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

401–410 of 488 posts

Re: Stable Diffusion is a big deal

#401

Post by DethNinja »

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…

It takes 3 seconds to generate 1 image with my GPU.

I can find a good prompt within 30 minutes to 1 hour.

My GPU can generate 100 images in 5 minutes.

Out of those 100 images, 10 is very close to what I exactly meant at professional concept artist level.

So, in this case Stable Diffusion only working 10% of the time is fine.

Future is already here, I’m already incorporating stable diffusion generated images to my professional work.

Re: Stable Diffusion is a big deal

#402

Post by belter »

prolyxis wrote: post #347
belter wrote: post #293

Earlier quoted context omitted.

You are looking at code like a production line. It's a semantic construction. Code 0.1% wrong, sends you to the Sun instead of the Moon, debits your account instead of crediting...

What's your metric for the "percentage the code is wrong"? Is it how many lines of code were wrong, or how many test cases the code fails? Presumably if AI-generated code passes every test case, but would fail on edge cases that some human programmer(s) did not anticipate in their suite of tests, the humans potentially might have made similar coding mistakes as the AI if they had had to personally write the code.

For AI to generate code for a test case we would need AGI.

Re: Stable Diffusion is a big deal

#403

Post by napier »

_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…

This is open source. Github Copilot is not.

Re: Stable Diffusion is a big deal

#404

Post by mullingitover »

stableskeptic wrote: post #360
mullingitover wrote: post #72

Earlier quoted context omitted.

Linux started out pretty mediocre compared to closed source Unix, and even today there are operating systems that are better, but being freely available made it a game changer. I don’t foresee Dalle2 being open any time soon.

I'm struggling with the connection you're making between Linux and Stable Diffusion. Linus didn't invent the core concepts of Unix. He copied the (arguably) good parts from an (arguably) closed ecosystem. His big innovation was leveraging the internet to create a new kind of community not really seen before it. The Linux bazaar gave smart developers excluded from the Bell Labs / BSD cathedral a place to be productive…

> I'm struggling with the connection you're making between Linux and Stable Diffusion.

Stable Diffusion is open source, and was being compared unfavorably to a technically superior but closed source system.

Re: Stable Diffusion is a big deal

#405

Post by homarp »

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…

>we need the ability to persist a design

isn't that Textual Inversion (https://textual-inversion.github.io/ ) ?

It's more or less implemented in some forks (e.g. https://github.com/lstein/stable-diffusion#personalizing-tex... or https://github.com/hlky/sd-enable-textual-inversion (discussed previously ( https://news.ycombinator.com/item?id=32643564 ) )

Re: Stable Diffusion is a big deal

#406

Post by tstrimple »

DethNinja wrote: post #401
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…

It takes 3 seconds to generate 1 image with my GPU. I can find a good prompt within 30 minutes to 1 hour. My GPU can generate 100 images in 5 minutes. Out of those 100 images, 10 is very close to what I exactly meant at professional concept artist level. So, in this case Stable Diffusion only working 10% of the time is fine. Future is already here, I’m already incorporating stable diffusion generated images to my pro…

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.

Re: Stable Diffusion is a big deal

#407

Post by chucksmash »

galdosdi wrote: post #400
chucksmash wrote: post #364

Earlier quoted context omitted.

You're nitpicking one of the given examples without engaging the user on the point they were trying to make. To be nuanced, maybe they might have said, "cobblers are less in demand now that many people have moved from owning fewer pairs of shoes they make last through repair to owning more pairs of shoes that they tend to get rid of when they are worn out due to changes in construction materials used in production,"…

It's a nitpick, but a little bigger than that. It's as bad as including "bus driver" in the list. Cobblers just shouldn't be included in the category at all. Cobblers are in just as much demand in most of the world as they always have been. They only fell out of demand in car-dependent areas, which is a small minority of the world population (but a vast majority of the HN commenting population since most of the USA o…

> I don't know if it has anything to do with construction but doubt it. If you actually walk everywhere shoes don't last very long these days, especially shoes under $100.

By construction, I mean the material and design of shoes people tend to wear. I can't say I've ever met someone who takes sneakers or running shoes to a cobbler and these shoes are more common nowadays.

Re: Stable Diffusion is a big deal

#408

Post by Filligree »

michaelchisari wrote: post #9
canjobear wrote: post #7

Earlier quoted context omitted.

> Second, we need the ability to persist a design. I See yesterday's Stable Diffusion article: https://news.ycombinator.com/item?id=32643564

I haven't had a chance to try it yet, I'm optimistic but skeptical that it can have true persistence like I'm referring to. Especially since it requires training your own model which requires multiple images of the same asset/character/object/etc.

Textual inversion seems like it'll eventually cover this space. But bootstrapping is going to be harder than necessary, yes.

Re: Stable Diffusion is a big deal

#409

Post by ahmedbaracat »

sinenomine wrote: post #357
_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…

I don't think AI trained on my creative output (not just code, mind you) is a problem per se . What is a problem, in my opinion, is the tendency of large corporations and small circles on top of these to monopolize access to these models, and if some of the functionality gets available to the public, it's going through a very paternalistic, corporate, puritan censorship pipeline. If you train artificial intelligence…

This reminds me of Lex Fridman talking with Jaron Lanier about Data Dignity https://youtu.be/Fx0G6DHMfXM?t=2609

Re: Stable Diffusion is a big deal

#410

Post by ryandrake »

ChrisMarshallNY wrote: post #324
dsign wrote: post #64

Earlier quoted context omitted.

I agree, this would be an incredible tool. I can see how some of the outputs may help me improve a piece I'm working on, even if I would never use the model's output for my final product.

It looks like a decent way to produce concept sketches. Currently, working with creatives (including programmers) is a very iterative process for non-creatives. "I want this." "No, I meant this." "Can we try making that line longer?" "Eh, I'm not feeling it. Why don't we try brighter colors?" "Ugh. That looks obnoxious. Can we tone down the red?" etc. If the non-creative can use this system to reduce that iteration,…

Does SD have the ability to take this kind of micromanagement input though? From what I've seen, it works off of a general descriptive prompt. Will adding a very specific "but put the duck 5 pixels to the left" or a very vague "give it more pop" to the prompt actually have the intended effect?
Post reply on HN