#💬|general-chat
1 messages · Page 138 of 1
SAI has the ability to make a platform for OS dataset creation & distributed cloud training systems.
Its just not being done
they rely on datacenter training & dataset providers
personally i think there is a need for a fully OS platform for AI that allows crowdfunding
they are basically burning money
huggingface is missing that part, they allow sharing of datasets & models, but theres no system to let people create data on demand and get paid for it or have people contribute compute for a distributed training cloud
in minus
8 million of bills and yet only a fraction is being covered by the revenue
there are a bunch of AI coins that have distributed compute and token economies for it, but no way to pay for dataset creation
and its highly unrealistic to get the community to cover the rest of the bills
its not
especially long term
all it takes is good communication and transparency
well now its the time, nothing comes from either side
they simply needed those big investors and they didnt come up to Emad
he demanded 4 billion dollars before
How do we access the artisan channel
gotta sign up and pay
#artisan-faq @weak cosmos
SAI isnt in a position of a Adobe that has 19,4 billion dollars of revenue (2023) with products that dont rely on generative AI in the first place and can cover the costs and has a bunch of high profile investors on them
or Google and co.
as long as they're not acting and taking the opportunity to utilize the community, the community is going to diversify and spread out over different competitors
spot heal brush enters the chat
so the longer they wait, the less of the ~5 mil SD users that exist rn are gonna be willing to pay
my hope would be that they release a big 8B model at some point, and use the api/paid bot for revenue from all the ppl that don't have 16-24gb vram
so sd3 still going to be open sourced? or is this to be taken like the repeated phrase „in two weeks“
O_O
it will only be two week later once two weeks have passed
that simple
i think the people are diversified anyway, although the majority might be strictly FOSS
i wonder what their plans are for after SD3
since SD3 is supposed to be released open source as well
Hello everyone, I just joined and I am using comfyui + stable diffusion, SDXL models right now. I am still very new to this and am hoping someone can answer some questions or provide a good article / video to help clarify some of my understandings better.
- Are certain base models better for different types of images? I thought SDXL was just a higher end approach to SD1.5 but I see so many 1.5 models out there.
- I have read some prompting guides but I am still struggling to get the output or expected results I am hoping for. Is there a good prompting guide / template that is highly recommended that I can refer to?
I wonder if SD3 will be released first, or the ELLA finetune of SDXL
how do i use #1237459938901491852
SD1.5 was fine tuned a lot more than SDXL, this is partially because it came before, runs on cheaper hardware (cheaper to train, too) and the NovelAI leak means anime/realistic models have some overlap, this is why you'll find so many finetunes of it
(technically SD 1.5 was also a "leak" by RunwayML, as StabilityAI didn't initially seem to want to release it)
i thought for a second Artisan might be free lol
then as soon as i saw that it uses SD3 i already knew it will be credits based
Your plan will be billed automatically once the free trial has ended. You can cancel or change your plan at any time.
At some point this practice has to end
With comfyui as long as I am not mixing and matching the base models I should be able to just download and swap out the base / vae to use any models available?
I wonder if there is a plugin for comyui to limit what is available in the dropdowns based on the base model selected. I just started and I feel like it is hard to keep track of all the models in those drop downs :P.
yeah you just need to use a SD 1.5 VAE for 1.5, and a SDXL VAE for SDXL; if there is "ringing" look up some "SDXL fix" VAE as it's just some steganographic fingerprinting but it makes colors a bit screwy
Guys I when I add the mov2mov extension it doesnt show up. Can you help me
I don't use ComfyUI and instead Automatic1111's, but AFAIK you kinda create a workflow and you're set, I guess? A1111 is more like a GUI thing
ComfyUI is faster
Do 1.5 loras / vaes work with the variety of 1.5 base checkpoints? I see 1.5 LCM / 1.5 Hyper etc.
more versatile
1.5 LoRAs only work with 1.5 and not all models, but they're very general. SDXL LoRAs almost always need a specific model to work right
you can make subfolders in the diffrent model folders, like i have SD1.5 and SDXL folders in checkpoints and loras
Thank you, that is just what I wanted to hear
good to know, thank you. Going to look for the stable diffusion sd1.5 base models / vae to start. I haven't seen a link in the discord yet but is there a links sticky or wiki on here with some of the common links?
@neon oriole so... about this artisan rollout
What is the VAE I want to get from https://huggingface.co/stabilityai for sd1.5?
while i don't think there is a "right kind" of personality for anything, there are aspects of this where the manic part of the personality is showing. @neon oriole my feedback for the appropriate person.
@shut hare are you open to feedback about artisan?
vae-ft-mse-840000-ema-pruned.ckpt
Thank you, quick search found it right away on hugging face
I respectfully disagree with the pricing and implementation of Artisan. Inflation is hindering the entire planet. A billion people want to use this software, but survival comes first. And it still takes hundreds (if not a thousand) images to make a GREAT image. What is our time worth. Professional use isn't there yet, even with 3.0. This is just fun; a hobby. Industry marketing of AI is terrible. You guys need a major pivot and new ideology.
Takes people a lot of learning curve to get prompting down.
AI as we know it is literally only like a year and a half old
Imagine going back to 1979
when microprocessors were still in infancy
those 156khz processors look like kids toys now
same thing will happen with AI. In short: patience, padawan
its already been 100 years
while i don't believe there is a "right kind" of communications style, things evocative of fringe beliefs or larger-than-life dramas make it harder to distill what you say
I'm just a professional artist, business owner, Masters holder and retiree. I'm a nobody. Ignore my feedback.
what do you think my credentials are
can this bot be used in other discords in the near future?
Wow, he has a masters! Guys he has a masters!
i don't think we should bully the guy. that's not what i am saying at all
these stability folks have a few more shots
they have to go and tell their bosses something. they have a strong incentive to say the best things possible. of course, it sort of doesnt' matter, they have metrics.
i wouldn't have set something up without being 100% confident about the metrics
to keep something like this sustainable you have to be the best
i don't think this is true. the crown of best art generation model has been traded between a bunch of different authors
StabilityAI still has a lot of goodwill in my books. If they decide not to release SD3 weights for some reason, then I'll grab my pitchfork. If they want to make money off of the model they spent millions on training until then that's their prerogative. They don't owe anyone anything because we didn't pay anything.
A lot of people with actual skills and experience and wisdom leave this server. I wonder why.
nobody with experience wants to hang out in the pissy kiddy pool i suppose
I can pay for sd3 weights.
noone said we wanted handouts.
What I wont pay for are "cReDiTs" to use some dude's gpu
soo you agree i think
I want local and willing to pay for it.
open weights == best
well on the flip side, much like the VA clinic or academia, there is room for a variety of personalities, and those people can make things people like too
it's so hard
i don't want to generalize about that. i am just talking about this bot
Hello!
i was referring more to comments like this "Wow, he has a masters! Guys he has a masters!"
i feel like they don't have a good grasp on what their strengths are. making more money than it costs to do this is definitely one of the weaknesses, but that's true about many endeavours
anyway if they dont attach themselves to money soon they will be steamrolled
That's not their business model. Getting a few dollars from image generation is not how SAI survives
i often agree with michael douglas. coffee is for closers. greed is good.
succinctly, they should have a simplified pricing, and offer unlimited generations if they want subscriptions, or implement openai's credit-and-usage based model
i don't think they realize how anomalous midjourney is
MJ is a blip that will likely be acquired once they've saturated their market opportunity.
or....get this..... offer the option to use locally... that way the community can train and improve their models
it's a win-win
midjourney clicked into an audience of graphic and art professionals who are arbitraging an AI bot to do the grind of creating the art, and focusing on the more human-centered "selling art for SMB purposes" business
Same thing happened with the guy's previous company, Magic Leap
it's kind of like how bethesda releases half-assed games bc they know the modding community will fix their bugs for them
i didnt say that though, more like a OpenAI gobbling on some microsoft situation to comtinue to be relevant with a badly designed model
hand tracking was a blip. now its so readily available and can be done with cameras
my only complaint is that the bot seemingly does not work im dms
How do you know it's badly designed?
From what I've seen SD3 has a lot of potential
well, that is eminently a product decision. does midjourney work that way yet?
yep, you can use the MJ bot via dms
when 70b models can come close and beat a 1.8T model you have issues
Not really. The modding community is a small niche of palyers. The vast majority of bethesda players are on console or playing unmodded
i think there's a notion in their heads that they want to show the product off organically
ya.... cuz the masses prefer xboxes and playstations over PC
this makes some sense
what the masses prefer does not matter to those in a ... higher class
Oh sorry, You're talking about OpenAI. I thought you were talking about SD3. I kind of agree with you lol
garbage in garbage out, was trained on garbage
If modding were as big of a deal as people made it out to be, nexusmods would be publishing games and be a power house dev company
they're niche though
well you need the game for the mods to work
i think the customers of midjourney's customers are going to cut off the... customers
anything that is priced at $1,000/y is graphic designer / ad agency pricing territory. if you were building a real business on it, you would charge millions
Midjourney is a fun toy, but no one will be powering their business with it, which is where most of the money in software comes from.
nah i think stability is still on track with technology but will have money issues
there's only so much arbitrage
yeah i think it's (em)powering the bulk of creative agency labor to focus on sales
I think MJ will just become a common commodity and what makes them special will fizzle out. They're competing with Adobe in this realm. I don't think they'll last much longer against Adobe. Aquisition incoming
i wouldn't write them off, after all the model is good
which is what people care about
it creates the visual art that the customers of ad agencies want
it will all take a very long time
for a brand's 1 social media manager to start using midjourney directly
anohter year and many models will be as good as MJ5 was
instead of asking vendors to use midjourney for the brand, which is literally what is happening now, but at human-art-production prices
i don't know how much of midjourney's customer base this describes
ya see i dont think it was good, it shoehorend you into things it makes nice
good is highly subjective, yup
it is
chatgpt had the same effect, and indeed, there is less demand for copywriters, because brands (more broadly, anyone paying a vendor for writing services) can now get the writing they need directly from the source, instead of paying human authored prices for arbitrage
here's the rub
i am very opinionated about this
I don't even think MJ7 is that great tbh
SDXL in its current state isn't too far behind if at all
comparing MJ vs SD is like comparing Windows to Mac
i don't know if @shut hare and @finite cloak bosses are opinionated about whether or not brands, or creative agencies, or any particular player in this business, are actually worthy
they are just too different in many aspects to be comparable
"less demand for copywriters" yeah i'm not sure that's true. I think the people using gpt for copywrite are the same managers who would be lazy and not hire copywriters in the first place
which is what i hope they think about. another image generator API is a very unopinionated thing
i mean, maybe their special opinion is, "furries and hentai are worthy"
I guess this is true, but the differences between SD and MJ are quite important. Censorship, customizability, tuning, they're all important
it just doesn't make any money lol
I wish I could say SD has the censorship argument in its favor..... but it hasnt since 1.5
so maybe the problem with the bot, compared to civitai, is that it will not make furries and hentai
EVERYTHING is censored these days, NO EXCEPTIONS
which is kind of what makes their opinions special
Base SD is censored to prevent frivolous lawsuits. No one uses base SD because that's not the point of SD
if I were running stability, i would have different opinions. i'd be like, okay, i don't really care about branding agencies, or really, the vast majority of stuff that artists are paid to make in the first place.
Right but then it's the community uncensoring the model, not SAI
MJ was never competing with SD imo. It was always competing with services like Adobe. SD is pulling ye old blue ocean strategy. Open weights are a wholy different market strategy
it is competing with runwayml, which made SD 🙂
$
You seem to be taking away all the credit away from SAI which keeps releases open weight models they spent millions training
The point is to fine-tune their models
Do i have to pay?
Lmao, stable diffusion just turned into shit already
What a lame
They could avoid those potential lawsuits at any moment. All they'd have to do is train the next AI models on the results of AI images currently being made. Now there's no trademark or patent infringement.
kind of. It was a team effort. And then RWML went and closed up their offerings and won't release weights or research
so you will never see me supporting MJ as long as SD is alive
Training an AI using AI images would result in shitty quality in my opinion
i think they have to be opinionated about audiences
which is so hard for manic tech people
Potentially, yes. Beyond my knowledge tho.
manic tech people are only opinionated about the stuff quite literally in front of them, like their energy drinks and porn
Why anyone would pay for this shit, when you have models like DALLE-3 for free, they're all fucking stupid
lol
do you mean bing image creator?
i've never liked the hyperbole term "censored" to describe models with curated datasets.
If a studio was making a tv show and never intended boobies to be in that show, i wouldn't call that show censored. If a comany is training a model and culls hardcore pornography from the dataset because their intention is to create a base model and not YiffingXL, that's not censoring either.
i saw there's a new discord bot for SD but I don't see a way to make it run locally
well i kind of agree that you should never charge money for something a giant tech company can give away for free. however you can stay liquid for long enough that maybe that doesn't matter
is it just a boring api wrapper or is it actually useful
Training images are one thing, but the harmful lawsuits are if, for example, some cringelord decides to create certain types of realistic images of a child with Stable Diffusion. In court, SD can say their models as they released them aren't capable of that so they are not liable.
like dropbox got to IPO didn't it? even though its core offering, google can give away for free
i'm just not sure why it is so hard for some manic tech people to say, "well hentai is kind of a huge waste of time"
i've never believed the "anatomy" argument either. Adding pornography to datasets just makes images look like pornos. Studio lighting. Fake women. Too much makeup. Porn is fake. It's not reality. people forget that
"you shouldn't drink energy drinks"
USPTO has already ruled that it's the author and not the AI company.
You gotta be really stupid to pay for sd3 when you literally have sd 1.5 for free (hundred of models included) + models like dalle-3 created by openAI for free too
the prompt is the violation
and many though sd3 was coming today lol
the nihilist in me wants to say all media is a waste of time 😉
even i was like "maaaybe" when i saw the ping
amiwrong.gif
i don't know. there are definitely better kinds than others. it's important to be opinionated is my point
what is their opinion?
midjourney likes hollywood, clearly. enough to train on every movie there is
it is infused in everything they do
they like hollywood and AAA Brand Aesthetic
I agree. Forming your own opinions can be a difficult task in a sea of hyperbole
this makes sense, it is aligned with where some money is
Would it be possible to do your LoRa training and other more intensive tasks on a cloud provider and then once you have that use an RTX 3080 10GB to generate images locally? Or would that just be more hassle than it's worth
people need to be taught to reject exageration
sd3 is not coming
not open source at least, not free open source at least cause they're all fucking stupid, they think their product worth by itself but it doesn't, stability ai models have always been good because its open source and people work on it, nothing else
💩 💩
well the opportunity i personally engage with, in my ai business as it were, is for games
I think you can train loras on civitai now, but yes your method is viable
not sure if you're trying to say your hentai point is bad by comparing it with this one, or if you would say both things and assume most people would agree
which is a very different set of aesthetics and requirements
Lots of people do it that way
No more of a hastle than getting a finished lora from civit
Do you have any evidence for what you say or are you here to sell pitchforks
lol hi
lol i am making up a colorful character of a manic, adderall fueled developer who's main interests outside of work are hentai and energy drinks
has there been pain api stuff like this when sd 1.5 and sdxl launched?
Everyone at SAI has said openly that SD3 weights will release
🤡 < just two clowns looking at each other
the smaller weights will be released
which i believe they are training now
i'll wait and see. faith is hard
Again, do you have evidence for this?
Nah, you dont need evidence buddy
Just see what they're doing, instead of releasing it one month ago they still selling shit to everyone so they get their filthy pockets with money selling a model that it's totally shit atm, they're a funny joke
💩 💩
well, if you drink too many it's probably bad, but I never drink them and dude, it's like magic for a game jam
i think before emad left, they were going to release the only model that they trained, the one that they use for the API, as early as April
i feel like, maybe they will get a taste of money with this paid bot and api, that sd3 might not actually come open at all
then emad is out, and they changed strategy, and they don't want to be liars, so they are training the best some-B less than 8B model they can.
They were training two 8b models, sd3 and sd3 turbo. What you're saying makes no sense already
Greed is Good .gif
but yeah i agree. it seems that the strategy is no longer a full release
I remember the times when SD 1.3 was leaked, SD 1.4 was released, then SD 1.5 was leaked, then NAI was leaked, etc... everything from Stability was "soon™️"
Yea
They just want money, may be they're releasing it in few months but you still gotta have to pay or some shit they'll come up with
Stability AI is dead, periodt
kinda like they halted this release and put resouces into releasing this paid bot instead? maybe = /
Didn't SAI release SDXL, SDXL lightning, and cascade?
i wish i could reach through the screen and touch your heart, and help you chill about ideas llike "filthy pockets with money"
they did, SDXL was great
lightning is fron TikTok
@astral goblet although really, so many artists i know, they have a very love-hate relationship with money, that i think also makes the brand-creative-ad-agency-industrial-complex audience pretty shitty to sell to
and funnily enough, it's not non-commercial unlike turbo
everyone wants money. its not a bad thing. literally how commerce functions
🦀
it's kinda weird when companies like Facebook, ByteDance and other weird ones (like that Saudi Arabian fund for Falcon, Yandex for some other model) are promoting open source
something something something damn commies somethin something
Prove otherwise
Cause as it looks now, y'all are taking people's money on a model that's literally shit atm, why anyone would pay on sd3 when you have models like midjourney that are 100 times better
stability ai only has sense if their models are open source
SAI's business model is releasing an open weights model, having the community make it really good, and then charging businesses for using any of its fine-tunes
zuckerberg going "open source is the future of AI"
that would be genius if they did that, but as it stands it's Stability's base model, with open weights "whenever it's outdated, I guess" (like Grok)
"literally" means "figuratively" now. i can never take people who use it this way seriously though. literally feels like they're just regurgitating what they've been told by a youtuber
alibaba is doing spectacular things too, thats kinda what i was refering to before, no money you get steamrolled
reminds of that movie where, if u dont have enought credits, it lessens ur lifespan
and gaben going "linux is the future of gaming" back in 2013
gaben got skinny! he's not one of us couch lords anymore!! (jk. skinny gamers unite!)
SDXL took 3 months to release as well, and 9 months later it's still the best in its class and widely supported. Where is the doomerism coming from?
they should do it like they did SD 1, and SD 1.1, and SD 1.2, and SD 1.3, etc
i game on linux all the time. While its not the de facto standard, it certainly isn't a non starter anymore.
SDXL is best at some things, within the scope of open models, after a lot of finetuning by the community. When it had just come out it wasn't the best in anything
I hope SD3 2B or 4B releases this month
That's the point of SAI models, to be fine-tuned.
"best in its class" >laughs in 1.5
yeah seriously
comfy said that they might release smaller models first
Stability peaked when it released 1.5
stability released 1.5
SDXL takes a massive steaming dump on sd1.5
isent that kinda a red flag tho?
naw. sdxl was a way better base model. 1.5 has so much spite woven into it. RWML put some bad mojo into it, then ran off to only do closed source SAAS projects
I agree! that's why I think they should let the GPU rich finetune it, the GPU middle class run it, and the GPU poor rent credits.
open sourcing is at the core of what StabilityAI has going for it, it can't compete with MidJourney and DALL-E 3 on the base models they make themselves
SDv1 wasn't released to the public until 1.4
it had months of private buildup too
considering this bot release? isent sd3 like, far away now?
why release this bot if sd3 is coming soon?
well, I've lost interest in stability that's for sure
yeah tbh the way he phrased "well 8B is not low end enough" or whatever was just weird, people with 16-24GB exist, absolutely not the majority but we could still use it 🥲
If you're using sd1.5 and you think it's better than sdxl you're either coping because your graphic card is made from stitched together pcbs or you haven't been keeping up
I'll come back if they release something worth my time
2B or 4B will probably become the new standard
runway is competing for the same audience as midjourney. there are enough brand-industrial-complex artists in new york, that i think the two can coexist. but really, why focus on an audience like that at all
I wonder how 800M will be with something like a rewrite of FastSDCPU
got 24gb of big ol vram G
cant pay me to use XL
1.5 is still where the OG's dwell
i mean is it so hard to say, maybe we shouldn't make tools for people who take mom and dad's money and give it to disney, for nothing?
I have 24GB and I still run SD 1.5 more than I run SDXL, for quality reasons - upscale (hires fix type) goes a long way
well
stability ai is finally dead
y'all ruined the fucking company, still releasing models that no one cares like music shit or whatever you all selling
💩 💩
ultimate upscale has been meta
since its release
the thing is insane when used by skilled individuals.
CNET also revolutionized 1.5
when you have capital you can start your own company and not release music models all you want
i really don't think releasing a bot for subscription is the death knell you think it is. let me ask you something though, if you think everyone who charges money for stuff - except of course the thing that you are personally paid for - is an asshole, pray tell, why should they worry about your opinion
if stability fails today, they still burned bright and gave us all this free shit in the mean time.
why be mad?
on that note, is there ANY anime lineart controlnet that does a decent job with SDXL? just tell me which one and which model you get ok results for it
yeah base SDXL vs finetunes is like night and day (also take into account loras)
yes lol
sorry bloke I dont use XL
so I wouldnt know
not saying XL is bad
I just open my lips whenever someone says XL > 1.5
bc thats misinformation
thanks but no
in the meantime i'll see how they still selling things that no one cares abt, they'll realize at the end, that's for sure
"I don't use SDXL" "SD1.5 is better than SDXL". K then
contorlnets on sdxl exist, but they're VERY iffy. models tahat are extremely over trained don't work as well with the controlnet models
more parameters is more changes to account for
unfortunately, the best way to monetize hype is obscure ponzi schemes. maybe that's why emad was talking about crypto products
he was in a sense, a few steps ahead of his colleagues
r0fl. Do I need to drink pepsi more than once to know that coke is better?
same logic.
I used sd1.5 and sdxl and sdxl takes a steaming dump on sd1.5
dont use overtrained models
/tableflip tokenization!!!
(╯°□°)╯︵ ┻━┻
yeah well
if people feel very passionate about ai social issues
and like, don't have a business reason to do any of this stuff at all. if it's really just an alternative to playing more modern warfare or jacking off, i feel like your monetization opportunities change dramatically, and the highest yield audience is the kind that buys crypto or whatever
its a trojan horse to get the entire economy tokenized and microtransactionable
I use SDXL and SD 1.5. SDXL is great at photorealism, LoRAs tend to generalize a subject better, but it lacks something and I dunno how to put it. It's a lot less creative. Like sometimes a whole batch of images looks pretty much like the same idea.
There are some excelent models in 1.5 like Photon, but sdxl finetunes and loras are way more capable
i didnt agree at first when people say, that stability are doing weird stuff, but now im starting to understand more
Are you talking about the base models?
no, the finetunes with funky names that some dude in Nebraska trained
They shouldn't lol
obviously they don't give a fuck, they'll start caring abt others people opinion when they go bankrupt if they haven't already
cause no one cares abt some shitty music models or whatever they're selling now, almost 90% of the people that knows stability ai is because of models like sd 1.5 and just because it's open source, cause the base model is still shit
hope they'll realize in the future 💋
i pay for the stable members thing but it's too bad they don't maintain that community at all
i think some of the more "hyped" SDXL models arent the better ones
huggingface also could elevate its community above the fray
and people publish there
there's that parallel universe China Hugging Face, which is funny
coke vs pepsi is very subjective. you on the west coast of NA and it seems very obvious that coke is superior. you go to the east coast and suddenly everyone is like "coke? dude pepsi!"
both are just high fructose corn syrup with fizz
i just wonder how it's possible for the two of us, randos on the internet, could have a very frank and cleareyed conversation about these issues, in like 15 minutes. does this not happen at stability?
"just try again, spend more tokens"
so in a way ur right
or cane sugar, the refined gentleman's high fructose corn syrup
the coke vs pepsi debate has nothing to do with whose tastes better. It's both candy juice. It's really about who has the more effective marketing startegy
lmaooo
stability ai by itself is 💩
im sorry but thats the true, their models are nothing without the community
cane sugar is only in special limited edition products now a days. and because they use chemical fertilizers on them these days as opposed to back when the recipe was first made, the flavor profile of cane sugar is quite different now
I bet if StabilityAI let people fine tune the model, it'd eventually be good, like what happened with SDXL and SD1.5
yup
but like
but like yeah
of course it'll
but they won't let that happen
they want 💸 💸
stable cascade is very very good. from a scientific point of view, they have basically succeeded in creating a dataset diverse enough to train CLIP, allowing it to "generalize" (via recall) with performance approaching everyone else's LLM-based conditioning
how long for cascade
unless you live in a country where cane sugar is the norm 
just gotta wait for SAI to go bankrupt so that the restricted licensing on cascade becomes abandonware
did you know cane sugar is called sugar, the other ones are not sugar
a stability membership is $20/mo
people who want to do something innovative with it, for commercial purposes, face no obstacles
we need emmad back
😭
there's beet sugar also
and i really don't think these guys are going to be too greedy
That's not true. The license on cacade prohibits all commercial usage
Mexican sodas all use cane sugar and thats why i love them the most. I rarely drink sodas these days
"Jarritos" i used to always grab those. strawberry. mmm. great flavors.
this is true, but i think it will be under the same model as everything else
also, i guess nothing prohibits you from asking them
artificial sweeteners give me harsh migraines and nausea. one diet soda and i'll be wrecked for the weekend
no thanks
I do unless it uses Aspartame, which I think all coca-cola products do
cane sugar coca-cola > aspartame coca-cola
not to start a flame war but aspartame has been proven time and time again to be safe
ppl are obsessed with dumb conspiracies so the myths will never cease
i've never found a sweetener with zero calories, that doesn't cause my metabolism to flip upside down
i taste sweet, my stomache gears up for calories i guess
proven by the aspartame company ™
that is true. but being a sweet tooth, that's a sacrifice I'm willing to make
diet soda does satisfy that sweetness craving for me most of the time anyways
ya turns out you can gain more weight with artificial sweeteners crazy, i mean just look at your neighbor
i can even tell as soon as artificial sweeteners are in my mouth. they're WAY sweeter and taste a little metallic
while I am not versed enough to know, you can find articles as recent as 2023 from the new york times citing the WHO as saying it might be carcinogenic, and I just don't want the risk when other sweeteners exist
(NYT and the WHO are both a meme, but again, why risk using Aspartame?? there's like half a dozen sweeteners)
Guys when buying a graphics card, would you focus more on the generation of the card or the amount of VRAM? I am deciding between the RTX 3060 Ti with 16GB and the RTX 4070 Super with 12GB.
vram > everything else
vram is worth its weight in gold
speed? I can wait for gens
honey or regular cane/beet sugar for me
i think it teaches your metabolism wrong signals. sweet should mean calories incoming. but when they're empty, it may teach your metabolism that you are in a calorie defecit and it should pack on anything extra it gets just to be sure. i'm no biologist but i'm certain that the metabolism learns.
the gut has as many neuron connections lining it as your visual cortex has. They're just spread over the length of your guts, not crammed into brain wrinkles
Not sure where to ask about this because this isn't really a tech support question but the SD subscription terms state: We reserve the right to decline or provide a prorated refund if there are indications of potential abuse of our refund policy.
Apparently the $99/month plan gives you 12000 credits a month and still has the 3 days free trial, does using them all up within 3 days and then cancelling the subscription count as an abuse? And if yes, where is the line of how many credits I can use before it's considered abuse?
the gut is a learning machine
tbh human metabolism is shite to begin with
so by drinking diet soda ur just pissing in an ocean of piss
how many ppl in the world have diabetes. hundreds of millions
sounds about right
well thats obvious
aaaaand this is why you go local or get ripped off.
hopefully a lesson learned
can you imagine when people can digitize a brain but turns out the guy's personality was mostly in the gut? now they're gonna need gut.safetensors to be themselves again
i've wondered this for a long while. heard stories of heart and lung transplants changing people's personalities and they gain some of the transplantie's quirks
RTX 4080 equivalent when it comes to gaming performance and 4090 amount of VRAM but $200 cheaper than 4080
does our personality hold up in the brain only?
or is our mind and body one entangled system?
kinda
kind of why i'm annoyed with 3body problem. its hard scifi but ||they didn't put the dude's whole nervous system in the space shiP!"||
I am very curious about it, and I wonder if we'll find out. Technically speaking we're not far away from having the compute for it, but digitizing a brain is a whole different matter.
what is this, the philosophy channel?
neuro science is applicable methinks
atleast its not religion so its okey
technically it is
And speaking from a perspective of a person with 24GB of VRAM, ever since Comfy added fp8 support going above 12GB has really diminishing returns when it comes to image generation unless we'll get more heavy duty models in future, only thing more VRAM is good for are LLMs and they'll improve over time
What is?
where is the seat of the one who says "I am" ?
that they will treat it as abuse if you do what u said
And I asked where the line is. When does my free trial stop being free?
i feel like my "i am" is in my butt
is artisan of limits
the left testicle
paid members?
thats for staff to answer, but just expect teh worst generally
That would mean that mean the free trial stops being free as soon as you use it at all and spend 0.25 credits by sending one message to the bot
thats has nothing to do with vram tho
they are but you sure would want 2x24GB for LLMs, I only have one 24GB and if the economy was better I would get another
higher FOV = more vram
no
Textures aren't the bottleneck
what
only a very low amount of games use vram barely at all
^
thats why I said modern
actually no, polygons are very cheap
not something you should priritize, when it coems to gaming
I also cant train dreambooth max settings without 24gb. Which I know, is not something the average joe does
aren't rockstar games the only ones where the vram usage is listed in the settings
But if you want futureproof, as you should, vram is the key
Nah there is also Apex Legends and many others
Tho mostly AAA
I almost never play them
thats what I meant by modern lol. but even a game like fortnite
if u go on max rendering distance
how much vram u need to render the entire fookin map
hint: a fuckton
u would buy a 2-3000$ gpu for rendering a large map on game like fortnite?
Anyways graphics quality peaked around Batman Arkham Knight for realisitic and Overwatch for stylized. With exception of volumetric fog maybe, ANY extra compute you might need is here because the devs cheaped out on optimization
btw
to the guy asking about the free credits
look in artesan 2 and count festivmans
those free credits almost instantly gone
@rare parrot
Lmfao
So it's 50 free credits
And you only get them if you give them your credit card info
seems like its just a few images
And an off-chance you don't cancel on time so that they charge you 9+ USD
yeah i would reccomend not doing that
Yeah I believe one image is like 6.5 credits
Verse of the day:
Luke 6:37
“Do not judge, and you will not be judged. Do not condemn, and you will not be condemned. Forgive, and you will be forgiven".
lmao
I think everyone forgives StabilityAI, but I think "being a sub-par midjourney/dall-e" is not their A-game
well theres a difference in, being so great people want to throw money at u, vs forcing people to give it to you, and it seems sd has chosen their way
Wisdom of the day:
Me now
"Everyone judges you no matter what you do or don't. You will be condemned if you do something worth getting condemned in their eyes. Whether you tend to forgive people often or not won't affect how much they forgive you all that much."
🌷
Its about God forgiving you..
Isn't inaction when someone does something wrong considered bad if you could prevent something bad?
But yea
No reason to bring the religion in here
if you think about it, everything you perceive isn't real, it's a construction by your brain, an artificial reality synthesized from inputs from sensory neurons, created just so that you can understand what's around you
Artisan's just another way to use the services also offered by SAI in other ways. Nothing too bad there. But that it seemingly replaces the old bot, where SAI couldn't even be arsed to announce it was gone forever, with the last announcement it would be back soon. BOOOOO. Also way to misrepresent what was asked by saying in the announcement that that this thing is by popular demand. I also feel it's kinda wild to charge a premium for what is essentially beta testing a model that isn't ready yet. I would have hoped SAI to cultivate the enthusiasts, but instead they seem to milk them for all they're worth
no one has any idea of what anything actually looks like, feels like, tastes like, or even is. in fact, nothing is verifiably real except your own artificial reality
Luckily the nature of reality doesn't change reality
Nah not really
🔥
religion talk always ends in unhinged lunacy talk
Depends on how you define it but yea not really
it's true, though
Did they also charge you credits every single command with the past bot?
I also feel it's kinda wild to charge a premium for what is essentially beta testing a model that isn't ready yet.
@shut hare for the purposes of collecting telemetry there is some truth to this. however i don't think you get enough explicit feedback for DPO, no? midjourney does the upscaling thing to force people to do that
No, lol
well, pay to work, who doesent like that lol
Well, here is your answer why the old one got thanos snap'd
gpus aren't free - even if you own them. they like electricty
And I have solar panels
tbf, talking about how to use generative models is often unhinged lunacy and people making stuff up, all the same
you have a solar powered data center?
no, the other thing is worse
the free trial does not give you the full credit count of the plan you chose, it gives you a small trial credit allocation
No, but I have a solar powerd RX 7900 XTX Linux desktop
Yeah I figured out from that other conversation that it's 50 free credits apparently
"give the peasant a small credit allocation"
sounds like an astopian horror future movie"
as a suggestion - if someone's going to pay for the plan right then, dont' drop them on the free trial - jsut give them their plan right then
astopian?
So that'd be... 7 images maximum from free trial
please do help me with english, im not sure what word i was going for
distopien?
dis...
dystopian
dystopia is usually the one people mean in this context ❤️
0.25 credits for sending a command, 6.5 for generating the image
50 / 6.75 = 7.4
thanks
we might bump that soon actually
no promises but yeah that's fewer than we should be giving
0.25 cents is for Assistant chatting
and i'm still stuck, @finite cloak
yeah, lets bump the free credit, lets give em 7 images instead of 6, no talking about sd3
fun fact, dystopia is supposed to be the opposite of a "utopia" but utopias are a greek pun. means both "perfect place" and "no place" since they recognized it couldn't exist
I believe the free credit allotment is lower than intended, will find out and hopefully retroactively give you guys some more
Oh wait my bad it's 0.1 credit per message sent
it should probably be closer to 1000 per day for sdxl turbo, and 100 per day of sd3
Yeah I feel like somebody assumed 1 credit = 1 image when setting that
May be! We share the subscription and credits but our services and costs are different
how can it be lower then intended? thats the most.......crazy thing ive hear any staff say
but you should cut the complexity of credits altogether
Well, the website states "0.1 credit for each message sent to Stable Assistant"
it should just be "unlimited" and there should just be one tier
if this is the most you can get from your bosses, for now
Very exciting to have SD on the discord BUT without weights I still fear it's not actually released.
you really need to NOT put people on the free trial at all if they are paying for a plan. you need a demo, but that comes before someone's decided to pay
every VC who looks at midjourney's pricing page says, "these are just all the same thing"
who makes this and then dont put any thoughs into the credit , that is wild
it's not released
there's also ideogram, which is just free
That'd cost more than premium plan and premium plan is $99, I really doubt anyone would pay that much without an intent to abuse the hell out of that unlimited compute
it's not practicable to abuse a discord bot
and anyway, they can have an 1e4 order of magnitude block if they so desire
stability is more popular and people were just creating new accounts to use their free tokens
Well yes. And that makes all these announcements so boring :(
chill, sd3 will never be released
So they should charge you the maximum amount one could generate within a month of non-stop prompting?
i don't know. listen all of this could be true
ideogram will likely remove the free tokens once they've got recognition
i really regret coming here today
I mean they promised it will be but like. look at the state of the company
cost leaders. common marketing strategy
@shut hare what you guys are making with segmented pricing instead of 1 price, you are paying a LOT in brain damage
its more then a lot, i woudl say
this is the next step to releaseing it.
They stated it will be, and I'm assuming it'll be weeks minimum, as soon as they judge that the attention they'll bring to themselves by releasing the weights outweights the monetary value they get from people using their API. Which might be a long time but shouldn't be for ever, depending on API popularity
i'm not saying yes or no to subs. i'm saying the price segmentation is going to be an endless source of brain damage
yeah this is what i've been saying. WEights will release in some form once the subscriptions tail off
the delay on SD3 weights depends on training progress, not anything related to money
training isn't finished yet
yeah sure
i've never known any for profit business to just walk away from money
what you see that every day
So why release it via API?
"training could be 1 week or 10months, we just dont know"- sd
typical discord users - nothing but negative whining
Nah I understand them and don't disagree
Just stating facts
you aren't the people i was refering to
"hey, how many free images should we give peopel to get a taste of sd3?" "idk, just put in a random number"
(A) to get some money lol (B) to let people try the unfinished beta version
if a business has an opportunity to make more money doing something else, they'll walk away from money that prevents them from doing something else. sure.
why would they walk away from money with no other plan to replace it ?
In my opinion stability should release sd3 weights and train a huge model to use only in A100/H100 gpus in their API like mj/dalle
And B could be solved by releasing the weights as well so A is the main reason and B is a side effect, like I said earlier
I'd reckon that B motivated the SDXL release and A motivates the SD3 release
training new models less than 8B params, yes
For those of the "it's all about the money and nothing else" mindset, do remember that the open release of the model also gets us money so it would be very silly to delay it to earn more money. When delaying it earns less money
the API model is there and live
So again what I said earlier
8B isn't finished either
i thought the open release only makes money if people are paying the subscription for commercial rights
Yeah that's the reason most likely
and yeah when we open release the best model you can download, a lot of people gonna pay for commercial rights
word
"best model u can download" oh boy
which is why i'm guessing the full 8b parameter model won't have a free and open release. Weights will only be available via licensing
that says something
Not whining, think of it like constructive criticism, old bot helped SAI and enthusiasts like me. SAI be getting them data by voting, enthusiasts by getting together around prompting, finding new things that works, expanding ideas, and such. With this extremely aggressive pricing, all that's lost (at least SAI didn't have the audacity to implement voting). I feel long term gains (cultivating the use of stable diffusion) are sacrificed for short term ones
well listen, you're a mature guy. you know that omission and generalization are both not "the truth." ultimately people want the 8b parameter model live in the API, and your bosses have decided that the 8B model will never be released, and the telemetry for DPO is not obtainable from a release, even though you are the author of an inferencing UI and could easily add it. I don't even know how important the DPO really is in this case.
says you wont need a super computer cluster to run it
thats the concern. There's a huge money oppportunity to not release open weights. open as in anyone can refine them and publish them
uh
licensing seems to be tangled up in it now that they've stopped using rail-m
what happened to the bot
says I genuinely believe SD3 to be better than SDXL or any third party model or etc
do you gotta pay now?
but your colleague also decided not to proverbially "come clean" on the issue of releasing the 8b model
no i ment that only some will be released for download, otherwise i dont think u would have put it like that
maybe, the higher one will be api only?
my 20 year old self brain damaged me plenty. I'm immune to it now
The plan is to release all sizes of model
nothing will come out, not the best, not the worst 
So is this going to be like MidJourney now where we get to “try for free” but then when we try it will say all slots are full and we have to pay?
(Not necessarily all at once mind, probably will release different ones as they finish up and are ready to release)
well thats atleast good to hear, you are just scaring people on here ><
things just seem up in the air with 2 interim co ceo's of the company right now. Bad timing to transition leadership during a hype period. i'm not surprised there is an air of doubt
the only air of doubt i've seen on this discord is the air you've been spouting
seems like it
k
yeah...
i mean listen. you are not in charge of this. but c'mon
the brain damage alex. this line is dealing a lot of it
What would you say are the odds some of the sizes will be released later than others?
which will undoubtely seed doubt more so
what is the meaning of ready to be released? it's released. you can inference using the API.
Nah I feel like the doubt will be erased if the small sizes get published at all, it'll mean they just want to limit the better models to API for a bit longer until they cash in some more on those
if it's good enough for paying users, surely, that's released
he alrdy said its about money
^
still really curious about the architecture of the model. If loras or embeddings or controlnets made for the 2b model work on the8b model. if they're not compatible between parameter sizes, we'll probably see the community latch onto one version and ignore the others
Yup just found it, when you try to sign up—3 day free trial then get your wallet ready
and who knows if downstream users even want a DPO checkpoint from API and bot data. i don't.
well yeah, but would you rather (A) we release one or two at first, and others later, or (B) delay any release until ALL are done?
The inevitable FUDD from mixed releases is probably better from continuing the standard FUDD from it taking a few months from announcement to release
I've said no such thing, I directly said the opposite - the time on release is not related to money
They're doing that with all the sizes currently, so if they decide to release the smaller model, they'll have the exact same reason to release the bigger model later down the line
What about here
Ah kk
what FUD? there's no FUD. as soon as emad is gone, it was obvious that the 8b model wasn't going to be released.
Thanks for specifying
oh-- yes the reason it's on API/bot early is for to get money before it's done, yes
that does not affect the release schedule for sd3
i just don't get how it's good enough for API users
ok
he was asking about the api bot, not release
but not good enough to download. what is that vanishingly small amount of telemetry really worth?
there is barely any human in the loop for API usage
API/Artisan isn't about any direct telemetry
They stated multiple times they'll release all SD3 weights to public later, even after emad left. It might take a long time but eventually they will, otherwise they'll really damage their reputation
Fear, uncertainty and doubt, while often used as a control mechanism to convince crowds through propeganda, they're also real. FUD isn't unexpected here imo. Just a natrual occurence of people being hyped without all the information. Emad quitting after dropping massive hype bombs wasn't ever going to help
then the only remaining rational reasons are strategic, not scientific
i mean, i already know all of this
didnt some staff in some artisan channel alrdy admit part of the reason for api bot was telemetry?
Yea we've already established they want to get some money from it by making the early access API-only
i am just wondering what is the challenge with saying it
it's not about some money
nobody cares about the kopecks from the bot
human preference data isn't really scientific. science is about removing human bias
not rubbing it in like a wax coat
the SDXL bot was about collecting that data directly, the new Artisan bot and API enables general feedback but no direct rlhf style stuff
you can tell by the lack of a voting button lol
There is none, they already said it
unfortunately we cannot read minds and therefore cannot determine if a human liked a result or not without having a button for it
i guess, nobody knows if it's going to be kopecks or wahtever, but i don't think sd3 weights would really reduce the value proposition of a subscription. the kind of people who are arbing bots for brand work can't run local models practicably anyway. they definitely can't run the full-weights sd3
and there's no way to standardized the criteria for a human to determine if they liked it or not
anyway. i do hope you release it
i hope you can figure out how to convince your bosses. i don't know if you or your colleagues are convinced
i guess well just wait a few months and we will find out
you make a UI that people like. why not sell that? i don't get it. figma was worth $20b!
Why am I seeing the "View and edit image 1" and '[...] image 2"? Not that I think it's bad to get user data when they agree to it and it's to actually improve the product. Did the guy in #artisan-1 ask the bot to gen 2 images at once?
listen if anyone in charge tries to tell me we can't release the models i will personally riot and cause problems until that's changed i promise you
make it in othe rpeople's products. make a photoshop plugin
SAI really needs a PR manager, apI/artisan should be announced as "Get an early preview of the new SD# model we're training for public open weights, see the model improve towards release as the api/bot gets progressively updated with new training!!" I'd still think the same about it, but it would sound sooo much better
hmm... but the 8b API model isn't getting released, at least not any time soon
it will probably be delayed indefinitely
i didn't join stability to work on closed source secret things, i joined to work on the cool open source free to download stable diffusion models
all model sizes will be released
inb4 alex get fired in a week or 2, jokingly said
okay, let me ask you this. do you know what lying by omission really means?
i am not trying to bully you
and i am not saying you are doing this
at this point idk when weights will be released
The UI people like is a1111 and it's derivatives, you won't get nearly the same amount of money
but it's possible to say that something "will be released" without technically lying
cause the buttons for other features (creative upscale, search&replace, etc) turn into a giant mess when there's more than 1 image to include them for
figma was worth $20b!
so i don't work there. but surely, you must see, they are not going to release the API model
Figma is UI, visual assets, user data and then possibly more
they can say they will, and indefinitely delay it
do you want specifically the current beta model on the API in particular? If so yeah no it's gonna be trained more before it's released
they can convert the company to a taco truck also
i'd worry more about improvements to the eventually released model, like that thing called "core" in the api
If they do that, some other open weights model will outshine them eventually
if it's good enough for the API, why isn't it good enough to be released?
you charge money for the API
just think about it
i understand that it's going to be "Trained more"
@broken cave you're going in circles
Sadly there´s no other local models/competitors
but sophisticated people are not bothered by downloading another file. users who will never monetize anyway are bothered by having 2 or more files
hold interest by having it on the API, many reasons
They already told you the models are API-only because they want to earn money off of that.
no
Out of other things
The models aren't released because they're done.
They're on API before they're done for money, and to let people try the beta
if it's good enough to charge money for...
listen i don't know
There's tons of early access games, SAI simply does that with an image-gen model... not seeing why that concept is so hard to undertand
We've already established you can let peope beta test the unfinished model by releasing the weights to the public
will you please stop?
its a normal concept to have a paid beta, but still, i think sd needs to be more open about stuff
I'm just reminding them about that part of the conversation
i am in disbelief that someone who works there doesn't know the API model is not getting released...
you guys must know!
just ask!
ask for a deadline!
i mean your bosses make you work on deadlines
okay well, please remember that they don't have to release anything to anyone at all. stop and get a grip
Seriously cut it with the intentional FUD/misinformation
ask them for a deadline. everyone works on deadlines in the absence of real management
Nah not really lol
how about you just to make images and cool down
listen, just ask
It's a research project pretty much
just ask for a deadline
You can't deadline research, unless it's for budget reasons
for your own personal edification
ok i did
yeah just armwrestle the boss, if u win, he tells u
@finite cloak from the looks of it, this guy isn't really old enough to be on discord. 13 and over, no younger allowed. you should probalby just report his account to discord and kick him
they are businesspeople. they work on deadlines. i mean even at a startup, believe it or not, there are deadlines. some real deadlines exist.
technically he's right, the thing currently in api won't be 🤡
lets not be so easygoign about kicking people out
if there was only one way to do things we would only have one thing
you're another one that's just here to cause trouble and should be shown the door
thats very rude
flat statement of truth
just because im a funny guy sometime doesent mean i should be kicked
Training this model is an uncharted territory, it's a research project. The only real deadline is the budget, or the estimated value they can get out of it, and they won't be open sourcing their bank account status most likley so you're free to stop bothering the devs
yoiu've done nothing since the annoucement but whine and try to cause trouble
you arent a modd
you wanna limit what im allowed to think and say?
whats the announcement?
which is why i can say bluntly what I think
There is a difference between being funny and being toxic about your demands without listening to counter-arguments like @broken cave is
sometimes toxic demands are pretty funny though
Kinda sucks that every embryo of some legit discussion gets drowned in "this is the end of SAI open releases" drivel
people should stil be given a chance, not something to be hasty about, but i get ur point
Flowwolf, doctorpangloss, and bravebydefault have done almost nothing since the annoucement but post in every possible channel with whining commentary and an attempt to force stability into doing something theyv'e been told bluntly isn't going to happen
all three are probably 12 year old kids
that's how they're acting
is this a movie theater? who lit up the projector
what was announced? (sorry been out of the loop last couple weeks, just catching up)
stable artisan, a bot that you can use the API with
read through #artisan-faq
I'd say when they end up annoying the staff/devs more than it's worth is a good point to start considering
basically like midjourney
you are just a silly man crystal, these silly personal attacks, u know who does that? kids
danke!
they've annoyed me enough that if they ever show up on my discords, they're banned before they type 2 words
the discord army is kind of irrelevant in any of this
fr
they arent herer though
in your case, you're not a vocal minority, you're just vocal
Unless we're talking about Helldivers 2, then it's the vocal majority hah
i don't think it takes a degree in businessology, or even a college diploma, to understand that if the model they are developing is good enough to charge money for, it is probably done in most of the ways that would matter for the community. perhaps the ways that the model needs to be trained for community usage is ill-defined or unknown, for this 8b model and apparently not the smaller ones. it seems more likely than not that it is a pure strategic decision, a perfectly valid one, to indefinitely delay the release of the valuable 8b model. @finite cloak this is all meant in good faith. i think sophisticated people will read what i am saying, and understand that the secular (technical, scientific, whatever, non-strategic) arguments for releasing the smaller models are too full of holes to be true.
i just wanna be able to use/train SD3 on the very expensive gpus I've already paid for lol!
- stop 2. stop 3. stop or see the door shut in your face
yeah those are rare events. like xbox one always online . "we sell the 360 for offline gamers" moment
but then everyone went out and got cheap smart tv's over the next 10 years that spied on them anyways.
And it doesn't take even a room temperature IQ to realize that when they literally say "we're keeping this API-only partially to earn some money from API usage before releasing the models to the public" it literally means what you are trying to say and pretend you're the only person who knows it, despite them clearly saying it already
or another one i like, when coke became new coke
i am only trying to resolve the issue, do people at stability sincerely think that the 8b model will ever be released?
or when fox sports made the puck light up blue or red for slap shots
my opinion is no.
once in a while, everyone agrees "we hate this"
it will never be released.
there's no issue to resolve. you are trying to manipulate and outright bully them into doing what you want
god that recent news made me so proud of Democracy
kindly take yoruself off this discord and find something useful to do with your life
And their words are yes, and we won't know until they actually release them so you're free to stop harassing everyone with your opinions.
well you guys don't make a video game. it's kind of an apples to oranges audience. i don't think that kind of social media fueled blowback is going to happen by just reneging on some promise.
shout at the sky
look, go use leonardo or kaiber, and leave
It's a huge win for digital consumers everywhere honestly, it'll be used an as example to motivate people from now onward to fight for being treated better
harassing with opinions. i'm not often one to call someone fragile but this feels a llllitttle bit ceramic
@bleak matrix @sudden ruin yall around^?
He's harassing and the reason why he feels it's justified is his opinions, is what I meant
reported to discord
I feel a strange tension in this room.
yes lmao
you have ur youtube listed in ur own profile, what can be harassing about that
i really don't think this is a very big deal
aimingfall, you want some? 🍿
🙂
Sure. Thanks
gonna just do some temporary timeouts on the trouble makers while we wait for mods to get here
I think they should release or train the 8b model as fast as possible, if they keep waiting to release it what could happen is that midjourney releases the v7 version and openAI releases dalle4 so even the 8b model would be irrelevant
i want dedicated server binaries to be the norm again myself. I still feel like gaming servers have become over centralized since xbox live went live. we used to be able to run our own servers along side publisher servers. valve games still work this way. nothing is stopping companies from going this way except policy and control over consumers
Yeah, some guy jumped in and complained about SD3 being API only pretending he's the only person who understands that it's to earn a bit of money before the full model release
i cry everytime i buy a new game with centralized MP 😢
There is a huge legal movement right now actually fighting for things along those ways, you should REALLY check it out, you might actually make a difference https://www.stopkillinggames.com/
still don't know why anyone could muster the enthusiasm to even be upset about that, if its all this is about?
dude yeah 100%
but ok
i wouldn't call it huge but yeah i'm into stuff like taht. i'm a bit of an extrmeist and think we should have the right to transfer software licenses.
esp. any "automated matchmaking" games... like no i don't want a rank ladder dealio all the time sometimes i just want to chill and join a server full of the same players i've played against 50 times before and have fun
The guy was impressively good at being delusional
Agreed, I really miss that in games
server culture has died out on modern games. i was thinking of this while playing 2042 the other night. remembering the old 2142 and 2 servers that had the same people every night
seems like a pretty common discord problem tbh
Well, it's the biggest yet and likely the biggest we'll ever have honestly, it's more-less now or never, tho if we win anywhere it might trickle down to other places over time
there are still some games out there that do the whole server setup right - minecraft, Arma 3 (the two most opposite possible games lmao), Ark, ..
you still get it on games like counterstrike. you have homebase servers people stick too. but 'm not good at counterstriek. its a very tryhard game
those games do way better at retaining my attention than matchmade/centralized ones too, and not just like matter-of-principle reasons or anything, just actually i have more fun and want to stay longer
TF2 is another example
TF2 players though. (spy main here. case in point)
me when TF2's playerbase is actually is ~75% bots
(Team Fortress 2, not Titanfall 2, tho Titanfall 2 playerbase actually managed to reverse-engineer the whole server backend)
idk why any game company would want to prevent players from forming social connections with other players
seems clearly beneficial all around
So that they can push them to buying the next game instead and spending more money, sadly
(other than when weirdos are involved, but self-ran servers have local mods to ban weirdos too!)
yeah! cheating was less of a big deal because if you homebased a server, everyone knew each other and the mod would deal with cheaters
you can get into the morass of "are those players under 13" legal issues
Unless self-ran servers themselves are by weirdos, but that risk is definitely not worth shutting it all down
server admin could just view player's POV and know they're obviously hacking
"Think of the kids" is bs when it comes to things the parents should be responsible for
whole lot of not thinking about the kids is why we are here
i was all for blockchains for a little bit because i thought they would save games. embarassing times.
still should be, its a wonderfull ledger
Digital reality is nowadays a part of reality whether we like it or not and if you're a parent then you should either not let your kids take part in it, monitor them while they take part in it, or if you don't keep them safe on the internet at all then you should be legally treated the same way as a parent who doesn't keep their children safe offline because the damage can be just as real
pay teachers doctor level salaries. theres your starting point.
give them the only license to beat children too
i'm a bit of an extremist
Beating children isn't necessary, tho beating parents until they learn to take proper care of their children would probably be more effective kek
the beatings must continue
Nah
I've never been beaten but I'd never do dumb shit or disrespect anyone with no good reason etc
Where are API users supposed to post about the API in this server?
So I'm a proof it can work just fine if the parents don't suck
#🤝|tech-support probably
lol what did i walk into
how to fix things
Someone saying they approve of beating children I think kek
I thought this was an Ilya quote
You mean society?
@true canopy nah it seems it's a flowwolf quote 
its an old meme from pre meme times. "the floggings will continue until moral improves" dont know where its from.
The proper education route is definitely the most promising way as you mentioned earlier
bread and circuses is fun too but that just placates us. Education an arts, that stirs things up
people are radically different, discipline can positively effect some and hurt others
one size fit all solutions only fit one size
The only solution that always works is to always search for better solutions
Can someone inform, what command(s) that can be use to generate videos using an image.
Image to video
AutismMix-Confetti 25 steps Euler A Normal CFG6 seems to become objectively better with Hyper-SDXL-CFG58-8steps at strength 0.1 for both model and clip
autism mix... what a name
Anything new friends?
If you're bored, we could always use #🔆|dailies submissions 🙂
huh, I've literally never gone in there before
ever watch someone else write code? that's a little bit how I see looking at other people's images
i'm a big believer in allowing people the opportunity to learn from their failures. i'll happily watch them code as i sip coffee behind them
or if they're better than me than awesome. chance to learn
ever watch someone wipe out on a skateboard? thats a little bit how i see watching people learn how to code
why watch when you can code yourself 
why code yourself when you can ask an ai bot...
sorry, that was a little troll-baited
cause it's fun
literally got a game of bitburner running right now. idle game based on javascript
what happened to open source AI here?
wasn't that a major focus for stability AI?
are you talking about sd3 weights? They'll probably release, but they're doing more finetuning and focusing on the lower param models now.
@finite cloak must have said at least 10 times today that the weights will be open sourced and that they aren't finished training yet.
okay, nevermind then, sorry
Using sd3 chat was fun, I especially liked the reuse as structure. Looking forward to porting sliders to sd3
so you're wondering if it can generate Emad? 
impossible
What is reuse as structure? That sounds cool.
It's an option on the discord bot
Hey all, I've been fooling around with SD a lot these past few weeks and am getting the hang of things. I've noticed how large the checkpoint sizes are. Is there an optimal few checkpoints I should have on hand that can be used with any Loras to save on drive space?

recently i discovered animagine xl 3.1, it covers a LOT of anime characters, etc, you almost dont need a lora haha, and it provides very high quality if you prompt correctly, the author has some prompting tips. juggernaut X is overall very good for photorealistic, also recently released Haveall XL seems to cover a lot, so just those alone will give you a lot to play with. not sure what to recommend you for nsfw stuff tho :3 @velvet slate
Thank you very much
⚔️
whats artisan?
the bot to make imgs
How am I supposed to use the bin files? i added ip face adapter into the controlnet folder but it does not show up on preprocessor
hi
大家好呀
some who has access , can they do a F-4 Phantom II ....wanna see if its capable yet before I commit.
Hello 🙂
Hi! Just followed you somewhere 😄
Nah ... I think it's youtube ... using his SVD Workflow ... if he's the real one ...
What extensions are you guys using for autofill keywords? Been using https://github.com/mix1009/model-keyword but I just noticed that it's over a year old. Made me wonder if there's more, and/or better ones
Oh right 😄 @grave forge got some ID? Kappa
He has Paypal 😄
宝宝们好
Jaja dat!
image.save("testimage.png")
buffer = BytesIO()
image.save(buffer, format="PNG")
imgstr = base64.b64encode(buffer.getvalue())
return Response(content=imgstr, media_type="image/png")
I don't understand why this is being done
when I can just do return FileResponse("image.png")
why all that bytesio stuff what is it for?
Probably my Neuralunk profile on Openart.ai
So where can I try artisan here please ?
Hi guys, I have a car part I want to train a model on. Anyone suggest a model to use?
Midjourney and Krea are pretty muuch all you need.
and also hyper sdxl
Training on midjourney?
I guess just the base xl model for that. many derived models will be biased toward people is my guess
hey guys, what time open source with sd3?
I'm very very very very very very very very looking forward to it
the 69th of kektember
No one knows as of yet 😄
They said soon so maybe 1-3 months?
I'm not into long distance sorry
... from 2,3 weeks ago...
there are a lot of impatience people who start to consume a lot of copium
sd help people to releast imagine, this is the most important thing....
like hell
Alright, so hopefully 1-2 months
what?
basically bragging about SD3 not going to be release and going to be paywalled
a lot of theory appeared here and there
That pretty much happens to any product that people are anticipating
no, i don't think so..
Just gonna stay cool and don't hope too much on any specific date
they usually have a treaty hold up to prevent anyone in the treaty saying anything too explosive to the community
The reason why sd is so popular is that the community contributes more than half of it.
If it is closed source, new open source models will emerge, just like Uber and Didi back then.
^
what is Didi?
I think the official integration tools, lora and other acceleration tools are the source of the open source model. This part can be closed source and reduce the impact on the community.
The cycle will repeat until enough people lose interest, and I doubt that'll happen. Even just the single aspect that personalized ai nsfw will keep selling just like real nsfw is selling
taxi application
ah OK
u know that
... but eh it is too big to closed source in my opinion
The only taxes app that matters: https://twitter.com/HD_Kirin/status/1638361179025477633
To a certain extent, compared with sd1.5, sdxl has almost reached a bottleneck in content creation. The rich community solutions add a lot of capabilities. Although they are complex, they are very effective solutions.

Anyways cool I think I just got dm'd by a scammer
So technically speaking, everyone is making breakthroughs. With the existing technical capabilities, SDXL is still far from becoming obsolete.
Let's see where this goes
wt?
that one called Alexa Bunny?
She does not seemed like a scammer by the speed she replied and stuff.
but yeah be aware
what happen?
I got a DM from her and she invite me to an AI art sharing platform server
You can't install artisan on your own server yet right?
You can remake it with the API access quite easily
- invite
and yes, but it will soon after a week
Is there an Automatic1111 that lets you also call the API
Or, whats the best web gui for the API to do batching etc
I have my own colab but wondering if theres something nicer
