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

No elephants: Breakthroughs in image generation

oneusefulthing.org

Post reply on HN
Print view View original

331–340 of 373 posts

Re: No elephants: Breakthroughs in image generation

#331

Post by shaky-carrousel »

exodust wrote: post #319
schwartzworld wrote: post #257

Earlier quoted context omitted.

AI is mediocre at a lot of things, but it makes a damn fine upgrade from stock photos. This is the art that’s going to get replaced by this tech, shitty low effort stuff. Images where you just need a picture of X because people are expecting a picture. It’s the same with code. I don’t think software engineers will really be replaced, but small web dev agencies have a good reason to be nervous. Why would you pay someo…

3-5 prompts doesn't get you a professional restaurant website. The key word is professional . A good restaurant website begins with taking good photos of the premises and the food. AI won't come around to your business and take professional photos. There's a lot of bits and pieces to a website for bookings, content management, menu updates, etc. HTML templates and themes have been around for a long time. AI can basic…

No, but you can do the photos and have the AI fix them or make suggestions about them.

Re: No elephants: Breakthroughs in image generation

#332

Post by ZoomZoomZoom »

gcanyon wrote: post #34

It's interesting to hear people side with the artists when in previous discussions on this forum I've gotten significant approval/agreement arguing that copyright is far too long. As I've argued in the past, I think copyright should last maybe five years: in this modern era, monetizing your work doesn't (usually) have to take more than a short time. I'd happily concede to some sort of renewal process to extend that p…

> I think copyright should last maybe five years

If we imagine for a moment that "copyright" is something that works in the interests of a creator, than 5 years is nothing.

A painting can sit fifteen years before it gets to an exhibition with sufficient turn-over and media coverage to draw attention to.

A music album can be released with a shitty label and no support, years later be taken by a more competent one and start selling.

We're living in a world where worth art is constantly flying under the radars so limiting potential even more isn't helpful.

Re: No elephants: Breakthroughs in image generation

#333

Post by sfn42 »

justinator wrote: post #142

But the annotations are still wrong, https://substackcdn.com/image/fetch/w_1456,c_limit,f_webp,q_... (nice URL btw) The room, the door, the ceiling are all of a scale to fit many sizes of elephants.

The lines don't really make sense either, like the one above the sofa should probably go along the corner between the floor and wall.

Re: No elephants: Breakthroughs in image generation

#334

Post by sfn42 »

Der_Einzige wrote: post #133

The #1 reason that this technology won't proliferate more quickly is that humans are a bunch of COOMERS! We get Stable Diffusion V1.5 and SDXL and what does the community go do with it? Lmao see civit.ai and it's literal hundreds of thousands of NSFW loras. The most popular model today on that website is the NSFW anime version of SDXL, called "Pony Diffusion" (I'm literally not making this up. A bunch of Bronies made…

I don't see the problem. Most people don't know this stuff exists. You pretty much have to look for it to find it. So what if someone wants to create rule 34 stuff? Let them, who cares? It doesn't hurt anyone. There was already a large market for artists drawing weird fetish stuff, AI doesn't really change anything.

Re: No elephants: Breakthroughs in image generation

#335

Post by fc417fc802 »

dudeofea wrote: post #316
fc417fc802 wrote: post #306

Earlier quoted context omitted.

> The people who built this technology needed to use hundreds of millions of images without permission. It remains unclear if they needed permission in the first place. Aside from Meta's stunt with torrents I'm not aware of any legal precedent forbidding me to (internally) do as I please with public content that I scrape. > They regularly speak explicitly about all the jobs they plan to destroy. A fully legal endeavo…

"Data Laundering": Commercial entities fund (either with money or compute tokens) academic entities which, in turn, create AI models which the commercial entities sell. https://waxy.org/2022/09/ai-data-laundering-how-academic-and...

Again, it's unclear how exactly that's against the law. Provided that the data was obtained legally, of course.

Most of the larger commercial entities seem to be doing the work themselves and being quite upfront about the entire thing.

Re: No elephants: Breakthroughs in image generation

#337

Post by fc417fc802 »

__loam wrote: post #312
fc417fc802 wrote: post #306

Earlier quoted context omitted.

> The people who built this technology needed to use hundreds of millions of images without permission. It remains unclear if they needed permission in the first place. Aside from Meta's stunt with torrents I'm not aware of any legal precedent forbidding me to (internally) do as I please with public content that I scrape. > They regularly speak explicitly about all the jobs they plan to destroy. A fully legal endeavo…

> I'm not aware of any legal precedent forbidding me to (internally) do as I please with public content that I scrape. Because all the litigation is currently ongoing. > A fully legal endeavor that is very strongly rewarded by the market. Yes let's sacrifice all production of cultural artifacts for the market. This is honestly another thing that's being litigated. So far these companies have lost a lot of money on ma…

Precisely. So when you say they used the images without permission, you are knowingly making a false implication - that it was known to them that they needed permission and that they intentionally disregarded that fact. In reality that has yet to be legally established.

Who said anything about sacrificing production? The entire point of the tooling is to reduce the production cost to as near zero as possible. If you didn't expect it to work then I doubt you would be so bent out of shape over it.

I find your stance quite perplexing. The tech can't be un-invented. It's very much Pandora's box. Whatever consequences that has for the market, all we can do is wait and see.

Worst case scenario (for the AI purveyors) is a clear legal determination that the current training data situation isn't legal. I seriously doubt that would set them back by more than a couple of years.

Re: No elephants: Breakthroughs in image generation

#338

Post by lupusreal »

exodust wrote: post #319
schwartzworld wrote: post #257

Earlier quoted context omitted.

AI is mediocre at a lot of things, but it makes a damn fine upgrade from stock photos. This is the art that’s going to get replaced by this tech, shitty low effort stuff. Images where you just need a picture of X because people are expecting a picture. It’s the same with code. I don’t think software engineers will really be replaced, but small web dev agencies have a good reason to be nervous. Why would you pay someo…

3-5 prompts doesn't get you a professional restaurant website. The key word is professional . A good restaurant website begins with taking good photos of the premises and the food. AI won't come around to your business and take professional photos. There's a lot of bits and pieces to a website for bookings, content management, menu updates, etc. HTML templates and themes have been around for a long time. AI can basic…

Most restaurant websites don't have photos of the food. Certainly not professional ones. Pan around randomly on google maps then zoom in and find the nearest strip mall, then go down the line checking the websites of restaurants there. Most are generic crap and you'll be lucky if the menu online is even complete. If they have food photos it's probably smartphone pictures taken by the owner's kid.

I do this a lot, far more than I actually go to restaurants, because I like adding small business details to OSM. There are a few that have their shit together but the overwhelming majority do not.

Re: No elephants: Breakthroughs in image generation

#339

Post by lupusreal »

__loam wrote: post #170
Gasp0de wrote: post #121

Earlier quoted context omitted.

I love using LLMs to generate pictures. I'd call myself rather creative, but absolutely useless in any artistic craft. Now, I can just describe any image I can imagine and get 90% accurate results, which is good enough for the presentations I hold, online pet projects (created a squirrel-themed online math-learning game for which I previously would have needed a designer to create squirrel highschool themed imagery)…

If you use this technology, you're actively harming creative labor.

I was never going to hire a professional artist to sketch shit up for me. I have replaced MS Paint, not harmed "creative labor".

Re: No elephants: Breakthroughs in image generation

#340

Post by Zr01 »

I'm more interested in the technical details than the publicity. Pretty much anyone these days can learn what a diffusion model is, how they're implemented, what the control flow is. What about this new multimodal LLM? They have no problems with text, they generate images using tokens, but how exactly? There's no open-source implementations that I know of, and I'm struggling to find details.
Post reply on HN