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

251–260 of 488 posts

Re: Stable Diffusion is a big deal

#251

Post by shinycode »

systemvoltage wrote: post #138
michaelchisari wrote: post #6

After using SD heavily for a week, I half agree with this. It is incredibly disruptive, and it's wild how much it accelerates the creative process. I'll give you that. But two things I've noticed: First, artists will still have a massive advantage over non-artists with this tool. A photographer who intimately knows the different lenses and cameras and industry terms will get to a representation of their idea much fas…

I always wondered if we massively overestimate human creativity. Maybe it is ingrained in our culture and our very being. I’ve never heard counter arguments that humans are not that creative. Creativity demonstrated by Alpha zero chess engine blows Magnus Carlsen’s mind (from his recent interview with Lex Fridman), I wonder if at some point in the future, we’ll finally throw in the towel and get out of the denial pha…

Without trained models from human creativity, what can AI do ? These AI emerged because of human creativity. Picasso created it’s new art form from its own creativity. He created something no one ever though of before. Now AI are fueled with Picasso’s drawing and can produce art that looks like his maybe. But what about creating something entirely new that has never been fueled into the engine. Could the AI invent something I’m about to dream tomorrow and this be the exact same copy ?

Re: Stable Diffusion is a big deal

#252

Post by dannyw »

seibelj wrote: post #50

I am a contrarian by nature. Wearing my investor hat, I continue to be unimpressed with AI tools, including the latest image generation enhancements. I do find stable diffusion interesting, but I don't see the disruption. Similarly, I don't see the disruption from Github Copilot considering I run a company full of highly paid and extremely skilled developers and not one of them uses copilot. How often have you wanted…

How successful are you at investing?

Re: Stable Diffusion is a big deal

#253

Post by Fricken »

pnt12 wrote: post #225
planb wrote: post #175

> Stable Diffusion has been trained on millions of copyrighted images scraped from the web. My brain has been trained on even more copyrighted material. Every book I read, every tv show I watch, the toys I played with as a child. It's hard to imagine that I could come up with anything that is not inspired by copyrighted work.

Two objections: 1: you cannot produce thousands of detailed pictures in a day, this program can. The argument gets pretty clear if you transpose it to other objects, ie why is it fair to ride a bicycle in the sidewalk but not drive a car? 2: copyright laws. You may not see a picture and imitate it. How do you know this AI didn't just imitate one of the million pictures it saw? And if you distribute it and the author…

It's a lot of work for a human to produce an image, but they can think thousands of images in a day quite easily. The artistic process often begins with some daydreaming.

Re: Stable Diffusion is a big deal

#254

Post by slyall »

rhdunn wrote: post #169
LewisVerstappen wrote: post #5

Earlier quoted context omitted.

Yes but we have a common law system and there's already tons of precedent that training AI systems is transformative. It's also quite obvious just by looking at the generated images that it's clearly transformative. The images generated are unique and you can't trace the original copyrighted image from what's generated. You really don't need a judge to see that Fair Use covers Stable Diffusion.

What happens if you give an image prompt like "mona lisa", "daffodils van gogh", or similar designed to describe an image the model was trained on. Will it generate that image? Or for written works, start with a sentance from a copyrighted work, or part of licensed code. Will it start reproducing that work word for word (like code pilot can do with the GPL license)? Getting these to generate copies of GPL'd, company…

What do you picture when you see words like "Elvis Presley" , "Game of Thrones" or "The Google logo"?

Explain how you are not a copyright violating machine.

Re: Stable Diffusion is a big deal

#255

Post by nivenkos »

endorphine wrote: post #120
thirdlamp wrote: post #104

Is anyone training networks to detect deep fakes? As in you feed it an image and it gives you if it’s a real image or a deep fake

And then someone will train a network to create images undetectable by the networks that detect deep fakes.

That's how GANs work...

Re: Stable Diffusion is a big deal

#256

Post by dannyw »

hackerlight wrote: post #235
andybak wrote: post #222

Earlier quoted context omitted.

> All the photorealistic attempts are bad. I wish people would stop exaggerating or making statements without some insight behind them. I've seen hundreds, if not thousands of photorealistic result that range from acceptable to remarkably good. One example it took seconds to find: https://lexica.art/prompt/3fbc30ee-ca0f-42f6-8ea5-6890d469ab...

I misspoke. I meant to say photorealistic faces. I haven't seen a good one yet from DALL-E 2 or Stable Diffusion. The systems that are custom built just for faces do a good job at photorealistic faces, though.

They are bad because they explicitly filtered out faces from the training data.

Re: Stable Diffusion is a big deal

#257

Post by SyzygistSix »

amelius wrote: post #160
Fordec wrote: post #11

Earlier quoted context omitted.

For every tool, a better output will always be created by people who specialize in a craft. Just as photoshop revolutionized photography, to this day you can tell the difference easily between and bad 'shops. In video games upon release everyone is bad and its an even playing field. But as people practice and gain experience they improve their usage, refine their approaches. You eventually see metas develop and best…

Yes, but art school just turned into a one-semester course.

No, it didn't but people who took only one semester of art classes may remain under the delusion that art is about making pretty pictures.

People who need illustration or graphics with no particular style can meet their needs with this tool but that is far from art. This replaces commercial illustration, not artists.

Re: Stable Diffusion is a big deal

#258

Post by thrown_22 »

sigmoid10 wrote: post #179
amelius wrote: post #160

Earlier quoted context omitted.

Yes, but art school just turned into a one-semester course.

Not yet, but I can definitely imagine a future where these tools get more capable and refined, to the point where all the shortcomings listed above will be overcome. Knowledge about cameras and scene composition are already encoded in the networks to some degree, it just needs to become more accessible. There's probably also a better way to seed new images than by starting with random noise, so we could get similar v…

You need to give some information about the scene to the network.

Camera settings is just a short hand to describe the field of view and depth of focus (at the very least). If you make that implicit you'd still need to give the network the steradians, focal length, circle of confusion, etc. etc. etc. that you want your image to use.

You'd need to understand everything in Hecht's Optics to tweak all the parameters of an AI generated image.

Re: Stable Diffusion is a big deal

#259

Post by wruza »

stdbrouw wrote: post #233
wruza wrote: post #205

Earlier quoted context omitted.

In my opinion it’s just a technical side of ethics, which has much less value than the main side: what it does to an author. If they suffer from these copies (morally, financially, spiritually, etc) without a way to offset it, then it is not ethical.

Well, but that's hardly a universal rule either. If we have a magic cure for bad eyesight, that'll be pretty terrible for optometrists, lensmakers and so on, but nobody would think it unethical that we're taking away their jobs.

To me it’s a different situation, because no previous hard, unique, “ownership” work of these people was used without consent to make this cure.

Think of this instead: a doctor collects a big volume of symptoms and analyses and creates a statistical way to cure people more easily. They publish many examples of their work without licensing anyone (legally and morally) to use it freely. Now some algorithm collects their data and many others data and transforms it into a better method. A doctor suffers from going out of business. Is that ethical? On one hand, the algorithm invented something new and easier to access. On the other, it basically stole parts of their and similar researches on a previously unthinkable scale. We humans copy ideas all the time and this is somewhat normal, but this enormous at-scale capability was never a thing.

Personally I don’t care for optometrists, uber drivers or designers. Nature will find a way. But when we talk about fundamental social contracts like property or accumulated knowledge protection, I think it is unethical to break them, regardless of technicalities. If it’s such a great advancement benefiting everyone, why can’t AI creators just ask permission for 2.3B of datapoints they used?

Re: Stable Diffusion is a big deal

#260

Post by sarchertech »

ekianjo wrote: post #239
spaniard89277 wrote: post #190

Earlier quoted context omitted.

I've tried too, and honestly I'll hire artists again (I did several times). It's easy to come up with nonsense, so in the end I think this is a tool in the hands of artists more than anything else. For threatening artists it should be much better. Maybe it's not the model, maybe it's the interface human-computer that fails, but in the end it's what we have. Maybe someone with a lot of time in their hands can iterate…

Thus is just the beginning. Imagine 5 years later what tools we will have.

That’s possible, but my gut feeling is that this turns out to the the same as the DARPA grand challenge moment for self driving cars.

It was an absolutely amazing accomplishment. I legitimately thought I could hold out long enough that my next car would be fully self driving. Truckers were an endangered species.

But here we are 20 years later, and we’re still almost there.

We’ve made amazing progress and I love the self driving features I do have on my car, but how many jobs have been replaced by self driving cars?

Post reply on HN