#💬|general-chat
1 messages · Page 116 of 1
🙏
Sure it have its limits. We all know about DALL E 3 and Midjourney.
It is not like I care how's SD going on, more like turning SD into 3rd-force in the market against other two.
media is going wild now, making headlines
Somehow I doubt SD3 will be open source
Would be happy to be proven wrong
I just doubt it
but if lets say SAI failed and fallen, I expected other smaller projects will rise up in the effects of SAI developer refugees.
Clearly open source is not great for making money
Then I guess we'll have to make to with Cascade XD
@low moon Wasn't Cascade trained on a much smaller set of images than SDXL?
So basically cascade was a research project
And SDXL remains their flagship
depends on their commercial licensing and how it works out. lots of companies do open source stuff and make money, the secret is that people who make free improvements to it are the product
Perhaps
Cascade seems better than sdxl on release...
training A, B and C.
None of the AI companies are making money right now except maybe google with their search engine and medpal
open source itself isn't a business model
That's fair, if no AI companies are making money, maybe open source is then double bad
thats a big claim,got a source?
you have to look at success stories like how Meta does it, they fund Pytorch, React, and released Llama 2, but Meta has a bunch of revenue streams established first
I think SAI just needed more time to cook. It took amazon like a decade to become profitable
Yeah, that's fair
Firefox gets paid by Google for search placement, etc
Maybe SAI needs to take ad deals, and coke logos pops up in random images lol
In my opinion SAI should do what midjourney is doing, but continue releasing open source models too
https://twitter.com/EMostaque/status/1771400218170519741 new tweet who dis
"Small" models for the community (than ran on nvidias XX80-90 and lower) and paid big models that ran on an a100
no, but have you heard of this little outfit called linux? it doesnt have to be about money, or corporate interests
Supposedly MidJourney is profitable
But they are also tiny, and do only one thing and do it well
meta wants you to use pytorch, because they use it in their toolchains, so they provide it to you for free so you will hopefully report on the silent incorrectness issues that I seemingly run into on every single project I start with it. they also want to have that market share so that people aren't expanding on their competitor's libraries (Google's, TensorFlow and more recently JAX), because anyone who's used both knows that better supported JAX would wipe pytorch out of existence
https://www.youtube.com/watch?v=XZ3w_jec1v8 watch this, about open source...talks a lot about where it works, and where you just get Jeff Bezos'd (i.e. SaaS'd)
I've literally been through this myself
what does that tweet even mean ....
My interpretation is that they control the board but emad controlls the shares?
yes but...
Developers still support him and resignation?
yes, they have a tiny staff and were revenue+ from almost the start, got a compute grant from daddy Emad to get early model out
Clearly lighting investor money on fire is great for the community but not for the investors lol
so he has the majority of the shares, so he has voting power over what exactly? or since hes not on board he cant make a single decision?
corporate governance is really weird
This looks pretty big. 100% chance SD3 is open if emad has the power
More like "had" Investors aren't goint to put any $$$ unless they get shares
I wonder if there is something in the VC deals that restricts his power over the board
The balanced approach imo is to release the previous generation's model. If companies did this we'd have Midjourney V5 and GPT-3.5 which, while outdated now, are still competent models. They can still make money offering the flagship model as an API. Seems fair consider it's our data they're training on and selling
SD3 lore is real on this one
From Emad https://twitter.com/EMostaque/status/1771400218170519741
"As my notifications are RIP some notes:
-
My shares have majority of vote @ StabilityAI
-
They have full board control
The concentration of power in AI is bad for us all
I decided to step down to fix this at Stability & elsewhere
Will be sharing more soon
Exciting times"
lol so he stepped down to get more control over what happens at stability ai?
maybe?
That's not my understanding
I bet forced out by investors
Like I like open source as any other dude
how can he be forced out if he has 51% shares?
But clearly this is not making money for the investors
Maybe there is hope for SD3
I was confused before, but now....wha....?
is this some american thing i dont understand?
Honestly, I'ma let Emad cook
finally, crypto ai lol
Step down or no more funding?
he literally said he stepped down,is it that hard
Many different ways a VC can force you to step down
um... how to phrase this
SAI has billions in capital, they're not running out of funds. Currently spending 8 mill a month and earning 3 mill IIRC
as long as he lets me cook SD3 🫡
There are some companies out there with majority shares holder unable to control the company's governance
its a pretty relaxed tweet, ill say that
lmao, cant wait for this to brutally fail work out
I do not believe that
that part is easy, the why could have certain ramifications though, dont you agree?
surely emad knows what he's doing 
Think about this, 5 million per month spend, if they have 50 million left that's 10 months run rate
Exciting time
a CEO stepping down != guy on a shipping dock leaving his job
1 billion dollar valuation doesn't mean 1 billion dollars in the bank
emad? knowing what they're doing?
That valuation is just that, a valuation
they got $100m in funding for 10% of the ownership, that's all the $1b valuation means, it could mean they own banks $250m
If I take a company, and I say I'll pay 1$ for 0% of the company, that company is basically valued at infinity according to me
yea maybe due to closed source stuff
Sure, but you're not a bank with a 100 million dollars on the line. While I agree with your main point I think the analogy is off
ya'll never watched Shark Tank?
I'm quite confused by your point
or due to numerous other things we can make up, aquisition, company HR issues, infinite list, hence the need for more disclosure
Valuation = (how much you paid)/(ownership percentage)
It does not necessarily mean that you have any hard cash
yea he said he was gonna share more with us soon,so its better to wait for what he has to say
If a bank invests 100million for 10% ownership, it means the company has enough to be worth more than 1 billion dollars. Though I agree they don't have a liquid 1 billion dollars
gotta be honest, im more positive now after the tweet, hes a madman
If that was the case, softbank would be profitable
And yet, softbank has basically lost enough money to struggle raising money for vision fund 2
is chatpgt worth something, absolutely....but founding members leave, CEO leaves...that's a wtf moment
I'd like to imagine this is all 4D chess but 🤷♂️
or harry potter chess 😄
I've been forced out of companies before, I'm sure Emad must be pretty upset right now
... or is he?
But such is the poison you accept when you take VC money
he writes a bit like elon, like hes 1 day from becoming completly batshit crazy
like Freon said maybe Emad and founder members is playing a chess with whoever in the company
If he has a heart of steel then perhaps not
But I've seen similar things happen to many friends, it is never fun
I think it will be ridiculous if my prediction really happening
doesent this seem more like its emad who is the one forcing things? haha
and not him being forced out of anything
Let's hope Huggingface save the world :<
Only if you take his tweet at face value
I'm glad he's put on a brave face
Clem (ceo huggingface) tweeted : Should we acquire Stability and open-source SD3?
tbh I hope a bigger company doesn't acquire stability
Hugginface would be a good choice, but I wonder if they have the money to actually buy SAI
this doenst seem like that, they all would have gotten golden parachutes
he must be pretty e-mad
You are a menace, too many puns!
with their valuation i guess they could raise a little more to do it 😄
honestly, i think there's reasonable non-doomer explanations still so I'll just wait
They could I bet, but it isn't clear if SAI is profitable. If it isn't, raising money to buy SAI will be tough
Scenario A:
He resigned due to the pressure
- Get pushed out due to failure in addressing HR and financial issues
- Vote-out so to make SAI closing their gates ( unlikely in my opinion )
Scenario B:
He resigned for personal reason / willingly
- "maybe other people can solve this better than me"
- Resigned so he can be in another position like Linus Sebastian from LTT
messaging sounds a lot more like B
hopefully not an OAI 2.0
My problem with scenario A is that it was right before a critical PR release for them
option C) Skynet
I admire people that can see the plus side after all the reporting from fortune magazine and resignations of researchers
Scenario C:
He resigned so he can be prepare for April Fools
- BRO!
hrm, I dont
i can see emad downloading sd3 onto a floppy drive and walking out
hmm.
with sunglasses on, on a cloudy day
8 billion parameters is 32GB at fp32 size
You could probably do it with a fat flash drive
Yeah, it's odd for Emad to randomly throw his company into a percieved hell spiral like that without any sort of PR


You kind of have to set proper expectations before randomly dipping as CEO
Fortunately, I've scrapped every single pantheon & showdown prompt from 1 day after SDXL 1.0 bot launch to bot closing
Roughly 230 thousand prompts
60gb
Wait, how?
so, many floppydrives?
or even further, it should never be random outside of a wtf moment
text encoders
I assume those are untrained (so standard T5/CLIP), so only critical weights would be 32GB
nope!
they trained every tenc
all 3 of them
I'm sure if you went back there have been all sorts of issues that have plagued CEOS, steve jobs had legit advanced cancer for example
what if elon musk buys stability
Wait, seriously?

I didn't read the SD3 paper, so I assumed CLIP + T5 were untrained and frozen
then its doomed as in really doomed
There's a post on reddit about thanking emad: https://www.reddit.com/r/StableDiffusion/comments/1blk17x/thank_you_emad/
Grok Diffusion
Yes.
2.7t model
why?
I wonder if we're still in the days of AI like the early days of computers, where one 1MB computer took up an entire speckless room. Our weak AI take up so much harddrive space, and copium predicts they'll get smaller and better over time
i dont want to answer that
he's an open source proponent?
8B is not big
Elon Musk is pretty... uhhh
smart?
I think he kinda have a Zuck mindset
dont even mention his name, its like cursing sd3
I swear Elon is bipolar
I remember in the summer of 2017 during Tesla 3 ramp
Mr. Muskrat knows how to run a business but dude really needs to take his meds
Elon was unhappy every day, until he was really happy, we saw him and wondered why, turns out he broke up with his GF
His assistants are all super models (seriously)
Men & women
Dude is like Ironman without the charisma or cool superhero suit
Just from my personal experience, seeing billionares with super model assistants is not uncommon...
he was this, he was that, he's a brilliant entrepreneur - check, he was involved in this company - check, he wants it to be opensourced - stated publicly, and lawsuit even - check, where are things grey here?
Elon or Emad?
brilliant entrepreneur
Cuz Elon is a shitstain
his company?
lost 8b on twitter
elon
Elon do be a bit of shitstain. You can like his business ventures but let's be honest for a moment
And I used to really respect the guy
u mean the ones he bougt into and sued into making him a makebelieve "creator" ?
When I was at Tesla, he would pretend to be an engineer and would force the team to work on dead end stuff, and someone would always have to secretly fix it
he is literally suing this company, stability, over opensourcing their content
I dont get the argument
lol rumors
He's suing OpenAI
I'm not aware that he's suing stability
if u have a lot of money, u can make smart people make u a lot more money
yah, my bad
it was openai I was thinking of
Ah, a simple mistake. See he's suing OpenAI who doesn't actually make open ai but he's not suing Stability who do actually make open ai
Most of the top langugage models are close source, and the top open source LLM is chinese
maybe its because he wants to be noted as a founder of openai? like he did with tesla?
I think we'll eventually get to the point where our AI tech gets so good we can have top-tier LLMs on our computers
If you have like 4x 4090 you can run the top open source LLM (Qwen 1.5 72B) on your computer, and it is ranked higher than ChatGPT 3.5
4x4090 is only like $6000
$6400
only
or just use chatgpt
chump change
I've moved over to Claude 3, the only problem is the limits are too low
Unfortunately Qwen is censored. Hopefully we can unravel that eventually
You can run LLMs on AMD too now with Rocm for Windows, so it should probably be less expensive
did they finish rocm on win?
Yeah it is, but I don't think uncensored LLMs/image models will ever be a thing
I can use LLMs with Kobold.cpp on AMD Rocm
Base models, probably not
We just gotta wait for some good natured autismo with millions in hardware to fine-tune Qwen
It seems that not for SD yet, only LLMs
oh, how?
do they not have pytorch yet
I'm not sure, it seems that it will need MIOpen or something like that
Rocm is such a lost opportunity
send a repo link it it exists
mlir is almost same perf iirc
i dont have an amd card (for a good reason) to test
What does it tell you when Intel Habana & Huawei Ascend both have better native support than AMD Rocm
I don't know it
I really hope AMD gets their shit together
either amd gets their shit together and we have 3 competitors
i would give it 5 years
or we have to bet it all on battlemage (intel Bxxx)
battle mage is def not gonna be competitive with nvidia in performance but pricing will probably be good
in ai, vram is KING
pricing will make or break it
SD3 when?
intel is probably selling at max 16gb cards
1+ months
never atp
Rip
limited access this week, if emad's words are true
Read the comments above, everyone is still very confused tbh
limited access? didn’t they already start discord invites?
We should do a bet
Well
I bet that not all SD3 models will be open weight in the future
tldr: there are at most a dozen non-sai people in the sd3 server
Bet those beta tester have signed NDA of not posting any leaking
emad said that this week, hed give out more invites
Define not all
to influencers/researchers
All model sizes and variants
i bet SAI will collapse before sd3 or pivot to being closed
Do you want to bet cash?
Let's do $100 to a charity of each of our choices
we are betting the same thing
Nah
People betting SAI bankruptcy before analysing its financial situation lol.
We aren't
I don't believe SAI will collapse
But are you down for cash donations of $100?
How can SAI collapse when there is booba to be generated
😢
Let’s hope Clem buys it
We can donate to the palestines or jewish charities of your choice
nah that’s what my wallet is for
lmaooo why those
pay up.
Eh, I just like watching people fight in the middle east
damn that’s really forward uh
wild in general chat LMAO
its a good day for sure in here
they fight here in the US too, just wait for november race to heat up
Yes, our government is also a big fan of turning people in the middle east into skeletons if I'm quite honest
I hope they relese at least sd3 open source 😢
a sane stance is to equally hate or love everyone
interesting
Bro doing a ban speed run
A lockheed martin investor I see
Yes sir
honestly if sd3 doesn’t release i’ll be kinda happy cause i won’t have to hear ppl whining about open source or SAI anymore
I vote democrat/republican, no matter what I know there will be fighting in the middle east
Alright, enough of the jokes
pikachuface
people already whining SD3 8B won't run on their pocket calculator
What the government doing in the middle east is a horror show
or on their dollar store AMD card
It runs on mine. My pockets are huge.
I think that goes for every government involved in the middle east
Yes, but SDXL boobas make life better
And SD3 booba will be even better
general chat is wild in every direction yo
On a serious note, I have been donating to Médecins Sans Frontières and some peace causes
"you mean I have to train booba into the model myself? NYEHHH!"
"How do I put boobs into a checkpoint file?!"
so I’ve never done any of this stuff before but like, yeah I was typing and that summarizes it
I donate to games done quick, usually MSF charity
where do I input files to train!
MSF = doctors without borders
correct
its alternating MSF and prevent cancer between the events
I had no idea that games done quick had a charity drive
games done quick is entirely a chairty drive
I suppose one can argue that SAI is a charity drive for people of culture
two a year, summer/winter, I think they alternate MSF and Prevent Cancer foundation? something like that
Well, I mostly started donating to MSF after the recent news where MSF medics were killed in the gaza war
y’all should give me money and ill buy SAI
me too? we can share
1 billion dollar valuation
I'll give ya tree fitty
Better start collecting donations now
1 bil, thats nothing
especially to someone like me
i got it
I'll spend $25k to teach SD3 boob textures
sd3 will be about to come out but then government restrictions will cuck us
and boob textures alone
cat is out of the bag, though
Yeah
Did you see the think tank paper on criminalizing open weights?
Serious*
And the recently Biden EA on know your customer for models above a certain size
Every home will have a licensed TSA agent retrained to prevent boobs from appearing on your monitors
We had a good run
there are some things that cant be stopped, booba being one of the few
Would be best in Nvidia hard coded booba bans into their driver
literally impossible
Boob generation detected? believe it or not, instant ban
since the dawn of time
the year is 2075. booba has been outlawed. quantum generation has been slowly creeping into the black market to surpass restrictions.
suddenly, we came to teh conclusion that it was booba all along, and no quantum generation
we’ve been deceived!
just the reflection of the eyes of someone making booba images
I knew the world was deceptively soft and bouncy
but but but but AI is just stealing other people's work!!!!!1!
waaaaah deal with it
something as pure as booba cant be taken by anyone, just like u cant rob someone of love
okay but what about the other side
Good question. I wonder what the ratio of male to female users of SD are
i don’t think were talking about the same other side
ive...seen some stuff on civitai to make me believe that there are quite a few female users
Ah, that other side
||(((Huge Ass:2.5)))||
moment
lets not start ww3 in here
Sir, the buttmen have launched an attack on breastcity. We must retaliate!
detailed ass > huge ass
fair argument that I would side with!
(gargantuan ass:2.5)
google homosexuality
WHAT? suprised pikachu face
Just use SD 1.5 twice.
SD 4.5
just leak sd3 before, end it with a firework
would anyone be willing to help me... 'restore?' the checkpoint dropdown menu that's (suppose to be) in the upper left corner of A1111. I'm not sure why it isn't there anymore. so I don't really know what to look for in terms or googling it.
emad did WHAT???😰
Emad will never get the credit he fully deserves and it's a bit tragic. While I do not agree with a lot of his decisions or things he's said, he saved us from a potential dystopian future, that was looking pretty bleak. Everyone was circling the wagons on AI, buying up researchers and organizations. Nothing was open, everything was closed off and done behind walls as they tried to figure out how best to monetize the tech, preserve the advertising and marketing ecosystems, and etc. What Emad did prevented that for awhile at least. The way he released Stable Diffusion, he created data scientist and programmers due to the sudden interest in this advanced technology dumped in our laps. It was magic that captivated billions in an instant. Programmers and computer scientists were born, more advancement started suddenly coming out of all the AI fields. It spread like wildfire and started a movement. It's likely one of the few buffers that kept the majority from being absolutely against AI. Could you image if Stable Diffusion never went open-source? The advancements that would have never came, the data scientists that never were, the software that was never made, we are so far more advanced as a species right now than we otherwise would have been, so it would be decades of advancement lost that we gained in just a few years.
Why can't I join the midjourney channel?

That's probably the wrong place to ask here.
just read the news. stability is falling apart, isn't it?
Not necessarily
We don't know anything for sure except that Emad left of his own accord
The CEO of huggingface also showed interest in aquiring SAI
Go into Settings, User Interface, Quicksettings, there add checkpoint
And then apply and reload
This would be the best case scenario I think. Or CivitAI 😅. That would be funny, And hmm 🤔. That could actually work well actually. CivitAI has a better setup for monetizing Stable Diffusion without being greedy or intrusive. They can have premium model market where users charge for custom models and lora creation, then they collect a percentage.
Recently a lot of events have been unfolding
- Stable Cascade team leaves
- Stable Diffusion and Stable Video team leaves
- StabilityAI CEO resigns
All in like 2 days, what the hell 😅
thanks, much easier than having to go into different tabs to do it. 😄
Meanwhile StabilityAI internally:
That's what I would guess 😂
I will re-send this in #🏞|general-with-images
Bruh
I mean, if hugging face successfully buys stability
The ceo promises SD3 to be open source
I wonder if they would change the license too lol
So it's like apache or something similar to mistral
Hi!
I saw a model named "ZDS", but I don't know what it means and where I could find it
what an farce, at least finish what yall started
i finished already
not yet bb
I think he's mocking Elon
I would love to be wrong
no hes buying it its real i saw him with a cloth bag full of money,the bag had a big dollar sign printed
def some good points here
yeah...
my thurs/fri work days are absolutely insane so i usually miss most of what happens for 48h
hey there everyone, i hope this general chat is the right place to ask. So i got forge from Git into a folder, batched it, and indeed the webui showed up, but it says error and connection error. Anyone knows what to do? i never installed stable diffusion before, maybe that was a mistake to start with the ui?
With the recent change of CEO, will SD3 be open source as planned?
cant wait for the SD3 NFT release 🙏
Which is better, stable cascade or playgournd 2.5?
neither is better, what matters is what you're trying to do
Are you a poet?
What the hell happened at Stability over the last month?
Yeah what’s going on? My guess is investor pressure to close down open source and push harder to make money
No clue, and I would say the last 2 days not month 😅
Emad has majority of vote, so it's believable it's his own decision to step down, let's hope for the best outcome
Oh yeah just saw this
https://x.com/emostaque/status/1771400218170519741?s=46&t=yqWvcZNctIPYbCGRleXgSw
Now I’m even more confused
Pg2.5 is phenomenal for a lot of things. It has aesthetics that are off the charts at times and it has by far the best color contrast out of all the sd models. I know they talk a lot about that part in their HF page for it. I don't do a lot of stuff involving people, so I don't know how well it fares vs cascade(feels like 95% of users just want waifu similators, but I dont what your goal is)
But for almost everything else, it's one of the best models I've used. Feels like 90% of the images it makes are save worthy
They have to sell. They have to. OTOY is going to drive the reputation of this tech into the ground, permanently, while emad enriches himself with some crypto pyramid vapoware. Everyone seems to forget OTOY pushed this great 'lightfield' tech for oculus then delivered.... nothing. And where is emad going? yes, OTOY.
Stability management/board is likely doing the best they can to salvage this mess. Coatue management already asked for his stepping down last year, they were already talking about selling. Of course they're going to sell. It's about salvage at this point, turns out 3d waifus and not answering customer emails wasn't a winning strategy after all, colored me surprised. I just feel bad for the researchers who had to jump ship in the last few month, no wonder the smart guys bailed - they didn't want their tech to become a crypto scam
makes sense, when someone hear decentralize first thing that come to the mind is crypto. Do you think this thing has something to do with the delay we're seeing with the sd3 release?
We never getting SD3 💀
The fact that emad is posting crypto scams now presumably means they're out of cash
Im not surprised at all for this fate of Stability AI
The OS community alone cant sustain a company
Or doesnt want to
they teased us for two months, hope we get an proper update soon.
Are they going to open source it in the first place?
Or will it be behind a paywall
Emad just announced he's stepping down.
We're talking about it 😅
a dev should anonymous leak it
And then what? ^^
Let SAI disappear?
Would basically mean the very last model released by them
Emad wasn't kidding then 💀
That there would be nothing past SD3
I personally doubt it, first they already said it's free, second everyone would jump back to finetune sdxl, and Hugging face said they would buy SAI and open source SD3. But I can't predict the future
lets hope emad leaks sd3
Would HF be able to sustain itself with SAI on its back tho?
if they can run demos for biug models
yes
Dont know what their annual revenue is
no cause HF only made a joke on twitter now all of the redditors are spreading that as real
Tbh i didnt follow it to that degree i only saw that screenshot lol
yea they only saw that screenshot and thought he was being serious
You dont acquire something you know burns your money like hell without giving the profit back
Plus investors have a saying too
He has full board control.
Despite stepping down.
There is an actual plan here, it seems.
Speculation is a waste. We have no clue what is in store for SD.
Ofc
oh now! I love sbalityAI! i love what they did !Amazing team! prey for the best!
Breaking news! Stability AI is acquired by Microsoft (or Adobe)

wait no its gonna be Trump Media group
bought by open ai
that team dont really work there anymore
I just read, Huggingface has a annual revenue of 50 million dollars
No way they will buy SAI
And investors are still refusing to participate on that
how much SAI costs?
Its not how much it costs, its how much it burns
Without giving the profit back
Which is 8 million dollars per month according to sources
Emad wanted to get investors that will bring him 4 billion dollar investment and he failed
anyone anyone know where I can find endpoint docs for comfy ui?
Hello:
I wanted to ask a question: are there SD models for specific tasks? I mean, I have Dreamshaper_8 and it's very good for defining characters BUT when I want them to interact with each other it doesn't "work", it even seems to "resist" with verbs like "fight", or "put hands on neck", "hug from behind", it's like they don't "know" those words and many times they don't include more people from the prompt. I have seen very good images made with other models and that's why I ask myself the question. Or I just don't know how to handle it properly.
Anyone else can´t train anymore with the last kohya_ss update? 🥺 I get errors and the configuration is the same as before
hi ! does "webui forge" is really better ? i got a11111
It is faster than a1111 but it looks exactly like automatic, I use it now
ok how much faster ?
In a1111 my gpu takes 19 to 20 seconds, in comfyui 20 to 22 and in forge 14 to 17
Is not thaat much but still
ok good to know thanks
🤗
and witch one is the more complete ?
Forge has more preinstalled options
i fall on a video that say A1111 is bugging when switching models ?
Should be fixed with forge ...
I didn´t test it, but if it works for a1111 i think it should work with forge
you can try it without deleting a1111
ControldNet already installed, didn't try with XL yet ...
hummm out of space must make a choice i can backup a1111 lol
You can use the models from you A1111 Folder ... no xtra download/space needed for that
yes , ok i ll try it
👍
thanks
in webui-user.bat you can set the folders ...
about that ... lol the Git folder ??? what do i put ?
The folders where all your A1111 Models are ...
You install forge ...
git will install forge ... better than a one-click-installer
Loras are usually the specific task model, they are used with your SD model. Other than that there's a specific model for SVD, there's a series of models of cascade. There's models meant for speed, there's Riffusion which is a spectrogram model for generating audio. I'd recommend going to civitai and check the models out there. Give you an idea and probably you'll find what you're looking for.
So no SD3... well, it's finally over.

Ah, then I have to learn how to use Loras. Thank you very much.
@loud solar i dm you ok
Of course some odels are trained for special tasks, too. Like photorealism, anime, etc. just have a look at civit.ai
OK
Ok, ok, thanks you.
]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]
So I can download the model I need and put it in the models folder and use it when I need it without it interfering with the others?
wait what
when
don't let the smoothbrain assumptions get the best of you.
how should i do can generate imagines?
Thanks for calling me stupid, I know that already.
now, it is "# 🔴 Bot is currently down! Bot channels are also unavailable until we are ready to return.
In the meanntime, try running Stable Diffusion locally or using Dreamstudio & Clipdrop."
how to solve this problem ? help
That's not what I did. I mean that its easy for people to be pessimistic right now if they ignore all the details.
You will always choose the mode you wanna use right now ...
we don't know since currently the CEO stepped out of the company and nobody knows what is going to happen currently until someone from stability says something
I have 30+ models in my folder ...
I understand, and you use one depending on what you need. This is my first time using this and I'm pretty new.
That's correct ...
Ok, ok, other question, how do I do to obtain 4 pictures instead of 1?
The Loras are also interesting. They add additional styles als are way smaller
In A1111 or SD Forge it's here
"Batch count"
Tried to upload a picture 🙂
Here or in the checkpoint?
checkpoint or model is just the memory you load into your A.I. and you can use different user interfaces ... in A1111 or SD Forge you set "Batch count" to 4 to start with 4 pictures
Ok, I'm trying to upload a picture bau I can't before I could...
Every time I write a prompt, SD "learns" about it? I am doing a paper on the canary stick, a martial art from the Canary Islands but when I write prompts the model does not respond, "does not know", it is a little known discipline in the world. The more prompts I write, is SD supposed to learn more?
I'm not really sure about the learning. I think maybe it learns in a session you use it. But if you restart the A.I. it would be gone ...
I think this chatroom doesn't allow pictures ...
Uhm, got it.
Ok, thanks.
It's more that I sometimes got kinda artifacts from pictures I have created before in that session. But the real learning process AFAIK is training a model or LORA ....
Got it.
Once you've gotten a handle on models, you should check out workflows 😅. That's where some craziness truly begins lol.
for diffusion models, training is a whole separate process. it's not like chatgpt where it retains the knowledge of the conversation.
I'm looking for help generating a 3d video using the sv3d-u model, is that possible here, I'm trying to use comfyui but I'm not finding alot of video tutorials. I've tried google colabs too.
I think the closest analogy to having a conversation with chatGPT would be successive rounds of inpainting
sv3d is for vid -> 3dmodel btw
SV3D-P is for 3d model
SV3D_p: Extending the capability of SVD3_u, this variant accommodates both single images and orbital views, allowing for the creation of 3D video along specified camera paths. Or am i misunderstanding this ?
it turns image to orbital video
what the hell is this news? emad resigns and sd3 release in limbo?
yea basically
more sd3 invites supposedly today
gosh. hopefully, they will release it. this sucks. he also said that they had video gen using patches like sora
yup
resigned from roles. he still owns stability
Damn, how do I solve the problem with jsonmerge?
yes, majority. hopefully we will hear soon. i would think stability would put out some info
i saw the facade cracking and falling away after devs started ripping on other devs and calling new models like proteus "a virus" . basically internal people at stability just shitting on everyone else from their plateau. culture of elitism is bad leadership.
but thats all just educated speculation on my part
did they say this before or after Emad resigned?
shouldn't they be beyond proteus level anyway with sd3? that sounds crazy
i haven't used it yet
yeah sd3 should be SOTA currently in image gen
source?
😅
sota for consumers yes
when sd3 comes out, it'll potentially be sota. all the tooling and refinements of older models will still exist though
if sd3 comes out
rumors about the board wanting to hold it back and only provide access through monetizable means like blockchain services
rumors are rumors, it wont do that
You should not believe anything except messages by Stability staff or the official comms channels. Anyone can write a tweet.
their partnership with render network blockchain is pretty substantial
i tend not even to believe tweets of stability staff either. i put all the pieces on the board into play
Stability can calm everyone down in 5 seconds if they just say "don't worry, SD3 will be open source." here. The fact they aren't saying anything is a bit worrying imo
It's a Saturday. Most staff are not working.
hugging face acquiring the company i think is the best path i've seen. i need some of that spice melange to be sure thoughh
when blockchain?
that would be great
see what i mean. the speculators are already circling. its all gone to shit.
blockchains are basically just ponzi schemes for big businesses
Will we have airdrop?
came out of nowhere for me.. when did you get the hint?
i had complete trust that no legitimate startup would dabble in blockchain before this week. i figured stability was fully legitimate. now they're partnered with 2 different blockchain companies. so it blind sided me too.
i knew the internal culture was a little broken though for a bit now. maybe longer and just didn't want to admit it. comfyanon shitting all over the work of illy's forge ui and claiming he coded it all, was one of the first bricks to fall from the wall
We don't know anything. You can't make something up and then rely on it for another made up thing, and another, and another.
sure i can
litereally everything is made up and acts as the basis for other things people make up
https://twitter.com/GozukaraFurkan/status/1771315565182783611 another thing thats blind siding me this week
this goofball spends most of his day spamming links to paywalled content on public/free huggingface models and every discord he's on
His credentials as an academic have always been questionable to me, but i don't care enough to verify.
Thank you for clarifying, do you have a suggestion, I was looking into deforum on google colabs but it is having an issue with the K-diffusion model.
hey guys, I'm looking for learning content for using diffusers,
i've been using it for nearly a month, and yet I've never succeeded in using it to the fullest with Civitai models and its resources
afaik, you have to convert safe tensors to a diffusers format in order to use them with the diffusers library.
is actually my biggest concern with hugging face potentially buying stability is that they'll make diffusers into the standard rather than the tidy .safetensor files we all enjoy
Academic credentials tend to be relative to the academy, etc.
Regarding SUPIR, he has taken the Gradio demo and scripted about 41 versions of it for people
41 versions to what benefit?
He’s a DIYer, but he has provided accessibility pathways that are somewhat valuable to the average person
No real benefit lol
you know what, i'm not going to argue the merits of someone who sings qanon level conspiracy nonsense. good luck
😂
regard the models yes you are right
but regards using loras no it dosen't need i think
i've seen youtube videos , they only download the lora from civitai and uppload it to hugingface thn use its id
when i do that i only get errors like missing models and somthing like that
Closed Sourse Censor Diffusion 3 out yet? Lmao
All models are censored. Any kind of dataset culling is technically censorship. Change my mind.
sd3 is dead, i told you guys))
yeah the hype died out for sure.. but probably not for the reasons you were on about earlier
now it's going to be censored forever, RIP SD3
yeah seems like a good chance that SD3 weights will never make it to huggingface
hugging face seems to be in the running to buy stability. don't count the chickens yet. BUCAWK!
it seems like SAI is heading in a bad direction, so HF will need to make some big changes if they do buy SAI
how likely is it that hugging face won't become closed too?
seems likely that HF will stay open to me, that's almost all of their appeal
HF is quite established in their sector. Of course anything could change, but it'd be a black swan event if it changed today
I'd like to think so but we have seen Mistral, OAI, now even SAI maybe
Mistral technically hasn't sold out
Emad resignation means open source death, Stability AI is already dead
hugging face's primary purpose is to be a development platform. changing from an open platform to a closed one would be catastrophic for them. It would allow competitors to out perform them.
You're just a doomer
has HF developed any model yet? Maybe they will continue to have open platform but idk if they will release models
Yes, they released a 180b model a year ago
But their purpose is to be the AI repo as of now
a lot of code too
Leaks please 🥺
1.5 is over shadowed by a lot of better models already. It just lives on because its all poor people can manage
.....
Does SD offer a killswitch manager job? Willing to travel 🙂
What will become of open source once SAI is gone?
noone knows
Our last hope if you don't want to end up in controlled and cencored cloud subscriptions ...
that's what it will be
I would love to get banned on 30 accounts and pay hundreds of dollars trying to bypass filters tbh
Like genuinely sounds fun
open source will end still in mostly portraits
Big companies are already moving their services into the cloud ... not sure when local computinig will end ...
lol open source will live on. there are lots of teams out there learning all the new tricks . the cost of compute is always coming down.
we'll see...I hope you're right
local computing won’t end, but it sure will get paywalled like some adobe type shit
we're just getting some decent video stuff - then we see SAI crumbling
paywalled and, as we saw with gemini, censored
They will sell it to us as: "You don't need an expensive computer any longer. A display with internet connection will be enough ...."
right, and you don't need the images you want....we'll tell you what images to make
hi friends.
And pleasy pay monthly ... they will say less loud ...
pick from a list of 3 pregenerated prompts lol
first official post 🙂
Hi
A "surprise me" button will be enough ....
right
I am on the struggggggle with comfy and getting video with SVD
if comfy is to complicated try Stable Diffusion Forge ...
forge doesn't work with SVD iirc. You'll want SD.Next for that
Forge works with SVD 😁
comfy itself is pretty barebones and needs custom nodes
You can do a lot with comfy but it needs time to understand the workflows and get the nodes and models ...
Forge only the XL svd model works though
I come from a very video heavy backround
However, I must have done SOMETHING while working in photo to have it not want to run the barebones json workflow
cant paste pics... I Spose thats a good thing concidering lol
I was gonan share my workflow cuz I know you guys would spot something stupid in 2 seconds flat
My favorite way to work with animatediff lately has been automatic1111 1.8 and the extension. Forge has potential with the unet patcher, but the dev needs to rebuild it from the ground up and he's in the middle of classes right now, so that'll be a while.
I've seen a couple of dedicated UI's specifically for animation, but i've not tested any yet
nodegraphs tend to become rats nests fasts
This chatroom is without pics ... there's a general one with pics ...
So I had to sleep. Whats up is Satbility over? did they join the evil empire?
ill slide in there for a sec
Good luck!
you blinked ! emad resigned and dr furkan is a qanoner
Trump bought it to creaty content for his network 😄
dang
Ohhh... he ran out of money ... I forgot ^^
Error occurred when executing KSampler:
CUDA error: invalid configuration argument
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
I was right. lol.
well... lol it would help if I knew what that meant 🙂
I am just a user ... and less of ComfyUI ... no idea, sorry!
lol
Comfui. ... NO CODING REQUIRED!......
uh huh 🙂
I assume its something stupid I need to download and put in the right folder
I wonder if its possible to create an ok comic with stable diffusion, ive seen some people do it before but im not sure what techniques or tools they use.
it seems interesting, also i think i wonder too much.
its pretty much what im trying to do, but
Problem is to keep consistant charakters ...
its not just that, its other stuff as well im afraid
You can try using famous people ... or train a Lora ... haven't found another solution yet
Where can I find more info on the ksampler sampler names? The official Comfui community manual is under construction
even if u have consistant characters, its so hard to make anything out of it, emotions, movements
I know ive heard different method like seeds, ipadaper and such.
its at time like these si wish i had drawing skills but i dont also im impatient.
If you are able to sketch with your hands it might be easier 🙂
models that aren't over fit help a lot with emotions
by chance do you have a good example or nah?
a good comic should have pre work done to it too. its not a tall order to expect frames be pre planned and have a little open pose composition to them
you have a point!
cyberrealistic models i've been toying with lately. there author made wild card sets that work well with his models
u would need a rly good setup to make anything work, then with multple characters, its awhole other mess
im trying to make characters seperate, and then paste then into backgrounds but.. its hard
multiple characters in one panel isn't too difficult. just tricky. Regional prompting in latent mode allows for multiple loras
oh wild cards ive been wanting to try therm but they seem kinda daunting...
they're just text files that fill a placeholder in your prompt. super useful
A.I. is just another tool that needs a lot of work if you want to have good results ...
i should look into them and maybe make my own.
People want that startrek holodeck
Where vague prompts always do exactly what they want
Me too!
we all can relate.
Think i don't want my brain to expand into the entirety of the Enterprise D class? I got that D envy u know?
Fooooooooooooooooooooooooocus is for comic books
I'm learning to use A.I. 2 years now ... and I think it will be a neverending story....
Art is an adventure
we will plateau
An... artventure
why would foocus be better for comics?
like with video editing and photo editing, eventually every sing;le program will be good enough for the basics
I use stable diffusion ai here and there but my setup still stinks, but its ok.
AI wll reach that too when you just boot up whatever and u can make character slock them and play with them etc
¯_(ツ)_/¯
There's always ne stuff coming and while you learn about the new one you might forget the good old one like regional prompting ....
yeah lol i used that like for 3 days
noted
regional prompting is bae
As long as you have fun with it ... do like me and try to continue ... but some gifted people better use their hands ...
I littealy have to run off of cpu, because my graphic card shuts down my pc when i use ai stuff on the gpu.
that sounds kinda wise.
Many people just think: "Oh ... we have A.I. it can work for me and I don't have to do/know anything." but that's the wrong idea ...
yeah they are super wrong. i thought kinda like that at first but then i dident.
It's a tool like a pencil or brush ... you have to learn to control it .... if you don't want random pictures ...
driver problem maybe. or it's drawing too much power from the PSU. potentially solved by using MSI Afterburner and capping the power. Or doing a total driver uninstall and reinstalling
i am going insane
i spent 2 days trying to train loras to get fingers perfect
i think ive almost done it
i remember when i first started learning photoshop. Seemed like i could do anything. Then I realized the skill ceiling was way higher than i had thought
here its more of a skill/hardwhere cealing lol
takes years to get good at it
Ill look into that when i can i think its drawing too much power honestly. ive tried reinstalling the driver so much its no longer funny.
hardware issues are skill issues imo. That's just PC gaming life
don't got frames? git gud
are u cold? just run some workflows, BRRRRRRRRREEEEEEEEE
its like having a open fireplace
fires are better they give that full spectrum infrared heat instead of just convection
just light your pc on fire for that effect
ez
aw, civitai is down = (
ah dang
😂
up here. maybe a dns issue
maybe it was temporary
these days i use non default ISP DNS. Like google's or cloudflares
1.1.1.1
and the backup is 1.0.0.1
im looking up how to use msi afterburner is it better to overclock it or do the opposite?
depends on gpu
some gpus can go wild
i have an amd 480... so yeah.
other gpus are just stuck with +1.5% gains
i'd just limit the power use ot 80% and see if that helps
I'm using silend mode with my 4090 most of the time ... still fast enough for 90%
noted
if you're going to experiment with it, make one adjustment and test. then note that and do another and test. iterative
lol
i forgot what the reccomended settings were for a 3060
im using whatever
it runs everything
lol
do you offhand know what your psu is , and what your cpu/gpu are?
except cascade
had to use some technique to get that to run
using some github repo
load one model then unload then load other model
offloading
kinda my cpu in an intel i3 with 4 cores, my gpu is amd 480 (8vram) but beside that im clueless.
amd 480 huh. might just be a driver issue 😉
noted
i had that card. it's a gooder but it's and older one
what you mentioned is traditional offloading
maybe be able to squeeze some diffusioning out of it on linux
yes
load model a, run model a, unload model a; repeat for model b
Possibly but i find linux so hard to use.
oh cpu offloading sucks for diffusers
no we arenty processing it on cpu we process on gpu but store one model in sys ram
is easy once you're over the learning curve. never any problems with linux ever after that (don't believe me)
then load the other model in vram once other offloads
Doesn't Forge do something like that? Looks like it's caring about freeing memory (VRAM) again ...
Yeah it seems to do that
Note that although the above codes look like independent codes, they actually will automatically offload/unload any other models. For example, below is me opening webui, load SDXL, generated an image, then go to SVD, then generated image frames. You can see that the GPU memory is perfectly managed and the SDXL is moved to RAM then SVD is moved to GPU.
This is how I imagined SD3 8B running on 12GB, by offloading T5 and the MMDiT model to RAM while not being used for example
But someone told me that 8B MMDiT will take up a LOT of VRAM apparently, so I guess 12GB is out of the question
You will have to try ... even if it doesn't work soon ... it might work in a few days/weeks
New technology is coming, people try to make it work and later they will work on efficiancy ...
has anyone tried to (intentionally) make a photomosaic? i tried to upscale an image yesterday and didnt configure some settings properly, and ended up with a very interesting photomosaic. I still havent figured out whether or not its an intended feature or if i stumbled across something unintentional 😆
[Memory Management] Current Free GPU Memory (MB) = 11075.047607421875
[Memory Management] Model Memory (MB) = 186.4295482635498
[Memory Management] Minimal Inference Memory (MB) = 1024.0
[Memory Management] Estimated Remaining GPU Memory (MB) = 9864.618059158325
Moving model(s) has taken 0.12 seconds
Now ...
we are at a different stage
we are at the "what the hell is happening" stage
when comes after that stage
😮
They need to send updates more often
It's there ... maybe not for you or me ... even if I could send you a copy you couldn't use it ....
im not even sure a lot of people at stability even know whats going on with all the crazyness, resigns and stuff
Time is running ...
did e-dog write anything else on twitterinos?
Guys, I added some — argument to my SD, and it made my SD run so much better, things rendered so much faster
like, maybe, "psych, it was all a joke, got u", perhaps?
I didn’t know how inefficient thr standard SD UI was
It's hard work/learning ...
if this is true i was got hard
got me so good
"i got hacked lol"
jokes aside, can anyone actually check what emad latest tweet is? he said he would update us
No new tweets
X
thanks
I'm kind of doubting that the full 8B model will be released to the public. They will probably just release the 2B version and keep the 8 and turbo8 closed off for commercial purposes. If HF does actually buy them, that will take months to a year before they'd release anything
What arguments, if I may ask? And for which UI?
was about to ask the same.
Is there a 3D camera for framing imagery available or in the works?
speaking of UI's , im convinced that only masochists actually enjoy using comfy
Just the webUI, I think it’s called, I’m not on my PC at the moment, so I forget which one.
It wasn’t xFormers, it was the the other one, I think it has a D in it
https://twitter.com/DataPlusEngine/status/1771581518713950273 Datavoid working to do a base transformers model now. man with a plan.
automatic1111 is your ui im guessing?
I believe A1111 is indeed referred to as just WebUI
But as far as I know "WebUI" is also a collective term 😂
i tried calling it steak sauce for a while but no one knew what i meant
For all web user interfaces
Try porcupine quills next!
yeah, theres forge, comfy, a1111, foooocus, and i think a couple others
I think
Is it made with Gradio?
cant think of a performance argument that starts with a d, you will have to let us know when you see what it is. i run these for performance --xformers --cuda-malloc --cuda-stream --pin-shared-memory on forge
couldnt tell a difference either way when using --alwaysgpu(or was it onlygpu?, i forget)
anyone knows how to do facetomany type img2img generation on A1111?
its called "stable-diffusion-webui"
and the arg is "--opt-sdp-attention"
Then A1111 for sure
ah ok
It seems to result in performance improvements for AMD GPUs 🤔
I did a quick search
I wonder if it will do anytthing on an Nvidia
will run it in a bit on forge and report back. for reference, nvidia rtx a4500
its a different kind of memory optimization. like xformers, but it works on amd
what happens then when i run it on nvidia with xformers 😄
i dont think it'll use both. it'll use just one of the two
k will do comparison of x formers and --opt-sdp-attention
is there a tool that turns video into jpg?
I don't think I use even --xformers 🤔 Would it improve model loading times?
am guessing that it wont work with --cuda-malloc --cuda-stream --pin-shared-memory
Those are painfully slow
Likely a video editing tool can do this for you. Take a video, move to a frame, and export as image 🙂
^ davinci resolve. its free
bruv i mean i need all of the frames
and like at 15 fps so every other one has to be deleted lol
whats the opposite of interpolation
The question was unclear then.
i THINK you can also do that with davinci, not an expert with it
sorry
Vegas Pro offers that I'm pretty sure
thats crazy
actually im almost positive danvinci can, just dont ask me how. im surte some quick googling will get it answered
damn i think premire can do it too, export sequance as images, and i think u can set sequance fps
The second one is about DaVinci Resolve
Though this second thread seems rather unhelpful 💀
https://forum.blackmagicdesign.com/viewtopic.php?f=21&t=106236 I think this is better
It helped so much, things ran way faster, and there was no VRAM memory leaks
1024 images ran to the speed of what the 512 images were

it's not a vram leak (heh vram memory like atm machine)
all the vram is allocated legitimately , you just dont have enough.
a leak is when memory that isn't needed anymore is never released
mishaou assistant will make sure it gets released if youre haveing an issue with that
I wonder why it's called a leak if it's never released 🤔
yeah the terminology is not the best metaphor
ok well, im probably just doing something wrong but it seems that --opt-sdp-attention does not work with nvidia at all
it does. in the past i've tested both. i prefer xformers. just seems more reliable
tries to load xformers even if the argument isnt used
hook me up ..
my first installation was succeed to open SD on my browser (http://127.0.0.1:7860/)
but it's out of memory
"OutOfMemoryError: CUDA out of memory. Tried to allocate 40.00 MiB. GPU 0 has a total capacty of 4.00 GiB of which 0 bytes is free. Of the allocated memory 3.40 GiB is allocated by PyTorch, and 52.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
Time taken: 1 min. 13.7 sec."
damn i thought extras tab would let u do batch stable diffusions, but its only for extras. how sad lol
i think u need like 8 or 12gb to do stable diffusion stuff
^ ive heard of people using forge with 4gb.
so glad i bought a 16gb card. wanted 24 but the prices were too insane
ill start with a 4060ti, until 5090 is released
so, this may be a forge issue but, even when removing all args except the --opt-sdp-attention(yes i know, this one has the extra args, forgot to copy the log when running woithout, but result was the same)
Launching Web UI with arguments: --autolaunch --api --theme dark --opt-sdp-attention --cuda-malloc --cuda-stream --pin-shared-memory
Using cudaMallocAsync backend.
Total VRAM 20470 MB, total RAM 28652 MB
WARNING:xformers:A matching Triton is not available, some optimizations will not be enabled.
Error caught was: No module named 'triton'
xformers version: 0.0.23.post1
Set vram state to: NORMAL_VRAM
Always pin shared GPU memory
Device: cuda:0 NVIDIA RTX A4500 : cudaMallocAsync
VAE dtype: torch.bfloat16
CUDA Stream Activated: True
Traceback (most recent call last):
etcetcetc
import xformers
ModuleNotFoundError: import of xformers halted; None in sys.modules
Press any key to continue . . .
might be enabled in settings
i dont care enough to mess around with it much more 😄 happy enough with current results lol
To load target model SDXL
Begin to load 1 model
[Memory Management] Current Free GPU Memory (MB) = 17430.19991874695
[Memory Management] Model Memory (MB) = 4897.086494445801
[Memory Management] Minimal Inference Memory (MB) = 1024.0
[Memory Management] Estimated Remaining GPU Memory (MB) = 11509.113424301147
Moving model(s) has taken 1.57 seconds
100%|██████████████████████████████████████████████████████████████████████████████████| 20/20 [00:07<00:00, 2.62it/s]
To load target model AutoencoderKL█████████████████████████████████████████████████████| 20/20 [00:05<00:00, 3.14it/s]
Begin to load 1 model
[Memory Management] Current Free GPU Memory (MB) = 12529.618307113647
[Memory Management] Model Memory (MB) = 159.55708122253418
[Memory Management] Minimal Inference Memory (MB) = 1024.0
[Memory Management] Estimated Remaining GPU Memory (MB) = 11346.061225891113
Moving model(s) has taken 0.25 seconds
Cleanup minimal inference memory.
Miaoshouai boot assistant: Memory Released!
Total progress: 100%|██████████████████████████████████████████████████████████████████| 20/20 [00:13<00:00, 1.46it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 20/20 [00:13<00:00, 3.14it/s]
what gpu
rtx a4500. worth noting that model loading has been way faster with forge, however the integrated controlnet doesnt like to play nicely with some extensions
I just have a simple 4080
i wouldent call that simple
also if anyone knows how to get triton working with windows (without running a virtual linux environment or using docker) please share your knowledge
whats the best ai voice cloner
applio
What's an "AI Voice Cloner"?
its open source....
implementing it in a sting operation program
sting operation? ewww.
the 4080 is a beast. dont believe the hate for it. just youtube clickbait mostly
3090 is way cheaper and with more vram
5090 is coming this year anyways so all of those cards will be irrelevant
my body is ready
💀 what?
this makes me think you are a very weird indivual
why do you dislike sting operations?
you do realise who sting operations target right
sting operation makes me think of drug bust
lol see what i mean. haters gonn hate
ay going after those
🤔
pedophiles?
yeah by all means then proceed 😄
😆 i assumed drug busts
i would've regretted buying a 3090. it doesn't have very great fp8 support and that's been a huge boon lately
imagine if SLI was still a thing 🙂
sting operation makes me think of that song that ice cream trucks play. i'm an old
its exactly what it sounds like, instead of text to image or img 2 img its like text to voice or voice 2 voice, but using someone elses voice
So, text to speech 💀 But way more advanced
mhmm
text to speech is better than voice 2 voice
voice 2 voice is just 12 year olds using voice changers
The night is quite young, I should try it out 🤔
ugh. plans!? iiii dont wanna play. i just wanna stable diffuse all day
yeah im not too familiar with that tech, but im checking out the applio thing someone mentioned. i actually had an idea for a short animation i wanted to do using marlon brando's voice from the godfather, and i havent been able to figure out a way to make it sound good
oh i guess we can't call it stable diffusing anymore , since they're all but dead now
hmmm open pose is giving me a decent thing to work with, but it looks like SD doesnt understand open pose
marlon brando + my hybrid ronald mcdonald \ It . should be pretty damn funny if i can get it working
you need a preprocessor AND the correct model for the architecture you're using
make sure you arent using a preprocessor, and that youre using the correct model. upload your pose as reference image
meaning you cant use sd15 openpose models on sdxl
wtf is a preprocessor
can you not post images on this thing?..
processes the pose out of an image before diffusion begins
you dont need the preprocessor if you already have the pose, preprocessor just if youre creating a pose
