Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Stable Diffusion is a big deal

simonwillison.net

Post reply on HN
Print view View original

381–390 of 488 posts

Re: Stable Diffusion is a big deal

#381

Post by ShamelessC »

Karunamon wrote: post #348
_gabe_ wrote: post #307

Is it just me, or do the comments in this thread seem to be the exact opposite of the sentiment in the comments on similar Github Copilot threads? I just find it a bit ironic that programmers are irate about Github Copilot using their copyrighted material to train. However, if it's an ML model training off of copyrighted artists material, clearly its a transformative work. I just find the opposing sentiments for thes…

The difference appears to be that Github Copilot just reproduces, verbatim, snippets of code without actually doing anything transformative to it. You cannot describe a sort algorithm and have it spit out something other than a direct reproduction of someone else's implementation. With these artwork models, they can emulate general styles, occasionally known characters or bits of text will show up in the output, but…

> verbatim

It actually, by default I believe, now checks for exact collisions with any existing GitHub code (the training data) and removes them. These are somewhat rare in any case, is my understanding.

Not to detract from your other points. Just a common misconception I see a lot.

Re: Stable Diffusion is a big deal

#382

Post by omgwtfbyobbq »

burlesona wrote: post #297
dylan604 wrote: post #288

Earlier quoted context omitted.

>Code that's 98% correct is completely useless. Whoa! Really? Seems like modern software releases would be excited to achieve 98% with the incredible amount of bug fixes/patches released very quickly after the massive beta test known as release day.

Code has to be 100% correct or else it’s considered a bug (assuming it’s syntactically valid). Code that is 98% correct is actually much worse than no code at all. That’s the kind of code that will introduce subtle, systemic faults, and years later result in catastrophic failure when the company realizes they’ve been calculating millions of payments without sales tax, or a clever hacker discovers they can get escalat…

There's also unintended behavior in more subtle ways. Ideally stuff can be formally verified, but that's not practical for most things.

Re: Stable Diffusion is a big deal

#384

Post by jdthedisciple »

Hah! I uploaded a drawn picture of a bridge over cliffs and a river using the prompt "view of bridge over cliffs and river".

What I got was "NSFW content detected, please try a different prompt".

I must say it didn't exactly exceed my expectations.

Re: Stable Diffusion is a big deal

#385

Post by dsign »

peoplefromibiza wrote: post #339
Workaccount2 wrote: post #336

Earlier quoted context omitted.

I would venture to guess that the author doesn't have a full understanding of what the AI is doing or how it works. There are a gazillion artist who take clear inspiration from other artists, and there is no copywrite violation in doing so. The AI viewing his work isn't any different.

> I would venture to guess that the author doesn't have a full understanding of what the AI is doing or how it works. Actually he does. > There are a gazillion artist who take clear inspiration from other artists Inspiration is not the same thing Stable Diffusion does. We can't even define inspiration in a proper manner, but for sure we can say that if someone wants to draw comics in the way Tezuka made them, they ha…

I think that your and your friend's argument reduces to "the AI is not a human, but a machine-like, and thus can not be `inspired' but only `reproduce'". I don't know if it is a novel argument to be tried in a court of (copyright) law, but it is certainly a good one.

Re: Stable Diffusion is a big deal

#386

Post by adamsmith143 »

aronowb14 wrote: post #308

It's funny how hyped up stable diffusion is on HN right now: reminds me of when style transfer first started making it's rounds in 2017. https://news.ycombinator.com/item?id=13958366 I think as technologists we want to think that code can "solve" some of the problems in the art world... but I think we still have a really, really long way to go. I tried to get style transfer adopted at work (worked at a creative techn…

> Imagine deploying production code that only worked 85% of the time... would be a nightmare. I felt, and feel similarly about deep learning approaches to art. They're just so finnicky and unpredictable, for example, add a single extra pixel to that example in this article and the output would look completely different.

Don't think the metaphor works. Code that only works 85% of the time is obviously broken but Art is subjective so an 85% solution to a creative problem could be more than enough for most consumers.

Re: Stable Diffusion is a big deal

#387

Post by kergonath »

throwaway1851 wrote: post #199
kergonath wrote: post #82

Earlier quoted context omitted.

They can (and do) clone compressed electrical representations of everything they see. Everything is just stored in memories in their brains instead of memory on the cloud. In the case of these complex AI, you cannot really extract the initial works, right? They have all been atomised, mashed together, and very imperfectly encoded as weights and parameters. Yeah I am sure a lot of lawyers are going to have a lot of fu…

I see this type of reduction again and again to advocate for one position or another relating to ML. Human consciousness and thought processes aren’t “just” anything (math, electrical impulses, etc.) — the fact is, we don’t know what the brain really does and how it’s connected to our conscious experience, or even what that is! Deep learning is very powerful and impressive in its applications to date. However, it’s s…

> Human consciousness and thought processes aren’t “just” anything (math, electrical impulses, etc.) — the fact is, we don’t know what the brain really does and how it’s connected to our conscious experience, or even what that is!

There are a lot of things we don’t know, but it is not magic. There is no discussion that artificial neurones don’t have much in common with the real ones, which are very non-linear and much more connected. But in the end it’s all electrochemistry.

> Neural networks, despite their name, don’t model the brain.

But that’s not directly related to my point. My point that even in the case of a ML model, you cannot get an exact reproduction any more than you can get from a human’s memory. In one case it’s scrambled somewhere in someone’s brain, in the other on a hard drive but the difference is not really relevant. Subjecting an AI’s production to the copyrights of all the things it’s been exposed to is very similar to subjecting a painter’s production to the copyrights of all the painting they have seen.

Re: Stable Diffusion is a big deal

#388

Post by itronitron »

Karunamon wrote: post #348
_gabe_ wrote: post #307

Is it just me, or do the comments in this thread seem to be the exact opposite of the sentiment in the comments on similar Github Copilot threads? I just find it a bit ironic that programmers are irate about Github Copilot using their copyrighted material to train. However, if it's an ML model training off of copyrighted artists material, clearly its a transformative work. I just find the opposing sentiments for thes…

The difference appears to be that Github Copilot just reproduces, verbatim, snippets of code without actually doing anything transformative to it. You cannot describe a sort algorithm and have it spit out something other than a direct reproduction of someone else's implementation. With these artwork models, they can emulate general styles, occasionally known characters or bits of text will show up in the output, but…

Well, if you crop the image you are more likely to end up with an exact copy of prior work.

Similarly, Copilot can be seen to reproduce 'snippets of code' but would probably struggle with a full application.

The images generated from these AI models are not transformative. In fact they are not even generating art, they are only generating pictures.

Re: Stable Diffusion is a big deal

#389

Post by peoplefromibiza »

Workaccount2 wrote: post #352
peoplefromibiza wrote: post #339

Earlier quoted context omitted.

> I would venture to guess that the author doesn't have a full understanding of what the AI is doing or how it works. Actually he does. > There are a gazillion artist who take clear inspiration from other artists Inspiration is not the same thing Stable Diffusion does. We can't even define inspiration in a proper manner, but for sure we can say that if someone wants to draw comics in the way Tezuka made them, they ha…

We can go back and forth all day. But legally your friend has no legs to stand on. People do "like copies" of works all the time, and it's been ok'ed by the courts just as much. Unless your friend thinks the AI is copying snippets of his work into generated images, I'm not sure where they are getting these ideas from.

> But legally your friend has no legs to stand on.

My friend is just upset.

Legally he has every right in the World, he's the author for Christ Sake!

Will he try anything?

Of course not.

Are you okay with this?

Well, then you should reconsider your values.

> People do "like copies" of works all the time,

If those copies are authorized, I don't see the problem.

Try to recreate a Star Wars image and sell it on the Internet.

See what happens.

> Unless your friend thinks the AI is copying snippets of his work into generated images

Would you bet your life on the fact that it doesn't?

You know why we Europeans came up with the GDPR?

Yes, exactly, because processing of the data, automatic or not, must be authorized by the holder of the rights on that data.

We are not talking about artistic expression here, I'm not sure where you are getting this idea from, this is not inspiration or art or human expression, this is simply data processing leading to algorithmic replica of other people's works.

Without people's work, no model could replicate it.

Impressive, still completely dependent on some kind of source material, that they scraped, they haven't produced it by themselves.

There will be a day when people like you will realize that appropriation is not right.

Books are different, most of the works models are trained on are in the public domain, Shakespeare is public Domain, nobody will ever contest that, but if you think a living author has no right to have a say before someone process their work, you're the one with crazy ideas.

Re: Stable Diffusion is a big deal

#390

Post by anonAndOn »

lurquer wrote: post #374
earth_walker wrote: post #371

Earlier quoted context omitted.

I see what you mean about GPT returning copies. But is there really a difference if your model just happens to 'calculate' an exact copy instead? For example, if you enter in something simple and high profile, what it returns looks pretty close to an existing work... Try, for example, typing in "Banksy", "Brad Pitt" or "Starbucks". So if you type in "Photo of a Coffee", "Impressionist painting of a fruit bowl" or "Bl…

Copyright - at least for images - is going to become moot shortly. Copyright exists to allow an artist time to reap the benefits of his labor. Without this time, they say, no rational person would invest his labor into making art. Dubious… but, whatever. The point is, if you remove the ‘labor’ from the mix, there’s no need for copyright. If I can produce spectacular images with zero individual labor, there will be li…

> there will be little reason for me to copy someone else’s work

Can I interest you in some recently discovered DaVinci paintings? 10% off when you buy 2 or more.

Post reply on HN