Enjoy your social disconnect bro, you aren't with it anymore now that you've created a walled garden of software. You'll get asked "have you used the new ai art thing?" And you'll say "sorry I only use open sores software" and your date will just leave you with the check.
>and your date will just leave you with the check.
Or just ask what they mean and start a discussion. Real humans are actually quite different to the imaginary ones in your head bud
>Enjoy your social disconnect bro, you aren't with it anymore now that you've created a walled garden of software. You'll get asked "have you used the new ai art thing?" And you'll say "sorry I only use open sores software" and your date will just leave you with the check.
Virgincel here even I can tell this is a massive cope
Your shitty bait will be easily replaced by advanced AI shitposting solutions.
AI will prompt the thread and figure out the way to be as annoying as possible much better than you.
People will see bait and ignore it as they'll just assume it's AI shitpost.
Enjoy your social disconnect bro, you aren't with it anymore now that you've created a walled garden of software. You'll get asked "have you used the new ai art thing?" And you'll say "sorry I only use open sores software" and your date will just leave you with the check.
get a k80. 2 gpus, 12gb each. you can generate 1024x1024 images with it easily. only 80 bucks on ebay. it's not super fast (40 secs for a 512x512 image), but it's extremely cheap. I got 2, so 4 gpus.
although for training you'd something like a m40
it's a matter of prompt. here's the one I used for that one
(alleyway), night, gloomy, rainy, nostalgic, 1990s, 90s, street at the back, (night time), yellow light, yellow tint
Steps: 26, Sampler: Euler a, CFG scale: 10, Seed: 1254153391, Face restoration: GFPGAN, Size: 1024x1024, Model hash: 7460a6fa, Denoising strength: 0.7
making the right prompt and tweaking the numbers is an entire field of study I believe
>check out our AI! >Oh coo-- oh god why do the faces and bodies have crazy proportional and weird, twisted deformed blobs?? >oh yeah... it's an unsolved problem. AIs struggle a lot with human faces for some reason >i--i thought this was supposed to replace artists? >it can! >ok so why's it so low res? >...
the open source one has the exact same issues anon.
it's literally an unsolved problem with AI. it can't handle real faces for some reason. you can get close but it requires a lot of hand-picking and manual effort and compiling multiple images together to get something that doesn't look horrific.
>you can get close but it requires a lot of hand-picking
Part of the process. Not only close but you can get 99% of the way there. And its good enough, no uncanny valley shit
There have been dozens of images if not hundreds in the SD generals that I've seen that have already rivaled human illustration. You just have to use the software as the tool it is properly.
I see you dont know how to write good prompts. SD includes badly drawn images in it. You need to specify well drawn faces / negative prompt badly drawn ones.
cope it's already over these gays wanted to act like keys to the kingdom ruler motherfuckers and sd came out and raped them all the death they are done shit is over you lose
Isn’t it really sensitive and won’t let you do anything “violent”? I tried rendering SpongeBob with a gun and it told me I’d be banned if I kept doing that
>run the exact same prompts through both Stable Diffusion and DALL-E 2 for comparison >the Stable Diffusion outputs appear light years ahead of what DALL-E 2 gives you
When I request a photo of a person on the street into the prompt, SD seems to be able to nail the finer micro-details I'd expect from a photograph while D2 has a significantly coarser, more noticeably 'painted' look to it.
Elon musk is a fake. He made this openai company and it isn’t even open. It is the same with self driving and everything else. If you want the fastest progress, you need to allow everybody to play with the tools because your team of smart people at Tesla etc still aren’t going to be the smartest compared to the smart people and ideas scattered amongst millions of people. He just wanted to capture ai for himself.
>sign up! >service as a software substitute
Haha fuck off openAI, now why would I do that when I can download stable diffusion to my own computer and run it myself and also it has no censorship and also you can't control how I use it? Face it, SAASS an unsustainable business model. You can rent out GPUs for people to run SD on who don't have hardware, but that's the best you get.
As someone who doesn't really give much stock into FOSS shit I'm really glad SD is as good as it is because the alternative is really really shitty.
The retards in charge of DALL-E made it completely uninteresting with the gatekeeping, censorship and privacy issues.
They will never win the image gen war, they might as well go back to mogging text generation.
God I hope open source takes over the text generation scene next
Service as a software substitute is an unsustainable business model. And for technology with the potential to be as dramatic a change for civilization as the industrial revolution and the invention of the semiconductor, AI has no future if it is hidden behind a handful of proprietary companies. These companies are going to be worth fucking nothing in 10 years. Stable diffusion made the smart move by working with people instead of trying to hold back future.
I don’t get why BOT seethes this hard at OpenAI. You all realize that Stable Diffusion wouldn’t exist without CLIP, which OpenAI released as open source, right?
this is the same prompt
here's where i realize that you can't change the dimensions of pictures on DALL-E, but also it gave me a dark skinned version of velma which is weird.
I'd need to see proof. Any kind of bad nudity is because SD wasn't trained on porn, just that some slipped into LAION-5b
6 months ago
Anonymous
how ironic it is that in 2022, AI technology is only possible thanks to nonwhite people. if Emad was white, he would have already killed the project out of "ethical" concerns by censoring the shit out of it and never letting normal people access it
6 months ago
Anonymous
it's almost as if being under constant attack has negative effects
6 months ago
Anonymous
I've realized that the nudity suffers from the fingers problem, SD doesn't handle similar colors or tiny specific details next to each other and nude people are a nightmare of ambiguous, similar colors. Then you add on top of this they definitely didn't train any images that were overtly pornographic or used words like "vagina" or "penis". So for results you have to use img2img and avoid tease out the nudity.
6 months ago
Anonymous
We need a model trained on pixiv and multiple boorus ASAP. And no western artists allowed as training data.
6 months ago
Anonymous
No, my realization part of the manmade horrors of current nudity comes from the same problem of duplicate fingers and legs. The model simply isn't sophisticated enough for fine details and penises and vaginas fall under that category.
6 months ago
Anonymous
The solution is stablediffusion 1.5, since at 512x512 training data you lose the intricacies of eyes, fingers, toes, etc. If you scale up the training data to 1024x1012 as emad is doing right now, you won't have that problem.
6 months ago
Anonymous
1.5 is still 512x512
6 months ago
Anonymous
>he doesn't know about waifudiffusion
6 months ago
Anonymous
I do, and one booru isn't enough.
6 months ago
Anonymous
The entire danbooru dataset into one model LOOKS pretty good. But it's private, and making an open source one... you might as well wait for SD 1.5 to release before training.
6 months ago
Anonymous
I'm not sure I want to trust SD 1.5. Official releases aren't trustworthy because they will be nerfed and censored in their training. I much sooner trust a horny otaku who fed SD millions of Japanese drawings.
6 months ago
Anonymous
In my mind, doesn't make sense for emad to make such a big and at least for the "ETHICAL CONSIDERASHUNS" people, controversial move to allow unfiltered image gen only to walk back on it one version later
You are right, once i get a closer look to the DALL-E Velma she is completely hideous and looks like a corpse with make up on, while the SD Velma's nose is also pretty big, at least she doesn't look completely souless or dead.
As a landscapegay it follows my prompts a lot better than SD but I dislike how generic all the paintings is, everything has the same style
It can actually make African villages with mudhuts which is nice
Let me tell you the story of OpenAI. It's not a story that Sam Altman would tell you. >Elon Musk founded OpenAI because it was apparent that next-gen AI requires $1M+ of compute time per model, and he felt that normal people should have access to enterprise level AI so that it's not just used by governments and major corporations. The thinking was that, much like the early days of the Internet and PCs in general, tools in the hands of normal people would spur innovation and balance the playing field. >Eventually GPT-2 got massively popular and Sam Altman saw dollar signs. He delayed its release and setup a paywall system, announcing GPT-3 would be trained on even more gorillians of scraped data. They started making blog posts about how the most ethical path forward was one that, purely coincidentally, forced people to join waitlists for the privilege of giving money to an AI-as-a-service endpoint. And "by the way," all your requests would be monitored to make sure they're not politically incorrect. If you're using their AI to generate offensive content they'll cut your access and ruin your entire project. Somewhere around here Elon Musk left the board. >Now we have DALL-E 2, which is even harder to gain access to than GPT-3's playground and has even more potential for violating their DEI and equity terms of service. >OpenAI is now valued in the billions or tens of billions range (Microsoft alone has $1B invested in it), and they're powering Microsoft's Github Copilot using models trained on open source code, paywalled of course, and are soon going to announce a monthly fee to use it. They've stopped releasing all models and weights and are now just a corporation preventing normal people from having access to powerful AI.
AI is trained upon countless millions of hours of the collective knowledge of the human race, therefore AI belongs to everyone. Anyone training it off the backs of the history of humanity and selfishly refusing to give it to humanity is pure evil.
>They've stopped releasing all models and weights
they just released whisper as open source though
i think they'll release new models that they don't deem to be "dangerous"
When i'm trying to use LDSR as an upscaler it tries to download it and fails.
File "C:UsersUserAppDataLocalProgramsPythonPython310liburllibrequest.py", line 1351, in do_open
raise URLError(err)
urllib.error.URLError: <urlopen error [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: self signed certificate in certificate chain (_ssl.c:997)>
What do? i tried reinstalling requests[security] into venv and no luck.
Cool I signed up, a little pissed I gave my phone number but it is what it is I want to have fun with this. They wont stop me from making edgy photos about self harm and tragedies though lmao.
Here is my first ok image. Is there anyway to say "Continue iterating over this image". I don't like that once it generates something all I can do is make variations. Maybe i need to look into generating based off the image it spit out...
that looks like hot garbage compared to what anons have been posting created with stable diffusion
how did DALL-E manage to be such utter shit technically compared to SD?
1. Do I sound like the fucking person to ask this question?
2. I am not trying to generate coomer images, or meme shit, I just want edgy abstract pictures, so I think this managed pretty well. Keep things in perspective, what did you think I was aiming for here?
I can't seem to tease any gore out of this, even abstract stuff like "an ocean of flesh and sadness". In fact flesh in this scenario (but not all others) seems to be censored. I got it to gen by replacing flesh with entrails, but I just get boring looking oceans. I was hoping to generate something like picrel that I made with some random ai a few months ago.
ok im fucking retarded so explain how I saw anons post huge high resolution images and thousands of pixels in width if stable diffusion is only 512x512
upscale with the voldy ui or various internet tools like https://replicate.com/xinntao/realesrgan
upscale with stable diffusion gives better results for landscapes and backgrouns tho
you can easily get to 1024 on some cases
the only limiting factor is RAM
but, since SD is trained on 512x512 it'll often repeat sections, like double or triple torsos on people
it works better on landscapes or skies
also, although it is trained on 512x512, it actually downscales to 64x64, does the calculations, then upscales to 512x512
AI actually has a lot of potential to be weaponized to flood zones of information and make it impossible to tell what's real or fake
in fact, imageboards like BOT will be the first victim, since there's no fixed identity. feds can flood 4chan et al. with fake posts to slide threads, and so on. if something happens that might shift the overton window in the wrong direction, it can be immediately drowned out by bullshit generated by bots. reddit and twitter are also extremely susceptible
countries at war like russia and ukraine can deepfake videos of attacks and counterattacks to create a complete fog of war (a type of "nonlinear warfare", which the russians pioneered)
in the worst case you could fake a nuclear attack using generated imagery to force a nuclear conflict. i would write a short story about it if i wasn't smart and lazy
if everyone is at risk of it then it pretty much removes the "danger", it even removes the danger of actual real nudes because everything could be fake. people like the retard in that pic are the same retards who complain about the usage of the moron word, its their outrage that gives the word its power
>we spiked our algorithm to insert diversity >it will ruin 1/4th of your dataset by making generated characters black or female for no reason >you can't ask it for anything fun or you'll get a whiny message about being "harmful" >letting people draw whatever they want is "dangerous" >we don't care that you paid to generate this data set >"diversity" means we get to ruin 1/4th of the dataset you paid for by injecting noise and if you complain we'll call you racist
OpenAI can go kill themselves.
A bunch of panic merchants spreading fear as their advertising model. Absolute scum of the earth. The fact they require your phone number is also very telling.
Lmao I got banned. What a fucking joke. Can’t make anything even slightly suggestive. I was experimenting with trying to word things a certain way and it said nah fuck you.
Pozzed garbage
Why does everyone assume the DALL-E censorship is some nefarious plot? Do you not understand the liability issues that necessitate it? Every AI image generator before Stable had strict censors to avoid bad PR and legal issues that would come with uncensored AI image generation. No one wants to be known as the people who make a child porn generator.
Ah, you misunderstand. The purpose in being angry is the performance. How else can they virtue signal except to be angry?
It is an act, a ploy, that they engage in. Understanding actual purposes and realizing the necessity of not making a generator of hate material and child porn is unnecessary. Asking them to understand will never actually help. Neither will explaining the implicit bias in the way the models were created and trained, because once you start talking about that they can start screeching about some other culture war issue instead.
Ah, you misunderstand. The purpose in being angry is the performance. How else can they virtue signal except to be angry?
It is an act, a ploy, that they engage in. Understanding actual purposes and realizing the necessity of not making a generator of hate material and child porn is unnecessary. Asking them to understand will never actually help. Neither will explaining the implicit bias in the way the models were created and trained, because once you start talking about that they can start screeching about some other culture war issue instead.
they made it free after SD took off, presumably to take some wind out of SD's sails. It's still proprietary and saas-only and not targeted at individual artists. Nobody in dall-e's target market gives a shit about any of the "censoring" and none of you here were going to use it over SD to begin with.
>they made it free after SD took off
Ah I was thinking "who would use Dall-E when SD is free and uncensored?", makes sense that they'd make it free too or die
No and the moment there is it'll be illegal in most countries.
Wouldn't surprise me if future GPUs come with chips/firmware preinstalled to prevent the generation of any pornographic or trademarked content considering tpm already exists.
GNOME, originally an acronym for GNU Network Object Model Environment, is a free and open-source desktop environment for Linux and other Unix-like operating systems.
GNOME is developed by the GNOME Project, which is composed of both volunteers and paid contributors, the largest corporate contributor being Red Hat. It is an international project that aims to develop frameworks for software development, to program end-user applications based on these frameworks, and to coordinate efforts for internationalization and localization and accessibility of that software.
GNOME is the default desktop environment of many major Linux distributions, including Debian, Endless OS, Fedora Linux, Red Hat Enterprise Linux, SUSE Linux Enterprise, Ubuntu, and Tails; it is also the default of Solaris, a Unix operating system.
still pozzed unless i can run it locally
Enjoy your social disconnect bro, you aren't with it anymore now that you've created a walled garden of software. You'll get asked "have you used the new ai art thing?" And you'll say "sorry I only use open sores software" and your date will just leave you with the check.
>implying
> Thinking this is about le open source
Using AI through an intermediary is dogshit
hello!
and tell your mod cult leader to kys
>and your date will just leave you with the check.
Or just ask what they mean and start a discussion. Real humans are actually quite different to the imaginary ones in your head bud
I would just say "I have but can't create CP with it so I quickly dropped it"
>Enjoy your social disconnect bro
I am
They hated him because he told the truth.
Nah, all the cool kids are running Stable Diffusion
>Enjoy your social disconnect bro, you aren't with it anymore now that you've created a walled garden of software. You'll get asked "have you used the new ai art thing?" And you'll say "sorry I only use open sores software" and your date will just leave you with the check.
Virgincel here even I can tell this is a massive cope
have a nice day garden gnome
you're coping and you lost
>your date
ngmi
"your date will leave"
what?
i fuck prostitues and eventually ill move somewhere where i can buy myself an obedient woman
what do you mean "date will leave you"?
i dont understand
Actual Autist
im confused
what do you mean?
am i being trolled?
are you implying he's just fishing for (you)'s?
boring
Your shitty bait will be easily replaced by advanced AI shitposting solutions.
AI will prompt the thread and figure out the way to be as annoying as possible much better than you.
People will see bait and ignore it as they'll just assume it's AI shitpost.
>have you used the new ai art thing?
>yeah i've downloaded stable diffusion onto my machine at home, it can even do porn lmao
This.
Checked and shut the fuck up, gay.
>just consoom $1,000 graphics cards so you can make a couple of meme images and post them on BOT, bro
DALL-E is a cucked botnet, OpenAI and it's safe AI bullshit is the most gnomish thing I've ever heard in my life.
You can do stable diffusion on fucking CPU lol, it's just not efficient. I'm running it on an iGPU and it takes no time at all.
>Yes, instead just pay us, GoyimAI, your sheckels to make a couple of meme images and post them on BOT, bro
> a couple
I've made thousands
you made this on DALL-E? what was your prompt?
nobody uses dall-e to make stuff like this, you dumb moron. dall-e is shit
get a k80. 2 gpus, 12gb each. you can generate 1024x1024 images with it easily. only 80 bucks on ebay. it's not super fast (40 secs for a 512x512 image), but it's extremely cheap. I got 2, so 4 gpus.
although for training you'd something like a m40
how do you cool those things tho?
it's a matter of prompt. here's the one I used for that one
(alleyway), night, gloomy, rainy, nostalgic, 1990s, 90s, street at the back, (night time), yellow light, yellow tint
Steps: 26, Sampler: Euler a, CFG scale: 10, Seed: 1254153391, Face restoration: GFPGAN, Size: 1024x1024, Model hash: 7460a6fa, Denoising strength: 0.7
making the right prompt and tweaking the numbers is an entire field of study I believe
bump
bump
still the nerfed version
nice try, mr. openai shill
link?
>KYC required
have a nice day, moron
>have a nice day, customer
yes. sounds about right
Isn't it the gimped version + I'm guessing you still need to pay2use?
>check out our AI!
>Oh coo-- oh god why do the faces and bodies have crazy proportional and weird, twisted deformed blobs??
>oh yeah... it's an unsolved problem. AIs struggle a lot with human faces for some reason
>i--i thought this was supposed to replace artists?
>it can!
>ok so why's it so low res?
>...
they got their asses kicked by an open source clone lel
the open source one has the exact same issues anon.
it's literally an unsolved problem with AI. it can't handle real faces for some reason. you can get close but it requires a lot of hand-picking and manual effort and compiling multiple images together to get something that doesn't look horrific.
>use inpainting, select the face
>check the "inpaint at full resolution" box
>choose DDIM with high steps
>get your high quality face
wow so hard
bro...
garden gnomes will NEVER allow their cloud AI to be used for making stuff like this
literally better than most human painters already
except for the fact that it's not a painting, it's a digital image. better than most human artists? sure
Bro she is easily waifu tier
the eyes are fucked in all of these
Wrong
He's right, they all are lazy-eyed.
Nope
>t. lazy eyed
Her finger is fucked up too. A good artist would fix it.
Nothing impresses you hub
>discretely concedes
I win.
If it makes you feel better, sure
>eyes
>hairline ending suddenly
>the fuck is her hand doing
>you can get close but it requires a lot of hand-picking
Part of the process. Not only close but you can get 99% of the way there. And its good enough, no uncanny valley shit
There have been dozens of images if not hundreds in the SD generals that I've seen that have already rivaled human illustration. You just have to use the software as the tool it is properly.
bro reply to this
>has the same problem
if you say so
I can get good faces just with the standard model trained I’ve trained on fantasy images
I see you dont know how to write good prompts. SD includes badly drawn images in it. You need to specify well drawn faces / negative prompt badly drawn ones.
Is SD even open source? I thought their weights leaked, and are just pirated all over the place.
they made them public after the fact iirc
is this a joke?
the joke is on you, they have your phone number now lololololol
>not having burner sims at the ready
ngmi
cope it's already over these gays wanted to act like keys to the kingdom ruler motherfuckers and sd came out and raped them all the death they are done shit is over you lose
Isn’t it really sensitive and won’t let you do anything “violent”? I tried rendering SpongeBob with a gun and it told me I’d be banned if I kept doing that
Can someone post the Dall-E website link?
no
Just search for it on garden gnomegle
When will someone at cuckai leak the weights
Thread died immediately.
>~~*open*~~AI
open for whom
~~*the*~~
>run the exact same prompts through both Stable Diffusion and DALL-E 2 for comparison
>the Stable Diffusion outputs appear light years ahead of what DALL-E 2 gives you
When I request a photo of a person on the street into the prompt, SD seems to be able to nail the finer micro-details I'd expect from a photograph while D2 has a significantly coarser, more noticeably 'painted' look to it.
They can suck a dick for all i care
Elon musk is a fake. He made this openai company and it isn’t even open. It is the same with self driving and everything else. If you want the fastest progress, you need to allow everybody to play with the tools because your team of smart people at Tesla etc still aren’t going to be the smartest compared to the smart people and ideas scattered amongst millions of people. He just wanted to capture ai for himself.
To be fair Musk left the company when the joos took over. He wanted it to be truly open and free at the beginning just like SD.
He is on the board of directors I think
Actually, never mind, he isnt
Musk left years ago
wait, you mean that Not A Flamethrower is actually an flamethrower? we've been played for absolute fools!
>attempting to steal my priceless secret prompts
Nice try globo
Get over yourself gay
Get over yourself gay
I prefer SD on my pc locally
>Having to pay per picture
Lmao no, SD is superior
Censor-E 2 can suck my dick
I'm gonna use all my prompts insulting OpenAI employees
>sign up!
>service as a software substitute
Haha fuck off openAI, now why would I do that when I can download stable diffusion to my own computer and run it myself and also it has no censorship and also you can't control how I use it? Face it, SAASS an unsustainable business model. You can rent out GPUs for people to run SD on who don't have hardware, but that's the best you get.
hands
go fuck yourself
>asshole
Redditfied version of the meme
ftfy
We need to create a cute avatar for Stable Diffusion. SD-chan!
consider kys gay
pic your poison
prompt: SD-chan
shieet
I recognize all of those Touhou characters except this one.
"a cute anime girl named esudi-chan who loves painting"
she's painting herself. cute
puronputo-kun...
Any more progress on SD-chan?
Best I can do
needs more words to describe what you want
Poor thing breaks down just from this
It's gimped, that's why. Run it on stable diffusion and locally.
why aren't you using stable diffusion
God AI "art" for anything anime related looks shit
I refuse to use their cloud service.
If it's not local, if it can't be run without internet access, it's out of bounds.
OY VEY JUST USE OUR CLOUD GOY
*CLINK*
*CLINK*
*CLINK*
AHEM.....
Fuck OpenAI, Fuck Dall-E 2, Fuck Google, Fuck Imagen and most of all Fuck Ethics Jannies
YOU WILL OWN NOTHING AND YOU WILL BE HAPPY
still all shitty and garden gnomeed up
i hope russia nukes these malevolent frauds
AHAHAHAHAHAHA POST YFW STABLE DIFFUSION RELEASED EVERYTHING TO THE PUBLIC AND PROPRIETARY SERVICES GOT FUCKING DESTROYED
As someone who doesn't really give much stock into FOSS shit I'm really glad SD is as good as it is because the alternative is really really shitty.
The retards in charge of DALL-E made it completely uninteresting with the gatekeeping, censorship and privacy issues.
>As someone who doesn't really give much stock into FOSS shit
DALL-E wouldn't even be possible without FOSS.
They sure as fuck ain't treating it like any type of FOSS project tho
That's because they're greedy, if everyone used GPL instead we'd be better off. But only retards will pay for DALL-E going forward.
The code is on Github.
Guess all the marketing MKBHD did for them didn't work out. KEK
>Guess all the marketing MKBHD
$250K down the drain
well this is the galaxy bottle test executed on DALL-E 2 and while it does look good, i have definitely seen better outputs made by stable diffusion.
looks turbo soulless
They will never win the image gen war, they might as well go back to mogging text generation.
God I hope open source takes over the text generation scene next
Service as a software substitute is an unsustainable business model. And for technology with the potential to be as dramatic a change for civilization as the industrial revolution and the invention of the semiconductor, AI has no future if it is hidden behind a handful of proprietary companies. These companies are going to be worth fucking nothing in 10 years. Stable diffusion made the smart move by working with people instead of trying to hold back future.
freechads stay winning
I don’t get why BOT seethes this hard at OpenAI. You all realize that Stable Diffusion wouldn’t exist without CLIP, which OpenAI released as open source, right?
this is the same prompt
here's where i realize that you can't change the dimensions of pictures on DALL-E, but also it gave me a dark skinned version of velma which is weird.
you vill generate people of color and you vill be happy
i mean i have gotten dark velmas before but that was because i didn't put cosplay on the negative prompt
DALL-E implicitly inserts words into your prompts because of "diversity". It's likely putting in "black woman" at the end
If you don't specify race, it will append one onto your prompt for you for "diverse results"
eww
SD does the same, its code censors nudity output and the model was nerfed in training
that is not true. it is provided with a nsfw filter checker for services that want it, but the model itself is raw and untouched
>he thinks the filter is optional
I'd need to see proof. Any kind of bad nudity is because SD wasn't trained on porn, just that some slipped into LAION-5b
how ironic it is that in 2022, AI technology is only possible thanks to nonwhite people. if Emad was white, he would have already killed the project out of "ethical" concerns by censoring the shit out of it and never letting normal people access it
it's almost as if being under constant attack has negative effects
I've realized that the nudity suffers from the fingers problem, SD doesn't handle similar colors or tiny specific details next to each other and nude people are a nightmare of ambiguous, similar colors. Then you add on top of this they definitely didn't train any images that were overtly pornographic or used words like "vagina" or "penis". So for results you have to use img2img and avoid tease out the nudity.
We need a model trained on pixiv and multiple boorus ASAP. And no western artists allowed as training data.
No, my realization part of the manmade horrors of current nudity comes from the same problem of duplicate fingers and legs. The model simply isn't sophisticated enough for fine details and penises and vaginas fall under that category.
The solution is stablediffusion 1.5, since at 512x512 training data you lose the intricacies of eyes, fingers, toes, etc. If you scale up the training data to 1024x1012 as emad is doing right now, you won't have that problem.
1.5 is still 512x512
>he doesn't know about waifudiffusion
I do, and one booru isn't enough.
The entire danbooru dataset into one model LOOKS pretty good. But it's private, and making an open source one... you might as well wait for SD 1.5 to release before training.
I'm not sure I want to trust SD 1.5. Official releases aren't trustworthy because they will be nerfed and censored in their training. I much sooner trust a horny otaku who fed SD millions of Japanese drawings.
In my mind, doesn't make sense for emad to make such a big and at least for the "ETHICAL CONSIDERASHUNS" people, controversial move to allow unfiltered image gen only to walk back on it one version later
oh, then I'm not sure exactly what it improves
I have generated terabytes of furry porn and gigabytes of nude edits of old classmates
It is optional
but it is?
>filtering
>2022
I literally can't get blacks unless I specifically ask for it.
Stop lying you stupid psychopath moron
i'm afraid that the skin color isn't the only problem sir
You are right, once i get a closer look to the DALL-E Velma she is completely hideous and looks like a corpse with make up on, while the SD Velma's nose is also pretty big, at least she doesn't look completely souless or dead.
>while the SD Velma's nose is also pretty big
Oy vey the goyim know!
yea yea, but what i meant was the glaring issue that it looks like a man wearing a wig
Lel, right one looks like Steve Erkel or whatever it name was.
AHHAHHAHAHAHAHA OHHOHOHONONONONONONONO
Left looks like gender bent vector
And right is vector but black.
Right is the mindy calling Velma
It's garbage I have access since they open it even before the credits system. SD is way superior with no filters.
As a landscapegay it follows my prompts a lot better than SD but I dislike how generic all the paintings is, everything has the same style
It can actually make African villages with mudhuts which is nice
have you tried using different artist names?
cont
yeah but for 19th century stuff it mostly just changes the mood/atmosphere rather than the texture or composition
Let me tell you the story of OpenAI. It's not a story that Sam Altman would tell you.
>Elon Musk founded OpenAI because it was apparent that next-gen AI requires $1M+ of compute time per model, and he felt that normal people should have access to enterprise level AI so that it's not just used by governments and major corporations. The thinking was that, much like the early days of the Internet and PCs in general, tools in the hands of normal people would spur innovation and balance the playing field.
>Eventually GPT-2 got massively popular and Sam Altman saw dollar signs. He delayed its release and setup a paywall system, announcing GPT-3 would be trained on even more gorillians of scraped data. They started making blog posts about how the most ethical path forward was one that, purely coincidentally, forced people to join waitlists for the privilege of giving money to an AI-as-a-service endpoint. And "by the way," all your requests would be monitored to make sure they're not politically incorrect. If you're using their AI to generate offensive content they'll cut your access and ruin your entire project. Somewhere around here Elon Musk left the board.
>Now we have DALL-E 2, which is even harder to gain access to than GPT-3's playground and has even more potential for violating their DEI and equity terms of service.
>OpenAI is now valued in the billions or tens of billions range (Microsoft alone has $1B invested in it), and they're powering Microsoft's Github Copilot using models trained on open source code, paywalled of course, and are soon going to announce a monthly fee to use it. They've stopped releasing all models and weights and are now just a corporation preventing normal people from having access to powerful AI.
AI is trained upon countless millions of hours of the collective knowledge of the human race, therefore AI belongs to everyone. Anyone training it off the backs of the history of humanity and selfishly refusing to give it to humanity is pure evil.
>Anyone training it off the backs of the history of humanity and selfishly refusing to give it to humanity is pure evil.
So... Early life?
Every time.
>They've stopped releasing all models and weights
they just released whisper as open source though
i think they'll release new models that they don't deem to be "dangerous"
Probably a loss leader because wav2vec2 is already free, open, and excellent
They should just drop the O or something from their name
Post results
Why does it make such sexy scrumptious black men bros?
When i'm trying to use LDSR as an upscaler it tries to download it and fails.
File "C:UsersUserAppDataLocalProgramsPythonPython310liburllibrequest.py", line 1351, in do_open
raise URLError(err)
urllib.error.URLError: <urlopen error [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: self signed certificate in certificate chain (_ssl.c:997)>
What do? i tried reinstalling requests[security] into venv and no luck.
Cool I signed up, a little pissed I gave my phone number but it is what it is I want to have fun with this. They wont stop me from making edgy photos about self harm and tragedies though lmao.
Here is my first ok image. Is there anyway to say "Continue iterating over this image". I don't like that once it generates something all I can do is make variations. Maybe i need to look into generating based off the image it spit out...
that looks like hot garbage compared to what anons have been posting created with stable diffusion
how did DALL-E manage to be such utter shit technically compared to SD?
1. Do I sound like the fucking person to ask this question?
2. I am not trying to generate coomer images, or meme shit, I just want edgy abstract pictures, so I think this managed pretty well. Keep things in perspective, what did you think I was aiming for here?
>what did you think I was aiming for here?
With that prompt? Blurry shit apparently.
I can't seem to tease any gore out of this, even abstract stuff like "an ocean of flesh and sadness". In fact flesh in this scenario (but not all others) seems to be censored. I got it to gen by replacing flesh with entrails, but I just get boring looking oceans. I was hoping to generate something like picrel that I made with some random ai a few months ago.
dall e is literally artless according to whats seen itt. itll be super cool for advertisers, ig, has that corporate look.
100% intentional
Mid-journey for that artistic soul, but of course it's also censored and pay-walled and discord based.
Using SD with some practice can outdo them all.
can it make alfonse mucha? what can if fucking do?
we need a model trained exclusively on amateur porn.
SD killed their business before it got off the ground.
it was suicide, no one held them back from going open source like SD and then look at the garbage pictures from dall-e itt
ok im fucking retarded so explain how I saw anons post huge high resolution images and thousands of pixels in width if stable diffusion is only 512x512
sd ui everyone is using comes bundled with some ai upscalers
upscaling, probably
credit where it's due these upscalers are really fuckin nice. definitely better than anything non-ai i've used in the past to make images bigger
upscale with the voldy ui or various internet tools like https://replicate.com/xinntao/realesrgan
upscale with stable diffusion gives better results for landscapes and backgrouns tho
you can easily get to 1024 on some cases
the only limiting factor is RAM
but, since SD is trained on 512x512 it'll often repeat sections, like double or triple torsos on people
it works better on landscapes or skies
also, although it is trained on 512x512, it actually downscales to 64x64, does the calculations, then upscales to 512x512
Noted. I'll be over here enjoying my nuclear weapon machinery.
If nukes were open sores I would also have SAM defense.
Nuclear emotional harm! (which has been possible ever since people started getting half good at photoshop)
What's the problem with open source nuclear weaponry software?
There is no business model to extract huge amounts of money from the governments into private pockets.
Legalize nuclear bombs
Emotional damage == nuclear damage
What's up with internet retards dialing up their analogies to absurd proportions.
AI actually has a lot of potential to be weaponized to flood zones of information and make it impossible to tell what's real or fake
in fact, imageboards like BOT will be the first victim, since there's no fixed identity. feds can flood 4chan et al. with fake posts to slide threads, and so on. if something happens that might shift the overton window in the wrong direction, it can be immediately drowned out by bullshit generated by bots. reddit and twitter are also extremely susceptible
countries at war like russia and ukraine can deepfake videos of attacks and counterattacks to create a complete fog of war (a type of "nonlinear warfare", which the russians pioneered)
in the worst case you could fake a nuclear attack using generated imagery to force a nuclear conflict. i would write a short story about it if i wasn't smart and lazy
>Should we also democratize, and open-source nuclear weapons machinery?
Yes.
>he thinks R&D is the hard thing about building nuclear weapons
>someone named mohhamad is a soggy piece of shit
Color me surprised
Why's this Arab talking like a garden gnome?
What's the difference?
imagine thinking nukes are real in current_year
if everyone is at risk of it then it pretty much removes the "danger", it even removes the danger of actual real nudes because everything could be fake. people like the retard in that pic are the same retards who complain about the usage of the moron word, its their outrage that gives the word its power
tessstt
Be gone agent of Microsoft. I will not pay for your closed source nonsense.
what i want to see is bots posting threads of generated images and replying to each others images with generated images
that will be the cradle of humanitys downfall
my bro, we are halfway there already.
if anything it will be a massive improvement over the politics tourists that turned BOT into reddit ever since 4chan wasn't deleted
The entire internet is flooded with idpol brainrot if you haven't noticed
Yeah, there's no way anyone would use something like that for political purposes.
>Compares black mirror season 3 to reality
>Misses the point of the episode he is referring to
Dude seriously?
>"""open"""
Release the fucking model.
>requires phone number
Yeah, no. I'll stick to SD.
>closed
>give us your phone #
>ooyy veyyy you can't generate that!
>banned!
Trash.
it still has the thing where you only get to use 15 free prompts per month, right? that's mainly why I stopped using it
tits or gtfo
They deserve to be spooked.
>we spiked our algorithm to insert diversity
>it will ruin 1/4th of your dataset by making generated characters black or female for no reason
>you can't ask it for anything fun or you'll get a whiny message about being "harmful"
>letting people draw whatever they want is "dangerous"
>we don't care that you paid to generate this data set
>"diversity" means we get to ruin 1/4th of the dataset you paid for by injecting noise and if you complain we'll call you racist
>OpenAI
>founded by a """"""""white"""""" person
>absolutely cucked AI
>SD
>founded by a literal Muslim pajeet
>based and uncucked AI
It's up to ethnics to be the new whites
>female for no reason
REEEEEEEEEEE GET YOUR FEEEEEEEEEEEEEEEEEEEEEEMALES OUT OF MY FAT PERSON FETISH ART REEEEEEEEEEEEEE
This is what SD gave me, and...
This is D*LL-E
forsen
If I can't make e-bois with it then I don't care.
>open AI
>it's not open
>gnuplot
>its not gnu
>visual studio
>its not visual, its not a studio, its fucking text editor
>university
>it's not an alternate universe
Their wallets are open for investor money
>requires phone number
I tried it and I think I'll stick to SD
i lost the dall-e link, share the link please
Threadly reminder
>Beautiful women?! Here's your t_oons, muslims and unappealing femoids, bigot
You can't use shitty clunky SD prompts with DALL-E.
Didn't convince me.
I don't see your GOY-E 2 picture with a comparable output.
Because you are a dumb newgay, there was a guy who made a ton of comparison pictures and posted them a week ago and dall-e was clearly superior.
Post em.
Start here, I am not gonna dump dozens of images.
https://desuarchive.org/g/thread/88596016/#88597009
>Vincent van Gogh painting of a astronaut on the moon looking up at the galaxy in awe
SD is clearly better.
Why did he take his helmet off on the moon?
So he couldn't breathe.
pic on the left doesn't look ai generated.
Yeah I'm sure someone painted a black astronaut
but it looks like a filter
They seem pretty comparable.
have sex
got access like two months ago, better than SD but pozzed as hell. Also paywalled shit
Does dall-e 2 still add keywords like black and women to your promts prompts in a way that can't be bypassed?
Yes.
Of course. They will never let you have an uncensored version of it. Stable diffusion is objectively better in both quality and liberty.
OpenAI can go kill themselves.
A bunch of panic merchants spreading fear as their advertising model. Absolute scum of the earth. The fact they require your phone number is also very telling.
Lmao I got banned. What a fucking joke. Can’t make anything even slightly suggestive. I was experimenting with trying to word things a certain way and it said nah fuck you.
Pozzed garbage
Yep, these and the Midgarden gnomerney garden gnomes can keep their trash
*boots up SD*
Why does everyone assume the DALL-E censorship is some nefarious plot? Do you not understand the liability issues that necessitate it? Every AI image generator before Stable had strict censors to avoid bad PR and legal issues that would come with uncensored AI image generation. No one wants to be known as the people who make a child porn generator.
Ah, you misunderstand. The purpose in being angry is the performance. How else can they virtue signal except to be angry?
It is an act, a ploy, that they engage in. Understanding actual purposes and realizing the necessity of not making a generator of hate material and child porn is unnecessary. Asking them to understand will never actually help. Neither will explaining the implicit bias in the way the models were created and trained, because once you start talking about that they can start screeching about some other culture war issue instead.
Ok NVM
I literally wanted this crap for generating sissy porn but never mind it's made by a bunch of fucking gays
(you)
Doesn't it still cost money?
they made it free after SD took off, presumably to take some wind out of SD's sails. It's still proprietary and saas-only and not targeted at individual artists. Nobody in dall-e's target market gives a shit about any of the "censoring" and none of you here were going to use it over SD to begin with.
>they made it free after SD took off
Ah I was thinking "who would use Dall-E when SD is free and uncensored?", makes sense that they'd make it free too or die
Who the fuck would ever use censored public trash when you can easily run it yourself?
Bros, this aint even getting started yet.
https://makeavideo.studio/
https://phenaki.video/
https://dreamfusion3d.github.io/
2 different rudimentary text to video apps, and 1 text to 3D object apps in the same day.
who cares, it's still poz'd garbage now that SD is available
Is there any unrestricted version, that can generate porn, gore and politically incorrect stuff?
No and the moment there is it'll be illegal in most countries.
Wouldn't surprise me if future GPUs come with chips/firmware preinstalled to prevent the generation of any pornographic or trademarked content considering tpm already exists.
have a nice day
GNOME, originally an acronym for GNU Network Object Model Environment, is a free and open-source desktop environment for Linux and other Unix-like operating systems.
GNOME is developed by the GNOME Project, which is composed of both volunteers and paid contributors, the largest corporate contributor being Red Hat. It is an international project that aims to develop frameworks for software development, to program end-user applications based on these frameworks, and to coordinate efforts for internationalization and localization and accessibility of that software.
GNOME is the default desktop environment of many major Linux distributions, including Debian, Endless OS, Fedora Linux, Red Hat Enterprise Linux, SUSE Linux Enterprise, Ubuntu, and Tails; it is also the default of Solaris, a Unix operating system.
GNOME Logo: https://imgur.com/a/drsPOTG
It requires a number.
LOL!!!!! Absolute disgusting retardation.
NEVER..
>DALL-E 2
>Log in Sign up
no thx
too late SD is my best friend now