Hacker News Reader

Comment by version_five | Hacker News Reader

This appears to be just looking for the nearest neighbors of the image in embedding space and calling those the source data. This by definition would find similar looking images, but it's not strictly correct to call it attribution. To some extent all of the training data is responsible for the result - as an example, the model is also learning from negative examples. The result here may feel satisfying, but it's overly simplistic and it's misrepresenting what it is to call it attribution.

(The silly story they have on the site doesn't really score any points either, it reminds me of RIAA et al)

View on HN

So it is the opposite of attribution. It just makes up a plausible-looking story of attribution and passes it off as the truth. If you passed it a hand-painted image from 1850 which was not in SD's training dataset, it can happily declare that it was inspired by someone's piece from 2019. There's no actual causal inference going on.

It has more in common with using an AI language model as a "bullshit generator" than it has with human attribution. Which isn't a coincidence, since nearest neighbour similarity searches ARE a type of machine learning, just a much simpler one than SD.

Anti AI people who are upset about attribution, should learn the technology and try to create an actual attribution model: one that has a notion of causality, which could say who influenced who.

Causal model requires causal assumptions, but you can probably get far with the simple assumption that "works from the future do not influence works from the past".

You can literally take a photo right now, upload it, and it will find similar photos from the SD dataset.

As an obvious example, I uploaded a photograph I took of a painting I have by Julie Hagen-Schwarz* that's been in private possession since 1882 and never put online: https://www.stableattribution.com/?image=7730efac-bf52-4077-...

It still finds ostensible "source images" for the art. So, yes, it's clear this service is pretty much bogus.

* https://en.wikipedia.org/wiki/Julie_Wilhelmine_Hagen-Schwarz

Why do you close the space for people who are pro-ai AND pro-attribution? The entire ai space wreaks of this divisiveness, and is likely why it will continue to die out as another "art-fad". There is seemingly little willingness to integrate into the existing art world in good faith.

I close the space for liars, whether they are pro or anti, and this service lies about attribution. Lying about attribution will just make everything worse.

I carefully did not mention this service, but rather pointed it to your sentiment which AGAIN seems to be binary in this pro/anti idea which is unnecessarily rigid and resistant to furthering discussion on the topic.

Yea it feels misleading. I think finding attribution is genuinely a hard problem and a solution in this space would be valuable.

I don't think such a solution can work just based on the resulting image though. It probably needs to input prompt to have any chance of working.

> Yea it feels misleading. I think finding attribution is genuinely a hard problem and a solution in this space would be valuable.

Wouldn't the attribution be basically all the dataset with various attribution probabilities? It's kind of like a reverse neural network.

I'd like to say AI trained on say patent database would prevent many patents that shoudn't be granted in the first place but my guess is that lawyers would quickly adapt to writing them in a way that doesn't trigger AI

Yeah, it seems like it'll just as happily 'attribute' human-made images as well, which calls the whole thing into question. If it's really showing the images that stable diffusion has 'stolen from', and it'll do the same for humans, does that not mean people are equally guilty?

Maybe Stability AI is playing 4D chess and made this website themselves as a sorta "false flag" to help demonstrate that what SD is doing is no different than what humans do, to help them win any legal battles.

How is what any of these image generators are doing any different from myself when I (try to) make art? I draw on my experiences and senses and try to reproduce a picture and those experiences include natural things I've seen as well as art others have made.

More so how are these image generators any different from text generators like ChatGTP? I feel like if first tool out from these AI gates was a bot that wrote good-enough-to-use code, no one would have batted an eye. Everyone would just go "yeah we told you that those pesky programmers would eventually automate themselves out of jobs", but since it is rendering pictures which most of the populace can appreciate and it is a skill that is easy enough for literally any child to pick up, but requires a lot of dedication to master all of a sudden this is theft and should be illegal.

I really hope this lawsuit or whatever doesn't go anywhere, because it will not change anything. The genie is already out of the bottle as far as image generators are concerned - however it means that we the people won't get whatever comes next. Whatever comes next will be tightly held by big corporations and they alone will reap the benefits, whatever they may be.

> More so how are these image generators any different from text generators like ChatGTP?

I've spotted this pattern a couple of times and this sort of circular reasoning seems concerning. Whenever one of (Stable diffusion, Copilot, ChatGPT) comes up in a discussion, their legitimacy seems to be swiftly justified by existence of the other two, even though they're all uniquely problematic in how they wash away attribution and licensing.

The thing is that no one seems to have these concerns with ChatGPT. When DALL-E came out it was touted as end of artists since now everyone could "just make their own art", but most people just see ChatGPT as a toy.

This is not circular reasoning, more observation of what people value. Since everyone can "Google and gather information" ChatGPT isn't valued enough, but since most people don't know how to draw DALL-E and Stable Diffusion are seen as industry destroyers.

It’s interesting to read your perspective. I feel like I’ve seen more doom-saying about co-pilot and ChatGPT than I have about DALL-E and friends. But that may be selection bias.

I think there's plenty of doom saying with both. LLMs and diffusion models are both highly disruptive technologies, and I think it's fair to speculate on what the impact of that will be.

The difference is I don't see book authors and news writers trying to sue OpenAI for "stealing" their articles and not providing attribution, or creating websites that try to divine which specific book chapters or news articles ChatGPT used to generate any particular response (as if that's at all representative of how GPT works).

Well, yeah, copilot had bunch of clueless people yell how it would make programming so much easier, but it was pretty obvious for anyone who used it that it was no where near ready do replace anyone. It was more like a good generic LSP

Yep straight up just uploaded a picture from my camera phone and it confidently found a bunch of strikingly similar images.

That in itself isn’t an indictment of the website. The tool is built on the assumption of being provided an image generated from Stable Diffusion. If you violate that assumption, it isn’t surprising that the tool fails your “test.”

By comparison, a “dog vs. cat” classifier that has 100% accuracy on the dog/cat task will, nonetheless, tell you that a slice of pizza is a dog... or a cat.

(You could possibly interpret the results as “if SD had generated this image, it would have drawn most from these sources in doing so”.)

It's an image similarity search engine, slapping on the tag SD attribution for marketing purposes.

Note that just because an image is similar (to human eyes) doesn't mean that it played a more significant role than a seemingly more dissimilar image. It could even return similar images that SD wasn't trained on at all. Even conditioned on providing an SD-generated image, it fails.

(Something doing what it claims to do, as opposed to naive image similarity, would actually be pretty cool and useful.)

I fed it several very different images and yeah, that was my experience as well. There were many, many details that weren't present in the 'evidence' it provided.

Reply on news.ycombinator.com