News for hackers. Stuff that matters.
Re:AI Real-Time Human Full-Body Photo Generator by gwern on Wednesday August 23, 2023 @12:17PM (#37238316)
If you're wondering how it's so fast and cheap and they can generate variants so easily, it's because they're using GANs (see the footer). GANs are way faster than diffusion models because they generate the image in a single forward pass and their true latent space encoding makes editing a breeze.

(And if you're wondering how it can look so good when 'everyone knows GANs don't work because they're too unstable', a widespread myth, repeated by many DL researchers who ought to know better, GANs can scale to high-quality realistic images on billion-image scale datasets, and become more, not less, stable with scale, like many things in deep reinforcement learning. See for example BigGAN on JFT-300m https://arxiv.org/pdf/1809.11096.pdf#page=8&org=deepmind , GigaGAN https://arxiv.org/abs/2303.05511 , Projected GAN https://arxiv.org/abs/2111.01007 , StyleGAN-XL https://arxiv.org/abs/2202.00273 , or Tensorfork's chaos runs https://gwern.net/gan#tensorfork-chaos-runs . 'Computing is a pop culture'...)

Re:AI Real-Time Human Full-Body Photo Generator by motoboi on Wednesday August 23, 2023 @01:05PM (#37239020)
What I’m really wondering is how can this be free. What is the business model here?

Are they using me to refine the model in some way?

Re:AI Real-Time Human Full-Body Photo Generator by htrp on Wednesday August 23, 2023 @01:08PM (#37239059)
Human evaluation is super expensive so yes. Seeing which sessions you discard and keep alone is worth the compute time, especially if it's GAN based.
Re:AI Real-Time Human Full-Body Photo Generator by irrational on Wednesday August 23, 2023 @01:29PM (#37239348)
It says free for non-commercial. If commercial, contact us. I assume they plan on paying the bills with commercial work.
Re:AI Real-Time Human Full-Body Photo Generator by brucethemoose2 on Wednesday August 23, 2023 @01:11PM (#37239088)
There was a whole community around ESRGAN img2img finetuning kinda like Stable Diffusion LORA community... albeit a much, much smaller one.
Re:AI Real-Time Human Full-Body Photo Generator by lacoolj on Wednesday August 23, 2023 @01:20PM (#37239222)
maybe it was fast 58 minutes ago but apparently it is now at peak capacity. even if you don't get rejected, a new image takes minutes
Re:AI Real-Time Human Full-Body Photo Generator by syntaxing on Wednesday August 23, 2023 @01:26PM (#37239291)
> everyone knows GANs don't work because they're too unstable

Is that a wide spread myth? I thought it’s widely accepted that GAN is really good generating these artificial pictures (it’s what started DeepFake after all) when you know your model’s “button”. Similar to how this uses GAN since they have a model “boundary condition”. While humans are diverse, we have a set of repeatable features (two legs, two arms, etc). Diffusion models are great because you can control the latent space with something way more generic, like text hence why it’s been so much more mainstream.

Edit: actually I might be misremembering, I think Deepfake used VAEs?

Re:AI Real-Time Human Full-Body Photo Generator by gwern on Wednesday August 23, 2023 @04:00PM (#37241349)
It is very widespread. You will see people in this very thread dismissing GANs as fundamentally failed, and hotly objecting to any kind of parity, even if they have to fall back to 'well ok GANs do scale, but they're more complicated'. I also have some representative quotes in my linked draft essay from various papers & DL Twitter discussions. (Another way to put it would be: when was the last time you saw someone besides me asserting that GANs can scale to high-quality general images and are not dramatically inferior to diffusion? I rest my case.)
Re:AI Real-Time Human Full-Body Photo Generator by ShamelessC on Wednesday August 23, 2023 @06:10PM (#37242520)
Sounds very important to you that you don’t have to change the premise of your essay or ever admit you’re wrong. No one is dismissive of GAN’s here without justification. They’re fine. They don’t beat diffusion, but they’re fine.

You come across as severely, _severely_ biased and reactionary.

Re:AI Real-Time Human Full-Body Photo Generator by ShamelessC on Wednesday August 23, 2023 @10:52PM (#37244495)
> Another way to put it would be: when was the last time you saw someone besides me asserting that GANs can scale to high-quality general images and are not dramatically inferior to diffusion? I rest my case.

Get off of Twitter, maybe? I don't know but I have seen a ton of work done on large scale GANs. The incentives are pretty fucking obvious - they are faster to use. Of _course_ people want them to work for the same usecases as diffusion.

Maybe they will someday! But they aren't currently. You're referencing a zeitgeist that I am not familiar with and therefore come across overly defensive.

Re:AI Real-Time Human Full-Body Photo Generator by nbardy on Wednesday August 23, 2023 @01:55PM (#37239735)
You’re overstating the simplicity of a scaling a GAN well.

GigaGAN is the best quality out of those and requires 7 loss functions and is incredibly complicated.

Sure GANs can scale, but Diffusion models are drastically easier to scale.

Re:AI Real-Time Human Full-Body Photo Generator by gwern on Wednesday August 23, 2023 @04:07PM (#37241407)
No, I'm not. BigGAN did fine on scaling up to JFT-300M with basically no changes beyond model size and a simple architecture. This is also what we were observing, even with a buggy BigGAN implementation. GigaGAN is the best quality, but that's mostly because it's also the biggest; as Table 1 shows most of the gains come from various kinds of additional scaling. (And this is moving the goalposts from the usual assertion that "GANs can't scale" to "they're harder to scale"; note the self-fulfilling nature of such assertions. Considering how there is next to no GAN scaling research, these results are remarkable and show how much low-hanging fruit there is.)

Diffusion models are only 'drastically easier to scale' because researchers have spent the past 3 years researching pretty much nothing but diffusion models, discovering all sorts of subtle issues with them and how to make them scale, which is why it took them so long to become SOTA, and why massive architectural sweeps like https://arxiv.org/abs/2206.00364#nvidia were necessary to discover what makes them 'easier to scale'. If this level of brute force and moon math is 'easy', lord save us from any architecture which is 'hard'!

Re:AI Real-Time Human Full-Body Photo Generator by GaggiX on Wednesday August 23, 2023 @05:32PM (#37242118)
Researchers have spent several years try to even create a GAN that can fit well a distribution made of aligned faces (resulted into StyleGAN1/2); with a simple unet with e-objective and cosine schedule you can fit much complex distributions, still using one loss: L1/L2.

Reading your comments make me feel like that you believe that just every researchers (even extremely smart dude like Karras) just switch to diffusion models because they are idiots, they should have instead focus on GANs and today we will have GANs that are as powerful or more than the diffusion models we have today and also work one step; this is just a weird delusion. Diffusion models are just simply much easier to train (just a L1/L2 loss in most cases), write (for example your buggy BigGAN implementation), they usually work out-of-box on different resolutions and aspect ratios, you can just finetune them if you want to create an inpainting model; and for what is right now you just need much less compute to reach a good image coherency or maybe just reaching a coherence that as not been achieved by GAN models; like I would be curious even on a small scale experiment what a GAN (with ~55M parameters) would be able to perform after a 1-day/2-day GPU time of training on Icon645 dataset, because my diffusion model I can assure is much better than I could have imagine while being trivial to implement (I just implemented a Unet as I remember one, nothing rigorous and of course no architecture sweep).

Re:AI Real-Time Human Full-Body Photo Generator by sillysaurusx on Wednesday August 23, 2023 @09:09PM (#37243963)
There are so many holes in this, but to pick just one:

> for example your buggy BigGAN implementation

The bug was that the batchnorm gamma was initialized to zero instead of one, so all the model params were being multiplied by zero instead of one. Literally any model that uses batchnorm is susceptible to this bug, and it certainly has nothing to do with GANs.

I think it’s too easy to spout dogma, and calling gwern delusional is something I’ve learned from experience will prove foolish as t approaches infinity.

Re:AI Real-Time Human Full-Body Photo Generator by GaggiX on Wednesday August 23, 2023 @09:56PM (#37244213)
It has to do with GANs as the added complexity makes the model more difficult to debug, the other day a friend of mine had a problem with the normalization layers of his diffusion model, it was fixed almost immediately, if something goes wrong you have less to worry about.

Also no I don't take "he is gwern" as an argument ahah

Re:AI Real-Time Human Full-Body Photo Generator by sillysaurusx on Wednesday August 23, 2023 @11:17PM (#37244614)
Bookmarking this for when time proves you wrong after a couple years. I’ll drop by to say hello.
Re:AI Real-Time Human Full-Body Photo Generator by ShamelessC on Wednesday August 23, 2023 @06:05PM (#37242468)
> Diffusion models are only 'drastically easier to scale' because researchers have spent the past 3 years researching pretty much nothing but diffusion models

This is what tends to happen when you find a superior method.

GAN’s are fine, they have plenty of promise for tasks requiring rapid inference. But diffusion models beat GANs on robustness and image quality every time.

Re:AI Real-Time Human Full-Body Photo Generator by dragonwriter on Wednesday August 23, 2023 @02:24PM (#37240153)
> If you're wondering how it's so fast and cheap and they can generate variants so easily

I assume its cheap because they are burning money to build a business, its not fast at all, and the quality... sucks.

> And if you're wondering how it can look so good

I’m not.

I'm wondering why they’re trying to get people to use something worse than using a decent photorealistic SD1.5-based checkpoint with some basic prompt templating.

Not saying GANs can't be awesome, just that this site isn't what I’d use to make that case.

Re:AI Real-Time Human Full-Body Photo Generator by mrguyorama on Wednesday August 23, 2023 @04:18PM (#37241437)
Looking at the poses, it feels optimized for generating porn, but one example someone showed had a child's face (good god please don't let your "AI" system generate child anything if you want to sell it for porn purposes), and another user noted that their attempt error'd out because it "detected nudity", even though other users get given a nude model by default.
Re:AI Real-Time Human Full-Body Photo Generator by dotancohen on Wednesday August 23, 2023 @05:03PM (#37241820)
I don't know if the thing in the crotch is a penis or a scrotum, but it is definitely NSFW:

https://images.generated.photos/0wV1dBnZ15hGneEfqfZT7SdEIill...

My prompt was simply "Standing in front of a rocket.".

Re:AI Real-Time Human Full-Body Photo Generator by userbinator on Wednesday August 23, 2023 @08:11PM (#37243568)
It's neither, because this AI doesn't care for gender norms. /s

That looks more like a machine, an android perhaps, than an actual human.

Re:AI Real-Time Human Full-Body Photo Generator by oniony on Wednesday August 23, 2023 @05:15PM (#37241926)
That image is so full of wtf
Re:AI Real-Time Human Full-Body Photo Generator by fouc on Wednesday August 23, 2023 @08:58PM (#37243875)
Oh no, no, no, I'm a rocket man
Re:AI Real-Time Human Full-Body Photo Generator by keyle on Wednesday August 23, 2023 @07:36PM (#37243334)
Good thing you didn't ask for sitting.
Re:AI Real-Time Human Full-Body Photo Generator by taneq on Wednesday August 23, 2023 @07:13PM (#37243139)
I’m going with that guy is a reptile in disguise.
Re:AI Real-Time Human Full-Body Photo Generator by brvsft on Wednesday August 23, 2023 @10:01PM (#37244239)
Hah. I've been playing with StableDiffusion locally and have gotten a few abominations like this myself.
Re:AI Real-Time Human Full-Body Photo Generator by dontupvoteme on Wednesday August 23, 2023 @11:32PM (#37244695)
Whatever you do don't take all those long prompts that include (((aesthetic))) and such and put them in the negative prompt with "human" in the positive prompt.

It's not nightmare fuel it's a nightmare engine

Re:AI Real-Time Human Full-Body Photo Generator by notmeanywhere on Thursday August 24, 2023 @12:34AM (#37245035)
I had one generate a fully nude, spread eagle male model w/ a very vivid vagina when I typed in "on couch". Tried it again w/ other randomly generated male and female models with nowhere near the same results w/ any of the other tries...
Re:AI Real-Time Human Full-Body Photo Generator by redox99 on Thursday August 24, 2023 @02:29AM (#37245692)
It's interesting that the image has a "mirrored along the X axis" kind of look.
Re:AI Real-Time Human Full-Body Photo Generator by the-dude on Friday August 25, 2023 @08:23AM (#37260889)
You asked for a phallus symbol, it gave you phallus symbols.
Re:AI Real-Time Human Full-Body Photo Generator by nomdep on Wednesday August 23, 2023 @08:23PM (#37243655)
An AI-generated image is just a drawing, not a real human.
Re:AI Real-Time Human Full-Body Photo Generator by mgbmtl on Thursday August 24, 2023 @06:32AM (#37247202)
Re:AI Real-Time Human Full-Body Photo Generator by 93po on Thursday August 24, 2023 @11:50AM (#37250298)
I believe anything depicting CP is illegal in the US as well
Re:AI Real-Time Human Full-Body Photo Generator by MacsHeadroom on Saturday August 26, 2023 @03:12AM (#37270522)
The justice department defines it as anything depicting a real identifiable minor. Fictive persons are notably exempt from this definition (regardless of realism of the depiction). Canada is different in that regard.
[dead]
Re:AI Real-Time Human Full-Body Photo Generator by tempestn on Wednesday August 23, 2023 @05:57PM (#37242389)
To expand on what I take as your implied argument- Some (small) percentage of people are pedophiles, meaning they're attracted to children. Presumably they can't help that, just as others can't change their sexual preferences. Clearly acting on this urge with an actual child is wrong. That's true whether it's directly assaulting a child, or consuming child porn, as that market encourages others to exploit children to generate it. However, if it is possible to produce CP without involving actual children, it could provide an outlet for those desires that would reduce demand for actual CP, and thereby reduce incidents of children being abused to produce it.

One could argue that such an outlet could even reduce incidents of direct sexual assault of children by pedophiles, but there is also a counter-argument that it would instead serve to "whet the appetite" and encourage such behaviour. And of course there are other counter-arguments; it could make actual CP more difficult to detect, for one. Finally there is the argument from the perspective of fundamental morality, that depicting children in a sexual manner is wrong in and of itself, and therefore the various potential effects are irrelevant. (Much like it's wrong to murder an innocent, even if you could harvest their organs and save five others as a result.)

Re:AI Real-Time Human Full-Body Photo Generator by DharmaPolice on Wednesday August 23, 2023 @07:05PM (#37243060)
>Finally there is the argument from the perspective of fundamental morality, that depicting children in a sexual manner is wrong in and of itself, and therefore the various potential effects are irrelevant. (Much like it's wrong to murder an innocent, even if you could harvest their organs and save five others as a result.)

I don't think those two scenarios are usefully similar. Murder is wrong because of its effects. This would be more like if you could murder someone without _anyone_ dying.

Re:AI Real-Time Human Full-Body Photo Generator by thegabriele on Wednesday August 23, 2023 @07:14PM (#37243156)
Murder someone without anyone dying is the plot of many, many videogames since the 80s
Re:AI Real-Time Human Full-Body Photo Generator by wkat4242 on Wednesday August 23, 2023 @07:48PM (#37243410)
In the 80s you were "murdering" a handful of pixels. It's really not comparable. Neither In the 90s or 00s even though ever more pixels were involved.

I have to say though that this argument starts making a lot more sense with current VR tech. Sometimes it's really disgustingly realistic.

While I'm still not in favor of banning violent games, I do think a valid argument is emerging. I've never truly had a feel of fright playing on a 2d screen but since VR the spiders headcrabs of Half Life Alyx really make me panic sometimes.

On the csam side I'm just not sure if this would diminish or increase their interest. I would leave that up to someone who studied for this kind of insight.

Re:AI Real-Time Human Full-Body Photo Generator by hackinthebochs on Wednesday August 23, 2023 @06:14PM (#37242547)
It's interesting to notice when utilitarian arguments are accepted and when they're rejected. The argument offered in favor of abortion without limits tends to be that women will get abortions regardless, they will just be dangerous. Presumably the greater good is served by allowing abortions despite the moral issues surrounding killing fetuses/unborn children. I have no trouble imagining many people supporting such a utilitarian argument for abortion but not for generated CP. Though I have a hard time making the distinction intelligible.
Re:AI Real-Time Human Full-Body Photo Generator by stavros on Wednesday August 23, 2023 @06:42PM (#37242849)
That's a good summary, thanks. I think AI-generated will lead to actual child porn not making financial sense (hopefully, anyway). I also don't think that the "whetting the appetite" argument is true, from other areas I've seen (eg playing violent games doesn't lead you to becoming a murderer), but I have no data on that.
Re:AI Real-Time Human Full-Body Photo Generator by katabasis on Wednesday August 23, 2023 @06:22PM (#37242629)
1. You are potentially giving a shield of deniability to people who create or distribute real CSAM because now they could claim that the images are just AI generated and therefore "harmless"

2. Efforts to stamp out real child abuse may be undermined by a flood of AI-generated false positive imagery

3. When people see something over and over again they start to think that it's normal. AI generation of this kind of material (something which can be done at a huge scale) risks normalizing the sexual abuse of children.

I'm sure there are many other arguments beyond these.

Re:AI Real-Time Human Full-Body Photo Generator by lemper on Thursday August 24, 2023 @12:51AM (#37245130)
ok, you've convinced me about it. before, I didn't have any opinions regarding llm and cp. I didn't even think people could generate cp with llm. thanks mate.
Re:AI Real-Time Human Full-Body Photo Generator by NikolaNovak on Thursday August 24, 2023 @08:48AM (#37248150)
I see #3 as potentially valid, but also as applying equally to murder, drugs, and car speeding, which are prevalent in movies, animation, and video games. Heck, tv shows catering to kids have variations on all 3.

Don't get me wrong. I have a built in revulsion toward CP. But I wonder how much is that built in revulsion helping vs hindering rational discussion - just like it took me like 15 years of reading and pondering to remove the indoctrinated deference I had toward religion, and I hope I can have a more stable, less knee jerk discussion about religion. I think CP is wrong, but I also Think murder and torture are wrong. The latter two are amply represented in entertainment media. What is the Venn diagram, what is the delta that makes cp different?

Re:AI Real-Time Human Full-Body Photo Generator by ThrowAway1922A on Wednesday August 23, 2023 @05:58PM (#37242391)
> What exactly is the argument against AI-generated child porn?

Currently? The fact that all the models need training data and the law will see that as victimizing the people who were used in the data set be they adults of children.

Overall? The fact that it's disgusting and pedophiles deserve things which I can say IRL and everyone agrees with, but on HN will get me banned.

Many countries ban underage anime porn too. Children and their likeness are off limits.

Re:AI Real-Time Human Full-Body Photo Generator by bruce511 on Wednesday August 23, 2023 @07:10PM (#37243120)
It's natural (normal) to think child point is disgusting.

But its not hard to find many other sexual acts that at least some people, if not most people, find disgusting. So I'm not sure disgust is a useful criteria here.

Most of what makes pedophilia so disgusting is the fact that a child is involved. Remove the child from the equation and it's more in line with other sex preferences.

People can't really control their preferences. We've grown to understand that there are healthy, and unhealthy ways to satiate those preferences. Most of those revolve around the consent of those involved (taking as fact that children cannot consent.)

One could argue that AI allows for a healthy outlet for a pedophile.

You mention children being off limits. In other societies homosexuality is still criminal. Even where it is legal many consider it "off limits" or "disgusting". Laws, and emotions, in this space are not always logical.

There will be second-order arguments (think - the gays are trying to turn my child gay) but most of them fundamentally misunderstand how human sexuality works.

Re:AI Real-Time Human Full-Body Photo Generator by wkat4242 on Wednesday August 23, 2023 @07:58PM (#37243482)
> You mention children being off limits. In other societies homosexuality is still criminal. Even where it is legal many consider it "off limits" or "disgusting". Laws, and emotions, in this space are not always logical.

In fact in a lot of those countries where fully consensual homosexuality is highly illegal, marrying (with all its implications) what we would consider a child is legal and "normal" and so is forced/arranged marriage (again with all that implies). I'm really lucky to have grown up here.

Re:AI Real-Time Human Full-Body Photo Generator by bruce511 on Wednesday August 23, 2023 @09:56PM (#37244215)
Certainly yes, different cultures have different norms.

As a complete aside, forced marriages are not the same as arranged marriages. In many cultures arranged marriages (which seem completely bonkers to someone grown up in a western mindset) are both common, work well, and often good.

Clearly neither arranged marriages, nor "choice marriages", are guaranteed. The divorce rate in the west bears this out. Both work well, both work badly.

Again arranged is not the same as forced - clearly forced marriages are not something I consider a good thing.

Re:AI Real-Time Human Full-Body Photo Generator by danbruc on Wednesday August 23, 2023 @07:25PM (#37243245)
The fact that it's disgusting and pedophiles deserve things which I can say IRL and everyone agrees with, but on HN will get me banned.

Would you say the same about homosexuality? Nobody chooses their own sexual preference. Pedophiles are unfortunate victims having a sexual preference that they can never satisfy without causing others harm. Even if people could pick their sexual preferences - which they can not - among the countless possibilities, children would be one of the worst choices.

Re:AI Real-Time Human Full-Body Photo Generator by richie_adler on Wednesday August 23, 2023 @06:55PM (#37242971)
Far from me to defend pederasty, but I'm quite sure I would disagree with the thing you wouldn't want to publish, RL or not.
Re:AI Real-Time Human Full-Body Photo Generator by caslon on Wednesday August 23, 2023 @07:48PM (#37243405)
I agree with you that porn is disgusting, and that the people who view it are immoral. However, there are a lot of arguments to be made against capital punishment, murder, and other forms of vigilante violence. No nation with capital punishment is a stranger to wrongful execution.

https://en.wikipedia.org/wiki/Wrongful_execution

Encouraging state violence based on the fact that something is "disgusting" is not a good idea. Law needs to be grounded in concrete harm. The lines currently seem ill-defined, and existing laws in this area have been used against children (see State of Washington v. E.G., where an autistic child was charged and convicted with distribution of child pornography for taking a selfie and sending it to an adult).

Re:AI Real-Time Human Full-Body Photo Generator by dragonwriter on Wednesday August 23, 2023 @05:38PM (#37242179)
> What exactly is the argument against AI-generated child porn?

As something you generate in a photorealistic image generator you are building a business around?

The fact that it is a serious crime in many jurisdictions and, even where it isn’t, photorealistic child porn images that get noticed anywhere or going to result in uncomfortable conversations for everyone involved in the process of establishing that they aren’t evidence of a crime.

Re:AI Real-Time Human Full-Body Photo Generator by CapitalistCartr on Wednesday August 23, 2023 @05:12PM (#37241902)
The argument against it is that the police and prosecutors don't care about your arguments.
Re:AI Real-Time Human Full-Body Photo Generator by masfuerte on Wednesday August 23, 2023 @06:21PM (#37242618)
Under English law creating a rough hand-drawn child porn sketch for your own amusement is a serious crime. I don't understand the rationale for this, but people should be aware that if they use a porn generator and it spits out an image that looks like CP then they will have committed an offence in England.
Re:AI Real-Time Human Full-Body Photo Generator by blitz_skull on Wednesday August 23, 2023 @05:20PM (#37241994)
Probably moral depravity if I had to guess. Not sure why we even need “an argument” against it. It’s pretty self-evidently wrong.
Re:AI Real-Time Human Full-Body Photo Generator by wizofaus on Wednesday August 23, 2023 @06:07PM (#37242490)
What if it turned out it were the only effective way to prevent people engaging in real paedophilia?
Re:AI Real-Time Human Full-Body Photo Generator by blitz_skull on Wednesday August 23, 2023 @08:33PM (#37243724)
You’re suggesting that perhaps allowing people to explore an urge that should be fought and suppressed might dissuade them from pushing that urge further resulting in real-world action? Fascinating, do tell me more.

Blatant disregard for children aside, the larger issue here is the presupposition that we should someone permit all urges. Not every thought that passes through your brain should be “explored”; some thing are just bad and should be fought.

Re:AI Real-Time Human Full-Body Photo Generator by wizofaus on Thursday August 24, 2023 @12:56AM (#37245165)
Not suggesting either, just putting forward a hypothetical possibility. The idea of using AI to generate kiddy porn is off-putting to me in the extreme, and I can think of plenty of reasons why it's almost certainly a bad idea to permit it, but I'd be wary of relying primarily on my instinctual aversion to it to be so certain it should be banned outright.
Re:AI Real-Time Human Full-Body Photo Generator by blitz_skull on Thursday August 24, 2023 @06:17AM (#37247122)
While I agree it’s probably a good idea to avoid instinctual aversion in many things, I believe some things in life are too sacred to rely on “data”. Some things we just know to be true, and have to trust that whatever story data might tell isn’t a story we’re interested in.

Keep in mind that science can tell any story you want it to tell as long as you interpret the data in the way you need. That’s why some things, such as our children, cannot be left to slippery slopes such as this.

We’ve already got more than enough data proving that pornography is absolutely destructive and a net negative on all of humanity. I don’t think we need additional data to prove that bringing kids into the mix would somehow make it better.

Re:AI Real-Time Human Full-Body Photo Generator by dragonwriter on Wednesday August 23, 2023 @07:05PM (#37243070)
> What if it turned out it were the only effective way to prevent people engaging in real paedophilia?

Paedophilia is not synonym for child sex abuse, and to the extent that conjecture hasn't been ruled out by study already, that's at most an argument that its an appropriate function for tightly-controlled research-supporting tools, not a general public business.

Re:AI Real-Time Human Full-Body Photo Generator by wizofaus on Thursday August 24, 2023 @01:03AM (#37245207)
No real disagreement there. Ultimately my biggest concern is that real children are not exploited and damaged by experiences they're clearly not ready for - if using AI-generated pornography turned out to be a particularly effective way to achieve that (with no obvious negative side-effects, and to be clear I doubt that's particularly likely to be the case), then I would think we shouldn't dismiss it out of hand as something that must be banned just because it intuitively seems so awful.
Re:AI Real-Time Human Full-Body Photo Generator by Jolter on Wednesday August 23, 2023 @06:52PM (#37242949)
But it’s not.
Re:AI Real-Time Human Full-Body Photo Generator by trehalose on Wednesday August 23, 2023 @06:41PM (#37242843)
That's a very big "what if". What data could demonstrate that to be true or false?
Re:AI Real-Time Human Full-Body Photo Generator by wizofaus on Thursday August 24, 2023 @12:57AM (#37245175)
I did have exactly that thought when I posted it. It was more of a thought experiment than anything.
Re:AI Real-Time Human Full-Body Photo Generator by JackFr on Wednesday August 23, 2023 @06:45PM (#37242885)
Prisons?
Re:AI Real-Time Human Full-Body Photo Generator by jacurtis on Wednesday August 23, 2023 @10:43PM (#37244454)
> it feels optimized for porn

So I was relatively unimpressed with the first few images, they looked incredibly unrealistic. But then after reading this message i decided to add "topless" to the notes just to see how crazy the result was.

I was actually impressed, it returned an image that actually looks lifelike once I gave it an adult request. Which reinforces the idea that this was optimized for porn.

For those curious, this was the result I was provided. NSFW of course, topless female gender (ai generated): https://generated.photos/human-generator/64e6c26e190809000fb...

Re:AI Real-Time Human Full-Body Photo Generator by jprete on Wednesday August 23, 2023 @11:42PM (#37244773)
That image is really off-putting. It looks wrong in several ways I can identify and some that I can’t.
Re:AI Real-Time Human Full-Body Photo Generator by mkoryak on Thursday August 24, 2023 @12:54AM (#37245149)
I can count a number of things wrong with that image on her right hand. 7 or 8.
Re:AI Real-Time Human Full-Body Photo Generator by fock on Thursday August 24, 2023 @01:50AM (#37245460)
7-finger porn!
Re:AI Real-Time Human Full-Body Photo Generator by speed_spread on Thursday August 24, 2023 @08:01AM (#37247813)
We're not judging.
Re:AI Real-Time Human Full-Body Photo Generator by DaveExeter on Thursday August 24, 2023 @07:24PM (#37255854)
Re:AI Real-Time Human Full-Body Photo Generator by jmkni on Thursday August 24, 2023 @10:28AM (#37249123)
I made the mistake of opening this in work and had to instantly close it
Re:AI Real-Time Human Full-Body Photo Generator by ricardobayes on Thursday August 24, 2023 @03:34AM (#37246070)
I am also a bit disappointed how it's not photorealistic. We had better quality 3-4 years ago.
Re:AI Real-Time Human Full-Body Photo Generator by GaggiX on Wednesday August 23, 2023 @02:24PM (#37240154)
>they're too unstable', a widespread myth

>See for example BigGAN

I remember when you try training a BigGAN model on anime images, the quality was bad. Now look at this example, one single GPU, 1.5M images with a diffusion model: https://medium.com/@enryu9000/anifusion-diffusion-models-for... ,the difference in quality is absurd, you can say this or that is not true but the quality speak for itself, obtaining good quality on complex distribution is much easier with a diffusion model than a GAN.

For example in the case of the site linked they have conditioned the model on poses because you're not going to get anything close to be coherent without them with a simple StyleGAN as they say they're using.

Re:AI Real-Time Human Full-Body Photo Generator by gwern on Wednesday August 23, 2023 @03:58PM (#37241323)
> I remember when you try training a BigGAN model on anime images, the quality was bad

Because there was a bug in the code, in a part unrelated to the GAN itself.

> the difference in quality is absurd

Yes, it does help to train on anime with code that isn't buggy. (BTW, Skylion was getting good results with GANs on anime similarly restricted to centered figures like those samples, he just refuses to ever publish anything.)

Re:AI Real-Time Human Full-Body Photo Generator by GaggiX on Wednesday August 23, 2023 @04:45PM (#37241666)
So you believe that without the bug you would be able to come close to the quality of the diffusion model I have linked? I'm not even asking about using the same compute (1 GPU for ~1 month) but if you just believe BigGAN can come close to that in general.

Also the bug is probably related to the added complexity of training a GAN model in comparison to a diffusion model.

Re:AI Real-Time Human Full-Body Photo Generator by Tyr42 on Wednesday August 23, 2023 @02:52PM (#37240530)
Broken link?
Re:AI Real-Time Human Full-Body Photo Generator by GaggiX on Wednesday August 23, 2023 @03:05PM (#37240709)
Fixed thx
Re:AI Real-Time Human Full-Body Photo Generator by sebzim4500 on Wednesday August 23, 2023 @05:28PM (#37242072)
>And if you're wondering how it can look so good

I don't think anyone is wondering this, especially if they are used to playing with diffusion models.

Re:AI Real-Time Human Full-Body Photo Generator by jug on Thursday August 24, 2023 @08:59AM (#37248237)
Yeah, honestly this whole comment thread is a testament of how quickly generative AI has moved the past year. Even in a hacker community like this where we ought to be close in terms of realizing where current tech stands, we're on different pages on what's possible. Photorealism is pretty much already here and this website is a poor example of current generative AI. The most it has going for it is that it has an approachable UI.
Re:AI Real-Time Human Full-Body Photo Generator by tavavex on Wednesday August 23, 2023 @05:57PM (#37242386)
While there's discussion on the topic here - are there any resources that can explain the exact mechanism of how a GAN works for image generation? I have a rough idea of how diffusion models work, but I'm still no AI researcher.
Re:AI Real-Time Human Full-Body Photo Generator by dublin on Wednesday August 23, 2023 @06:08PM (#37242496)
How it can look so good? ROFL!! It just created a guy with a hand coming off his left ankle in place of a foot, and toes or fingers or something poking out the end of the show on his right foot! https://generated.photos/human-generator/64e682308448b8000c5...
Re:AI Real-Time Human Full-Body Photo Generator by notahacker on Wednesday August 23, 2023 @07:43PM (#37243373)
Four hands is much better for climbing industrial installations so you can hang in mid air next to them
Re:AI Real-Time Human Full-Body Photo Generator by wussboy on Wednesday August 23, 2023 @10:23PM (#37244343)
If it isn't my old friend, Willie McGreg. With a leg for an arm and an arm for a leg!
Re:AI Real-Time Human Full-Body Photo Generator by bravura on Wednesday August 23, 2023 @11:12PM (#37244589)
Hi gwern, asking earnestly and without snark about this:

"And if you're wondering how it can look so good when 'everyone knows GANs don't work because they're too unstable', a widespread myth, repeated by many DL researchers who ought to know better, GANs can scale to high-quality realistic images on billion-image scale datasets, and become more, not less, stable with scale, like many things in deep reinforcement learning."

My interpretation from how this is written is that you are saying: "Researchers say that GANs are unstable at inference time." Is that the correct reading? If so, where have you seen this sentiment? More commonly I've heard that people criticize GANs at inference time for mode collapse and monotony (which comes from training), not instability.

Or are you saying that researchers say GANs are unstable during training, which is a common criticism. Don't you feel this is the case? A lot of different tricks are used for getting generators that are roughly balanced with more powerful discriminators so that the adversarial game is balanced, like TTUR and encodec's weight balancer etc. In this case, are you saying that GAN training is as straightforward as diffusion training?

In my experience, GAN training involves an unholy number of heuristics and best-practices are still quite murky.

I am eager to hear your response.

[edit: my experience with GANs and diffusion models isn't about taking image problems with existing working models and scaling or refining them, but about applying GANs and diffusion models to different domains (audio) and on novel problems with different kinds of conditioning. I would love to see more controlled experiments comparing weak and sophisticated generative backbones as the experimental control and varying the training regime across different GAN + diffusion flavors.]

Re:AI Real-Time Human Full-Body Photo Generator by justinclift on Thursday August 24, 2023 @07:44AM (#37247702)
Trying it out just now, it's still getting major anatomical features wrong, and hands/fingers are still a noticeable problem.

eg (NSFW!): https://generated.photos/human-generator/64e741e6c4c0f80009e...

Wrong number of limbs, wrong number of fingers on hands, etc.

Re:AI Real-Time Human Full-Body Photo Generator by MasterScrat on Thursday August 24, 2023 @04:25PM (#37254086)
I’ll take this opportunity to mention our research scaling StyleGAN 2 to larger datasets (using LAION) on food images, leveraging free TPU compute through the TRC program.

We trained for 36 days on a v4-8 on 558k images.

https://nyx-ai.github.io/stylegan2-flax-tpu/

We were hopeful GANs would beat diffusion models when trained on specific domains. But we’ve now switched to Stable Diffusion and Dreambooth training which has proven more much efficient for this purpose.

I still have hopes for GANs! I miss their insane inference speed.