Stable Diffusion is a big deal • Hacker News Board

Re: Stable Diffusion is a big deal

#211

Post by nvln » Wed, Aug 31, 2022, 9:59 AM UTC

My son is an artist and we have conversations around AI generated art almost every week these days. Our general consensus has been:

1. Artists that use SD etc., will save a lot of time and have a big edge. Example: Generating references instead of searching for them, Generate random inspiration(s) for abstract ideas, Variations of a particular idea to compare and contrast.

2. Non artists who have powerful imagination but lack the artistic skill will stand to benefit as well. For example, I look forward to generating visual companions for my poetry, essays or fiction. I have the choice of commissioning an artist if my piece is successful.

3. Artists that are purists and masters of their craft with a very unique style will continue to thrive in their niche. These people have to lobby and come up with a license that explicitly prohibits feeding their images for learning. But a skilled artist can replicate the style and feed it. There is an ethical line here that may require creative tweaks to existing laws to protect people that fall in this category.

Super exciting tech and I can't wait for this to move over to 3D models. I know 2D image to 3D model is already pretty close to being real(nvidia ominiverse, nerf etc) and SD can be the starting point of that pipeline. Prompt -> 3D world that can be decomposed to meshes, textures and USD will change the game development landscape quite a bit.

Re: Stable Diffusion is a big deal

#212

Post by nlh » Wed, Aug 31, 2022, 10:04 AM UTC

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…

I’m going to partially agree and partially disagree with you.

Agree that Stable Diffusion seems a bit over hyped. It is very cool! I am definitely going to play with it. I’m not sure really how much it’s going to change the world - there will be some neat apps and it may put a certain class of concept artist out of business (note I am not saying it will eliminate them - there will always always always be human artists creating wonderful art. But as a business model / way of earning a living, that might change).

I am going to disagree with you on GitHub Copilot. It has radically changed the way I do software engineering on my personal projects and it has increased the efficiency of the engineers at my company by at least 10%, maybe more. You should check it out. It eliminates an entire class of puzzle solving, and the least efficient kind (“how do I drop a column by condition in a pandas dataframe again?”). Simple answer via Google, but maybe 3-5 minutes of reading. 10 seconds via Copilot.

Re: Stable Diffusion is a big deal

#213

Post by ManuelKiessling » Wed, Aug 31, 2022, 10:04 AM UTC

I've started an open source implementation of a Discord bot that turns your prompt into images using Stable Diffusion [0].

It would be great if this would turn into a community-driven chat-based text2image offering - which is certainly challenging as the required GPU-powered server instances aren't exactly cheap. Maybe this could grow into an open network where people provide the GPU power of their personal PCs? Or we find a way to cover hosting expenses through a credit system or some kind of sponsoring?

Feel welcome to join the community [1]. You can easily test drive the current Bot implementation on this server.

[0] https://github.com/manuelkiessling/stable-diffusion-discord-... [1] https://discord.gg/nsfeutx35z

Re: Stable Diffusion is a big deal

#214

Post by Kiro » Wed, Aug 31, 2022, 10:07 AM UTC

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…

What have you been using it for?

Re: Stable Diffusion is a big deal

#215

Post by ccozan » Wed, Aug 31, 2022, 10:12 AM UTC

I am repeating myself from another SD post here in HN [1]

"I mean, what stops us a life-imitates-art system where we have speech2img like in the Westworld ( the narative creating scenes ) ? I guess, I hope someone reads this and will pick up this. Maybe coupled with a VR set?"

Now I really hope on this :) Imagine giving this power to kids ( they cannot write but talk! )

[1] https://news.ycombinator.com/item?id=32647792

Re: Stable Diffusion is a big deal

#216

Post by egeozcan » Wed, Aug 31, 2022, 10:16 AM UTC

Now there are so many very successful description -> image research and tooling, I find it surprising that most of the image -> description/labels solutions are closed source and/or paid.

BTW generating images using Anaconda on Windows using my RTX 2080 Super was very easy (followed the instructions here: https://rentry.org/SDInstallGuide ). The only thing was, because it has just (!) 8GB of ram, I had to use --H 256 --W 256 to limit the output image to its minimum size, which barely fit into its memory.

Re: Stable Diffusion is a big deal

#217

Post by selestify » Wed, Aug 31, 2022, 10:17 AM UTC

noobermin wrote: post #195

solumunus wrote: post #102

Earlier quoted context omitted.

I fail to see how the brain isn't doing a very similar thing?

Is my brain's floating point calculations subject to round off error? In the (actual) neurons, is there a representation of real numbers? Where are the numbers in the brain stored? I feel like people who assert this neither understand what neural networks are and how brains work.

> Is my brain's floating point calculations subject to round off error? > > In the (actual) neurons, is there a representation of real numbers? Where are the numbers in the brain stored?

None of these details are relevant to the bigger picture similarities of non-hardcoded learning from training data. None of these details change the ethics of what's being discussed here.

> I feel like people who assert this neither understand what neural networks are and how brains work.

That's just your bias showing.

Re: Stable Diffusion is a big deal

#218

Post by Somatochlora » Wed, Aug 31, 2022, 10:17 AM UTC

batmenace wrote: post #206

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.

I would assume that the point is that you had to pay - in one way or another - for access to this material, which is the intention when it is published and shared. You bought your books, your toys, your TV shows (through ads). SD, by scraping the web, gets all the benefits of prior work without necessarily having adequately compensated the work's creators.

Besides the point that these are freely viewable images anyways, this is also clearly not people's real objection. If you were making a novel-writing ai, it would not placate anybody if you bought a single copy of each book before incorporating it into the model.

Re: Stable Diffusion is a big deal

#219

Post by andybak » Wed, Aug 31, 2022, 10:18 AM UTC

barrenko wrote: post #196

Still looks like crap, in the same way that a CGI sequence in a movie jumps out to someone who is a gamer. (In my very modest modest opinion).

When witnessing a huge leap forward in human technological achievement, doesn't it feel a little strange to respond with "looks like crap"?

Even on your own terms, you're engaging in hyperbole. I've looked though thousands of these images and the harshest I can come up with is "assuming the result isn't obviously broken (i.e. some generated anatomy) then the results are sometimes a little bit uncanny or 'off'")

Re: Stable Diffusion is a big deal

#220

Post by andybak » Wed, Aug 31, 2022, 10:21 AM UTC

paulpauper wrote: post #30

Why do Stable Diffusion images all sorta look the same. They look either deformed or like pastels . I tried it a few times on some free sites, and it was not accurate. I would put in 5 keywords and maybe it would get 2 of them right. These are common search terms.

> These are common search terms.

It's not a search engine. It takes a while to get a feel for what prompts work and how to phrase things. I suggest browsing https://lexica.art/ and using some prompts from there as a template.