got images from for their dataset).
The important bit is that even Getty Images acknowledges that falls under fair use, the simile here would be Getty Images producing a large amount of chess games, and Stability AI using them to produce a model that can generate chess games. The copyright of Getty Images goes out the window and you can see Stable Diffusion not only doing its best to replicate Getty's stock images, but also their logo:

(note images of much better quality can be obtained with better text prompts, this is just an example)
Since Stable Diffusion was released Getty's sales have been cut it half, turns out half their clients preferred to generate the images instead of buying stock ones.
All fine till this point, except that Stability AI is selling access to Stable Diffusion at their huggingface domain for a profit. Yes, the software is open source, everyone is free to download it and use it at home, and there are free sites where you can use it like dezgo, but some people just want to give their money away.
This is what Getty has issues with, if they are profiting from a model that was trained with their images they want a slice of the cake, too, if the service was provided for free they wouldn't have done anything. This is a real life case we don't have to imagine, you can use whatever you want to train your NN and it won't infringe anyone's copyright, if it did Getty would be suing for THAT.
But if you are doing it for a profit, you could be violating the terms of use of the content you're scrapping, to comply you may need to get a licence and share the profits (or not, if Getty loses the case).
The way things happened contrast a lot with all the patent talk on this thread, it's clear there has been a change in mentality. In the old mentality, you want to create something unique that only you have, patent it so nobody else can use it, and sell it to maximize profits. In the text to image generation world, OpenAI developed the technology and showed how to do it with the world, and the latest incarnation is like Stockfish, anybody is free to set up a web service to sell it if they want, and people are willing to pay just because installing it at home is such a hassle.
But it's clear the way technology is headed, open source has won the war.
Your beliefs create your reality, so be careful what you wish for.