While human painter may repeat style of fragments of work of other artist as homage, computer software does it more as plagiarism.
This isn't much different than issues surrounding Microsoft GitHub Copilot.
While human painter may repeat style of fragments of work of other artist as homage, computer software does it more as plagiarism.
This isn't much different than issues surrounding Microsoft GitHub Copilot.
It is? I may be cursed with a good memory but absolutely no "AI art" that my friends passed on to me looked original. In some cases it was painfully obvious what it's based on, in other cases it just felt like i've seen it before.
Legal it may be (although there will be conflicts about machine generated derivative works) but it's also boring.
People are just enthusiastic because it's "do it with a computer". If a young human artist came up with this stuff you'd tell them they're good technically but now they need to develop their own personality.
Inference takes 4 seconds on an A100, I’d say that makes it useful for a wide range of applications that weren’t even possible before.
I do agree that the outputs will lack a sense of communication with the viewer that art has.
Do you have an example of this?
Ideally with a link (for DALL-E 2/Midjourney) or prompt and seed (for Stable Diffusion) to show that it was generated from an incidental prompt - as opposed to uploading the existing image and just using the AI for inpainting/fine-tuning.
But I definitely remember being shown something that was the submarine from the beatles album cover with different colours and ... something else that i forgot what it was ... pasted in. The text prompt included submarine but not beatles.
It is 100% legal for me to blatantly copy the artistic style of Ilya Kuvshinov. I could even sell such work as my own - because it would be. I'd be seen as a derivative or knock-off artist but it can be done. As far as I can tell the real issue is that the AI does it far better - and much faster - than people can. Creating heavily derivative works that, if nonfamiliar with Kuvshinov's work, could possibly be mistaken as being theirs.
That the AI has learned that artist signatures/watermarks are a "feature" of some artwork and so sometimes adds its own has only reinforced people's mistaken belief that it is stealing art when that simply isn't how neural networks like this work at all.
Speaking of - how many people doing style studies of Ilya's work reached out for permission first? I'd wager zero. Because that isn't a thing artists tend to reach out and ask for permission to do. So why should an AI be expected to ask permission first?
[0] Quite literally only to illustrate my point.
https://www.youtube.com/watch?v=FrvZTeLP96E
https://www.youtube.com/watch?v=_DadJAu0Ogs
https://www.youtube.com/shorts/gc8DhwMZDKU
https://www.youtube.com/watch?v=7U4RSCK-Tn0
https://www.youtube.com/watch?v=KPWOy7NxYho
https://www.youtube.com/watch?v=1nc--w5NfCM
https://www.youtube.com/watch?v=a0K3F9al2Js
https://www.youtube.com/watch?v=K5l8y1HZ1zc
https://www.youtube.com/watch?v=sDRS9im1q4s
https://www.youtube.com/shorts/4zBoHcIo4mc