3 ms·

The situation with AI art is significantly complicated by the fact that we can't easily distinguish between works produced with vs. without it; with cameras/pho

The situation with AI art is significantly complicated by the fact that we can't easily distinguish between works produced with vs. without it; with cameras/photography that wasn't as much of an issue, which prevented it from being a true encroachment (not to mention the potential subject matter of photography is a small subset of what's more generally possible in human-created art)—with AI art toes and much more will be stepped on and people will be pissed. (Photoshop I see as a separate issue; people take it better because it still resembles traditional methods to such a large degree.)

px43 4y ago
> we can't easily distinguish between works produced with vs. without it If you've played with any of these tools for 5-10 minutes, you'll know what to look for to quickly identify purely generated images. Have a look at some of these people https://thispersondoesnotexist.com https://thispersondoesnotexist.com After flicking through 10 or so of these, you'll start to become more sensitive to the subtle inconsistencies in symmetry, colors, and the unnatural way high contrast transitions take place, etc, and it's the same with anything that comes out of any of these art generating tools. With time, those particular aberrations can be smoothed over, but I think it's kind of like how when we saw Jurassic Park in 1995, lots of people were blown away about the realism of the dinosaurs, but when you watch it now, they just seem like plastic toy figures. Over time, we've become more sensitive to recognizing computer generated animations. Subtle issues with lighting, hair, shadows, etc. When new technology comes out, it takes some time to recognize how to spot these quirks, but over time it becomes pretty obvious. I see some amazing artists using midjourney etc, and then clipping together fragments and painting over them using classic digital art tooling, with some astounding results. As with any artistic tooling technology, the real winners are going to be the artists that learn to embrace the tech.
- forgotpwd16 4y ago
  Someone should make an AI that finds out whether a work is AI generated.
- westoncb 4y ago
  > If you've played with any of these tools for 5-10 minutes, you'll know what to look for to quickly identify purely generated images. I've spent tons of time with a wide variety of tools/models, and I disagree. It's easy to spot when anomalies show up, like yes, this weird swirl on someone's ear is clearly a DL artifact, sure; but the images that people will actually take and use elsewhere will be artifact-free (unless that's the purpose of their usage). Take a look here: https://lexica.art/ https://lexica.art/ I would bet anything you could take a handful of current generation stable diffusion-generated images (handpicked for the highest quality, no visible artifacts), mix them with human-created images in shared genres, and in a controlled experiment people would do no better than random in identifying their sources.

px43 4y ago

> we can't easily distinguish between works produced with vs. without it If you've played with any of these tools for 5-10 minutes, you'll know what to look for to quickly identify purely generated images. Have a look at some of these people https://thispersondoesnotexist.com https://thispersondoesnotexist.com After flicking through 10 or so of these, you'll start to become more sensitive to the subtle inconsistencies in symmetry, colors, and the unnatural way high contrast transitions take place, etc, and it's the same with anything that comes out of any of these art generating tools. With time, those particular aberrations can be smoothed over, but I think it's kind of like how when we saw Jurassic Park in 1995, lots of people were blown away about the realism of the dinosaurs, but when you watch it now, they just seem like plastic toy figures. Over time, we've become more sensitive to recognizing computer generated animations. Subtle issues with lighting, hair, shadows, etc. When new technology comes out, it takes some time to recognize how to spot these quirks, but over time it becomes pretty obvious. I see some amazing artists using midjourney etc, and then clipping together fragments and painting over them using classic digital art tooling, with some astounding results. As with any artistic tooling technology, the real winners are going to be the artists that learn to embrace the tech.

forgotpwd16 4y ago
Someone should make an AI that finds out whether a work is AI generated.
westoncb 4y ago
> If you've played with any of these tools for 5-10 minutes, you'll know what to look for to quickly identify purely generated images. I've spent tons of time with a wide variety of tools/models, and I disagree. It's easy to spot when anomalies show up, like yes, this weird swirl on someone's ear is clearly a DL artifact, sure; but the images that people will actually take and use elsewhere will be artifact-free (unless that's the purpose of their usage). Take a look here: https://lexica.art/ https://lexica.art/ I would bet anything you could take a handful of current generation stable diffusion-generated images (handpicked for the highest quality, no visible artifacts), mix them with human-created images in shared genres, and in a controlled experiment people would do no better than random in identifying their sources.