3 ms·
Artist is dominating AI-generated art. And he’s not happy about it
- anigbrowl 4y agoMason encourages any artists who don’t want their works in the data set to contact LAION, which is an independent entity from the startup. LAION did not immediately respond to a request for comment. I'm very much pro AI but also think it's going to be badly disruptive for the creative community, and the people saying 'it will enable new forms of creativity' are failing to get this in a big way. It enables non-artists to generate or construct custom work : I think that's great as I'm a pretty mediocre artist and something like Stable Diffusion gives me all kinds of new toys to play with. but if you're a professional artist and you already have the ability to make art, perhaps in a variety of styles, is it going to help you much? No, no more than a sampler is much help to a guitarist. But the professional artist is going to find themselves competing with mechanical interpretations, not merely reproductions, of their own work. I was at first mystified by the number of prompts I would see with 'artstation' included until I realized it was essentially a way of getting art in the particular style you like without having to pay a commission. Mason encourages any artists who don’t want their works in the data set to contact LAION, which is an independent entity from the startup. LAION did not immediately respond to a request for comment. Licensing terms and optouts as a legal figleaves are a very American way of shrugging off responsibility for the thing you built which is definitely going to be used in the same way you used to market it to begin with. The ability to copy style so perfectly is part of the hook for all these tools, but in some ways a shortcoming. They can reproduce small scale visual features incredibly well - you can get different versions of the same ideas as they might have been painted by this or that famous artist (though I really think they should have limited themselves to dead ones for the input). So far these models seem to be less successful at larger features like compositional balance, though that's a very subjective assessment on my part. It might be that in the rush to impress they've optimized on solving the less interesting problem.