The "cat is out of the bag" position is IMHO a rather short-sighted and somewhat unimaginative admission of defeat in regards to the way current models are using data from non-public domain images, without obtaining any license from the artists who got scraped. One could very well imagine a situation where providers of…
This does not work. Since you can train the AI with these images anyways. And the dataset is free available content of the internet. Nobody would ever think of limiting Photoshop to draw just vertical lines. Why do you want to do this with AI? It always boils down that you want to regulate the tool and all its users…
Well, to be fair there is at least one thing which I believe is ML-based and is working really well IMHO, without competing with its own dataset : the Adobe auto background removal tool. I think this is a great example of something using ML that is indeed a tool rather than a replacement. It's not running locally though,…
@Tiles But we've already shown examples of pixel-per-pixel (more or less) copies of prominent images, in one of the other AI threads. If "The main concern here so far was that "these basterds uses my data", which they don't. Not a single grain of copyright is touched. They use it for training by simply looking at it.…
The tech is here, nobody will take it away. But there are a few things that need to be fixed. The first and foremost would be making sure how its sourcing its inputs. Using an open source library that was made to study, in a commercial sense will have to stop or there will have to be systems in place to compensate. I feel…
You nailed it. I will not throw common sense and facts overboard just because some hateraging luddies wants me to. When they want to hate me for my view at the matter, shall they. The discussion is over anyways. We have gone in more than one full circle now. I have nothing more to say to the matter here. Tough one.…
one in which the unauthorized use of imagery in a data set used for image generation would be illegal. But why should it be illegal? This is still the casus knaxus ^^ It will definitely be interesting what interpretations a court will follow in the future. But months are gone, and no case yet :)
Furthermore, a proof that it understands what "Ghibli" means, even though the studio explicitely opted out of image generators (which is of course impossible at the deepest level, meaning that MJ for instance only blocks Ghibli as a prompt, but retains the training made off the back of imagery from the studio) : Yuck.
A few questions? Is A.I. trained and can work without Image Libary because it has some fundamental understanding when i mean "chainmail" or lookd it every time for reference pictures? Some AI. works looks fantastic. Looks like a dozen Screenshots of a movie BUT never was the Character or Monster etc... the same everytime…
I think the copyright issues of the training data is an interesting and problematic area, but people and companies are going to make their own datasets. If Star Wars wants a new rebel ship that fits within their universe, they can have an AI fed with their tens of thousands of pieces of concept art, movie stills, merch,…