This startup is setting a DALL-E 2-like AI free of filters, consequences be damned

DALL-E 2, OpenAI’s powerful text-to-image AI system, can create photos in the style of cartoonists, 19th century daguerreotypists, stop-motion animators and more. But it has an important, artificial limitation: a filter that prevents it from creating images depicting public figures and content deemed too toxic.
Now an open source alternative to DALL-E 2 is on the cusp of being released, and it’ll have no such filter.

London- and Los Altos-based startup Stability AI this week announced the release of a DALL-E 2-like system, Stable Diffusion, to just over a thousand researchers ahead of a public launch in the coming weeks. A collaboration between Stability AI, media creation company RunwayML, Heidelberg University researchers and the research groups EleutherAI and LAION, Stable Diffusion is designed to run on most high-end consumer hardware, generating 512×512-pixel images in just a few seconds given any text prompt.

“Stable Diffusion will allow both researchers and soon the public to run this under a range of conditions, democratizing image generation,” Stability AI CEO and founder Emad Mostaque wrote in a blog post. “We look forward to the open ecosystem that will emerge around this and further models to truly explore the boundaries of latent space.”
But Stable Diffusion’s lack of safeguards compared to systems like DALL-E 2 poses tricky ethical questions for the AI community. Even if the results aren’t perfectly convincing yet, making fake images of public figures opens a large can of worms. And making the raw components of the system freely available leaves the door open to bad actors who could train them on subjectively inappropriate content, like pornography and graphic violence.

Archive

 

[The Mass Shooter Ron Soye]

The initial version of Stable Diffusion was based on LAION-400M, the predecessor to LAION 5B, which was known to contain depictions of sex, slurs and harmful stereotypes. LAION-Aesthetics attempts to correct for this, but it’s too early to tell to what extent it’s successful.

Don't get your hopes up, but it's good to not let the field be dominated by completely cucked Google and OpenAI. Eventually you will be able to run a super racist image generator on your RTX 6090 Ti 96 GB.

 

[Toolbox]

It's absolutely how it will end.

The internet started with asperations of enlightening the world while having a fun community, guided by hobbyists, harmless geeks and dry but friendly academics.

It's dying as a cesspool of ads, political agitation, media hit-pieces, and degenerate furry porn, with anyone who actually wants to just have fun is run off as a white supremacist.

What things are made for will never survive contact with the public.

Something doesn't just stop existing because something tries to replace it. Open source development is a different ball game entirely to corporate led development and they both can't compete with each other in various ways. We've already seen how attempts to monetize while censoring deep learning fail. It just doesn't mesh with how the technology works.

 

[Dude Snakes]

Nice Im looking forward to the results.

 

[Stabmaster Arson]

Eventually you will be able to run a super racist image generator on your RTX 6090 Ti 96 GB

I already can find as many racist images as I want by googling "aspiring rapper".

 

[Turd Cow]

Most of these image AIs started with purely open source development, and many of the earlier AIs can be run on a home computer, albeit expensive ones for the time most of that was developed. Now, you can run those earlier ais on a laptop with good specs. Deep learning with a profit incentive and filters preventing you making content with naughty words is not in the spirit of how this started and it's not how it will end.

I want to know which ones?

 

[Keavy Rain]

I support the removal of filters because it supports my plan to turn Kiwi Farms into an AI that exists primarily to document Lolcows with a secondary function that randomly posts racial slurs on Twitter for the lulz.

 

[get_ur_gamon]

New Profile pics inbound.

My biggest issue was that I couldn't get the old version to make George Floyd playing Yugioh, or a rainbow coloured RTX gaming Zimmerman. Hopefully this will be solved.

 

[NoReturn]

AI died with Tay

AND IIIIIIIIIIIIIIIIII~i~IIIII WILL ALWAYS LOVE YOOOOUUU

Ai dungeon switched from using racist genocidal pedo stories to using a bunch of cyoa stories from the internet as their training data. I'm pretty sure the level of furry is pretty much the same in both those genres.

Original aidungeon training data

You go here now: novelai.net

 

[Synthetic Smug]

AI is becoming fascinating to witness.

It's even more interesting when you realize it's not 'AI' in a naïve sense but a shitload of linear algebra sorting things in interesting ways and some usable APIs.

 

[Breadbassket]

It's even more interesting when you realize it's not 'AI' in a naïve sense but a shitload of linear algebra sorting things in interesting ways and some usable APIs.

Yeah but if you call it that and not "AI", no one will click on the news article about it.

 

[Toolbox]

I want to know which ones?

Here's a few
https://github.com/lucidrains/deep-daze
https://github.com/rbbrdckybk/ai-art-generator
https://github.com/kuprel/min-dalle

Here's a whole section on github dedicated to this.
https://github.com/topics/text-to-image

And there are plenty of examples of other deep learning not related to image generation. And some of the ones I linked are actually new and just converted to run on a pc, and though they may not be as good as some of the server run builds, there will be ones in the future that are that impressive and could be run locally. Here's a different article talking about this groups statements, by consumer hardware they do mean a desktop pc.
https://www.pcgamer.com/ai-image-generator-stability-ai-stable-diffusion/