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

431–440 of 488 posts

Re: Stable Diffusion is a big deal

#431

Post by dymk »

TchoBeer wrote: post #421
dymk wrote: post #335

Earlier quoted context omitted.

If someone crashes into your house, do you sue the car or the person?

What if AI is controlling the car? Do you sue the passenger?

Yes, I don't get a free pass for harming somebody with a tool I'm operating just because the tool is "automatic" in some way. If the person operating the car is the passenger, then you sue them. If it's Google operating their Waymo fleet, you sue them.

Car manufacturers in the future may offer to take on the liability themselves for autopilot mistakes, but that's not yet the deal offered.

At the end of the day, there's no legal magic or loopholes. Somebody is ultimately the operator of the vehicle, even if their hands aren't on the steering wheel.

Re: Stable Diffusion is a big deal

#432

Post by ipaddr »

Morgawr wrote: post #54
ipaddr wrote: post #36

Earlier quoted context omitted.

Images generated have a watermark that can be used to filter.

The images I generate don't have a watermark. Same applies for a lot of other AI generated images. The watermark is the most trivial part to remove and is definitely not a defining feature of what makes an image AI generated.

If you use one of the ui or don't remove the watermark (99% of people won't) then search engines will be able to filter those stable diffusion images.

Re: Stable Diffusion is a big deal

#433

Post by ipaddr »

simonw wrote: post #45
ipaddr wrote: post #35

Earlier quoted context omitted.

The watermark.

What watermark? The model is open source and free for people to make their own modifications. I don't see how any watermark can survive in those conditions.

All images from stable diffusion have an invisible watermark unless you modify the source to remove it.

Re: Stable Diffusion is a big deal

#434

Post by scarmig »

kixiQu wrote: post #356
lelag wrote: post #342

Earlier quoted context omitted.

Indeed. A difference I see between the GPT-like language models and those image generation models is that it feels like the language models actually hold full copies of a lot of sample of the training dataset (hence their ability to recite existing content), whereas the image generation models clearly do not: Stable Diffusion is 4Gb something, yet it can draw anything. That's the amazing part: the training dataset co…

Haven't we seen a lot of examples of these things vomiting up Starry Night on demand? Isn't that a clear case of memorization?

Midjourney did this one:

https://i0.wp.com/www.technollama.co.uk/wp-content/uploads/2...

Not really a copy. I'm sure you could cajole it into something very similar to the actual piece, but at that point it's more the model + the extensive prompt instead of the model itself.

Re: Stable Diffusion is a big deal

#435

Post by tstrimple »

DethNinja wrote: post #422
tstrimple wrote: post #418

Earlier quoted context omitted.

I'm using 512x768 as the default, but a quick test shows only a marginal difference in speed between the two. I'll have to give Windows a try to see if it's the driver holding me back. Do you have any tips or resources for up-scaling the image after?

Currently this library can generate multiple images and upscale them through RealESRGAN: https://github.com/hlky/stable-diffusion If you are not using this library already, give it a shot. Also, I'm using Nvidia Studio drivers though I'm not sure if that would make a difference.

I've been using the main fork. This even has GFPGAN built in! Looks very useful thanks.

Re: Stable Diffusion is a big deal

#436

Post by tstrimple »

bitshiftfaced wrote: post #423
tstrimple wrote: post #406

Earlier quoted context omitted.

What kind of GPU are you running this on? My 3080 seems to take about 30 seconds per image with 50 passes. I'm wondering if I'm missing out on some optimizations. Could just be the quality of Linux NVidia drivers.

I'd recommend trying a different fork. Perhaps you're using the the official one. I believe that one still "ramps up the system" on every image generation. Other repos do the ramp up only once.

Yeah, this might be the problem. I was on the main fork, but going to try switching over to this: https://github.com/hlky/stable-diffusion

Re: Stable Diffusion is a big deal

#437

Post by amelius »

idontpost wrote: post #391
Workaccount2 wrote: post #312

Earlier quoted context omitted.

This is first generation tech that is dropping jaws compared to just 2 years ago. A career typically requires 40 years of job security.

> A career typically requires 40 years of job security. 40 years ago was 1980. There's a vanishingly small number of fields that have had continuous job security from 1980-2020 (even ignoring Covid). You might as well say that careers are over for everyone, and have been for a while.

Doctors and lawyers are doing great, still.

Re: Stable Diffusion is a big deal

#438

Post by planb »

thegrimmest wrote: post #424
odessacubbage wrote: post #368

Earlier quoted context omitted.

you also didn't flip millions of illustrations into a commercial product without paying the creators or rights holders to license them. this is theft plain and simple.

I think the point is that he did, and that we all do , when we train our meat neural networks (on copyrighted material) in a way that makes us better at providing a commercial service (our job). If a human has a right to view and learn from a copyrighted image on the internet, why shouldn't an AI?

Yes, this is exactly my point. And I’d go even one step further: creating real artificial intelligence cannot be possible without learning from the real world - which means learning also from copyrighted material. So a law that prevents this (even by just requiring to pay a minuscule fee) would effectively ensure that such a technology will be developed in a country with different laws.

Re: Stable Diffusion is a big deal

#439

Post by planb »

dymk wrote: post #431
TchoBeer wrote: post #421

Earlier quoted context omitted.

What if AI is controlling the car? Do you sue the passenger?

Yes, I don't get a free pass for harming somebody with a tool I'm operating just because the tool is "automatic" in some way. If the person operating the car is the passenger, then you sue them. If it's Google operating their Waymo fleet, you sue them. Car manufacturers in the future may offer to take on the liability themselves for autopilot mistakes, but that's not yet the deal offered. At the end of the day, there…

Absolutely. If my hand brake breaks while I’m not in my car, the neighbor parking behind me will still sue me, not the car manufacturer.

Re: Stable Diffusion is a big deal

#440

Post by TulliusCicero »

selestify wrote: post #226
TulliusCicero wrote: post #186

Earlier quoted context omitted.

That depends on how easy it is to use the new tools. Getting higher level languages and various libraries/frameworks didn't make coders obsolete, because there was still tons of work to do at a "higher level" that required algorithmic/engineering-style thinking. Yes, previously difficult tasks got a lot easier, but that just meant that even more ambitious projects could be tackled, and that was useful. In contrast, i…

Is the same not true for programming? There's a lot of non-programmers that create programs in Excel, or use no code solutions. A relative newbie can still get a Wordpress website up and running, you only need professional SWEs for the heavy duty stuff.

Agreed, but the rub is that there was a ton of demand for "heavy duty stuff"; indeed, the improvement in code tooling actually increased that demand, because software programs became so much more powerful.

It's not clear that you'll see something similar here, that easier-to-produce high complexity art will greatly increase demand for said art.

Post reply on HN