Visual Artists x Imagen | Google Lab Sessions 

Ratatouille says it better.. not anyone can be an artist.. but an artist can emerge from any place

That doesnt mean anything

People have always copied established artists. This is nothing new. Just the tools are changing. Art by a computer won’t be worth much. Art by humans always will. That’s why prints are worth so much less than originals. Artists, don’t worry, we will always value you.

@@mechafractal Thats nice, but unfortunately rubbish Im afraid.

@joekletz2705 You are sounding like a troll just prodding people. Are you are of that?

thanks!

If you think artists are just about slaving away at individual brush strokes to be useful, you don't think very much of the potential of artists, do you? That's like mandating programmers slave away at raw 0s and 1s instead of the richness of higher-level programming languages we have today. The scope and scale of the vision these artists will be able to achieve will be like nothing ever before.

Such is life, you can't fight against AI progress, and remember , video killed the radio star

It's NOT art or style! It's a.i promoting! Don't get these mixed! 🤦‍♂️

And, please do tell, *why* do you need permission to train on someone's work? For 99% of human history, we had *no* concept of erecting walled gardens around the things we create. We only created copyright because the printing press made it trivially easy to make *exact duplicates* of works and undercut the artist's channel of *distribution*, so we made a small protection for specifically *that*, and *that* only. Don't try to warp that protection into some magical, all-encompassing thing that protects anything and everything about a work, or that governs how works may be *used* after being legally obtained. Those are *not* things we as a society have already agreed upon as good things, and they *easily* can end up being naive forms of rent-seeking that not only hurt society at large, but also the individual creating things in the first place. "Think of the artists!" is the new "Think of the children!" where people try to just abuse it as some "get out of jail free" card to try to bully others to let them get their way. How about we *actually* think of the artists, and understand what tools we want to build for them, and artists-to-be, to achieve things of a scale never before possible. Harry Potter isn't fundamentally rendered obsolete if an LLM was trained on its works and can produce alternative young adult entertainment; it stands side-by-side strongly on its specific combination of merits.

@@Aurora12488 Thanks for the reply! I didn’t think people cared about comments on videos like this so I appreciate hearing what you had to say. People need permission to train on someone else's work because that's generally how composite works have been handled for decades. You’re absolutely right, by the way. Copyright law hasn’t been with us for most of our existence on this planet, among other things, but let’s not pretend it’s some brand new idea that’s only been around for a few years. True, it was originally made to protect authors from British printing presses (~234 years ago, only about 14 years after the U.S. became a country.) It has existed for all of digital art as we know it, and far before the concept of machine learning. Suggesting that copyright law should not address emerging technologies is, quite frankly, completely absurd. Copyright law has always adapted to protect artists from mistreatment of their work by new technology. It’s intended to do that. Preventing opportunists from freely scraping any and all literature, illustrations, videos, or music they can get their hands on to train a dataset is hardly a stretch. It would have been more relevant to cite composite works in defense of this idea due to conceptual similarities; I would have suggested the generative AI works could pass the threshold of originality despite containing samples of other people’s works by way of the utilized dataset. Despite what some people may believe, obfuscating somebody else's copyrighted work in a new digital format does not nullify its protections. There’s already a clear precedent for using someone else’s copyrighted samples too: the author needs to get permission from the original creators to do so. If everybody were perfectly complicit in getting permission from every artist whose input was used for their AI generated output (which is highly improbable as it currently stands), then I would have no objections. But that’s not what’s happening. The platforms that host generative AI tools do very little, if anything at all, to enable anybody to do that in the first place. In fairness, however, I’m not suggesting previous works will be “rendered obsolete” by works created using generative AI. To be honest, I’m not completely sure how to really apply the concept of obsolescence to a children’s fantasy series in the first place. (Then again, I’ve heard strange things about Harry Potter …) That’s not really what was at the heart of my original comment to Google anyway. I apologize for not being very clear before but admittedly it was a kneejerk reaction. I genuinely expected it to go ignored. As a matter of fact, I actually completely agree that this technology has great potential to spawn its own creative movement, and artists should use generative AI to create great works in the future! For the most part, I find the AI art with the most editorializing to be the most interesting pieces, and I don’t otherwise have an issue with them - besides the fact that they should not rightfully exist in their current state. At the end of the day, the fact is that people have uploaded their work to the internet for years before this. They had near zero expectation that their work would be used for this purpose, and for good reason. If all infringements involved were resolved today, the Gen AI tomorrow would be incapable of producing anything. The current trajectory is infeasible, and it isn't the fault of artists. It is the AI developer's responsibility, and based on the recent wording it at least seems Google might somewhat agree. Even if it's just to protect their products from litigation. It's probably a logistical and financial nightmare to fix - it should be fixed regardless.

@@OopsWrongButton-d1i Thank you as well for the reply; I think it's good to discuss these things openly and honestly, and I appreciate the time you took in your response. I brought up the timeline of copyright law, not to argue that it's something recent and to be trivially discarded and changed due to that recency, but to paint the larger context it operates in. Before copyright was formalized, it was considered largely by Western culture that creative and scientific efforts were not something to be gatekept, but something that naturally spread as people immitated and built off the things they were exposed to. The expectation with creators was that you would get in essense a first-mover advantage, but no inherent legal protection to your creative endeavors (beyond things like fraud, forgery, literal theft, etc.). Copyright was created to address a very *specific* technical concern. The printing press essentially broke the ability for creators to be able to compete in the distribution of the specific work they created, as now it was trivial for others to simply wholesale copy and distribute the very same work, undercutting the original creator. So, we created laws to give limited time, well-scoped protection. Limited time was important, as we still valued the pre-copyright spirit I described above as the true "de-facto" state of things, and this being just a very temporary concession. And well-scoped is also extremely important: the rights given were *not* : "as a creator, you get the exclusive right to dictate everything about how society interacts with your work". Rather, it was *very specifically* that you held exclusive rights to the *distribution* of copies of your work, portions of that work, or trivial and naive transformations of your work. These are *very* different things. Even today's copyright law reflects the latter, and yet people just *assume* it encapsulates the former. It does not. The law does not make claims as to how works may be consumed, just about what properties works that are distributed may have. And notably, those prohibited properties are all aligned with the expectation of rendering the author's distribution channel for the original work redundant in some form. And it's mostly this *distribution channel* that is what I'm calling out as being "rendered redundant", not the work itself; I should have been more clear. Pirated copies of Harry Potter trivialize the author's *fair distribution channel* of Harry Potter, but not the work itself. AI models trained on the work of Harry Potter trivialize neither. As to whether the distributed AI models themselves and the distributed works they produce exhibit these properties: I see very little evidence they do. These models do not at all contain samples of the original dataset; an image generation model is around 6GB usually, which is about 1/4th the capacity of a single-layer Bluray disc. What they store is weights in a neural network that encode patterns it picked up on across billions of individual works (well beyond hundreds of TBs) it was exposed to as part of training. These patterns are as fundamentally removed as you can get from the actual works themselves, and outside of *extremely* rare, overfitted circumstances, you can't actually get the original work out of the model. There's not enough data there. Now, certain things like specific character designs *will* be encoded, which you could argue being able to reproduce is fundamentally IP violation (somewhat separate from *copyright* ). That's something we'll have to figure out; is it okay to distribute something that could be prompted to create what's akin to fan art/fanfiction? And is that okay as long as you don't distribute it? It's uncharted territory to have essentially a "digital brain" that encodes these sorts of abstract patterns completely independently of the individual works themselves. But that's a separate discussion entirely from our main topic as to whether unconsented training is moral; you could scrub this sort of thing from the model if you really tried hard enough. There were a lot of aspects to your response; I tried to address the spirit of the most pressing ones, and am happy to drill into any specifics, like how these models fundamentally work, in a later response.

@@OopsWrongButton-d1i (Splitting this up because YT for some reason won't let me post the full response.) Thank you as well for the reply; I think it's good to discuss these things openly and honestly, and I appreciate the time you took in your response. I brought up the timeline of copyright law, not to argue that it's something recent and to be trivially discarded and changed due to that recency, but to paint the larger context it operates in. Before copyright was formalized, it was considered largely by Western culture that creative and scientific efforts were not something to be gatekept, but something that naturally spread as people imitated and built off the things they were exposed to. This was not just a limitation of our legal systems, but actually something considered a fundamental good. The expectation with creators was that you would get in essence a first-mover advantage, but no inherent legal protection to your creative endeavors (beyond things like fraud, forgery, literal theft, etc.).

Copyright was created to address a very *specific* technical concern. The printing press essentially broke the ability for creators to be able to compete in the distribution of the specific work they created, as now it was trivial for others to simply wholesale copy and distribute the very same work, undercutting the original creator. So, we created laws to give limited time, well-scoped protection. Limited time was important, as we still valued the pre-copyright spirit I described above as the "de-facto" state of things that everything should return to as soon as reasonably possible. And well-scoped is also extremely important: the rights given were *not* : "as a creator, you get the exclusive right to dictate everything about how society interacts with your work". Rather, it was *very specifically* that you held exclusive rights to the *distribution* of copies of your work, portions of that work, or trivial and naive transformations of your work. These are *very* different things. Even today's copyright law reflects the latter, and yet people just *assume* it encapsulates the former. It does not.

Ever heard of sample based music like hip-hop and house music ?

Or collage art ?

@@weppingpresent to sample other peoples music, you need to obtain permissions/a license from the original creator, this is basic copyright crap dude.

@@weppingpresent you also can’t just grab a bunch of art that other people own and put it into a collage without permission/a license.

​@@gravity_mxk5663 this isn't even collage art, I can get an artist to draw in your style in third world countries then use it to make images.