World News

That's not very nice.

If photography is art, so is AI image generation. If one can see something in the natural world they had no part in creating, and get an idea, a spark of creativity, and then choose a camera, choose the angle, choose the framing, set the configurable aspects of the camera such as shutter speed, exposure time, what type of film, what lens to put on, and produce a photograph, perhaps several, perhaps even a dozen attempts to get it just right, and the final result can be placed in a gallery alongside paintings and sculptures and Jackson Pollocks without a single modern art snob batting an eye, how then is that any different from someone with the same spark of creativity tuning a prompt for a model they've become deeply familiar with, seeking to bring the inspiration in their mind's eye into the real world where others can see and experience it too?

I'm sure you've heard it before, but it bears repeating for those who haven't - photography was not initially considered a form of art. But photographers didn't seem to care too much, and neither did the layperson, so here we are again, having the same old argument about another new art form made possible through a technology that invokes Clarke's law.

People falsely conflating fiddling with prompts with being an artist are clowns I agree but I think AI has a legitimate place as a secondary part of an artistic process. I remember the uproar about autotune in music but in reality it’s often used much as a guitarist might use an effects pedal, there’s nothing wrong with transforming art you created yourself with AI I think as long as you’re upfront about what you’re doing and not trying to pass off skills you don’t have. I think the dodgy parts are people being dishonest about what they’re doing artistically and of course using models built from training sets of questionably licensed IP such as in the case of DeviantArt.

Just to be an irritating pedant (I work in this space, not AI art but on a software team making heavy use of our proprietary AI toolset) the scraping of people’s art only happened once when these models were created to begin with. When they generate images from a prompt they’re not accessing a perfect database of people’s artwork rather it’s more like each artwork that’s part of the training set influences the shape of an enormous mountain by a tiny shovel’s worth of rock and the prompt throws a ball down it in a given direction once the mountain has finished being shaped by billions of tiny digs; the output is the path that ball takes. That’s why the vast majority of AI art kind of looks like arse; it’s like trying to learn music by listening to an entire orchestra at once rather than each instrument individually. The AI knows nothing of the intermediate steps in creating a piece so it can only try and imitate the finished original.

I think the solution is to train AI art models only using public domain licensed artwork and perhaps Creative Commons themselves writing a license that specifically excludes scraping for AI models for artists who wish to publish under a CC licence but object to AI art models for whatever reason.

You do not have to dwell on it at all. What you could usefully dwell on was learning how latent diffusion and generative imaging in general work. Then maybe you wouldn't post such factually incorrect and asinine takes