Live data from Hacker News

Stable Attribution

stableattribution.com

11–20 of 365 posts

Re: Stable Attribution

#11
I was taught to paint by instructors, and then refined my abilities by studying paintings of the old masters, right down to their brushwork and core techniques visible in the paintings to all who see them.

Now I go and create a painting called Sunflowers. Does Van Gogh's estate own some of my work?

Re: Stable Attribution

#12
To actually accomplish something like this purports to be (the linked tool only searches for similar images and doesn't tell you anything about how information ended up inside the model), you could try removing individual images or sets of images from the same artist from the training dataset to see what outputs the resulting model would lose the ability to create. It would be expensive to do that for more than a few images, but given how helpful it would be for the debate about copyright and AI, I think it would be great if some researchers could try it.

Re: Stable Attribution

#13

Am I the only person who looks at this and thinks, “So what?” I mean it’s technically impressive, I guess. But why would anyone care or pay for this product?

I suppose it depends on how much you care about attribution for art.

Re: Stable Attribution

#14

Even if this did work, why would people want such a piece of software?

Linking training data to the output of an ML model is an extremely important area of research. It potentially takes away much of the "black box" aspect by understanding what a model is basing it's decision on. For example, sample attribution can be used to inform data collection, to see if a model is operating correctly, or to decide whether the output should be trusted - there's currently lots of talk about chatGPT making stuff up and not sourcing it's output. This would be the answer.

The page in this thread doesn't do attribution, as I said in my other comment. But if it was possible, it would be very important to the field

Re: Stable Attribution

#15

To actually accomplish something like this purports to be (the linked tool only searches for similar images and doesn't tell you anything about how information ended up inside the model), you could try removing individual images or sets of images from the same artist from the training dataset to see what outputs the resulting model would lose the ability to create. It would be expensive to do that for more than a few…

Your idea is the gold standard in explaining the influence of training data. People may be interested in this paper and more modern variations: https://arxiv.org/abs/1703.04730

It attempts to do as you suggest in a tractable way, to understand which training data is most influential.

Re: Stable Attribution

#16
I wonder how this Super Altruistic Startup is going to monetize outside of becoming the art equivalent of patent trolls, wildly throwing out lawsuits at anything that their algorithm pegs as AI-generated along with a positive “similar” outcome from a reverse image search.

I have yet to hear firsthand from any professional artist a single incident of Stable Diffusion causing them harm or lost revenue, but I have heard from a lot of armchair lawyers salivating about the idea of demonizing/criminalizing anybody that uses a piece of novel software.

Re: Stable Attribution

#17

I was taught to paint by instructors, and then refined my abilities by studying paintings of the old masters, right down to their brushwork and core techniques visible in the paintings to all who see them. Now I go and create a painting called Sunflowers. Does Van Gogh's estate own some of my work?

No one is necessarily saying that you owe Van Gogh a cut. What they _are_ saying is not to claim that you didn't train on Van Gogh or to pretend that you don't know what you practiced on.

Re: Stable Attribution

#18

I was taught to paint by instructors, and then refined my abilities by studying paintings of the old masters, right down to their brushwork and core techniques visible in the paintings to all who see them. Now I go and create a painting called Sunflowers. Does Van Gogh's estate own some of my work?

No, they should clearly own all your work because you used something learned from them.

Re: Stable Attribution

#19

This is a great website, but not in the way the authors intended. Based on some of the examples they explicitly provided, it is clear to me Stable Diffusion creates novel art. Here's a random example https://www.stableattribution.com/?image=a2666aee-0a1a-411b-... I will admit this is a nice tool for verifying the creations of SD aren't pure copies, so I think it will be useful for a time. But as AI-generated images s…

[flagged]
Post reply on HN