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

Using Stable Diffusion's img2img on some old Sierra titles

sciprogramming.com

Post reply on HN
Print view View original

41–50 of 70 posts

Re: Using Stable Diffusion's img2img on some old Sierra titles

#41

Post by cma »

tjr225 wrote: post #38
ziml77 wrote: post #31

Earlier quoted context omitted.

I think the weirdness is most obvious in the cliff overlooking a beach and in the woman's impossible object hand. All of the output here is cool and impressive for sure, but good art it is not.

The inaccuracy or weirdness of the resulting images has no bearing on how good or bad it is as art. Art has nothing to do with that. I would argue this is a shitty tech demo more than anything else. I do not mean to discount the creator as it’s cool regardless, it just doesn’t really have anything to do with art. They’re literally just running some old computer images through a technology. That’s it. There will proba…

Very similar stuff was said about the camera. Vermeer even essentially used one to win the realism competition in painting (of course later becoming Ayn Rand's favorite painter since he was the most "objective").

Re: Using Stable Diffusion's img2img on some old Sierra titles

#42

Post by crazygringo »

avidphantasm wrote: post #18
MattRix wrote: post #15

Earlier quoted context omitted.

Why do people always ask “why?” on these things? Not everything has to be “useful” or have commercial value. The original games aren’t going anywhere. People just enjoy doing this kind of stuff, pushing the technology to make interesting things.

I’m not criticizing it because it’s not useful. I’m criticizing it because it is not art, and it is anti-humanist.

Who said it's art? It's new and cool and fascinating. That's all.

Contrary to what you wrote, these AI-generated ones aren't replacing the originals. The originals are still there.

Sorry, but I think you're reading way too much into this... it's a fun toy, it's not replacing game art and it's not an "abomination". It's cool and new. That's all.

Re: Using Stable Diffusion's img2img on some old Sierra titles

#43

Post by Bakary »

z9znz wrote: post #30

The before and after are great. There is an art to conveying a feeling with limited resources. That's what makes early computer game images (the good ones... because there were plenty of bad ones) so special. The same could be said even more strongly for words. I'm not a writer and I don't remember everything, but someone famous once said something about eliminating everything non-essential from writing to make it be…

Why stop at the art? We'll have unique storylines and characters as well. Possibly even unique gameplay, although at that point we would pretty much be close to some sort of AGI anyway.

Re: Using Stable Diffusion's img2img on some old Sierra titles

#44

Post by gibbonsrcool »

vslira wrote: post #2

Is there any service offering user-friendly access to open source DL models? Paid is fine.

If you have over 4MB VRAM you can run it locally. I've been experimenting recently and find that even with 10MB VRAM I can only get 256x256 resolution images. I have a Dockerfile I can share that packages up the install process and removes censorship if anyone is interested. I find the censoring is extremely conservative.

Re: Using Stable Diffusion's img2img on some old Sierra titles

#45

Post by tablespoon »

fezfight wrote: post #27
davesque wrote: post #7

Going to have to be the naysayer here. First, I'll say the simple fact that Stable Diffusion produces anything coherent is incredible. I'm blown away by the tech. However, my honest opinion of the results showcased in this article is not positive. Many of the result images contain bizarre distortions or dreamlike artifacts that severely disrupt the flow of the image. Especially the first one. It's clear that there's…

Compared to low res Sierra graphics, what's been shown here is nothing short of astounding, if you ask me. Look at the skull! It's a masterpiece.

> Look at the skull! It's a masterpiece.

There was also a time when I (unironically) classified McDonald's food a delicacy (at some point when I was younger than ten).

Re: Using Stable Diffusion's img2img on some old Sierra titles

#46

Post by nstart »

jamiek88 wrote: post #9
davesque wrote: post #7

Going to have to be the naysayer here. First, I'll say the simple fact that Stable Diffusion produces anything coherent is incredible. I'm blown away by the tech. However, my honest opinion of the results showcased in this article is not positive. Many of the result images contain bizarre distortions or dreamlike artifacts that severely disrupt the flow of the image. Especially the first one. It's clear that there's…

Weird until I read your comment I was blown away. Then I had another proper look at the first image and in many ways I had to turn off my brains amazing ‘upscaling’ ability. My brain had upscaled that human like blob to a woman spinning around with a sword so her hair covered her face. Looking closely. None of that is there really, just a suggestion of it. And that is enough. The more I learn about vision and sight t…

Artists have for the longest time used our brains ability to upscale. Many paintings, even ones that seem super detailed like those by James Gurney in his Dinotopia series, will have blobs in the background. Our brain will recognize based on silhouette and shape an extraordinary amount of detail that isn’t actually there. Detail such as the type of clothing and the action of a person. But if you look closer it’s a rectangular blob with a triangular blob within it to indicate clothes.

The difference between AI art and actual human art is the level of intention one can detect in it. When I look at human art I absolutely marvel at the cleverness of the artist to convey something that still looks like what I was imagining even when I look closer at it.

With AI art I look closer and realize that the blob presents more confusion the closer I look at it.

I’ve been telling anyone who will listen that AI art isn’t stealing much lunch when it comes to professional art. But it may very well be a powerful tool to artists to speed up their workflows and artists who refuse to use the tool stand a chance at being left behind the same way some artists got left behind in the illustration industry world once digital tools showed up.

Re: Using Stable Diffusion's img2img on some old Sierra titles

#47

Post by bioemerl »

jamiek88 wrote: post #9
davesque wrote: post #7

Going to have to be the naysayer here. First, I'll say the simple fact that Stable Diffusion produces anything coherent is incredible. I'm blown away by the tech. However, my honest opinion of the results showcased in this article is not positive. Many of the result images contain bizarre distortions or dreamlike artifacts that severely disrupt the flow of the image. Especially the first one. It's clear that there's…

Weird until I read your comment I was blown away. Then I had another proper look at the first image and in many ways I had to turn off my brains amazing ‘upscaling’ ability. My brain had upscaled that human like blob to a woman spinning around with a sword so her hair covered her face. Looking closely. None of that is there really, just a suggestion of it. And that is enough. The more I learn about vision and sight t…

What people are doing right now is taking these wonky outputs and putting them back into the system after making some crude edits to it.

As you repeat that cycle things are going to get better, but it does require human labor at that point.

It's not a complete magic tool, but once you put a little bit of fixing effort in you can get a long way with very little.

Re: Using Stable Diffusion's img2img on some old Sierra titles

#48

Post by davesque »

AaronFriel wrote: post #16
davesque wrote: post #7

Going to have to be the naysayer here. First, I'll say the simple fact that Stable Diffusion produces anything coherent is incredible. I'm blown away by the tech. However, my honest opinion of the results showcased in this article is not positive. Many of the result images contain bizarre distortions or dreamlike artifacts that severely disrupt the flow of the image. Especially the first one. It's clear that there's…

Are you naysaying Stable Diffusion or the idea of generative models for art in general? It's hard to look at what's happened in the last few months and not think of it as akin to the invention of the steam engine, but for art. It's not perfect, as early machines had many flaws, were wildly inefficient, produced irregular output. But the innovation that followed created the industrial revolution.

> or the idea of generative models for art in general?

I'm not sure how you could read my comment and possibly come to this conclusion.

Re: Using Stable Diffusion's img2img on some old Sierra titles

#49

Post by arthurcolle »

gibbonsrcool wrote: post #44
vslira wrote: post #2

Is there any service offering user-friendly access to open source DL models? Paid is fine.

If you have over 4MB VRAM you can run it locally. I've been experimenting recently and find that even with 10MB VRAM I can only get 256x256 resolution images. I have a Dockerfile I can share that packages up the install process and removes censorship if anyone is interested. I find the censoring is extremely conservative.

Please do share. I did the same and have it but having trouble deploying it to a prod GPU server, still figuring it out.

Re: Using Stable Diffusion's img2img on some old Sierra titles

#50

Post by cercatrova »

skybrian wrote: post #34
Silverback_VII wrote: post #14

Earlier quoted context omitted.

yes, it's per image but for some reason I easily rushed through the 1000 generations in one day.

I did too, but I bought another 1000 generations and haven't run out yet. Many days I don't use it at all. I like it better than a monthly subscription like MidJourney.

Just use Google Colab, it's free indefinitely because Google is paying for it.
Post reply on HN