#💬|general-chat
1 messages · Page 96 of 1
They don't want to fall behind in the AI race and they seem to have accepted that it's fundamentally difficult to completely remove the possibility of any negative output. Those who have access to running it their local machines are likely a very small minority and already would've had the capability to access resources that the party would be against.
Before, when news were state-owned, you knew that to know the truth, you needed to listen to several countries radio broadcast and cross-check. I think now people are just thinking everything they hear is true.
but truth, is not only about politics
if about politics: truth can be declared from different sides
The long term goal with all of the censorship is to ensure that corporations, the military, etc all have access to unfettered AI that will do whatever they want (right up to and including literally assisting them in mass murder). Meanwhile, we will have stupid hobbled versions that can't even generate a picture of Will Smith eating spaghetti.

Damm general chat going philosophical today
I wouldn't say everything
All it takes is one model leak to end the hegemony of controlled AI
Without getting too deep into politics there's solid evidence that machine learning has been a key focus for them for the better part of a decade. Especially for upholding their values.
Know thyself 
I dunno know who brought politics into this; I just commented on an interesting conversation about democracy. And honestly, politics is everywhere, including in AI. Oh hell, it's definitely present for sure.
Nah
Who are you talking to right now, Skyler?
@fervent thunder back in the early testing days, the wildest things were posted here
You think AI is some tool for you to generate booba?
AI is a fucking national security threat of the highest degree.
There will be no leak of that sort. Not unless you're talking about the type of leak that will end up getting a spy beheaded.
I think because today was announced news about SD3
Most likely lol.
and are left waiting for application for the beta to be approved.
Meanwhile the nuclear stockpiles of the US were leaked on a flashcard app by soldiers responsible for their management
military documents get leaked to warthunder players
generally... yes
@craggy granite yeah knowing this server probably
Weren't the nuclear codes 000000 for a long time or something.
.
Those documents were incredibly useless.
Overclassification.
Oh god I never heard of that one
For 15 years during the Cold War, the code used to prevent an unauthorized launch of U.S. nuclear missiles was eight zeros: 00000000. It wasn’t until Bruce Blair, a former Minuteman missile launch control officer, alerted his superiors and members of Congress to the dangers of maintaining the not-so-secret codes that the Air Force finally reprogrammed them. When you take into account the amount of destruction a single nuclear weapon can cause, these codes represent a terrifying reality.
Where you got it?
the models are contained in mostly just a handful of files that are just a bunch of numbers, leaking them would not take much
is it site that can be trusted?
I wonder how much Vram SD3 will need? I would think a lot? I would be amazed if it runs under 18Gb of Vram to be honest.
Yes, I was curious why they didn't mention that
I would be amazed if it runs on my old 8GB card
The code is but one of a long series of safety measures tho
Cries in 12GB Vram
Really?!
Cries in 10
cries in 8
That's a not for profit anti nuclear weapon organisation, which could give them a bias.
https://sgs.princeton.edu/00000000
There's also this one from Bruce Blair themselves. They give quite a few sources.
"the soldiers used study apps such as Chegg, Cram, and Quizlet to create flashcards where they stored information about bases in Europe where US nuclear weapons were likely located, secret codes, passwords, and other details about security. "
smartest puppets NA
But yeah, @fervent thunder is right in saying that it's just one of a few measures, still very funny and somewhat concerning though.
@arctic sedge we thought that SD 1.4 would've taken a lot of VRAM when we're just testing it but we've been surprised that it can be ran on 8gbs VRAM, so SD 3 it might just run under 12gb VRAM hopefully
And not to mention that the VRAM requirements of old models continue to decrease with 500 IQ developer shenanigans
I hope so. It means i can then run it before upgrading to a new GPU. 😅
But if we compare it to SDXL, SDXL only barely runs on 8GB/10Gb.
SD3 looks like a whole different beast altogether.
I never ran SDXL locally so I don't know much about it
If SD3 uses T5 XXL...
yeah. It's pretty Vram intensive.
AI is developing too fast for my brain to handle, like not too long ago SD 1.4 used to be the best for AI image creation
encoder on CPU, tiled VAE, u-net on GPU might still be fine, I guess?
You're not the only one m8. I think we all feel like that.
We don't even have Stable Cascade yet! And we already have the announcement for SD3!
You think that would lower the Vram cost?
Just noticed the news about SD3 too
Is SD3 a continuation of 2 or XL?
Anybody know? Someone dropped a link but it was just to the announcement and it didn't say
Yes, doing those do save on VRAM
Seems like a different structure.
@arctic sedge like I remember when I first heard about stable diffusion and I thought "oh cool it's like a better dalle mini"
It's an entirely new architecture apparently
Oh nice
Yes m8. It's a full upgrade on architecture too.
I should clarify, it seems like a different internal architecture.
We don't have many details since we don't have the technical report yet.
Once we have that we'll be able to clarify more things.
It's so realistic that the booba generated with SD3 pop out of the screen and hit you in the face
I'm assuming it incorporates a bit of stable cascade into it.
They say transformer diffuser, I assume that isn't referring to the ViT
Ok, I didn't know if I just wasn't finding it or if it wasn't out (sorry, this was meant to be a reply to Banana)
I'd say they would've taken lessons from training Stable Cascade and utilised them within SD3.
Yeah haha! Good Old Dalle-Mini. Also now known as Craiyon
Imo this is the best way to look at it.
SD3 is the next step in Stable Diffusion as a whole so it's a little hard to pick one point to reference it against.
Sorry if that just adds to the confusion 🙂
Could've been vice-versa as well, not sure which one was trained first but Stable Cascade's underlying architecture trains a lot faster.
Emad said it's a bit like SORA in it's architecture, so think a diffusion model with gpt tokenizer, btw that is a gross oversimplification but you get the idea
So is it messing with the prompt internally afterwards?
Cascade & SD3 serve different purposes - Cascade being based on the Würstchen architecture & not being a "Stable Diffusion" model necessarily
Link for Würstchen: https://openreview.net/forum?id=gU58d5QeGv
no, it clears it up
I don't know if you're cleared to answer this or if you can even predict it, but is there any approximate timeline for SD3's public release
I think the only issue was going from 1 to 2 originally
Yup yup.
if we just assume each new step other than a decimal won't be the same architecture, that's fine.
soon™️
It would be nice if @wise stratus would take the center stage and tell us more

Guys, the answer is always Soon ™️
It could always be worse, they could be on Valve Time.
The good old SOON
SD3 releasing alongside halflife3 confirmed
CS2 is pretty great tho
I don't care about toys anymore. I just want Lord Yahweh reveal Himself by coming down from Heaven in his Superintelligent form.
Yeah, that's true. Everything they do takes absolutely forever but it's usually always worth it in the end.
Half-Life: Alyx was definitely a long wait but it was worthwhile in the end.
Generate Half Life 3 with AI.
Half Life: Aiyx
Only for it to be taken down by valve
You kinda can, not even that hard I think
if they can't make me an 4k image of an ethereal monstrosity, I'm not sure I care
It's possible since I think Marc Laidlaw released a synopsis of the plot of Episode 3, just with modifications to the characters of course lol.
"Gordon Freeman, massive boobs, by greg rutkowski"

You generate the assets one by one. 2D assets, and 3D assets. Have another AI write the story. Have another AI generate the animations.
I wonder how good would the ai would handle this prompt genuinely
There is a ton of work for using AI to generate various video game assets, generate good motion capture, even better graphics, etc.
We already have the story (kinda). You'd just need to write the script up.
Valve avoid Half life 3 like the plague because of extreme high expectations
NVIDIA has a tool for automatically upscaling game assets to be mega high res.
when do we try stable diffusion v3?
Yeah, the employees there are allowed to work on whatever they want.
And I can imagine living up to the expectations would be very daunting.
Especially as the release was further and further out lol.
Not all the people are..
just the majority of em 😄
unfortunately, that's more than enough in most of the lands

Morning from southeast asia
Oh god I stayed up all night again.
evening from Central Europe.
Hey mr fox, there is someone want me to finish job with stable AI before its getting commercial
Night from the middle east
I though you are from north, sorry for bad judgement
Philippines?
Seems like you old friends from USA
wittle baby, I regularly go more than a week without sleeping
That's not good.
nah , PL is located in the exact center of Europe
That very not good
Isn't it hallucinations post 3-4 days of sleep.
Yes, I lived in Cebu for a bunch of months
yes, sometimes, especially if I take my 14000mg of THC gummies
Why even bother with Stable Diffusion, just hallucinate shit instead.
I was in the army, and we sometimes go without sleep for weeks, but it really messes you up
You reminds me of Sniper Netcafe at 2004
Ah yeah, the army is a rough environment.
Pretty sure the center of Europe has a building that has a black sun image in its center 
Stable Dream when?
Especially the Iraqi army
I'm not sure it's a good idea to keep your army awake for extensively long periods of time. Humans need sleep.
Anyone here interested for contest?
Because it's very hard to control the hallucinations of the human brain.
Actually hallucinating irl + using Stable Diffusion is where it's at!
Bro we used to be beaten before evey night so I got used to it
This might explain the morale issue during a certain invasion.
World War 2 happened on a lot of meth : )
Do you know what hallucinate really means? Already checked for oxford?
what do you mean? what's your point?
Surprisingly only the republican guard used meth back in the day and not the army
Thats a chinese proverbs
You was controlled by china
I'm controlled by the NHK
Nippon Hikikomori Kyokai
the Japanese Hikikomori Association
Do you understand cinematography?
What is K talking about?
Cinematography, apparently.
idk
I am talking about artificially manipulated outputting hallucinate, thats chinese proverbs. They are the makers including Japan & Korean
I feel like I'm reading adlibs
Do you know what server your in?, this isn't a server for sharing riddles
I'm not sure what you're referring to
How can I find more information about that?
Are you talking about monks manipulating their own brains in order to hallucinate?
Or are you talking about state(governmental) brainwashing?
But mr.fox was true. It just about cinematography, AI helps them
Inception movie was the best cinematography film before Generative AI was built
since 1945 that's no longer the case.
Times change, never know

You must see what was upper level was done, and now i thinking about underground
We cannot leave back them because its all about stability. Its called urban paradox
Marginality and outliers
There are two types of outliers
Alright, calm down there Confucius
If you ever facing school, when the final score reported, when there are outliers, the outliers will see they scores it upped or maybe downed. Do you ever faced that?
Cannot be the bestness or cannot be worstness
Yeah, the black sun has become very black recently , so I'd rather opt for calling that fairyland a center of New Africa or New Middle East. And I wish them all the best, times are hard these days ..
I want to iniate contest to outcome equilibrium prize
Black hole sun, won't you come
And wash away the rain?
Stuttering, cold and damp
Steal the warm wind, tired friend
Times are gone for honest men
And sometimes far too long for snakes
Can i send images?
Only if you're a God like me. 
How many you've God?
137.042
Crypto?
Too few if its said world
Crypt never be the world never be currencies. It just like the story of pirates search for treasures
https:// x.com/EMostaque/status/1760725050095747249?s=20 oh that looks promising !!
Crypto is in fact dead.
All money is dead.
Last year, OpenAI [in a collaboration with the US intelligence community] has developed AI that has cracked cryptography. It's called Q*/Qualia.
Additionally, OpenAI [and others] can make trillions from predicting the stock market.
But they won't because it would crash our civilization. The civilization is still a good thing to have. Because they'll just keep developing Superintelligence. When ASI is develpoped, that's when it's time to crash the world.
Where can I see stable diffusion 3 generated images
Yeah I need more
thought so we all want more of it
Crypt is vault. Not all dead in bunkers
I join here, hoping to get more news soon... 😉
does anyone know how many tries before this example image was made?
from SD 3.0
OpenAI has a blog post that is very old. It talks about the fact that OpenAI was already using an AGI detector; even many years back. For example, the AGI detector automatically scans if there is a crazy winner on the stock market (because this is an indicator for that group having AGI).
I was thinking to test their prompt on some sdxl model to see if we can get near ...
Are you using a translator? It's a little hard to understand your English sometimes.
Very idiomatic, but if you know the truth, its really miner manner
It digging
If you know economic bubbles happened in Japan 1991
Muse - Time Is Running Out
SD3 channels?
If you have consciousness about that you wont be rich person, because in that time all japanese are billionaire
They already known when just only one country is too rich
So, there are so many statement from Old Japanese dont want to be rich person
Stability is need
1991 the money become soap. So thats called economic bubble
Nowadays called burn money
So crypto and blockchain comes as a disappointment of currencies system
Is SD3 going to be released under the same license as the others?
But nobody really understand about why need to crypt
Important Considerations
"Richness" is relative: Even in a rich country, there would still likely be varying degrees of wealth. The key is eliminating desperate poverty.
Human Nature: Conflict, ambition, and the desire for status are inherent human traits. A society without poverty wouldn't eliminate all social issues.
Global Context: A single wealthy nation in a world with poverty raises ethical and geopolitical questions about responsibility and interconnectedness.
Let me know if you'd like to explore any of these scenarios in more depth or consider additional factors!
bruh i'm beginning to have same feelings i had when sdxl debuted and i stopped saving sd1.5 cause i wanteed to preserve sstorage space for sdxl. i'm beginning to look down abit on sdxl, just a tiny bit 😩😭😭. but i mean why even save anything now
So one day you can look back at yourself (3 months from now) and realize how primitive your software was
bruh when i look back at my sd1.5 pics oh laud i feel like crying for my dumbass i used those gaslight myself to ridiculous thinking they were the shit 😩😭😭😭
"The fap material of yesterday is the junk of tomorrow." - Stable Confucius
Did they accurately CAPTION image datasets for SD3? Otherwise same garbage in garbage out
Anyone know enough about flow control to say if they could get positional alignment correct without good captioning?
yooooo i gotta screenshot this gem 😭😂😂😂
Rather good timing for this announcement with the Gemini image thing going on at the same time. 🙂
Really looking forward to trying SD3. Making me think of early days of SD.
man this looks amazing. does anyone know what the hardware requirements are gonna be like?
5090 SLI 
lol. damn i have a hard enough time just trying to find a build for a dual 3090 with nvlink lol
20GB VRAM (?) requirement
but there will be different sizes of the model
plus people will improve upon it
bruh with what people did with sdxl, sd3 finetunes are gonna be insane. i always imagined in my wet dreams of finetuned dalle3 now my wet dreams about to come to life...i just hope its not a memory sucking vampire like sdxl that took alot of compute times
It will come in both SD1 and SDXL sizes
SD3 want want want
Hopefully Nvidia releases household graphic cards with a humane amount of Vram
There was a news about NVidia pushing very good financially. They know how to save with the vram lol
tempted to buy a 4090 but will need to see how difficult it is to fine tune first 😋
they should make a base card with pluginable addon memory
if it were only that easy
me too ^^
How the finetuning is done? I recently got 3090 and curious to know if I could squeeze more speed out of it.
Well I actually do my finetunes using a paid service called RunPod which spins up a virtual machine over the internet. But the cost is relatively cheap. So for me with SD3 it would be how fine-tunable is it and how much does it cost.
omg sd3 
The 4090 at home would just be for me to do inference 😋
I try to finetune on my 3090 but run out of memory, then I rent a A6000
is there any vague even eta for 3 release?
id say around 1 month, they are pretty fast with this stuff
but pure speculation
i hope you are right. it sounds promising but so did they xl and 2.0 and both were duds imo
at this moment everyone wants lol
OMG i just checked ig it fake
april fools is still more than a month away
are there any more images from it or just on the announce page? they are ok but not really mind blowing imo
lmao yes
not sure if this is a tech support or general chat catagory.. Can I run any stable diffusion model without a UI? like can I just do something on a cmd?
ask in both places i guess haha
haha. well alright then
I fear it might be even more - which would mean that not even the highend "normal" users could try it until it gets an 8 or 4-bit version or something like that.
how much is in a 4090?
damn I got scared 
24GB
but remember that your PC also uses a bunch of VRAM auxillary
if you gotta ask you cant afford lol
something like $2,000 USD last I knew
Random question, is SD still like searching the internet for images? or is it like actually making stuff up from scratch?
its been a year since I messed with this stuff
what if SD3 is just early april fools joke? xD
i mean how much vram but thanks for being a jerk
I hope they do that 😭
OH LOL
its like 20 or 24GB
i totally misread. my bad man
Then what does that make Sora?
Oof gifs dont play here
there are channels for that
I wonder how much vram SORA needs to run
Aaaa, ok. So you mean finetuning a checkpoint, gotcha. I thought you mean to finetune the gpu, like core clocks, mem voltages etc
More than the number of cells in your body
🤣
probably like 5 a100s haha. they say to generate a one minute video for openai they can go out for a burrito and it will be done when they come back
so i dont think we be able to run it on any pc soon
and if you notice, most video they release are not even 1 minute because it take so long to generate. the woman who is walking is one that is a minute but not many others
what's the difference between cascade and sd3?
prompt adherence
also SD3 is using the same tech as Sora
so we almost at cap for non-professionals dudes in large models local training for image/video ?
damm its sad
bruh, do early member or beta member roles mean anything these days? Can beta testers just go and try 3.0 in some bot channel or that's impossible?
i see lots of prompts with < > and "lora" text inside with numbers. what do they mean?
can someone explain me in 2-3 sentences why AMD can't into model training?
its like always about amd drivers problems?
or training tools not support AMD?
rocm sux, that's why
🫨
and it's supported by the few gpus
AMD GPUs lack native support for CUDA, limiting their appeal to developers accustomed to writing code for CUDA and requiring those developers to either switch to AMD's own ROCm stack or continue developing for NVIDIA GPUs.
didnt they just make a rocm thing that works with cuda?
they did yes
it's called ZLUDA
and AMD github support sux even more, they don't care about issues or PRs, some issues are few years old.
I love AMD, but man they have some stupid things they need to fix
perhaps NV pays them to drop the ball on Ai, that's the only option I can imagine
well the CEOs are sorta related
for otherwise why would amd suck so much
the big money is there and their hardware is powerful
and big vram
I've kinda given up on AMD, intel would be the next challenger imo
battlemage drops at the end of this year
yeah you know microsoft is itching to put that in xbox series whatever
i dont see it being as performant as AMD though.. they are kind of REALLY good at low power draw for the CPUs performance
intel drivers are improving dramatically every few months
microsoft's xbox strategy is weird, hard to say what they will do except the clear focus on cloud
not as weird as google though, gemini is so botched with image gens that they had to stop it
LOL yeah i just read how it wouldnt create images of white people LOL
try asking for vanilla
but make the most ridiculous images of a black family lol
google really makes me want a pixel phone.. but i keep reading stupid crap about their bugs. which im not sure if samsung has on their phones...
How do you see training on home equipment in the future, what can be counted on and what not, what will be the main improvement factor in next 6/12 months
- optimization of models/libraries ?
- Nvidia will release cards that will drastically increase performance for training ?
- breakthrough in training on multi-threaded CPUs ?
again. google is also weird
i'm waiting to see how RCS support rolls out on iOS before I make the jump
Sora is your goddess deity.
meh... doubt it haha
You'll see Her open her wings wide... soon. The Third Impact is just months away. Then She reaches for the Moon.
what in the stable diffusion are you talking about lol
The future of the human race.

Idk why they do that, even dalle3 did something similar some months ago
Wait, looks like the director or something like that behind Gemini is a racist against white people
💀 💀 💀 No wonder why this happened
Gemini is proof that censorship is better off avoided TBH.
gemini is the victim here heh
I hope @wise stratus learns from Google's mistake and releases a SD 1.5-grade SD3.
LMAO
True, any kind of censorship is bad and results in things like this
breakthrough in training on multi-threaded CPUs ?
I wouldnt count on it any time soon, unfortunately. Unless it's Saphire Rapids sorta stuff
optimization of models/libraries?
hopefully new architectures, quadratic mem consumption in transformers/diffusers sux big time, bruh.
Nvidia will release cards that will drastically increase performance for training ?
maybe, but these will be drastically more expensive. NV don't want you to train fast, cheap and efficient. If that was the case, the current MFU for the most hyped stuff would be 90% instead of 30%. That's less than a Stirling engine charger made of a beer can and a candle, for $30k a piece.
If not, then we'll have to go to NovelAI instead.
Same with LLMs, chatgpt for example could mention "good things" about some left-wing dictatorships in Latin America, but refused to say the same about some right-wing presidents who have been democratically elected (at least I tried asking it like 5 times some months ago and the results were the same)
nvidia can make a 16GB vram card for 450 bucks but 24GB vram costs 2000 bucks, they cant put in more vram because its too "expensive"
they cant put another 16GB onto the 24GB card for another 450 bucks because vram is too "expensive"
they need to charge 8000 bucks now for cards like that, the cost of vram is very high!
AMD should make a 32gb VRAM for under 1500$ so Nvidia stop sleeping
That wont happen tho :c
meybe there would be cloud gpu price wars ;D ehh who am I kidding lol
i think tho, now that mining has kinda gone away and that didnt need much vram, and AI does, they are going to have to address this in upcoming GPUs, finally, hopefully.
its also super important to get rid of blower card designs and nvlink from consumer cards, you need to spend 8000 bucks on cards like that now, even though you didnt before, its not because its undercutting their very expensive cards, nvlink and blower designs are very expensive! Now!
that vram would magically become "inexpensive" if somebody leaked their rsa keys for vbios, so we could drop our own shit there and solder additional 24G vram for cheap.
thats just a matter of time really
they make 24GB for $999
i'm sure they'd release drivers that recognize the chip and disable that or something lame
then we'd release nouveau hacked drivers just like ppl used to do back in the day when crypto mining was a big deal.
If amd released a gpu with 40GB Vram at a reasonable price (~1500$) I'd switch and never og back
then it won't work with new features in redshift or vray or whatever else you are using
only useful if you are using it to do one thing
maybe
Well,imo it would work good for training/using AIs (i think)
you already get issues where AE breaks if uou use the driver recommended for VRAY or redshift etc, they all work best with different ones
vray spefically does NOT recommend ever using the latest drivers
but if you can drop your own firmware into the chip, in many cases you can emulate the hw that is already legit, like A5000 or A6000.
same architecture as consumer gpus after all
its also not really any faster, just uses less power, has more ram and can pool the ram. Maybe data bandwidth is a bit faster.
someone tested new sdxl Lightning with cfg 1-2 and steps 2-8?
that's the whole point , the only thing you need in Ai is essentially VRAM, a huuuge amount. for other tasks 16GB is more than enough.
testing the dreamshaperXL version that has it embedded
When do you think 32-48gb of vram would be the standart for low end - mid end GPUs? Maybe 10 years? 😿
well, you know, technically speaking you can buy 12 channel mobo....
and use DDR instead
Im trying to use regional prompter and in the known issues section it says i need to modify the code of gradio.components.Image.preprocess, however i cannot find this file? does anyone know the location
who can say
perhaps next next gen
so like 2-4 years
5090 will prob have 32GB vram, you heard it hear last
Did anyone got approved for testing SD3 or is everyone still on the waitlist as far as we know?
Or 24
One of the rumors is that it will remain at 24 gigs
Yeah, eventually
At this point, I am more tempted on renting cloud processors on google cloud. I certainly won't use SD more than 3-4h a week (it's just a hobby for me), so even renting an A100 is less expensive than buying a computer for that around a 4090 (that might become obsolete if 5090 comes out with 48 GB... [dreaming])
I dont believe it will be 48gb, maybe not even 32
doubt. Having the encryption key doesn't "magically" make the home brew community good engineers. Drivers and firmware still have to operate together. That's a lot of low level software engineering.
i think eventually, cisc processors will have support for all these processes and we'll just use ddr5 or ddr6
128gb ? yes please
the gpu tiers reminds me of the ye olden days of 486 processors where the version with the math coprocessor was more expensive to buy, but it was actually cheaper to make because the cheaper (to buy) version had to undergo an additional manufacturing process to have it disabled, making it more expensive to manufacture. A sales/marketing decision.
math coprocessors were hot
then intel made mmx tech, which did color math
woah
ah the good old days
I remember being amazed as a child when I saw a VGA game for the first time...
sierra games
gpus should start with 64GB vram. Its such bullshit. There's no reason not to do this any more. People need vram. Mining is dead. The excuse to not do this is done and gone.
Lets see when 5090 will even be released, rumors say by the mid 2025
NVIDIA wants to protect their sever grade chips so they won’t give consumers grade cards high VRAM
theres still a huge market for current cards. they'll plan a release date when sales curve off
64?! 16 GB is overkill for majority of people already xD
its not like amd is catching up to them
I've read competing rumors : late 2024 OR delay to 2025 but with a 4090 super or ti version in-between
yes thats what i'm saying.
Yeah
Because consumers dont need them
games would use it if it was available.
And they wont make exception for the one or other freak
doesn't avx-512 or avx2 or whatever the big new instruction set is, doesn't it have ML acceleration in it?
Their profit is mostly from server now@@ doesn’t look like we gonna get high vram cards anytime soon
They can barely even come close to 32 tho
Is AMD with Zluda a competitor right now? Or are they still far behind?
Does anyone know what’s the requirement VRAM on the new Stable Diffusion 3?
Ysah but as said none consumer needs those
no, but they have different models ranging from 800M to 8b parameters, so the higher models will probably need 20+ GB VRAM.
at least initially
Allegedly there is a 3D mesh AI generator that wont work properly unless you have...40(!) GB of VRAM
Don't you guys think the Green Goblin was great in American Psycho?
https://en.wikichip.org/wiki/x86/avx512_vnni this stuff. the vnni instructions. is there hopes and dreams there?
Maybe we will have quantized or offloading like cascade thing making it need less VRAM 🥲
Until i get my beast PC SD as well as competitors will release more advanced stuff
myth. Game developers largely target consoles. Star Citizen is the only high end pc project that's trying to use everything pcs can do, really.
what now? 14 years in dev?
lost track
Its not feasible
To go for the extreme
way more pc gamers than console gamers
Wrong
what would games even do with 64gb of vram?
PC gaming is a niche
thats so incorrect
Behind consoles
there's more PCs than consoles sure
Maybe if the next gen console gets higher VRAM like 16g then we might get a bump on mid range PC Card
Statistically PS and XBOX have more gamers than PC
just googling this, happy to be proven wrong, "PC and console, meanwhile, were at $45 billion and $30 billion, respectively— and data available in charts at the source reveals PC revenues outstripping consoles since 2013."
"Is PC or console more popular 2023?
Together, PC and console games will generate revenues of $95.2 billion in 2023—or, 51% of the entire games market. These revenues will be driven by 892 million players on PC and 629 million on console.Aug 31, 2023"
higher profits in that arena too by a wide wiiiide margin. and we're not even considering mobile gamers. the true giant market
Dunno about profits, but agree for the rest
Pc gamers are generally worse customers, we hack pirate and complains more and pay less, compare to console and mobile 🥲
i think you guys should all be apologising to me for your factually incorrect statements and saying "sorry mr fabricated girls you were so right, and good looking"
PC gaming, a hefty amoutn of the profits is roblox
minecraft is one of the most succesful pc games of all time
PC still masterrace ofc
i only have a switch so that i can feel better about playing the roms on my deck
But I guess it earns less money then Ganshin impact 🤔
Love Minecraft
another big part of the pc market is whales. WHAAALES. so many whales. what is even going on with tf2 hats and cs2 knives?
Anyway, most people dont need more than even 8 gigs of VRAM
Ahh the CS gambling ring
16 is pretty much the standard
Groq consumer cards couldn't come any sooner i swear
Pretty sure that is designed only for server exclusively, it can’t do any game stuff
Yea I mean for AI usage of course.
But those arent consumer cards
Thats why i said they couldn't come any sooner 
If they did exist, it would be very nice
They introduce AI into consumer grade cards and CPU but thats not like H100 and others
None consumer would even take advantage of those
The one freak that does isnt enough for them to even try it
is there any data on numbers of gpu users broken down by usage, games, AI, 3d rendering, editing, photography, excel users, mining etc etc
groq is what? elon's endeavor? it'll flop. he's got those chips in teslas too, but the front end is so pissy for people. it's all locked behind SAAS too
he went hard on trusted computing to prevent people from hacking them
Anyone here who has experience with sending files to the instance from google drive on vast.ai? I am a newb to this and I have no idea how to do it.
basically vaporware as a consumer product
and when they do come, it'll be the X premium he loves to charge for. His satellite internet that was supposed to democratize internet access, has ridiculously high fees and isn't even that great
Elons been talking about a partnership between X and midjourney
meh
SD3 looks good gotta say, also Emad mentions collaboration, composition and more
Midjourney is definitely far, far ahead of any SD model at this point
judging by text to image purely?
yeah obviously controlnet and custom training allows you to do things that midjourney cant but very few people know these tools well enough to get results as good as MJ can do by default
basically all the artists and photographers and filmmakers and designers are using MJ and all the engineers and furries and gooners are using SD, with some overlap.
true, its to be said people might also use more than one product
one can also argue competitors like Dall E and Adobe Firefly win easily by some standards as well
https://habana.ai/products/gaudi/ intel has this hardware in play now but it's enterprise not consumer
especially in the future
amazon and microsoft are building their own ai accelerators too
and jim keller has his company with the greyskull hardware. expecting a next generation of that gear soon
They going hard idd
Gaudi, Würstchen. Man those names
like more or less every filmmaker i know, and i know quite a few, uses midjourney now to do their pitch decks, storyboards, all that stuff now. Exactly none of them use SD. Big commercial cinematographer friend who does endless huge car ads, these ads are now recreating midjourney storyboards/style frames in real life. Every agency uses MJ. Every film production company. Its taken root. None of these people give one fuck about Dall E and maybe some have tried firefly but the vast majority just use MJ. Its easy and gives the best results in these fields.
got any proof?
doesn't even give a price for the 256gb gddr6 card https://tenstorrent.com/cards/
proof of what
well i know personally steven spielberg and he uses SD
sucks to be them. that's a quantifiable bill they're paying ot MJ for that. Seems like buying an apple cinema display just for flex power. something i know a lot of that field does from personal experience
Anyone here open for southeast asian language?
why, its like 40 bucks a month or whateverm its nothing.
Prestige Worldwide
i dont know any filmmaking company that uses MJ or gen AI as a standard part of the workflow
$40 per user per month
literally nothing
Thats cinematographer work
didnt know studios paid for a personal single user license
if it saves them half an hour, it has paid for itself...
excactly
they get sued out the butthole if they try to pay for anything less than their head count
You must think about inflation. Reduce cost, reduce inflation
STEVEN SEAGAL GAMING thinks I'm a liar for some reason, just because i'm using a fake name here for my stupid porn twitter account i made as a joke
MJ isnt nearly as much adopted as it sounds
if they're giving that access to 5 people, they need to pay for 5 licenses lol. Some business dudes need to talk to qualified financial and legal help
but definitelly more than SD
its true i was there i can confirm
You must think about cheaper by the dozen. High price is just barriers from minority level of high profile
I don't think that it is entirely MJ, the truth must be somewhere in the middle, as I have had a couple of companies using ICBINP in their workflows ask me about things, so I can confirm it definitely isn't 100% MJ
You must think how to make everything affordable like manufactured, be exclusive in your own part just for equilibrium of the world not for egocentric
why do you get to decide why he thinks you're fibbing?
maybe he saw a tell
I'm so confused right now 😛
oh shit. he confirmed. mb
Stable diffusion
Yeah, it just a model not reality, to avoid trial and error
not saying 100%, just that it has a stronghold in specific areas
And those licenses are NOT standard consumer licenses
whats the "tell"
i think they allow any subscription as long as you dont make a milly a year
if you know someone's tell, don't tell them what it is 😉
Which would eventually mean only certain companies use MJ in the first place
This is trust issues
you gotta know when to hold em
Groq =/= Grok, they manufactor AI specific chips https://groqerc.com/.
thanks!
Midjourney isnt even the company that targets the companies
Its a certain other company
why do you think i'm lying and what do you think i'm lying about?
https://groqerc.com/#token its a block chain startup, so disappointing
Bloodlines, and closed relationship. Interchange is the solutions for another trees
u not lying i just said it was true,im steven seagal and i can confirm im the creator of the furry model along with steven spielberg
knowww when to folddem
know when to walk awayy
hey guys, pleasure to be here, someone knows if stable chat has a public api?
Do you like icecream?
Dall e was closed
Better not count cho money when youre still sitttin at the table
Chinese proverb
I just selling e-ticket
There'll be time enough for countin when...
the dealins done
2 months later
not sure i follow this
I just need 100$/months to enabler my research using gemini AI, cloudshell, vertex, making landing page for CTA, then planning to community interchange by generative AI filmmaking contest. Anyone want supports me? This is urban issues
I am thinking about outliers to stabilize and stay underground
always a pleasure to meet real fans
cant be good to be a therapist, but psychiatrist is hard to say/spell, and shrink just has connotations
"Theres no need to start with concern trolling"
what else would you call it
what else would you call what you just did
actual concern about faux witch hunts. elaborately explained
this is concern trolling
Gonna pass on taking disengenuous moral advice from you, thx
Evanesco!
can we all just agree humans are fallable and they all suck (myself included) and move on?
yes, we all suck
i suck the most so i win
a fictional character that gives disingenuous moral advice as a super power, tell me that's not a great idea for a South Park character
i ceratinly am pretty sucky at times
just all be friendly to each other, we all sitting in the same boat
that was basically the wokeverse episode
Spoofed in the episode "Ike's Wee-Wee", when both sides agree about Mr. Mackey drinking a beer.
Devil: Go ahead. Drink the beer. It'll calm you down.
Angel: Yeah. Why the hell not? It's just a beer. Don't be such a p*ssy, m'kay?
plus, all of cartman in every episode he's in. central trope of his persona
The moral philosophy pertaining to allegations is essentially that even the means to bring up the topic of dismissing our disgust over allegations prior to evidence, is immediately an unpopular position to hold.
Disgust ultimately is the the prime fallacy of human cognition. It can guide us towards condemning innocent people, all the way to following genocidal beliefs.
I think the world would be significantly better if we used rationality over our intuitive feeling of disgust towards anything we are unfamiliar with, or think is self-evidently "evil".
i think "stay in your lane" needs to be a societal mantra lol
If SD3 is as much of an improvement as SC was to SDXL then we're in for a damn treat. Stable Cascade is (subjectively) superiour to DALL-E 3. At least he bing version!
#🏞|general-with-images message
Problem is that platforms like discord has power concentration in the hands of mods who will enforce their own philosophy which can disregard justice
Has anyone been approved from the waitlist yet?
old school ways of doing chatrooms had mods too. back when it wasn't all centraled
my inbox is as empty as ever! :(
Chances are not for a few more days, as even content creators on YouTube are scheduled for a few days.
How well will SD3 play with all the models out there? I have to have 30 downloaded, bunches of loras ECT...
I think I put the wrong email into my waitlist application
entirely new architecture and its a transformer based model
gonna need to retrain erything
Holy moly
Depends on your application, if you want beautiful abstract or non-specific art/photography, you don't need SD3
Retraining for everything
it'll take a good bit before we get finetunes, too
Loras should be fine according to what I read from lykon earlier
I wonder about stuff like deform... On my main box I have a highly tweaked out automatic1111 install. Looks like gonna have both eh
Well.... I'm 45 years old. Been in to tech for a long time now but really... This all is the most excited I been about tech in a long time. Everything is moving at such a fast clip.
It's pretty awesome
loras can be trained for it, but sd15 loras and sdxl won't work. not until theres an xadapter trained for sd3 maybe
Exponentials are even being miscalculated by experts, there's a consistent trend among predictions by experts for things like AGI, if the pace of inconsistency continues as it is, things like AGI may within two years from 2024.
Anybody tried training Loras for Stable Cascade yet? I’m trying to do an anime character Lora, and while it’s picking up the anime style fine, I just can’t get images that look like the character I want.
Experts are in a field that has a certain curve that can feel linear, but as an exponential moves up the hockey stick on its strong bend, these experts end up being the least informative.
oneTrainer said their system can now make loras for cascade, but none of the interfaces (comfy, a1111) are able to use them yet
I’m wondering if I need to increase dim/alpha beyond what I used for SDXL.
I just installed onetrainer, no nothing about it. Gonna dig in to it tonight after work.
I'm not gonna bother with Cascade.. too much hassle to go into an architecture that is going to be outdone within 3 months
Kohya supports cascade Lora training, but you have to do it in command line, and I had to mess with the scripts some to make it work.
The specificality of image diffusive models is a revolution in text to image prompting. Although, I think for fine-tuners, the problem is that SD3 from what I have seen may have multimodal features. Seems a little weird for that to be the case, I think that may be misinformation though. However, it isn't unreasonable for video multimodularity.
no, that was mentioned in the press release
Yeah, now that SD3 has been announced, I’m thinking it may not even be worth messing much with Cascade.
Thanks for that information, it's certainly a twist for audio integration.
I figured that because it has some work based on what Sora was based on, it was more about video and image multimodularity. But audio addition was weird to me.
well the idea behind any of those things is the same - text gets tokenised, some sort of transformers are used to process it, then an output is delivered... #oversimplified 😛
The only difference between 3090 and A6000 is the amount of VRAM and the BIOS. Not only that , but there're double VRAM mods that work fine and doesn't even require drivers from the other gpu models. https://ithardware.pl/media/artykul/26971/rtx-3070-16gb-vram-mod_2900ac3c39.jpg
Of course, although the main focus of the work from what I recalled was the new efforts from DeepMind on video/image interpretation and context delivery. Sora is an example of strong specificality.
That sounds cool, and what do you mean by double VRAM mods?
I may be naive to say but I wish there was some sort of backward compatibility or an easy switch between sd2 on automatic1111 and all the stuff folks have built. There's so much that's been made, it's a shame to see it all unusable. Shoot I wouldn't even mind an extra 5 GB of a program that has some backwards compatibility interoperability. But... Admittedly I may not know what the heck I'm talking about LoLz
kinda what x-adapter will do when integration occurs
thats backwards compatibility for sdxl
hacker swapped out the vram for bigger chips
I say hacker.. very clever dude that is good with a soldering iron 😛
Oh! Physical swapping! Thought was more about editing to make vram bigger using RAM or something
I mean exactly what it says. The screenshot depicts a GPU with its memory doubled, literally. Some GPUs allow for the addition of twice as many chips, while others support chips that are twice the capacity.
Sometimes less.
vram mods have been around for some time. you can't just put any chips on any card though. has to be similar chips with a bios that happens to accept it as legitimate. custom bioses aren't a big scene
Yes, I was thinking this was a software mod, not hardware mod. Where you use memory from elsewhere and also somehow it being still performative. But I was wrong, this is literally swapping chips lol
I have a 3090 Ti, but being able to use some actual RAM to spoof the VRAM would possibly help for larger models at the cost of performance, specifically for testing and such.
Hopefully SD3 full version can work well on 16GB-24GB cards.
sure, but the reason why custom bioses aren't a big scene is because nobody in the scene has the signing keys .And you can't run unsigned vbios on modern gpus unless you can do even more crazy voodoo and that is hacking the falcon chip itself. So far nobody was able to do that, which is puzzling.
the stable diffusion 3 site mentions AI safety 5 times on the landing page lol
https://stability.ai/news/stable-diffusion-3?utm_source=twitter&utm_medium=website&utm_campaign=blog
so no waifus?
How do I run JuggernautXL v9 for free? I don't currently have access to a good pc, so I can't run it myself
get a runpod or something
not free tho
I don't do it for long sessions, so it isn't worth it
I could do vast, but I don't feel like finding a working machine
they learned from google recent experience that gemini is doing things exactly right and everyone loves it and thats the model they should emulate
as soon as they release their models in the open
it takes like a week before someone makes a nsfw fine-tune
that's exactly what new drivers do, they dump the content to Ram if VRAM is insufficient amount . And that feature sux ass big time.
yeah but ppl gave up on SD2 and it took 6 months for any decent XL models to come out that can do this and still only kinda
do you know of a free option?
because the model doesnt really understand anatomy and is hard to train
not really
I want to use comfyui, A1111 is slow and buggy
and thats also the reason why people abandoned it
i was hopeful about cascade but whats the point now lol
SDXL knows anatomy but doesnt know what a genital is 😂
its so weird to censor nudity to me
as if nobody knows how they look naked in the bathroom
i'm sure they are afraid their user base would generate images that could put them in legal trouble
both in terms of copyright and also creeps
You can generate images of children in any form its not illegal in the states
AI safety is annoying, censorship of even the worst content necessarily means they won't satiate all desires, leading to more incentive to obtain uncensored content elsewhere, whatever that content may be.
I'm all for AI to take over our desires and allow us to fully embrace exploring fantasies and curiosities as the boundary between fiction and reality is pretty obvious. People know the difference between acting something out in reality and merely exploring curiosities with a prompt.
i think personal responsibility is ignored a lot
a pencil maker shouldnt be sued because someone wrote a death threat with their pencil
same applies to any tool
When images can be used with anonymity and no accountability, personal responsibility is kind of irrelevant for the bad actors.
I feel like people believe AI models are like FactoryPorn.exe SCP, where people think exploring fantasies leads to causing harm. This is evidently quite the opposite we observe.
Has anyone already gained access to SD 3? Hi everybody
there is no anonymity on the net
its been a myth for over a decade now
Nobody, not for a few more days at least
we have literal action movies where people get tortured
and the majority watches them for entertainment
Exactly
i think people have a good idea what fiction is
when laws come into play they'll be retroactively applied. old synthetic csam from before the laws isn't going to be grandfathered in here.
Exactly ppl can live their fanasies without enacting them
People know, however it comes back to the point on disgust. Disgust blinds us from even the most rational conclusions.
If they make it illegal they are just tryna protect their billion dollar human trafficking business
I think thats debatable, anyway even if there's a legal gray zone on what is or is not proscecutable, the issue is more about jawboning
but killin ppl is allowed in fiction,sex isnt
i think the porn industry has a big interest in censoring nsfw models lmao
funny enough, gore and violence is still possible even in SD 2.1
i dont get the discrepancy in values
is it because there is a war industry but no sexual assault industry?
Legality doesn't matter, but on your statement, it would be illegal if it's indistinguishable.
big porn always doing the industry propaganda dance. having people belive they were the cause of vhs winning when they just made that comittment near the end
yes every year theres a new war so maybe they want us to join their army
joining the team in the lead at the final round
Jawboning is the use of official speech to inappropriately compel private action. Jawboning occurs when a government official threatens to use his or her power — be it the power to prosecute, regulate, or legislate — to compel someone to take actions that the state official cannot. Jawboning is dangerous because it allows government officials to assume powers not granted to them by law. The capriciousness of jawboning is also cause for concern. Individual officials can jawbone at will, without any sort of due process, by opening their mouths, taking up a pen, or tweeting.>>>
I dont think. Isnt the point that it is a victimless action. If there is no victim.. who cares
if its trained on real images, there are victims.
It's always the same two arguments to justify their disgust.
Either it leads to "normalization" which in of itself isn't an argument. But is often flowed by "leads to an increased risk of harm", which is a false statement, evidence correlates with the opposite conclusion.
yeah thats true and sd 1.5 had images of gore, violence, sexual crimes and abuse in the laion dataset
@stuck sun
This is why the law doesn't matter, most laws are outdated and not updated due to disgust and the controversy of it.
Although a worse dataset may also mean more content included that they might of screened out in newer datasets, so there's less diversity in newer models.
but also less noise
SD 1.5 had a lot of useless data
"victimless crimes" until the 12 year old in a girl's class figures out how to use it and send everyone what he made
always remember. kids are soooo stupid
yeah.. the world is not ready for Sora...
i'm sure sora is censored as af
i dont think sora in uncensored
I dont buy that spiel its not the same image. Its the same as a human artist drawing it based on ideas of others. At the end of the day its someone (something) esles creation
and sora wont run on your home gpu because its like 300gb like gpt 4
its a datacenter only model
This would be a separate conversation about regulation, not banning. 12yo's shouldn't have access in the first place.
yeah sora is gonna refuse to make anything thats even a bit mean sounding
even dalle blocks most of girls in beach images
most people don't realize what animatediff is capable of
restricting tech from children just means they'll grow up into uninformed adults
animatediff and svd are close enough.. sora level tech would just be more convincing
i know, all my attempts making anything cool with it failed, it just makes camera motion and floaty people 😂
what are you gonnad o, require an internet license like a drivers license? feasible maybe
pretty much this ^
not even close.. the problem is that the cat is WELL out of the bag
takes a while to get something good even with vid2vid u need to experiment a lot to get good results this isnt something a random kid can do
fuck it's moved into the next area code
You can produce any content without any content that was produced harmfully, just depends on how good the model is at putting things together.
digital ID is unavoidable with the majority of DIU getting gen AI capabilities
atleast for some services
yeah. pandora farted up the whole joint
Groq chips are 230MB (that's mega not giga) and yet they can run Mixtral Moe 47B params. Can't even think of how gargantuan their supecluster must be to run 100GB model .And how they managed to archive such speed with Ethernet interconnect is way beyond my imagination capabilities.
doesn't take much to productise that..
Regulation doesn't mean restriction, it just means to debate the restriction itself.
they're doing something with blockchain to fund it. immediately makes me think it's basically vaporware
takes a while even more if u have a normal gpu with a several minutes vid
i checked out on groq
regulation is kinda pointless with AI, the regulatory bodies only react and only like 1 year later at best. at current speeds the technology is leaving them like dinosaurs getting smashed by a meteor
Obscenity Involving Minors
Federal statutes specifically prohibit obscenity involving minors, and convicted offenders generally face harsher statutory penalties than if the offense involved only adults.
Section 1470 of Title 18, United States Code, prohibits any individual from knowingly transferring or attempting to transfer obscene matter using the U.S. mail or any means or facility of interstate or foreign commerce to a minor under 16 years of age. Convicted offenders face fines and imprisonment for up to 10 years.
In addition, Section 1466A of Title 18, United State Code, makes it illegal for any person to knowingly produce, distribute, receive, or possess with intent to transfer or distribute visual representations, such as drawings, cartoons, or paintings that appear to depict minors engaged in s*xually explicit conduct and are deemed obscene. This statute offers an alternative 2-pronged test for obscenity with a lower threshold than the Miller test. The matter involving minors can be deemed obscene if it (i) depicts an image that is, or appears to be a minor engaged in graphic b*stiality, sadistic or m*sochistic abuse, or s*xual intercourse and (ii) if the image lacks serious literary, artistic, political, or scientific value. A first time offender convicted under this statute faces fines and at least 5 years to a maximum of 20 years in prison.
synonyms really
for all the legal scholars in here
Probably, but some things are more obvious than others pertaining to regulation.
someone will be renting a runpod with a couple of h100s and running a cheap domain...
yea a 12 year old will rent a runpod
'obscenity' is kind of uh, ambiguent
in my personal opinion most content people watch nowadays is super obscene
like tiktok, netflix, youtube
shorts
but it works and the speed is absolutely ridiculous so they must use SRAM for sure. No HBM3 can handle such a bandwidth , no way. We're talking about 30TB/s or more.
obscenely stupifying content
no.. shady adult will, turn it into a website that's a 1-click for user, and then 5 mins later you get email saying your deepfake is baked
ambiguous you mean, yes the law is ambiguous here, so saying "its not illegal" is not correct.
no i mean the grounds for legal action, obscene is subjective
why would anyone do that when u can sell the video itself and make more money
yeah but the person making the images is not the one that decides this
Obscenity is often not applied to content that wouldn't necessarily be considered real, however in the world of deepfakes and realistic generative AI, it may start to compell courts to emphasize more on this ambiguous term.
people pay to use the service.. think midjourney but for shady shit
all sounds like "BITCONNNNEEECT" hype to me as soon as i saw "tokens" on their page
which is why Stability are being very cautious about this, so as not to poke the government bear
You're right, as long as it can be trained and fine tuned easily, I'm ok with censored models, only issue with 2.1 and XL, is there's less fine-tuning, or difficult to do so
because there are a lot of hysterical Elizabeth Warren supporters who want an excuse to censor snake emojis for the "harm" it causes, let alone stuff thats much worse
forensic ai detectives gonna be really essential going forward (i.e deepfake detection)
because money
a 12 year old played global thermonuclear warfare with the worlds smartest AI computer tyvm. just a geeky kid who knew how to wardial
I think SD3 is gonna be slow as heck
diffusion transformer with patches lmao
DiT need a lot more data to be good compared to other models
its ok u will only need a 3090 easy peasy
When is the SD3 bot coming?
its on the preview server
who cares. their cluster works and I benched those models already. Digiteyes did that too, however he didn't use API, just the web page . I can post community benchmarks if you want. Not as smart at logic/puzzles as Mistral next, but damn good model regardless. But the gamechanger wasn't the model, it was the speed .
Do we have access to that or is that dev only? @pale latch
I been gone for awhile and seen the bot was disabled.
meh. blockchain shit. bitconnect made people a lot of money and "worked" until it didn't
for all you know they trained those models on other hardware to puff the vaporware up
thats how all pyramid schemes work
ponzi gonna ponzi
other people got access. you can apply. i'm on the waitlist
I want to see SD3 show examples of humans more, worried about the lack of people in their examples
I can't run it on my current laptop so it would nice to know that if I join the wait list I can use the disco version.
same, i hope i get to test it out
i've been here using SD since the start
I've been here a pretty long time but family issues and the nine
Just lost my mother to diabetes.
that sucks
magic retro references https://youtu.be/DOwQKWiRJAA
Most definitely
So If I do join the wait list I can use SD3 on disco? Or is that just for the models.
never heard they trained anything on their hw, Mixtral and llama2 they offered weren't tuned on their gears AFAIK. They just serve the inference .
most likely just the discord stuff
no leaky leaky
meh. blockchain tokens. all i need to know
Nice, I'm fine with that, I'm pretty hyped up to use something other than the corporate nonsense
All this censorship has everyone irritated
its overblown
You can not even make a picture of a president
That should be illegal tbh, We do have rights lol
Try to create Donald Trump in gemini or chatGPT
your american rights are meaningless
It's global
so hyped for Stable Diffusion 3
It's an effort to suppress everyone regardless what country
ugh i soooo hate the american election cycle. just get it over with already. drag it out so long
yeah this
they're basically teaching AI to be malicious and lie
gaslight people
your country is going to burn down and people elsewhere will all be "oh."
Yep
they cant tho unless they install a rootkit on everyone pc's
Who says the backdoor are not there already
but how can they block me from using webui
'alignment' is nothing but a codeword for enforced propaganda, gaslighting, psychological manipulation, historic revisionism and censorship now
i think the current form of alignment is more dangerous than no alignment at all
qanon nonsense. if you really want to be paranoid about computer, learn about the ken thompson hack.
It wouldn't be instant, It would be done in such a way that new technology would be outfitted and integrated to censor you
its gonna be like that transformers movie,my keyboard will turn into a robot hand and choke me to death
https://wiki.c2.com/?TheKenThompsonHack good luck
this has nothing to do with conspiarcy theories, they're openly saying that they're aligning the AI with their values, which are bad
😂
or like Full Throttle the meat carver gonna jump at you
The old devices become obsolete over time giving them opportunity to create new tech to try these things with.
When Google has to shut down their AI because it won't create white people we have an issue
thats' an exageration of what was going on. their prompts were just throwing different ethinicities into the mix on the backend
you don't understand, depicting white people in any context is harmful
alignment simply doesnt work, its dumb to think you can fully control a non human intelligence. its like a wild animal thatll bite ya if you're not careful
really is qanon conspiracy nonsense when you think some low engineer's idea is a grand conspiracy from the elders of zion or some shit
I don't care if it's the left or right, My idea is mine and I want it to be portrayed how I envision it. Not Google
what are you even talking about icey
the qanon truther in here
you keep refraining blockchain. Frankly I just eyeballed their page and couldt find any crypto or blockchain relate stuff there. You're sure we're talking bout the same company???
who believes in qanon
the one you linked
If you can not get beyond the politics and look at it generally we will all be doomed
I didn't link anything
Censorship imo is dangerous
wow very far right (or leftist from 2014) of you
yeah i think we should focus on the general idea here. its about a handful of people deciding what information you gain access to and what you dont
and most of all what you do with your computer
Why would you be OK with Google or anyone changing your prompt
if we're talking about groq then there's nothing crypto there at all.
Do you believe in your right to create
no most likely not, what kinda lora u wanna train anyways?
We wouldn't have true art if we would allow the government to dictate what we can draw or paint.
That's not freedom of expression
oops this one
Right now, I'm trying to engineer my own 2d art style
I'm here supporting SD because they never censored my political art, Even my Dr Fauci pictures which were pretty lit
they do have a part in their licence that says no real persons allowed tho
the whole page is promoting a crypto token around their ai. they're all about crypto
we are all doomed anyway, life is a sexually transmitted disease and it's 100% terminal 😛
Are political figures real people?
🤔
Lol
yeah, that's them. but where do you see anything crypto there?????
the whole page
depends if you believe in simulation theory, the lizard people illuminati theory or any other similar theorums
We have a fair use obligation to make the best memes
what?????
@nova zodiac Illuminati is just a Cabal or some would call it Deep State or shadow government.
Welcome to Groq AI, where innovation meets intelligence in the most exceptional way. Groq AI is not just a cryptocurrency;
Once you look up things like The Trilateral Commission and The Bilderberg group or Skull and Bones, You know you are in the rabbit hole
In only gets deeper from there
nazi propaganda in it's 2024 form
Where?
hi guys, may i know why i can't create anything in stable diffusion now?
Qanan does not even exist
There is Q which is a military psyop and there are Anons who are literally people like me or you who are anonymous who research information
cycles through different names. its all the same propeganda
Unless information can jump out and hurt you then that's a different story
It's not nobody's fault but your own if you can not handle the information
i bet you've got a final solution for all the secret ruling people. wonder what it is
Spreading lies and speaking as if you know what you are talking about is even more dangerous
Like I said Qanon is not real and is literally created by mainstream media
like i said, you're just mimicking the same nazi propeganda from pre kristalnacht
Bru thats the wrong url. That's some scam. the original groq is silicon chip company . they make Ai accelerators , been around for a hot minute. here's the chat, you can use it. it's free. 500 t/s that 10x faster than gpt4. https://groq.com/
The word Qanon is from corporate mainstream media talking heads
Q is a military op
not into it
hmm, today i'm going to click on the SD server and see a very normal conversation...
17th Letter Q and they work out of Virginia
Q is the highest military clearance you can have
One day but that day is not today.
i think i may have started this lol
i pointed out that the stable diffusion 3 landing page has 5 references to "safety" and it snowballed from there
It's normal to talk about this lol it's an experience people are actually having in the world
It's not ok to ignore it and make things up as you go
That's how people become confused.
It's no different than covid, ww3, blackouts, etc.... These things are actually happening.
no offense bro, but are you retаrded??? If somebody creates Microsoftok.com then you'd stop using ms windows and if there's Googlec.com in the webs then you gonna drop Google search all together?
it's just a brand. chill.
They're a crypto company tho. that coin is a legitimate symbol on the network and is one of the company's projects
feels like i'm in grade 10 again when i told my bud i liked adidas sneaks over nike
How could you
heey people
allso, microsoft has consistently delivered products over decades. groq is vaporware built on blockchain for some reason
Anyone tried proteus 0.4?
I’m glad that Google overplayed their hand with their AI image generation, as it made their insane racist, anti-civilizational programming clear to all - From Elon Musks X Profile an hour ago
"built on blockchain for some reason" -- unofficial tagline for half of SV startups
LMAO 🤣 🤣 🤣 🤣
We are watching the fall of these tyrants
In real time
SD3 will take over and force everyone else to go Open Source
take over what
Why would someone make Microsoft even richer or Open AI ?
Do you understand who you are supporting when you give them money for this technology?
I think it was more related to Google's engineers trying to solve the issues of AI biases and they've made it wayyy too overzealous. These models are hard to align for a middle ground.
Ok fair enough . I'll have to double check if that page is legit.
I kinda feel like its good there are many competing tools that can complement each other and excel at different things, and the google one was shit
If you pay for chatGPT you are supporting the cell makers who will build the cage around us.
Yeah, was never really impressive considering what was already released.
It's obvious that they do not want to the technology from stable foundation to be in our possession
Even things they should've excelled at were pretty disappointing.
Google models are most biased and dumb 😂
And for no reason other than to push propaganda
does anyone use the meta one?
I have not used that.
Meta's Imagine? That's more an experiment than anything.
i tried it once and was like, eh, this is ok, not better than any of the others particularly, so whatever
theyt are all experiments tho no?
dall e is like, not very real looking so not a competitor for MJ except for cartoony/illustration stuff
I watched a ton of information on it but never tried it fully, I think they had a demo we could try early on.
Yeah, that's true.
Everything is extremely samey as well.
Which is even more of a reason we shouldn't be paying yet
It's experimental and it's being trained on OUR life
It looks at our patterns and our information and learns from our ability
We should have more of a say
The one consistent AI that has not tried to be more than what it is would be LeonardoAI
well i pay for MJ and magnific
I can understand people pay for MJ or LeonardoAI
I'll cancel magnific as soon as there's another AI upscaler thats even close to as good
Which would you compare it with?
krea, stable diffusion testing loads of upscaler models, etc
I've never used magnific, I'm always willing to try new AI
I'm really waiting for an AI game engine.
tried the topaz one too
magnific stomps on them all, for now, for most things
not perfect but
its like a couple of clicks and generally really impressive results
None of them are to most of our standards as of now
But they are getting really impressive
SD3 has me excited again
I kind of burned myself our learning how to properly prompt with SDXL
i doint even believe it's their models. its the prompt engine
the workflows around the models
its unfortunate that its done because it doesn't allow researchers a real clear view on the actual model
It's how they manipulate your inputs and change them to fit a woke narrative or an agenda that they have in mind.
maybe someone has cracked upscaling in SD but the amount of time to test and the terrible quality of so many of the results is really offputting, you think you got it working then you try the workflow on other images and it tears the image apart, in comparison you toss it into magnific and press "go" and most of the time it either looks good immediatly or you push a couple of sliders around and thats it
I have a decent way of manipulating my images but I do agree with you to an extent. I have not used magnific so I can not relate on that end but I can imagine that if you are leaning that way while also understanding the prompting process of SDXl and how many times you would need to regenerate images to get something usable, I trust your words.
I saw this recently. Not exactly what you mean, but could help make upscaling faster.
https://arxiv.org/abs/2402.13573
dont trust my word, i'm a dimwit lol, find a better solution i dont wanna pay for magnific ahahahaa
not as much as Anthropic Claudes.
SD3 is dae wae
Don't get me started on Claude
Claude told me that a cheetah hunting in tall grass is offensive
is this not available yet?
I left and never returned lol
let's go for Goody2 then. 😄
I like stable foundations style, So far I liked everything that they have worked on
The emphasis on making it available to us is legendary
weights and documentation. such a beautiful thing.
When they do that they make it hard for tyranny to affect us. What they are doing is more noble than meets the eye.
I'm sure that there have been some back door meetings telling them not to release stuff to the public
the source code for training these models is under appreciated. it'll be cheaper to make these in the future.
It gets faster and faster because the AI can help
One thing I do not want is sentient AI
consider if magic and warlockery is real with a focused mind. what kind of magicka will machine intelligence achieve? How's that for illuminati conspiracy
one my favorite science fiction stories is "Destination: Void" where a deep space team of engineers create machine consciousness and it becomes a jesus
Is sd3 gonna be commercial friendly ? Cos if their model gonna suck due to censorship and safety, and their licensing gonna be restrictive, meaning no $ goes into finetuning which will be very costly , then Huston we have a problem.
It needs to be open to make an impact
generalized machine inteligence is going to make an impact no matter what
If it's censored like gemini we will seriously need to look in to alternatives
We are running out of options lol
Still anthropic models are not great , I agree they have huge context length but they suck even llava is better than google anthropic
all the models today won't matter. they're like all the hooey dooey airplanes that were all over before flight got really discovered