News for hackers. Stuff that matters.
Re:Stable Diffusion 3 by btbuildem on Thursday February 22, 2024 @09:06AM (#39467226)
That's nice, but could we please have an unsafe alternative? I would like to footgun both my legs off, thank you.
Re:Stable Diffusion 3 by viraptor on Thursday February 22, 2024 @09:50AM (#39467909)
Just wait some time. People release SD loras all the time. Once SD3 is open, you'll be able to get a patched model in days/weeks.
Re:Stable Diffusion 3 by SV_BubbleTime on Thursday February 22, 2024 @10:46AM (#39468782)
A blogger I follow had an article explaining that the NSFW models for SDXL, are just now SORT OF coming up to the quality of SD1.5 “pre safety” models.

It’s been 6 months and it still isn’t there. SD3 is going to be quite awhile if they’re baking “safety” in even harder.

Re:Stable Diffusion 3 by viraptor on Thursday February 22, 2024 @10:55AM (#39468925)
1.5 is still more popular than xl and 2 for reasons unrelated to safety. The size and generation speed matter a lot. This is just a matter of practical usability, not some idea of the model being locked down. Feed it enough porn and you'll get porn out of it. If people have incentive to do that (better results than 1.5), it really will happen within days.
Re:Stable Diffusion 3 by Der_Einzige on Thursday February 22, 2024 @10:56AM (#39468938)
Due to the pony community the SDXL nsfw models are far superior to SD1.5. Only issue is that controlnets don’t work with that pony SDXL fine tune
Re:Stable Diffusion 3 by SV_BubbleTime on Thursday February 22, 2024 @02:37PM (#39471938)
I am slightly aware of the pony models.

I wish I had something more clever to comment on it. I know what they’re doing which is cool and why which is, IDK, live and let live and enjoy your own kink. It just a little funny some of the most work put into in the fine tuning models.. is from the pony community.

So all I have is…

:/

Re:Stable Diffusion 3 by int_19h on Thursday February 22, 2024 @07:46PM (#39475452)
It's not just them. For example, 4chan is a surprisingly good way to get the most recent scoop on good models (both text and images), setup guides etc - if you can tolerate the inevitable, well, 4chan-ness of it. And the reason is exactly the same: a lot of people there really, really, really, want to generate porn and to chat with sexbots, and they're putting a lot of effort into getting the best (and least censored) results out of the resources that they have.
Re:Stable Diffusion 3 by Magi604 on Friday February 23, 2024 @05:02AM (#39478850)
4Chan was/is an amazing resource for finding ways to bypass Dalle restrictions.

Sadly the elite ai hackers have shuffled off into private discords.

Re:Stable Diffusion 3 by AstraliteHeart on Friday February 23, 2024 @12:22AM (#39477268)
> IDK

We are just trying to satisfy your values though ponies and friendship.

Re:Stable Diffusion 3 by makomk on Thursday February 22, 2024 @07:47PM (#39475459)
Apparently most stuff that relies on the weights being resonably similar to SDXL don't work - control nets, LORAs, commonly-used inpainting patches, the lot. It seems to go well beyond fine tuning and be substantially retrained to the point it's a good chunk of the way to being a different model entirely, and the amount of training time is on the order of what went into SDXL originally too from what I can tell.
Re:Stable Diffusion 3 by BeFlatXIII on Saturday February 24, 2024 @08:33AM (#39491362)
I hope they call it Horse Diffusion.
Re:Stable Diffusion 3 by dougmwne on Thursday February 22, 2024 @09:26AM (#39467525)
Since these are open models, people can fine tune them to do anything.
Re:Stable Diffusion 3 by politician on Thursday February 22, 2024 @09:43AM (#39467798)
It’s not obvious that fine-tuning can remove all latent compulsions from these models. Consider that the creators know that fine-tuning exists and have vastly more resources to explore the feasibility of removing deep bias using this method.
Re:Stable Diffusion 3 by dougmwne on Thursday February 22, 2024 @09:51AM (#39467924)
Go check out the Unstable Diffusion Discord.
Re:Stable Diffusion 3 by SV_BubbleTime on Thursday February 22, 2024 @02:37PM (#39471943)
The vast majority of images there are SD1.5, even the ones made today.

Which goes far more towards the idea that safety isn’t a desirable feature to a lot of AI users.

Re:Stable Diffusion 3 by ttul on Thursday February 22, 2024 @05:41PM (#39474332)
I suppose you could train a model from scratch if you have enough money to blow…
Re:Stable Diffusion 3 by wokwokwok on Thursday February 22, 2024 @09:47AM (#39467852)
How would that be meaningfully different to SDXL?

I mean, SDXL is great. Until you’ve had a chance to actually use this model, isn’t calling it out for some imagined offence that may or may not exist seems like you’re drinking some Kool-aid rather than responding to something based in concrete actual reality.

You get access to it… and it does the google thing and puts people of colour in every frame? Sure, complain away.

You get access to it, you can’t even generate pictures of girls? Sure. Burn the house down.

…you haven’t even seen it and you’re already bitching about it?

Come on… give them a chance. Judge what it is when you see it not what you imagine it is before you’ve even had a chance to try it out…

Lots of models, free, multiple sizes, hot damn. This is cool stuff. Be a bit grateful for the work they’re doing.

…and even if sucks, it’s open. If it’s not what you want, you can retune it.

Re:Stable Diffusion 3 by Sharlin on Thursday February 22, 2024 @09:27AM (#39467548)
"I should have an unrestricted model because I myself am not going to use it to harm other people in the form of deepfake porn or other things, I promise! I'm too self-centered to consider the consequences of such a model being available also to those who do want to harm other people!"
Re:Stable Diffusion 3 by btbuildem on Thursday February 22, 2024 @09:37AM (#39467697)
Consider the impact of the two scenarios below:

1. A mass-distributed LLM (hosted by google or openai or whoever) that's been neutered and twisted into political correctness in a haphazard series of kneejerk meetings of small groups of people who are terrified big investors will walk away or that some powerful political group will denounce them. Effectively they create an enormous bias of falsity and incorrectness, for billons of people to use and embed the results throughout all their intellectual output.

2. Some wacko with an expensive Nvidia GPU makes deepfake porn of a popular politician. Or goodness forbid, of some weird kink where if this was actually a scene filmed with real people, there would be serious ethical issues.

Which scenario do you think is more dangerous, long term, and in terms of broad impact on society in general?

Re:Stable Diffusion 3 by spencerflem on Thursday February 22, 2024 @10:13AM (#39468261)
I think you are seeing things from a bubble. Most people in the country are in favor of efforts to correct for historical injustices and are worried about AIs repeating biases in their training that could have a material impact on the world.

Case in point: large advertisers and entertainment companies "pander" to these sorts of views, because it is broadly popular

This is not the work of a shadowy cabal

Re:Stable Diffusion 3 by shapefrog on Thursday February 22, 2024 @10:36AM (#39468634)
I think you are seeing things from a bubble.

Most people in the country dont give a shit about historical injustices and are worried about how they are going to pay their rent or put food on the table today.

Re:Stable Diffusion 3 by spencerflem on Thursday February 22, 2024 @10:46AM (#39468787)
A lot of people have the capacity for both

I gave some evidence that being socially progressive / "woke" is broadly popular. Advertisers are not activists, they're following what people like

Re:Stable Diffusion 3 by SV_BubbleTime on Thursday February 22, 2024 @02:39PM (#39471968)
If you go by online, popularity, yes, a lot of people do want to erase history in favor of feelings. That can be your opinion too.

> Broadly popular

Are they? Or are the loudest voices asymmetrically affecting discourse?

> This is not the work of a shadowy cabal

And how would you know if it was? What would the clues be? If you were in a bubble that was designed to impart an informal religious view (and it is a religion sometimes, just one with a screen instead of a book) that encompass politics and morality, how would you know?

Re:Stable Diffusion 3 by jrm4 on Thursday February 22, 2024 @09:43AM (#39467800)
I appreciate the sentiment, but, like -- this never works. Not in software. Might as well let it go.
Re:Stable Diffusion 3 by Sharlin on Thursday February 22, 2024 @11:26AM (#39469409)
I just cannot comprehend how some people cannot see the incredibly obvious moral responsibility in releasing something that could be used for a lot of good but also to do bad things. There's no reasonable moral theory in which you could just shrug and go, "well, somebody is going to do it anyway so why should we even try to keep our conscience clean and avoid making it easy for them", it's fundamentally amoral and antisocial.

If someone invents a lockpick capable of opening any door, they have a moral responsibility in preventing it from falling into wrong hands, whether they want it or not. And it's absurd to complain when someone who could create an universal lockpick, refuses to do so, never mind release the technology to the wild, and only agrees to sell simpler picks capable of picking simpler locks. Do these people also complain about things like work against nuclear proliferation? After all, North Korea got nukes anyway, so what's the point?

Re:Stable Diffusion 3 by jrm4 on Thursday February 22, 2024 @01:15PM (#39470917)
Your lockpick example only works if there's e.g. only one and it's feasible to keep it hidden from the world.

Software doesn't work that way. I agree that you should be responsible about dangerous tech, but you also have to be realistic about what the best way to do that is, which is pretty much never "keep it hidden."

(and, of course, this is not even considering the question that should probably go here which is -- how dangerous is this exactly? Given the moral panic we saw a while ago about e.g. Photoshop, I'm not entirely convinced that this is much to worry about.)

Re:Stable Diffusion 3 by samatman on Thursday February 22, 2024 @10:03PM (#39476464)
Here, knock yourself out https://covertinstruments.com

They don't have a moral responsibility to do jack squat other than sell an honest product and honor their warranties. You don't like that, tough.

Re:Stable Diffusion 3 by huggingmouth on Thursday February 22, 2024 @09:40AM (#39467744)
This wouldn't be as silly a counter argument if there weren't already unrestricted models out there.

Unfortunately, this is a very silly argument. I'm so glad that people like you weren't around when they invented the personal computer.

Re:Stable Diffusion 3 by Sharlin on Thursday February 22, 2024 @11:26AM (#39469419)
On the contrary, the argument that "someone will do it anyway, so you should just let it happen and take no moral responsibility because I WANT MY SHINY TOY" is… not merely silly, but incredibly absurd, selfish, entitled, and amoral.

One could argue that if $BIGCORP doesn't want their thing to be used by bad actors, they should just refrain from developing the technology at all, and while it's a somewhat defensible position, that would also result in techbros not getting their toy, so it doesn't really apply here.

Re:Stable Diffusion 3 by huggingmouth on Friday February 23, 2024 @04:03PM (#39486053)
The argument is that somebody already did it. Continuing this nonsense only harms legitimate users.
Re:Stable Diffusion 3 by HeatrayEnjoyer on Thursday February 22, 2024 @09:45AM (#39467826)
The willful ignorance in these threads is maddening. They know why models have these restrictions, they just think the rules shouldn't apply to them and want to play the victim. It's the same libertarian attitude as people who whine about driving speed laws.

If there's one lesson from the 21st century, it's that's you shouldn't release massively impacting technology without strong ethics controls around it.

Re:Stable Diffusion 3 by Fervicus on Thursday February 22, 2024 @01:52PM (#39471368)
Nope, sorry. We can't allow you to commit thought crimes.