Everyone overlooks the fact that it will still take someone (i.e. a graphic artist) to produce great AI imagery.
First, AI generated art is random and disposable. Yes, you'll get a great image that you can use once, but then what? You can't build a campaign on it.
Second, AI generated art can't be copyrighted, so knockoff competitors are free to use your AI-generated marketing images.
At the very least, you can seed the AI with a paid graphic artist's work (seed-based AI images can be copyrighted). But that artist will do it better than your unpaid intern.
Mmm, I don't know about this. At the very least AI lowers the bar for how talented a graphic artist needs to be to produce professional work, which means it'll be easier to undercut them, which means it'll get much harder to make a living as a graphic designer. It amounts to the same thing as killing off the profession, as seen from the perspective of someone in the profession as opposed to someone without skin in the game. It's like saying push-button elevators didn't hurt the profession of elevator operator, because somebody's still got to push those buttons.
I think AI in general, across almost every industry, will shift value away from technical proficiency and toward creativity and taste. Implementation of an idea/vision will be commoditized, but having a great idea, a unique insight, the taste and ability to identify top-tier work will still be highly valuable. This could well remain true post-AGI.
In graphic arts, the overlap between people with technical proficiency and vision/taste is probably quite high, but it's not one-to-one. There are people with excellent taste who can identify great art or design when they see it, and who can perhaps imagine incredible masterpieces in their minds, but cannot draw a convincing stick figure. On the other side, there are people who can expertly make someone else's concept real, but can't come up with a compelling concept themselves. AI will be great for the former, and bad for the latter (or at least force the latter to adapt).
Whether this will have the effect of concentrating wealth or distributing it more widely strikes me as a very difficult question. It may be devastating for certain professions, but could also enable a whole new class of entrepreneurs. I could see it going either way, or the two effects may cancel each other out and economic equality stays about where it is. We're in the realm of complex systems here, so I wouldn't put much stock in anyone's prediction.
> I think AI in general, across almost every industry, will shift value away from technical proficiency and toward creativity and taste.
The problem is that an artist still needs to eat in the 10-20 years it takes to develop "creativity and taste".
What AI will do/is doing is knock out the entry-level jobs. If you can't train humans on the entry-level, you will eventually have no experienced people.
It also raises the bar of what's possible. What counts as "professional level" changes each time some new technique emerges. A skilled artist will always be better than a random person.
The visual entertainment "supply" is not limited by the current state of tools. It's always limited by the skills of the top crop. Professionals are always ahead and hard to come by. The industry's self-regulating mechanism is novelty; what is abundant becomes fundamentally uninteresting and dies.
This is the march of progress. Digital brushes in Procreate lowered the bar for how talented an artist needs to be to create an oil ‘painting’. The camera lowered the bar for creating portraits.
> AI lowers the bar for how talented a graphic artist needs to be to produce professional work
I think it's a different kind of talent, and not automatically a lower bar. The key to being a professional artist is being able to offer variants based on given direction. Either way, it's much much more than pushing a button or holding a lever in place for a period of time.
> Second, AI generated art can't be copyrighted, so knockoff competitors are free to use your AI-generated marketing images.
No. First off trademarks exist and they found that work done solely by the machine couldn't be treated as a work for hire copyrighted by the machine and assigned to the operator. There is no reason to believe that work couldn't be treated directly as copyrighted by the human operator who has creative input nor is the matter with the images used to train the model truly settled.
>First, AI generated art is random and disposable. Yes, you'll get a great image that you can use once, but then what? You can't build a campaign on it.
You can already get variations on a them and text driven modification eg make the blank a blank or make the blank blanker.
There is no reason to believe that work couldn't be treated directly as copyrighted by the human operator who has creative input nor is the matter with the images used to train the model truly settled.
...Other than the USPTO and the federal court system issuing multiple ruling stating the opposite, including a decision last week which specifically stated that the output of an AI model is not copyrightable, upholding an earlier decision by the USPTO... (https://www.hollywoodreporter.com/business/business-news/ai-...)
Except for the part where the court didn't find that. It found that work only created by the AI didn't qualify. Had it asked if a work created by the AI AND the person qualified it would no doubt have qualified as is already clear from using photoshop not serving to remove your ability to produce copyrightable works. The case didn't ask that and therefore it wasn't answered in any meaningful fashion.
The act of prompting and customizing iteratively especially in systems which allow the user to submit a prompt that modifies the existing work for example "replace the human being with a monkey" "make the monkey pink" etc are clearly creative works that USE an AI not uncopyrightable.
If you want to argue that point you absolutely cannot do so on the basis of a case that literally never addressed that issue unless you would like to traverse the muddy ground between actuality and fiction.
The ruling stated that the Constitutional justification for copyright (and other IP) laws was to incentivize creators. AI does not need incentives, and thus AI-generated content cannot qualify for copyright. Under this line of reasoning, neither can patents (though note that trademarks derive value from the resources and effort spent promoting them, not from their creation, so trademarks are unaffected).
The act of prompting and customizing iteratively especially in systems which allow the user to submit a prompt ...are clearly creative works that USE an AI not uncopyrightable.
If you want to argue that point you absolutely cannot do so on the basis of a case that literally never addressed that issue unless you would like to traverse the muddy ground between actuality and fiction.
The case literally deals with the output of the AI model, not the input. But on that note...under existing law, code can be copyrighted but not its output. Thus, it is logical to reason that prompts to an AI model can also be copyrighted to the extent they are not strictly functional.
But with AI models and content generally, nobody cares about the prompts/inputs. The output is what matters. (For comparison: Deep Impact and Armaggedon were both the results of the same input: disaster movie in which a team of astronaughts has to go to the asteroid to blow it up before it destroys Earth. The "models" were different screenwriters and directors. Compare the outputs: one is a blockbuster classic, and most people don't remember the other movie.)
There was a statement on the prior thread that described the situation particularly well I'll reproduce it herein and link the original comment rather than trying to better it.
> The headline doesn’t seem to be what actually happened. The filer was arguing that the ai created the work on its own as a work for hire and thus the ai was the author with the computer scientist merely being the owner of the copyright as it was made for hire. I don’t think the argument that ai is a tool and the human operating it is the author was considered because the filer explicitly didn’t want to consider it.
> It makes it clear that the computer scientist doing the filing was trying to argue this was a work made for hire with the author being the computer. They wanted to argue that copyright can be assigned to non humans, but that just isn’t how the law works. The summary makes it clear early that it’s just taking their word that the work had no human input and was thus purely the creation of the computer.
In short
A: Computer generated efforts virtually certainly qualify for copyright.
B: Non-artists can in fact iterate and modify work not just randomly generate shit.
C: This will virtually certainly get much better over time.
You will still get much better work out of a professional who can both utilize such tools when desired, and actually create not just copy or prompt art. This thesis is supportable but we shouldn't build it on sand lest it look more vulnerable than it is.
Random variations aren't interesting, they just make something abundant even more abundant and secondary. Unless you have a model with sufficient intelligence that can create something conceptually original (at which point we're all fucked, not just artists or programmers), it's not going to fly. Text driven modifications imply conceptual human input; besides, they are inherently worse than higher-order input, just like text to image alone is worthless for anything meaningful.
There exist systems where you can describe not only initial scenes but successive textual modifications to existing images and furthermore variations aren't random. Successive selections are a way to zero in on a concept.
AKA tell me you haven't spent time with diffusion models, without telling it :)
I actually did figure out what works and what doesn't in real artistic use. Which is the entire point of the article in OP which nobody seem to have read - text doesn't work well beyond the basic use or amateur play, regardless of it being the initial prompt or editing; you need sketching and references (and actual skill) to do real work. I don't think anybody's using available methods of textual modifications for anything complex - they are cumbersome and unreliable, even worse than textual prompts. In fact, I haven't seen anyone using them at all.
Besides the implementation details, natural language just doesn't have enough semantic density and precision to give artistic directions, even for a human or AGI. That's a fundamental limitation. Higher order guidance, style transfer, and compositing is how it's done.
> Yes, you'll get a great image that you can use once, but then what? You can't build a campaign on it.
Checkout confyui, it has an incredible amount of composability that allows you to generate new images based on others. Like image to image but on steroids.
For example, you can generate a character sheet and use it to generate the same characters on different poses using controlnet. Or you can have a base image for an object and use that to generate the same object from different angles and/or different colours etc.
I agree with you, but the main problem is that illustrators are under-appreciated. We are in a world where management with no technical knowledge are having too much power and stealing paychecks.
Also people cannot judge great art or imagery. Unless you have had the training. But the average person? Nope. You can tell what you LIKE but that’s not the same.
I don’t have much training but it is not that difficult to spot AI arts which is pretty repetitive. The first couple are awesome but it gets old really fast.
AI generated art may be disposable but it certainly is very, very good. Midjourney makes plenty of impeccable art and photorealistic images that have no flaws. Also, even if there are flaws a week with some YouTube videos can teach anyone how to fix them, you don’t need someone with five years of deep experience.
First, AI generated art is random and disposable. Yes, you'll get a great image that you can use once, but then what? You can't build a campaign on it.
Second, AI generated art can't be copyrighted, so knockoff competitors are free to use your AI-generated marketing images.
At the very least, you can seed the AI with a paid graphic artist's work (seed-based AI images can be copyrighted). But that artist will do it better than your unpaid intern.