3 ms·

i strongly feel that 'in the style of' prompts should be outright blacklisted honestly. other cases might be more vague but if you tell an ai to make a painting

i strongly feel that 'in the style of' prompts should be outright blacklisted honestly. other cases might be more vague but if you tell an ai to make a painting like james gurney, it can only do that by training itself on james gurney's body of work. by extension the creative commons license needs to be modified so that artists and photographers can decide whether or not to give permissions for their work to be datamined by ai.

fxtentacle 4y ago
I agree. I wonder if using "james gurney" as a description here is a trademark issue. That said, I find things like "DALL-E 2 Variations" much more worrying. If you look at their flower shop example, it's difficult to argue that we still need stock photographers if you can get those variations for "free". But of course, the AI can only produce the variations, because some human took the original photo. So in effect the DALL-E variations then become a tool to obfuscate the original photographer so that you can use his/her work without paying. As for the datamining, I believe that's unrelated to the license. I believe in the US you are legally allowed to train your AI on any data, no matter what the original license was. Otherwise, Co-Pilot on GPL or MIT code would also not be permitted.
cmorez 4y ago
The latter might be feasible (Though I doubt databrokers will let the current laissez-faire data-rights landscape go without a fight). The former is both wrong and infeasible. Bigger models get better at copying style with fewer examples. At some point, it'll be enough for a big model to see sufficiently few examples of real people drawing, "X in the style of Y" to get a sense for how to draw a thing in the style of Y. Or an end-user could just feed a single instance of parody art in some artists style, ask for variations, in-paint what they want, and voila! Copyright free [citation needed] style transfer. It will always be possible to launder style, even while adhering to promises to never touch an artists work (directly). Training on an artists entire body of work is not the only way (and thanks to copyright, a not even desirable) way to do so. edit: Open source models will also mean blacklists are unworkable.
alexvoda 4y ago
> if you tell an ai to make a painting like james gurney, it can only do that by training itself on james gurney's body of work. How is that different from telling a person to do the same?

fxtentacle 4y ago

I agree. I wonder if using "james gurney" as a description here is a trademark issue. That said, I find things like "DALL-E 2 Variations" much more worrying. If you look at their flower shop example, it's difficult to argue that we still need stock photographers if you can get those variations for "free". But of course, the AI can only produce the variations, because some human took the original photo. So in effect the DALL-E variations then become a tool to obfuscate the original photographer so that you can use his/her work without paying. As for the datamining, I believe that's unrelated to the license. I believe in the US you are legally allowed to train your AI on any data, no matter what the original license was. Otherwise, Co-Pilot on GPL or MIT code would also not be permitted.

cmorez 4y ago

The latter might be feasible (Though I doubt databrokers will let the current laissez-faire data-rights landscape go without a fight). The former is both wrong and infeasible. Bigger models get better at copying style with fewer examples. At some point, it'll be enough for a big model to see sufficiently few examples of real people drawing, "X in the style of Y" to get a sense for how to draw a thing in the style of Y. Or an end-user could just feed a single instance of parody art in some artists style, ask for variations, in-paint what they want, and voila! Copyright free [citation needed] style transfer. It will always be possible to launder style, even while adhering to promises to never touch an artists work (directly). Training on an artists entire body of work is not the only way (and thanks to copyright, a not even desirable) way to do so. edit: Open source models will also mean blacklists are unworkable.

alexvoda 4y ago

> if you tell an ai to make a painting like james gurney, it can only do that by training itself on james gurney's body of work. How is that different from telling a person to do the same?