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

Try Stable Diffusion's Img2Img Mode

huggingface.co

Post reply on HN
Print view View original

141–150 of 166 posts

Re: Try Stable Diffusion's Img2Img Mode

#141

Post by lkois »

stavros wrote: post #105
lkois wrote: post #52

Earlier quoted context omitted.

Nah that's normal. It's why GPUs are the usual thing for AI. Any crap, old, weak gpu with 4gb memory would run circles around a cpu It's often easier to actually get models to run on CPU, due to simpler install configs and more available memory. Just painful to get a result out of it. Which might help keep the install simple, because it's not even worth optimizing

The GP is asking why it's only 20x slower, rather than more slow.

Ha yep I see that now, on a reread. My bad

Re: Try Stable Diffusion's Img2Img Mode

#143

Post by fragmede »

ilaksh wrote: post #110
comex wrote: post #13

If you have a GPU with >4GB of VRAM and you want to run this locally, here's a fork of the Stable Diffusion repo with a convenient web UI: https://github.com/hlky/stable-diffusion It supports both txt2img and img2img. (Not affiliated.) Edit: Incidentally, I tried running it on a CPU. It is possible, but it took 3 minutes instead of 10 seconds to produce an image. It also required me to hack up the script in a really…

Amazing. Anyone know of a fork where this is hosted on a cloud GPU? Or any existing hosting of this?

There’s a Colab implementation (Google hosted GPU) linked to from https://www.youtube.com/watch?v=Xur1JeRjjOI

Re: Try Stable Diffusion's Img2Img Mode

#144

Post by smcameron »

frozencell wrote: post #2

What are your best prompts so far?

I typed in "space nerds in space fighting the zarkons", And it gave me a picture that looked similar to NASA's astronaut group photos, with 3 astronauts, 2 human, and one 3-eyed ninja-turtle looking alien in a spacesuit. Behind them was a space scene with a huge, earthlike planet. The faces of the two human astronauts looked eerily similar to the face of a friend of mine, a friend who happens to be the 2nd most prolific contributer of code to my open source game called "Space Nerds in Space". One of the human astronauts was bald, the other had a mullet, and the long portion of their hair was not entirely unlike my own hair.

Re: Try Stable Diffusion's Img2Img Mode

#145

Post by deanCommie »

epups wrote: post #99
deanCommie wrote: post #20

Earlier quoted context omitted.

I'm not gonna lie, I'm quite disappointed by the wilful obstinance on HN whenever there is a discussion of anti-NSFW or anti-racism filters that are present on today's artificial intelligence research. First, the filters are only on the interface. The research is public. That's why there has been an explosion of new implementations. You are welcome to run the code yourself and make the horniest model you like. But mo…

I can generate many violent prompts right now, and the model knows the face of hundreds of celebrities (you would think they would garble those to prevent PR disasters). It knows what the 9/11 attack was, it knows about nuclear bombs, horror, gore. Yet somehow it doesn't know what a blowjob is. This is 100% about narrow-minded, prudish conservative values. And yes, most of them come from the US right now, because the…

Tell me, earnestly, what progress is being blocked by your inability to use AI to generate an image of a blowjob.

I don't mean this in a moralistic sense. I have no qualms about images of blowjobs. I have no doubt that the porn industry has already deployed these models and is experimenting with this without filters. As a monetizable way to reduce the human costs, it's an entirely logical step, and adult performers should be as nervous as artists, and talking to lawyers.

But PROGRESS???

Re: Try Stable Diffusion's Img2Img Mode

#147

Post by outdoorblake »

ilaksh wrote: post #110
comex wrote: post #13

If you have a GPU with >4GB of VRAM and you want to run this locally, here's a fork of the Stable Diffusion repo with a convenient web UI: https://github.com/hlky/stable-diffusion It supports both txt2img and img2img. (Not affiliated.) Edit: Incidentally, I tried running it on a CPU. It is possible, but it took 3 minutes instead of 10 seconds to produce an image. It also required me to hack up the script in a really…

Amazing. Anyone know of a fork where this is hosted on a cloud GPU? Or any existing hosting of this?

here is a serverless GPU template for Stable Diffusion hosted on Banana's cloud platform. template: https://github.com/bananaml/serverless-template-stable-diffu... setup demo: https://www.banana.dev/blog/how-to-deploy-stable-diffusion-t...

Re: Try Stable Diffusion's Img2Img Mode

#148

Post by epups »

deanCommie wrote: post #145
epups wrote: post #99

Earlier quoted context omitted.

I can generate many violent prompts right now, and the model knows the face of hundreds of celebrities (you would think they would garble those to prevent PR disasters). It knows what the 9/11 attack was, it knows about nuclear bombs, horror, gore. Yet somehow it doesn't know what a blowjob is. This is 100% about narrow-minded, prudish conservative values. And yes, most of them come from the US right now, because the…

Tell me, earnestly, what progress is being blocked by your inability to use AI to generate an image of a blowjob. I don't mean this in a moralistic sense. I have no qualms about images of blowjobs. I have no doubt that the porn industry has already deployed these models and is experimenting with this without filters. As a monetizable way to reduce the human costs, it's an entirely logical step, and adult performers s…

It's funny because you spell out a use case that can represent undeniable progress: artificial porn. Porn with no component of human suffering, yet realistic and personalized. A dream product, no doubt. By what metric is this not progress, except a puritanical/religious one?

Moreover, sexuality is perhaps the most important theme in art, historically. It is a very legitimate thing to want to include in a tool such as this one. Its censorship is akin to "moral codes" of the past, completely regressive.

Re: Try Stable Diffusion's Img2Img Mode

#149

Post by omnimus »

desindol wrote: post #132
omnimus wrote: post #102

Earlier quoted context omitted.

Yes, yes and invention of camera will kill painters. Sure everybody knows this. But there is important difference between someone stealing and some algorithm copy replicating anything from the past in instant. One is a remix that brings something new (even if author doesn't want) the other in static its conservation. It will create side effects that will impact our (visual) culture. But who knows what those will be.

You must have missed the last 50 years with that much fear of change. Yikes.

Not sure why i deserve so much toxicity. We obviously work in different fields and have different view of what art actually is. No i am not happy to jump on AI bandwagon that will surely mean even faster pace and more precarious working conditions because individuals surely wont reap the benefits of automation. It's easy to not care when you are not affected.

None of that mean that i deserve the hate. It's just a different opinion.

Re: Try Stable Diffusion's Img2Img Mode

#150

Post by deanCommie »

epups wrote: post #148
deanCommie wrote: post #145

Earlier quoted context omitted.

Tell me, earnestly, what progress is being blocked by your inability to use AI to generate an image of a blowjob. I don't mean this in a moralistic sense. I have no qualms about images of blowjobs. I have no doubt that the porn industry has already deployed these models and is experimenting with this without filters. As a monetizable way to reduce the human costs, it's an entirely logical step, and adult performers s…

It's funny because you spell out a use case that can represent undeniable progress: artificial porn. Porn with no component of human suffering, yet realistic and personalized. A dream product, no doubt. By what metric is this not progress, except a puritanical/religious one? Moreover, sexuality is perhaps the most important theme in art, historically. It is a very legitimate thing to want to include in a tool such as…

I'm not as sure that artificial porn is undeniable progress.

While I am entirely pro liberty in terms of consumption of pornography, it is not a replacement for human interaction, and can be used in conjunction with other tools to dehumanize and isolate individuals. Some individuals that then make negative contributions to the rest of society at large. These are not isolated incidents, and they've been rising: https://en.wikipedia.org/wiki/Misogynist_terrorism

There are many kinds of porn - some that exploit women, some that empower them in their production. And there are also many kinds of products - ones that focus on a human connection (whether vanilla or kinky) and ones that don't.

I would absolutely worry that artificial porn would be good enough to meet some of these demands but not most, and would ultimately be a net negative for society.

> Moreover, sexuality is perhaps the most important theme in art, historically.

I mean, it's demonstrably not, religious art imagery dominates by volume. But that's also not necessarily important - that's just who happened to be patrons of arts and had the means to commission them.

I'm not gonna deny that humans are horny and want to make lots of sexy art.

But you're also not going to get porn on basic cable. You have to seek that out extra for yourself, and that's going to be the case with AI-generated porn art also.

Post reply on HN