3 ms·

> 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

> 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.

orbital-decay 3y ago
Trademarks are different from copyright. 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.
- michaelmrose 3y ago
  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. You are about a year behind the state of the art.
  - orbital-decay 3y ago
    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.
gamblor956 3y ago
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-works-not-copyrightable-studios-1235570316/ https://www.hollywoodreporter.com/business/business-news/ai-...)
- michaelmrose 3y ago
  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.
  - gamblor956 3y ago
    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.)

orbital-decay 3y ago

Trademarks are different from copyright. 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.

michaelmrose 3y ago
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. You are about a year behind the state of the art.
- orbital-decay 3y ago
  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.

gamblor956 3y ago

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-works-not-copyrightable-studios-1235570316/ https://www.hollywoodreporter.com/business/business-news/ai-...)

michaelmrose 3y ago
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.
- gamblor956 3y ago
  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.)