💬｜general-chat

[pale bay]

hello all !

how to use SD3 please ?

[jovial wraith]

no one has access yet, sadly

[storm iron]

Looking for a full stack Co Founder with AI & Computer Vision interests. Kindly reach out if you would like to co build an
AI Rental Marketplace, Matchmaker & Self Broker for migrant students & professionals to US.

[winter pike]

soon, may be available this thursday, i hope so

[primal star]

I doubt it, they've been hiring hardware engineers. given their history they will probably be working on some kind of new interface product

[rich kestrel]

that's because we've all heard this one before

a model that overpromises and underdelivers

yawn

only for 1.5 to still be king

what we really need is a clone of 1.5 but just revamped

same prompt styles, same lora compatibility, but just better

[shell tendon]

Why do you say Thurs? Did I miss something

[winter pike]

Emad posted sd3 generated images with the date "29/02" on twitter and then said on reddit the discord bot is "inminent" since the training is done and were on the finetune step. Nothing set on stone, pure speculation

[shell tendon]

K cool thx for the update

[crimson belfry]

The world is waiting for access

[shell tendon]

Anyone train a Lora that hybridizes two faces, creatures, etc?

[crimson belfry]

Everyone is waking up checking their email lol

[karmic brook]

i got access

sry tht's mean lollll

[fair pecan]

question

why is the bot down?

[crimson belfry]

😢

[fair pecan]

and has been for almost a month?

[indigo ermine]

Do we know anything about the requirements for SD3 hardware-wise? SDXL already murders my 3060ti...

[crimson belfry]

@karmic brook share the wealth

I been checking my email nonstop lmao

[karmic brook]

originally the bot was used for testing and iterating on models
it's better to wait for this to align with that

[karmic brook]

check now

[fair pecan]

alright

[crimson belfry]

Let them cook everyone

[fiery comet]

hi

[crimson belfry]

I honestly can not wait to use it

[glacial nacelle]

nothing yet

[shell tendon]

Sadly stupid Google locked me out of my account for this email right after I hopped on the wait-list

And then demanded phone verification after I got it unlocked, so I typed in my phone number - same one I used at registration

"this number has been used too many times for verification"

[gentle plume]

Hello, for several weeks it is no longer possible to access stable diffusion bots via Discord to create images or videos for free. Does anyone know if it's possible to access stable diffusion for free and how? Thank you very much🙂

[nova zodiac]

Try civitai generator, happyaccidents.ai, or aiscribbles.com

Also bots are 99% likely to be back for sd3 release in next few weeks

[gentle plume]

hello ResidentChiefNZ, thank you for your answer... I don't know these image generators at all... I'll look 🙂 Do they work with discord?

[nova zodiac]

Nope.. they work on the websites, but they are all simple to use and free or have a good free tier

[gentle plume]

ok thank you very much ! :)))

Ah the stable broadcast robots should return to discord? that would be awesome ! 🙂

[crimson belfry]

Which one do you prefer?

[nova zodiac]

Easiest to use is aiscribbles - the civitai generator is pretty good, but doesn't have all of civitais models (work in progress) or upscaling or img2img - happyaccidents is more feature rich but limited due to free tier. Of the three I'd recommend Happy Accidents. Personally I either use https://tinybots.net/artbot which is connected to the same horde as aiscribbles, but has much more control over the image; or I use my paperspace and fire up a1111

or runpod/vast.ai as well

[crimson belfry]

I need to get a better computer because I like making art a lot, I see many people making art light-years beyond what I'm capable because of this.

I appreciate the breakdown as well

[trim nymph]

can you show us how good it works with 19th century photography, at the moment?

this would answer an thesis i have for sd 3

[karmic brook]

i'm trolling

i work there lol

[trim nymph]

i know

[clear oyster]

would having a better gpu make images look better
just thinking lol

[nova zodiac]

If you don't have a pc at the moment, it's worth getting runpod instead as the payback on a gpu is around 3000-4000 hours

[trim nymph]

thats why im asking 😄

[karmic brook]

[nova zodiac]

nope, just makes them render faster

[crimson belfry]

A 19th century photo of a dog and cat sword fighting

[clear oyster]

yeah thats what I thought lol

[pale bay]

Midjourney doesn't work very well atm BUT its aesthetic style is beautifull and powerful !

[crimson belfry]

I will check this out

[nova zodiac]

midjourney and dall-e are good tools, and will make you a pretty image out of a prompt. the case for Stable Diffusion is that you can control the image way more, and do way more with it (things like Regional Prompting, ControlNet, loras for characters etc)

[clear oyster]

plus you can run SD locally ofc

[nova zodiac]

yeah.. degenerates cant degenerate on those platforms

[lone hawk]

aye, i agree about aesthetics, but it's prompt understanding is nowhere near Dall-E 3 not to mention the upcoming king SD3

[pale bay]

I love MidJourney and Stable diff (my 2 favorites) i use Comfy UI and automatic 1111 for enhance details and finalyse and inpainting and upscale too

[lone hawk]

also SD3 will be finetuned by community, you know it will be beast on Civit 😄

[clear oyster]

dont know what that means lol

[pale bay]

i love ia art hahaha

[nova zodiac]

midjourney and dall-e are censored, people that want to make things that are illegal can do so on a local sd install unfortunately

[clear oyster]

when will sd3 come out?

[clear oyster]

oh yeah you mean that stuff

[nova zodiac]

2-4 weeks (source: stability worker on reddit)

[clear oyster]

I see
dont you work for stability?

[clear oyster]

yeah thats unfortunate, it gives local ai a bad name

[nova zodiac]

I wish - @karmic brook hasn't made that happen yet 😛

[clear oyster]

oh so the black name isn't an employee I see haha

who is fruit, she works for them?

[karmic brook]

who's fruit

[nova zodiac]

yeah they are staff

[clear oyster]

lol i think I remember you from a few months ago actually

[pale bay]

civitai website is full of adult stable-diffusion models lol

[nova zodiac]

yeah, over 70% of AI art usage is nsfw

one thing I've learned in this space is humanity is horny AF

and a LOT of people need to be neutered

[clear oyster]

LMFAO

yeah its sad that they're putting such a good tool to uses like those

[nova zodiac]

There is a lot of commercial use, and some absolutely beautiful art coming out of the tool, but it is dwarfed by a tsunami of smut

#brandnewsentence 😛

[clear oyster]

lol

so looks like fruit is in charge here is that the case?

[nova zodiac]

She's not the only staff member here, but I believe they are the community manager

[clear oyster]

how do you get the black role?

[nova zodiac]

Won a picture of the week contest a while back

[clear oyster]

oh I see

[karmic brook]

yes i am

[clear oyster]

arent all the base SDs filtered since 1.5 though?

[nova zodiac]

And a damn good one too 🙂

[clear oyster]

but then anyone can just train them

anyways, about the hands
Im sure its been asked before but why cant ai get hands proper lol

[nova zodiac]

the training datasets for sd2 and sdxl had most of the nsfw removed from them, and sd3 will be the same; people then have to train back in those concepts in those models

[clear oyster]

I see, thats good

[jovial wraith]

sd3 will be able to do hands pretty darn well!

at least according to some images a saw on twitter

[clear oyster]

that would be good if it really can

[clear oyster]

there are images of it already? link

and will sd3 have the same sys reqs as sd 2 and xl/turbo

[jovial wraith]

https://twitter.com/hashtag/SD3?src=hashtag_click

[nova zodiac]

hands - 27 bones that can be posed in several hundred thousand poses... not only that, when you change the angle of the hand, you lose sight of some of the fingers which then confuses the training

[clear oyster]

but arent there so many more complex things like faces that the AI nails

[nova zodiac]

hands will always be difficult without things like openpose hand controlnet

[jovial wraith]

https://twitter.com/Lykon4072/status/1761194245698756833 here's a post showing hands in particular

[trail lion]

there are way more images of faces on the web than hands

[nova zodiac]

faces are a lot bigger and less parts moving at different angles relative to eachother

[clear oyster]

https://cdn.arstechnica.net/wp-content/uploads/2024/02/dalle3_SD3_comparison.jpgis it just me or did it do text

supposed to embed lol

[jovial wraith]

yup, SD3 can do text as well!

[clear oyster]

whoa

so it seems like sd3 is a gamechanger

but with all these advanced things in it wont the system requirements and vram requirements increase a lot?

[jovial wraith]

i hope not, i can barely run sdxl as is lmao

[clear oyster]

same lol

[nova zodiac]

the nose will generally be at 90 degrees to the line between the eyes, whereas a first finger bone has roughly 90 degrees of movement, and the second bone another 90 degrees to that

[clear oyster]

sdxl is pushing it just barely haha

[clear oyster]

thats a good point, actually now it does kind of make sense

[jovial wraith]

what gpu do you have?

[clear oyster]

6gb 4050 LMFAO

still it generates 4 in 4 seconds so not bad

[jovial wraith]

i'm out here with a 1060 3gb LMAO

[nova zodiac]

yeah 6gb for sdxl is pushing it..

[clear oyster]

the only thing is it goes to lowvram mode

[clear oyster]

OOOF RIP LMFAO sorry lol

[nova zodiac]

the 1060 6gb is a usable card.. the 3gb version, yeah ooof

[jovial wraith]

i mean, it's slow, sure. but it's still better than no gpu

[clear oyster]

it spills over likd .5 gb to ram

[jovial wraith]

and i got it for a great price

[clear oyster]

facts
cpu only is atrocious even on good CPUs

[nova zodiac]

oh that's gonna hurt

[nova zodiac]

yeah, ~20x slower

[clear oyster]

yeah since AI started nvidia GPUs been pricy

[clear oyster]

I actually dont really feel a difference surprisingly
I mean cant really compain about 4 images in 4 seconds lol

[nova zodiac]

they got pricey since covid, never recovered, then AI slammed the nail in the cheap gpu coffin

[clear oyster]

yeah thats probably it

on the standard sd models however 6 is totally fine

they uses more like 5 max

have you tried the 2gb models for that 3gb gpu?

[jovial wraith]

i bought mine a little over a year ago second hand. it was a fully working pc, albeit a bit on the older side. i5-4670, 16gb ddr3 and the aforementioned 1060 3gig. 150 euros

[jovial wraith]

you mean the pruned models?

[clear oyster]

yeah from civitai

[jovial wraith]

if theyr'e available i use them yeah

[clear oyster]

more than a bit older but for the price good

how long does it take on that?

sdxl turbo is definitely faster ofc

[jovial wraith]

i've been using Juggernaut XL Lightning a lot, and it takes about 20s/it on 832x1216. but it's only 4 steps so it's still pretty fast

[clear oyster]

interesting, only heard of that juggernaut just now, 4 steps is an odd number haha

[clear oyster]

isnt that like 80 seconds total then?

[jovial wraith]

roughly, yeah

[clear oyster]

cant imagine that lol
im used to them coming out in a few seconds

4090 users probably spoiled af lol

[jovial wraith]

and the results are pretty good! i just posted some images in general-with-images if you're wondering

[jovial wraith]

true lmao

[nova zodiac]

image won't change quality on older gpu, just affects speed

[clear oyster]

i mean must come out like instantly almost lol

[jovial wraith]

that's true, i was more so saying it because it's only 4 steps

[clear oyster]

still, the free/lower version plans of the online image genereators are pretty slow

[clear oyster]

I see

apparently juggernaut is the mst popular on civit right now

[jovial wraith]

and for good reason

[clear oyster]

i should try it
looks like 6.5 gb so same as turbo

sdxl turbo*

[nova zodiac]

Juggernaut XL Lightning and Dreamshaper XL Lightning (or Turbo) are definitely head and shoulders ahead popularity wise

[clear oyster]

me when I forget to set sdxl turbo to 1 step and cfg scale 1 and thinking why the images are coming out broken 🤣

[clear oyster]

I see
are much better/faster

I have the 2gb dreamshaper and its not very impressive, still better though

[nova zodiac]

the quality to speed tradeoff is definitely minimised on those two, but you can apply the lightning lora to any SDXL model to get a good result in the style of that model

[clear oyster]

for photorealistic images is there much improvement? about how much is the speed tradeoff

i'm just going to download and try it

[jovial wraith]

good luck!

[jovial wraith]

can't really say since i've only used the lightning/turbo variants

[clear oyster]

whats the lightning one?

i have the pruned one and now the turbo one but havent heard of a lightning one

also downloading juggernaut

i think this is enough models for me to have on my smll 512gb ssd for now lol

[jovial wraith]

the juggernaut version that's based on SDXL lightning

[clear oyster]

oh so just the juggernaut one that is first on civit right?

downloaded that just now

[jovial wraith]

i think so, yeah

make sure to read the description. it specifies a specific sampler. can't remember which one off the top of my head

[clear oyster]

checking that, thanks!

[pale latch]

since mining madness revealed to nvidia that people would pay a lot mroe than they thought for gpus, they've adjusted MSRP. Also, nvidia realized they were the only ones taking hardware support for machine learning seriously so if people want those features they need nvidia, and thus more value. Why leave money on the table?

[hallow sable]

with Zcuda that might be less of an issue in the near future

[nova zodiac]

indeed, but still takes a slight performance hit

[pale latch]

That's been the noise from the AMD camp for a long time. "soon guys! soon we'll get cuda support! and it won't be buggy!!"

[hallow sable]

k, I think amd has been improving a lot over the past few years while intel and Nvidia got lazy

[rich kestrel]

theyre gonna be left in the dust if they dont add native cuda support

[hallow sable]

remember the same intel cpu that kept making for 10 years?

[rich kestrel]

AI is the name of the game moving forward, even more so than gaming

[pale latch]

fun fact. AMD could've engineered cuda support into their drivers a decade ago. But they didn't. AMD drivers my man. AMD drivers.

[nova zodiac]

the only issue with nvidia is the fact they keep castrating vram on their consumer cards..

[rich kestrel]

wont be the case with 5090 g

[hallow sable]

you can find problems with nivida drivers too so whats the point...

[rich kestrel]

32gb vram

[nova zodiac]

the 1080 ti had 11gb.. the 4060 has less than that

[nova zodiac]

still way too low

[rich kestrel]

lol and what does amd have?

thats the standard my friend

[hallow sable]

yeah for a 2k GPU they should give you all the VRAM

[rich kestrel]

above standard I would say

[pale latch]

lol you're a fan boy. intel didn't stagnate for 10 years. That's also not that much of a burn as you think. If they did just chill out and keep releasing the same product for 10 years, they're still on top. Says a lot.

[rich kestrel]

intel is tha bomb

[nova zodiac]

it's been 4 (maybe 5 if you include the 16xx series) since the 1080.. even if a stepwise increase the top nvidia should be a 48gb

minimum

[rich kestrel]

bru

24gb is already enough for 99% of shit I do

and I'm into AI

let alone the average consumer

[nova zodiac]

images yes, llm no

[rich kestrel]

llms still suck ass anyways and nobody gonna tell me otherwise

[hallow sable]

intel is losing to AMD now and has been for a few years. and no you can check how little has been done to intel cpu from 2nd gen to 9th gen... single core performance has jumped to 15 to 30 % over the 2nd gen

[pale latch]

its not always about vram. the ada architecture has a ton of instruction sets and hardware support for things that ampere don't

[clear oyster]

oh yeah totally forgot about mining

[hallow sable]

also new sockets every generation... they are still making AM3 cpus

[clear oyster]

mining was nvidia too? makes sense

[pale latch]

hopper floating point engine is abig deal coming up. fp8 support on ada is superior in so many regards.

[clear oyster]

amd and all the others are rip compared to nvidia right now

nvidia stock fcking mooned lol

[pale latch]

sorry, mb, its called the TransformerEngine

[pale latch]

i love that a video game hardware company has become a trillion dollar corporation

[clear oyster]

its insane isnit it!

they jsut got lucky af lol

[pale latch]

honestly i expected nintendo to get there first

[clear oyster]

who would have thought that video games, mining, and ai all need the same thig

[rich kestrel]

lucky my ass. you make a good product you have a good company

I fuggin luv my 3090

amd is hell especially with rocm. like coke vs pepsi

[clear oyster]

they were lucky that there gpu was whats needed for mining and ai and not AMD's

[pale latch]

not even lucky. they invested billions into ML research when nobody else was. It was a huge moon shot

[clear oyster]

AMD gpus are better for$ value for gaming

[clear oyster]

true I guess

[pale latch]

they did the work to get there

[clear oyster]

so it wasnt full luck

they took a risk and it paid off

paid off bigtime

[pale latch]

i actually exclusively bought amd gpus for years. YEARS. because i was just doing gaming and they were 100% fine and i didn't need RTX.

wasn't until i got into diffusion and machine leraning that i realized my next card needed to be nvidia this time

[clear oyster]

im not a gamer never was so i never had amd

actually this 4050 laptop/mobile is my first dgpu lol

hence why it for me is a lot while for most its nothing lol

[pale latch]

ngl, laptop gpus aren't dgpus in my book , but still i appreciate that shit lol

[rich kestrel]

dont go there

[pale latch]

dgpus gotta be like, a big ol card with graphics on it

mfw i opened up my 4080 box

[rich kestrel]

ppl will eat you alive here for dissing their craptops

[pale latch]

dude i mention i have a 4080 and i get eaten alive by people pissed off that i spent to much. oh well. gpu go brrrrrrrrrrrrrrrrrrrrrrrrrrr

[clear oyster]

lol

[clear oyster]

lol gamers are like that
gamers just realized they're GPUs are good for AI and thats how they got into it lol

[clear oyster]

the igpu is the one in the cpu thats for sure

[clear oyster]

technology evolved and gets more compact 🤣

[pale latch]

correct. your gpu is at least it's own chip with some memory wired directly to it

[clear oyster]

yeah

[shell tendon]

Ha I just bought a 4090 after spending 1500 on a screen after spending 2000 on a TV and will prolly set up a second system for my backup card (3080 12gb)

[pale latch]

STILL TECHNICALLY IS INTEGRATED THO

[clear oyster]

LMFAOO

[pale latch]

yeah its mostly the poor people who get jelly

[rich kestrel]

yeah ok thats the other extreme

[clear oyster]

if you take it literally yes but you cant take it literally as in "integrated" lol

[rich kestrel]

not all of us shit money

its about making the smart purchases

[shell tendon]

I don't either but

[clear oyster]

we went from 1 extreme to another lol

[shell tendon]

I don't go to the bar, I don't eat out and all my fav travel locations are within a days drive

[rich kestrel]

thts why I get defensive in the first place. cuz ppl complain when they cant get AMD to work or have to go through hoops. Like bruh. Sell that junk and buy a real gpu, green team

[clear oyster]

exactly

[rich kestrel]

not being to afford a 4k tv thats different

[pale latch]

i'm definatly not just shitting it out. i sweat that money out harrrd. i worked for this pc hard over 2022. if i was shitting it out, it was one of those where you gotta take your shirt off and have someone bring you a water mid shit

[shell tendon]

Used to spend prolly 70 a week between trivia nights and various takeout or other shit on average

That's almost 4000 a year

[clear oyster]

while some people may seem richer by having all the newest and most expensive stuff, in reality the "poor" looking person with only ths stuff they need is probably richer by not having so much debt lol

[pale latch]

same. 2022 i stopped buying snacks at gas station. right there was easily a $1000 in my budget

[shell tendon]

I just choose to make my frivolous expenditures in one category only these days

Yuuuup

[clear oyster]

i mean most people (99%) buy everything on credit lol

[shell tendon]

Pretty easy to cut back if you tell yourself "don't buy this coffee every day and you can buy a 7950x3d in three months for the amount saved"

[clear oyster]

then those who have no credit card debt look poor LMFAO

[clear oyster]

then the question is what do you need a 7950x3d for

(just general not asking you)

for AI you raelly dont need a cpu thats any good

[pale latch]

you've still gotta budget for credit

[clear oyster]

are you guys gamers here? then that would explain, they always need the best things lol

[rich kestrel]

bru ive put off so many games bc I've been waiting for a 4k 120hz tv

[mortal delta]

I use to be a gamer and i still am from time to time.

[rich kestrel]

dont want to ruin the experience with inferior settings

and now that I have the TV I had to move out and those games are just piling up

the struggles of being a gamer

[pale latch]

i'm a huge gamer but i've always built budget PCs for gaming my whole life. AMD cards were all i got ever. I never cared that i couldn't turn on the best features, long as i was playing the game.

[hallow sable]

big talk for someone who just makes AI fake porn

[mortal delta]

hey i have an amd card...

[pale latch]

good enough for gaming ngl

[mortal delta]

not good enough for ai though, i cant find a way to get stable diffusion to run yet.

[pale latch]

if all i was doing was gaming, i'd still be using AMD and probbaly still be on Linux since the drivers for AMD are superior there

[pale latch]

driver support. Linux

[mortal delta]

i come from a family that cant stand linux

[hallow sable]

I am gonna boycott logischeit... horrible software and every year worse hardware

[rich kestrel]

good hardware, bad software

thats logitech g

[pale latch]

onhhh its a family pc. dude thats like driving your moms car.

can't exactly put a performance turbo on your moms minivan can u ?

[rich kestrel]

how tf do you save porn on a family pc

dafuq

[hallow sable]

they made it so you have to use windows settings to change lighting instead of the software that they made... also no linux but rather mac support. braindead japanese decision making as always

[mortal delta]

no its not, its my own pc i just get shame whenever i try to install linux.

[shell tendon]

Im pretty sure that with training SDXL on a 4090 you get CPU bottlenecked

[pale latch]

lol fuckem then. Tell them "skill issue"

[shell tendon]

5950x isn't enough to drive the 4090 past 50% load

[pale latch]

embrace the linux smug if they're going to be windows smug

[mortal delta]

lol, your so right

[clear oyster]

lmfao

[clear oyster]

for training, yes

[mortal delta]

maybe i should get a virtual environment to try things.... also what interface would work btw?

[vale otter]

both coke and pepsi have very little difference and both are overpriced, defending one over another only demonstrates narrow mind

[rich kestrel]

lol tell me you dont drink soda without telling me you dont drink soda

id rather drink water than pepsi

[pale latch]

cant use gpu in a virtual environment

not unless thhe host is linux and is passing it through

[mortal delta]

multiboot?

[pale latch]

you can get a live boot usb for many distros to try it out. where it runs the system off your usb.

[mortal delta]

how are you so smart?

[nova zodiac]

can use nvidia gpu in wsl

[pale latch]

i never assume i'm smart and i always feel like i'm dumb and am always striving to get smart. never stops.

[mortal delta]

i sue to have a low end nvidia but now i have 8vram amd.

[pale latch]

the moment i decide i am smart, i'll probably get lazy and stop learning

[mortal delta]

that sounds really epic like a quote from a movie.

[vale otter]

it's not anyone's business what sucks for you, but if AMD provides more VRAM I go for AMD (or I go for nvidia if they provide more)

[mortal delta]

the only linux distro i really tried was pop os and linux mint mate.

[pale latch]

https://www.youtube.com/watch?v=iV3lnLE4-nI

[mortal delta]

before i click what is this?

[pale latch]

scientific fax

[pale latch]

pop an mint are both nice ones. for diffusion stuff i was using arch distros. through 2022 i used garuda and manjaro. i liked garuda a lot

[hallow sable]

I like terminal only OS

[mortal delta]

sounds painful like ms-dos.

[pale latch]

i like when os's are gooey

[pale latch]

ms-dos did at least have dos shell

[mortal delta]

true but im too young for that.

what interface for stable diffusion even works on linux + amd?

[pale latch]

ROCm

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs#install-on-amd-and-arch-linux

[vale otter]

I like that ComfyUI just uses system python and not annoying venv

[pale latch]

comfy i think has it's own virtual environment still. that would be haywire if they just dropped all the libraries inot global environment

not great security either

[rich kestrel]

why would any program dump libraries into global

[mortal delta]

i dont understand how people understand comfy.

[rich kestrel]

understanding it, easy. preferring over a111, yeah thats wild

[pale latch]

node graphs are in a lot of software. that's the hardest part to figure out imo. wtf is a node graph

[rich kestrel]

I use both but... I'd never give up my beautiful a111

[vale otter]

maybe it can have it but it pretty much runs fine with system pip

[hallow sable]

blender, houdini, a few game engines, substance painter use nodes

[mortal delta]

the only nodes ive seen are in blender and unreal but i dont use either.

[pale latch]

usually when things are able to be done in comfy and not in a1, its because the code isn't ported yet. not because of a benefit of nodegraphs.

There are a few things node graphs can do that you can't achieve in a1, but i honestly think many of these workflows are just pointless

people will gan scale an image then diffuse it again... why even?

[mortal delta]

so comfy works on linux + amd, right that is what we are saying?

[vale otter]

yes it should work

[pale latch]

video editors too. code editors. tons of technical editing and design software use node graphs. it's a common UI paradigm.

"no coding required" though is bullshit since node graphs 100% are visual coding

its like using explorer to look through a folder and declaring "no file management needed" since you're not using terminal commands

as if drag and drop makes file management a thing of the past...

[mortal delta]

now all i have to do is install linux, are you sure arch is a good one?

[pale latch]

no. there are 1000s and i don't know your skill level. you could fuck it all up and that'd be bad

arch sure is one though

[mortal delta]

Im so bad at linux if that gives you an idea im pretty much a somewhat beginner.

[pale latch]

i used more complete distros of arch. downstream distros with goodies and thigns added by default so i wouldn't have to add them later. garuda comes packed.

i was running games on linux better than they ran on windows

[mortal delta]

noted

[pale latch]

red dead redemption 2 was the one i played the most

[mortal delta]

for some reason i really like semi retro/old games, but someday i hope to make a game by myself but idk if that will ever happen.

[pale latch]

rdr2 is soooo old theres no cars only horses!

god damnit. this guy talking about thick latinas when i'm talking about linux and video games. now i'm a dweeb by contrast. FUCK. what a fucking troll.

[lone hawk]

too late, game industry is over... Sora is the future, all those game developing companies + Hollywood are already dead, they don't know it yet))

[vale otter]

How to contact mods?

[pale latch]

ai won't replace crafted art

[mortal delta]

that not what i mean i mean like games that came out on the wii or around that area or even before that.

[lone hawk]

already did

[karmic brook]

#1010934719455707218

[pale latch]

mfw the wii is retro 😦

[mortal delta]

kinda?

[gaunt pulsar]

I think Sora is too expensive to run tbh. I wonder if they can even do a mass release this year

[pale latch]

to me, retro is still space invaders and supermario 3 is the new shit

[mortal delta]

i think its different for everyone.

[wispy cloak]

What amd videocard I should buy to run SD3? How much vram?

[gaunt pulsar]

As someone with an AMD card, I recommend Nvidia for anything related to SD

At least until they add better support

[wispy cloak]

how much vram I need for SD3?

[still glacier]

https://tenor.com/view/elmo-who-knows-idgaf-i-dont-care-i-dont-know-gif-23016529

SD3 isn't released yet. Not much is known so far.

[charred mesa]

hopefully, some time after SD3 comes out, 8B will run on ~8GB, but that's just my theory

[hallow sable]

I dont think 8gb will work bruh

[charred mesa]

not even with fp8?

[hallow sable]

more and more games are using higher and higher texture sizes so more vram is need. gpu manufactures will need to keep up

8gb will be the low end

[gaunt pulsar]

Did anyone get access to the waitlist yet?

[charred mesa]

maybe companies could keep up and not put up impossible deadlines so that the developers can optimize the games for once

no but I'm talking about SD3

[warm junco]

I also want to see how much fp8 will change model sizes and inference time

[charred mesa]

you actually LOAD the models in fp8

[warm junco]

Could be a game changer maybe

[wintry chasm]

How do the NVIDIA RTX cards compare to the 40 series? Was looking at adding a card to run AI workloads to my NAS.

[charred mesa]

30 series to 40 series?

[wispy cloak]

And what Stable Diffusion 3 safety means? is generating realistic zombies for game gonna be possible?

[charred mesa]

hopefully zombies will be possible, I can generate bloody and gory zombies with SDXL no problem, and that model was apparently censored too

I think its mostly about corn which will be censored heavily

[hallow sable]

people are still gonna train their own models using SD3 as a base regardless if they sanitize SD3

[gaunt pulsar]

During the research preview only right?

[charred mesa]

if it's not SD2.X or stable cascade level censorship then sure

[charred mesa]

welp... 😬

Don't think so

[gaunt pulsar]

[charred mesa]

#📣｜announcements -> Today, we are opening the waitlist for early preview. This phase is crucial for gathering insights to improve its performance and safety ahead of open release.

to improve its performance and safety

so they want to actually improve ""safety"" a bit more actually

I really hope its just like hardcore corn or something man

[gaunt pulsar]

Yeah, I really hope so too

[charred mesa]

or maybe they are just trying to please the public and the investors by claiming "safety", so that stability doesn't get bad rep

if it's sdxl censorship we're fine if it's sdxl censorship we're fine if it's sdxl censorship we're fine if it's sdxl censorship we're fine

[gaunt pulsar]

I'm hoping they make it a censorship only in the base model that is very easy to be completely removed

[charred mesa]

ehhhhhhhhh

yeah that's how it goes, but completely removing it would take a lot of time

bunch of a100 gpu days

and even then I don't know if it will be "completely removed"

but if this better text encoder or prompt adherence helps adding new concepts much better, then probably

[pale latch]

when the shit hits the fan, there'll be a lot of finger pointing and public scape goating. Every company wants to crack AI and be a trillionaire, but no company wants the kind of brand legacy that comes from destroying the old society

[charred mesa]

Mistral 8x7B was simply uncensored and they had a censorship/safety switch that companies could turn on

it's the perfect compromise

but image generation is much more dangerous I guess

[pale latch]

also, AI safety is a big deal. Releasing it in responsible ways makes a lot of sense. You've got to let society and culture acclimate much like the human body has to in new environments

The human collosus is as much an organism as any individual

[gaunt pulsar]

This would be perfect tbh

I mean we all know it will happen after a few months. It will only delay it a bit

[charred mesa]

eh

[gaunt pulsar]

If they don't do that

[echo marsh]

https://huggingface.co/spaces/mvaloatto/HF2X ya boi is in the top 50!

[charred mesa]

I want to be optimistic, but I have to have my limits

I can't just think that 8B will run on mobile phones and won't have censorship

it will be censored

[pale latch]

with image generation, people's definition of censor seems to shift all over. people say SDXL is censored but a friend of mine demonstrated some pretty steamy loras he has for sdxl

[charred mesa]

it will probably require 8-12GB of vram (the largest model, if not more!)

[echo marsh]

there is always sdxxxl

on civitai

[charred mesa]

yeah sdxl wasn't as censored as SD2.x

if SD3 is as censored as SDXL then it will be just fine

[pale latch]

i'm okay with base models not having pornhub in the training data tbh

[charred mesa]

exactly, same

[pale latch]

i guess i'm a puritan

sd2 isn't even as censored as people claim. you name the right photographer and or painter and boobies come out of it just fine

its just unfamiliar latent territory to most people since it doesn't use the clip vit-h

[charred mesa]

I was one of the advocates of SD2.X, but not because I claimed it wasn't as censored

[vale otter]

no model is truly uncensored, trivial tests can show this, and seemingly pure "pretrain" just won't work because of lack of instruction-following

[charred mesa]

yeah it's a stretch, but compared to other models like llama its much less censored iirc

even as a base model

oh god and gemma

[pale latch]

sd2 is a great demonstration of openclip's versatility and effectiveness. but its such a different latent space from clip vit-h. unfamiliar. It's used as Stable Video if i'm not mistaken

[charred mesa]

I loved SD2.1

[pale latch]

technically, any amount of culling of the data set is censorship

[charred mesa]

I wasn't into AI corn so I had a lot of fun with it

#1045349359044280360 message

[pale latch]

yeah honestly, the ai corn communities are fucking obnoxious. it all sucks. the only good ai corn relies on real porn for any quality.

[pale latch]

exclusive link. access denied.

[charred mesa]

Ohhh

2-point-1 channel

maybe you need archive access

[pale latch]

you nailed it thanks

[charred mesa]

https://media.discordapp.net/attachments/1045349359044280360/1091358780610064384/04999-3351106486-RAW20candid20cinema20GigaChad20sitting20in20front20of20a20laptop20typing20on20the20laptop20remarkable20color20ultra20realistic.png?ex=65eb1791&is=65d8a291&hm=7cd213fff09a406e157a257f089c16bc411594f914541be3907d9bacbcde3454&=&format=webp&quality=lossless

Here's the image if you click on it

also, look at this, this was before LCM and Turbo https://media.discordapp.net/attachments/1045349359044280360/1084777879914676284/xyz_grid-0036-2717413535-RAW_candid_cinema_close_up_portrait_of_Dwayne_Rock_Johnson_wearing_a_cowboy_hat_in_the_wild_west_wooden_plank_buildings_in_the.jpg?ex=65eed620&is=65dc6120&hm=6a3a6e2851b99b821ad255e9e434fc4f9e6c58199de12dd7ce580046dee7044b&=&format=webp&width=1078&height=676

[pale latch]

hahaha check out the new playground 2.5 so you can admire it's solid black images and then make regular ones again

[charred mesa]

UniPC could do 1 step images lmao

zero finetuning

they were quite low quality, but it was possible

https://media.discordapp.net/attachments/1045349359044280360/1084605856072601640/xyz_grid-0045-769585753-RAW20candid20cinema20close20portrait20of20a20blue20fox20in20a20winter20forest20snowflakes20from20the20sky20remarkable20color20ultra20realisti.png?ex=65ee35eb&is=65dbc0eb&hm=901ea481b6a57ac3951bf14e15e1af4c0dbf16f110b7d9e127c403b88c809a32&=&format=webp&quality=lossless&width=1440&height=178

[graceful peak]

whenever i use openpose it always freezes at 50% for like a minute any reason why openpose is rlly slow?

[wintry sequoia]

Am I not able to generate via this discord anymore?

[warm junco]

For Sdxl or 1.5 ?

[graceful peak]

sd 1.5

[pale latch]

probably are going out of memory and instead of crashing, it's falling back on system memory and doing some swapping

how much vram?

[pale latch]

ohhh wait. i'm trying to use pg2.5 in auto and its no worky. they've changed something somewhere and the model just creates random noise

[warm junco]

You need to make sure that your using the 700mb Controlnet models.
Because the 1.5-5gb ones can cause the loading issues

Because they need much more vram

[lime mulch]

hello all

[hallow sable]

no!

I need to get into comfy ui

[shell tendon]

have you experienced the same issue

[fervent thunder]

yoooo whats good

🙂

my friends there are here some people that like drugs and want some ?????

[native cave]

anyone getting access to sd3 yet?

[pine fiber]

no

[hallow sable]

brah wait... sd 4 or something else will come soon

[fervent thunder]

damn yall didn't get an email yet?

guess they are bringing us in, in batches

[shell tendon]

care to generate something

how about... a cow is lying upside down on top of an alligator. the alligator is on top of a school bus to the left of a city bus. both are on the surface of the moon. a supermassive black hole is spinning in the sky with a UFO getting ripped apart by the event horizon.

[fallow veldt]

I didn't get much great results in CN tiles, but I didn't understood the process well.

I feel Ultimate Upscaler works in the same way, and now I'm getting very good results, just playing with the denoising. 0.12-0.18 for respecting the image, 0.2 or more to let it create or add more details (produces aberration/alucinations the higher you go, just like CN tiles).

I wasn't getting good results either until I played with the denoising. I got really good results lately, and I think it is better as with a tile of 1024 it can see more of the image, so less aberrations.

[high trout]

There isn't a channel to try prompts here? right?

[buoyant moss]

https://huggingface.co/spaces/PixArt-alpha/PixArt-alpha
Pixart is a diffusion transformer, but be warned, due to safety they trained on a dataset with most faces blurred out

[nova zodiac]

Nah.. best to try them over on HappyAccidents.ai

[buoyant moss]

I sure hope HappyAccidents is paying stability

[nova zodiac]

I'm not associated with Stability sadly

[rustic prism]

oh so the stable channel is just dead now, they dont support image generation at all anymore?

[nova zodiac]

#1047610792226340935 - we are pretty certain they'll be back for sd3 release in near future

like 2-4 weeks near

[dusk scaffold]

It's only the beginning, but many things are already censored and although they show something new and interesting in a short time, nothing really grabs people anymore due to these limitations and difficult access. It won't be long before it becomes a tool for a limited group.

It's been a while since I've played with her. I'm not going to burn up hardware so that in a few months' time something new and better comes along.

[quasi rune]

Anyone know what text encoder CLIP model Cascade uses? Seems like it might be a fine tuned half precision variant of CLIP Big G?

Strange that there’s so little information about it. Have looked through docs and release statement.

[nova zodiac]

I've read somewhere that it is using 3 text encoders....

g, l and something else

my memory is a bit shit 😛

[clear oyster]

dont hve a 4090 lol

[pearl ocean]

*yet

[karmic cedar]

Just curious if folks have a favorite model/lora for pixel/sprite art?

[finite yacht]

voxAIo saying hello to everyone! Stable Diffusion / ComfyUI fanatics.... Waiting for SD 3.0... uugh!!

[karmic cedar]

yep—these days, an image is a thousand tokens

(or more)

[fervent thunder]

forget the 4090, double up and get the a6000

vGPU? yesss

sign me up

[karmic cedar]

I’m fine with tunneling into massive diffusion engines like cyberpunk abraham lincoln on a train WEEEEEEEEEE

[clear oyster]

lol

[clear oyster]

true lmfao

[icy quest]

ey anyone know why the preprocessing tab is gone

bruh this Aitrepreneur hypernetwork tut is aggravating. He shows how to train the hypernetwork, then quits and tells you not to do it and never runs the hypernetwork. Is there any steps to apply the hypernetwork to your model? because he just ignored that

[rich kestrel]

hypernetwork??? wtf is this, 2021?

outdated as all hell brudda

go for lora, TI at worst

[icy quest]

the vid is from 2023

uh, won't hypernetwork create new poses and new objects, isn't lora just to customize a person?

[rich kestrel]

nope

nothing a hypernetwork can do that a TI cant do, nothing that a TI can do that a lora cant do

order of effectivity is finetune>dreambooth>lora>TI>hyper

[icy quest]

so how does dreambooth do the training 200 times faster?

[rich kestrel]

nope, it is just stronger so to speak

but does require more time and more vram

and regularization images IMO to really get it to do what you want it to

thats why most folks opt for lora since its quicker, smaller, can be used across models, less vram

I prefer the DB+lora combo for maximum accuracy but not everyone is so picky

[icy quest]

whats regularization

[rich kestrel]

essentially images where instead of the desired subject/style, you add a variety of images in the same vein

that way, the model can better distinguish between your subject and other subjects within the same category. makes your model more flexible. not required for lora, I recommend it for DB

[icy quest]

okay, ill try it. so i got this hypernetwork, is it like a model? can i just use a model merger to use it?

[rich kestrel]

the hypernetwork I think you just drop into a folder, forgot which one

it is technically a model

then you use it in your prompt with a keyword like my_hypernetwork

[shell tendon]

https://twitter.com/EMostaque/status/1762256072713158743 I like this a lot

"who needs bokeh" we've been heard

I don't think there's anything I dislike more than bokeh

Single subject images are boring as f

[warped shard]

does anyone have any model reccomendation for really good anime renders?

[delicate stratus]

helo

[nova zodiac]

what sort of anime... are we talking a flat 2d style, or more of a 2.5d type renderish?

MeinaMix, TMND Mix, AAM XL, Quiet Goodnight would be good starting point

RealCartoon-Anime too

[shell tendon]

Also depends sd15 vs SDXL

I'm not into anime art specifically but I really like the checkpoints for the weird art shit

BluepencilXL, deep blueXL, realistic visionanime... Good stuff from those

OdemXL if you want to blast your brains out with aggressive weirdness, be sure to throw "collage" into the negative prompt though

[sudden ruin]

ckeck out the #🍥｜anime channel for inspiration and maybe ask around there

[inland swan]

I just wanna install koyha_ss without having every issue int he world

but I'm not allowed.

[inland swan]

I'm really starting to hate python

[potent spire]

Why

[stone stirrup]

Hello Stable Diffusers ! You are doing a great job 🙂 So glad SD exists, discovered it a month ago. Happy creating everyone !

[inland swan]

because instead of there being one verison that everyone uses I hvae to now install 3.10 as well as 3.12 so forth

[nova zodiac]

you know about virtual environments right?

[inland swan]

yes guess what

it didnt work

as soon as you run the setup it tries to run 3.12

so I changed the setup bat to ensure its running 3.10

and appearantly it worked outside of a new error I"m now getting

[warped shard]

2D for sure

I’m also not sure why rendering take more time than before now

[flat sparrow]

I have a sideview of a face of a person and I want to construct him from the front. Is there a way to do this?

[nova turtle]

Hey SD Fam, i may have a simple question... iam looking for an workflow. I would likte to inpaint my face into a existing picture of an man. What is here the prozess. i think face swap is mosty not detailed enoigh and also my haircut is then missing in the picture. Can anybody help?

[jovial wraith]

you can do some photobashing and then run it through img2img on a low denoising strength

[manic viper]

Create a logo of "Pint". Pint is herbal infused water

/Create a logo of "Pint". Pint is herbal infused water

[broken smelt]

i refuse

[burnt jasper]

was the zero123 channel removed?

[tepid jolt]

Hello guys, is 2.8it/s good for a desktop 3090? on SDXL

I feel its slow for 3090

[sudden ruin]

Seems like it was archived yes

[willow ridge]

hello guys! I am new member in your community

[raven field]

Bot is down permanently ?

[jovial wraith]

Probably not. Give it a few weeks max. That's just speculation tho

[willow ridge]

I am 3d designer, how can i help you?

[wispy oasis]

what can you make

[willow ridge]

3d models

[wispy oasis]

can you make 3d models with ai

[jovial wraith]

Not very well. The technology isn't really there yet afaik

[willow ridge]

yes sure

[jovial wraith]

What you can do is create a 3d scene and use it as a base for an ai image

[wispy oasis]

ai to generate textures for models?

[shell tendon]

Depends on resolution and the sampler

[tepid jolt]

1024x1024 and DPM++ 2M Karras

Maybe it's because Cinema4D and Substance Painter are uses gpu in the background

in sd1.5, I get 12it/s

[shell tendon]

I'll let you know some things with my 4090 later

But yeah gonna be a hell of a lot slower with SDXL than sd15

[wispy oasis]

hey

[tepid jolt]

I was going to buy a 4090 but I'm a little late for that, I'm waiting for 5090 now 😄

[wispy oasis]

do you guys know where i can find out roughly how many images per second/minute i can generate 1024x1024 SDXL on a 4070?

[shell tendon]

Make a Lora of the side view, then spam 1000 seeds until you finally get the front view right by blind luck

Then throw that back into your training set and make a new one and spam a thousand more seeds to get an angle view

[willow ridge]

how to prepear

[shell tendon]

Just bought a 4090... Have a 3080 12gb on the shelg now

Second the 5090 is out so long as it's significantly faster or has more vram I'll get that too

[wispy oasis]

bruh finna spend a car on a gpu

i salute u

[shell tendon]

Lol

5 bucks a day on coffee = 1800+ a year

[wispy oasis]

i am trying to buy 4070

[wispy oasis]

ya i got some habits to kick.. lol

[tepid jolt]

can I speed up my stable diffusion workflow with dual gpu? Therefore, I will install the 5090 to my current setup with 3090

[shell tendon]

Eating out and/ or going to the bar twice a week = 80 x 52 = 4160/year absolute min

[wispy oasis]

i am worrying that the 4070 wont run SD3 since it only has 12GB vram. it might not be worth buying. thoughts?

[shell tendon]

Just build a second PC it's what I'm gonna do, build. Zen 5 box with a 5090

So right there that's how I justify flushing 6000/year on AI shit

At least GPUs aren't bad for my health like those other things lol

[tepid jolt]

In the country I live in, graphics cards are almost twice the price due to taxes, but this does not affect me much 😄

It is possible to achieve anything you want after working hard

[noble star]

You can run multiple GPU's in one computer

Saves alot of time having everything in one place

[winter pike]

id say 12gb is good but since the requirements of sd3 arent available waiting may be the best option for now

[shell tendon]

https://www.yahoo.com/tech/willy-wonka-inspired-experience-scam-020131173.html 🤣

[noble star]

Get a 3060 12gb if you want a 12gb card. They are like 200$

[tepid jolt]

My advice to you is that it would make more sense to buy a second-hand 3090, it has more VRAM and probably it will be more economic

[shell tendon]

Yeah used I guess... New 3090 are crazy

4090 shortage somehow got even worse in the last few weeks

[unique quiver]

hello

[sudden ruin]

The price for my 4090 is like 400 bucks higher where I am located right now compared to 6 months ago when I got it!

Its crazy

[shell tendon]

Nuts

I got a gigabyte oc for 1800 two weeks ago

Very happy with it

30 Steps dpmpp Karras for SDXL 1024x1024 took 3-4 seconds

That first generation I was like well, wow, that was totally worth 1800 plus tax lol

[rose oar]

Hey, guys! I'm a Photographer who is starting out with Stable Diffusion, and I want to build a custom solution that I can use for my portrait photos in studio. I'm looking for some one with experience and some time on their hands to help me shortcut some learning curve and help me with this fun project. 🤞

[severe vine]

Hello, guys, does somebody know about projects in consistency characters and environments? I heard about plugins that changes faces, loras, but all these are bad. Main problem even not character itself but background. If we will be able to create consistent characters in envirouments it will open new ways in comincs, games story telling etc

[fervent thunder]

hello

[sudden ruin]

I got the MSI liquid X one for 1800 as well iirc

[sudden ruin]

Try posting this in #1092446741984444416

[sudden ruin]

Seems like SD3 will have awesome prompt understanding, so maybe that will help with consistency too? We will see tho

[worthy bone]

I hope that with 12GB of VRAM I can run SD3 😄

[severe vine]

Yeah, same

[worthy bone]

What GPU do you have?

[severe vine]

3060

[worthy bone]

same

[wispy oasis]

ya im trying to buy a 4070 and i want to run sd3 too

[worthy bone]

I don't know if with 12GB it will be able to handle the full model but to be safe I would advise you to wait before buying that GPU, at least understand if it will be able to handle it well. However, I really hope that with 12GB you can handle the full SD3 🙂

[sudden ruin]

Unfortunately we have no info on specs needed yet I think

[worthy bone]

Ah ok I understand. Thank you for answering 🙂

[teal osprey]

what yall suggest for generating a crowded place? im having hard time getting detailed people when they are many in count

they are getting deformed alot

[white star]

Hello everyone,
I'd like to apologize if I'm asking this question in the wrong channel.
I would like to know if there is already a library, software or any other way of integrating an image editing manager with area selection. The idea would be to have a result close to adobe firefly (selection of the area to be modified and generation of a modification via a prompt). I'm interested in any information so don't hesitate 🙂

[shrewd linden]

Is it me, or is stable diffusion very far behind dall-e at this moment (even with XL models) to interpret prompts. For example, I just put in this prompt in Dall-E and SD (using Webui-Force, DreamshaperXL Turbo) : "create a picture of indiana jones wearing a swimsuit, cartoon". Dall-E immediately shows a nice, 2D cartoon of what I asked. SD gives me a woman in bikini with an indiana jones hat 🙂

[jovial wraith]

just wait until SD3 is released in a few days/weeks. it'll change the game.

[rich kestrel]

dall-e might be farther in some ways, perhaps in accuracy

with SD it requires a lot of customization

but dall-e does hit a ceiling pretty soon while with SD sky is the limit g

[shrewd linden]

IT's very far off right now.

[rich kestrel]

in the hands of an amateur perhaps

[winter pike]

thats the reason sd3 is being developed

[shrewd linden]

yeah well, I'm trying to compare just how it works. been learning a lot about SD the last few days and results are often very good but it takes quite a bit of knowledge to put the prompts correctly, dall-e doesn't seem to need that but I don't want online stuff, want to generate things locally

For example this prompt : a ufo over a typical american house, small boy in the garden holding a teddy bear is looking up to the ufo, nighttime, light shining down, movie poster, realistic

In Dall-E it's spot on, in SD it doesn't show the teddybear, or the bear is on the grass or something like that

[rich kestrel]

theyre just 2 different animals

two very different tools

[gaunt pulsar]

did anyone get access from the waitlist?

[rich kestrel]

MJ and dalle are made more to be one-click wonders

SD is more like a studio/set of tools, just from experience

issue with mj + dalle is that ur limited to what u can prompt, you can always inpaint, img2img, etc. with sd

train models

finetune

loras etc. u cant do that with dall

[shrewd linden]

I understand that, but it's so frustrating to have to put together a LOT of prompts and then not always getting a results. For example for the prompt of the boy/ufo, I made it in dall-e and then uploaded it to a controlnet/canny and then recreated it in SD. Then it works.

[potent spire]

@shrewd lindenSD works differently with prompting

which can be very frustrating

@rich kestrelyeah each one has its advantages and disadvantages

a professional is more likely to use the simpler but very powerful ones tho instead of having to fight through for most people unnecessary settings like VAE etc.

[eternal cloak]

no more AI ?

[potent spire]

huh?

[shrewd linden]

I'm pretty new at this , but have learned quite a bit of the last few days about lora's, controlnet, models etc but one thing that still seems very hard to me is getting the right prompts

Been trying out (maybe too much different) models, but it seems there's a lot of crap. Would I be better off just sticking to the default sdxl?

[potent spire]

yeah it works more hard with SD from my experience, was the case with MJ and co. as well but at this point MJ, DALL-E, Firefly and some others are much more straight forward

[potent spire]

depends

if you are a generalist then sure

a bunch of those models are trained too much on photography and when you then try some different styles the generated images tend to be heavily photography oriented

[shrewd linden]

also, is there a list of what different sampling methods do?

[shrewd linden]

yeah I guess so, indeed when I use SDXL1.0 and ask for a cartoon, it's mostly correct. Not so with other models like DreamXL Turbo

[potent spire]

in my case it was JuggernautXL

and i really dont prefer to have 100 models for each specific thing i want

thats tedious

but then again as the user above mentioned, with SD you have a lot of customization possibilities and barely any censorship

[shrewd linden]

yeah that's a good thing. is there a model you would advise that isn't heavily focussed on photography or let's say "cute" girls?

[hallow sable]

jugnuts take like a min to load for some reason lmao

[potent spire]

didnt try a lot of them, continued using SDXL 1.0 base model

but SD with my specs is tedious in general

[teal osprey]

no wonder majority still uses midjourney to generate architecture related stuff....

[hallow sable]

why not train your own model on arch stuff?

[shell tendon]

so i don't think that's too shabby. i'm getting 6.8 it/s with forge, dpmpp karras 1024x1024 sdxl on a 4090

[burnt jasper]

anyone else had issues running sdxl models in forge ui?

[lean nimbus]

CAN ANYONE HELP ME I GOT STABLE TURBO BUT FOR MAIN TURBO CHEACK POINT I NEED 2 STEP OR 3 BUT OTHER ONE I NEED 10 OR BLURY WIRED PICTURE AND WHEN I USE ON 1111 AND DO SD THEN I GET WIRED COLURE

[shell tendon]

NOPE I DONT ANSWER QUESTIONS TYPED IN ALL CAPS AND IM PROBABLY NOT ALONE ON THAT

[pine fiber]

that finetune of sdxl is not meant for cartoons

even sd3 will have trouble with things like that

[teal osprey]

is there a vertical moving version of infinite zoom? i want to get the same effect of zoom in or out but in a vertical move

[jaunty sigil]

Can i run automatic 1111 on 4gb ram Intel hd graphics without nvidia gpu? Locally?

[shell tendon]

that sounsd like hell

integrated gpu = bad enough as it is, 4gb, i don't think you can run any model at its native resolution, which means shitty results

[tepid jolt]

thank you for information 😄

[nova zodiac]

Have you tried Fustercluck? If cartoons are your jam, Fustercluck is the model for you 🙂

[nova zodiac]

You can, be prepwared to wait forever, as even a 4 step lightning image will take 3-10 minutes depending on resolution

[cedar quiver]

Does anyone know where is a good place to ask questions about SD WebUI Forge? I got a console output I'm not sure is specific to it.
But here it is just in case:

Hint: your device supports --cuda-malloc for potential speed improvements.
Hint: your device supports --cuda-stream for potential speed improvements.```

I've never seen those args nor can I seem to find much info Googling them
But FWIW I'm on a RTX 3060 12GB so I assume the Cuda stuff is relevant to that

[nova zodiac]

The info about those is on the front page of the forge github

In order, worth it, not worth it, not worth it 🙂

[cedar quiver]

Oh dang okay

I already turned off sysmem fallback so if the first one is related to that it's not relevant to me anyway

but let me go look

[shell tendon]

no prob. what are you using? a1111, forge, comfy, something else?

for all i know you've been around the block with all the software and have more experience than me, but just in case not, forge and comfy are way the F faster than a1111

[cedar quiver]

Pin-shared-memory says it only works paired with cuda-stream
I think these aren't worth using for me rn anyway

[potent spire]

for most people unnecessary and tedious

and some of the competitors of SD offer a option to train on your own stuff ofc without having to deal with all the stuff like VAE etc.

[hallow sable]

yeah but why dont you make your own models?

[nova zodiac]

Who is offering model training that isnt based on a stable diffusion architecture?

[potent spire]

oh for the one specific its for the future, my bad. The other ones indeed have SD architecture i think like Alpaca

the one that doesnt is Adobe Firefly

[tepid jolt]

a1111 currently, but I'm planning to switch comfy

[potent spire]

also im not sure about OpenAI with their DALL-E for example

[shell tendon]

forge is very similar to a1111

but way faster

lmk what it/s you get with that

[tepid jolt]

Im familiar with nodes on substance designer and blender, I really like comfys UI design and workflow 😄

[jaunty sigil]

Damnn ig gotta a buy a new laptop

[tepid jolt]

👍

[hallow sable]

ive started using comfy

[plush solstice]

Unstability AI: Your API has been experiencing frequent outages for the past week. It only comes back online for around 30 minutes before going back down again. Please fix your servers.

[wispy oasis]

stability.ai billing will not accept my credit card

[sudden ruin]

Please contact their support

[wispy oasis]

i have done so

[brittle maple]

The neural network does not understand how to make high-quality faces of people at a distance

if not very close-deformed

[lunar crystal]

Hey friends! Just started my path on using stable diffusion and im running into some roadblocks. Does anyone have some time to help me out please?

[pearl ocean]

A.I lives matter!

[wispy oasis]

welcome and dont ask to ask just ask

we will always help u if we can

[brittle maple]

we give you right roadmap

[lone hawk]

Stability what are you doing!?? Release SD3!!! New competition -> ideogram.ai

[fervent thunder]

How long do you think it will take Automatic1111 to be updated with SD3?

[winter pike]

if its not open source then theres no competition at all

[pale latch]

Juggernaut is mostly photorealism in my experience. Hard to prompt it to do anything else.
Dreamshaper older models used to be a semi-real 3d looking Dreammy looking style, but now they've gone full photorealism with their new funding and new turbo tech too... so meh. It's limited in flexibility by being a turbo variant and we've already got a highly capable photoreal model.

i prefer dreamshaper beta version that they released for XL and never updated. It leans into photorealism less and does that original artsy style i know dream weaver for still

often i find the base model has better prompt comprehension too. it's such a huge latent space though that the destruction refined models drop upon it is often unnoticed

might try playing with playground 2.5 today. it looks like a particularly sexulent base refinement that doesn't destroy a lot of the capabilities. Belly buttons on shirts is one of the first indicators i notice of over fitting nudes in the dataset

[lone hawk]

Just tested ideogram, it's model 1.0 is amazing at following prompt, as good as Dall-E 3! But it's quality suck... Lol

but still, i am impressed))

[pine fiber]

agree

probably same architecture (dit) as sd3 but perhaps smaller model size / less compute / worse ddataset

its also closed source censored so you have to rely on them for implementing thinks like controlnet and blah blah blah

[lone hawk]

let's use the ultimate test: female jedi with purple lightsaber cuts pizza 😄

https://ideogram.ai/api/images/direct/1fRLXSVYQuC_QCDhOG3GqQ.png

yep, it's good! Wow xD

Holy... this thing is a beast at prompt following, omg! It just needs a better dataset to be less creepy...

[winter pike]

that one is very cool

[pale latch]

https://ideogram.ai/pricing meh

i'm never impressed by proprietary and closed models. they'll be a relic of the past and forgotten

[potent spire]

definitelly not

especially considering some important factors

[pale latch]

They've got good copy on their TOS though. This should be boilerplate, but instead most SAAS operations use the "we claim ownership and give you SOME rights back" model instead.

User Input and User Output. As part of your use of the Services, you may be able to input, post, upload and submit information (“User Input”) to the Services, and you may direct the Services to generate and output new content based on your User Input (“User Output”). **We do not claim any ownership rights in your User Input or User Output, and we do not restrict your ability to use User Output for your own purposes (including for commercial purposes). As between us and you, to the extent we acquire any rights in any User Output, we hereby assign to you all right, title and interest in and to such User Output. **However, you acknowledge that due to the nature of generative AI tools, your User Output may not be unique and users of the Services may create their own User Output that is similar or the same as your User Output because the same or similar User Input was provided. Your User Input, User Output, and any other information, materials, or content you post, upload, submit, or make available through the Services, such as your Account handle, Account picture, and comments are collectively referred to herein as “Your Content.” You are responsible for Your Content, including taking all steps necessary to ensure that it does not violate any laws or rights of third parties or these Terms.

[potent spire]

there are several reasons why you are wrong tho on that claim

[pale latch]

If i'm wrong i love to know, but somehow i don't think this is about to become an honest conversation with that opener

Lets just lay our biases out on the table before we begin

[potent spire]

sure, i dont mean to insult or something

😄

sorry if i sounded like that

[pale latch]

No, you're just being a typical argument baiter

[potent spire]

damn i hear that one for the first time :X

argument baiter

[pale latch]

case in point, you still haven't given your point

[potent spire]

well first of all the way the trend goes open source is getting less and less accessible...at least when it comes to local usage of generative AI especially when it goes more expensive like text to 3D

closed models end up usually being part of some service

but you dont worry about your specs that much if at all

on top of that comes typically but not always the simplicity of usage

[pale latch]

that's not been the trend at all. projecting from 5 years ago.

like i said, i wasn't expecting an honest conversation

[potent spire]

well at this point in some cases 40+ gb of VRAM is required for some things

if we stick to AI imagery, Midjourney, DALL-E, Firefly dont die out either and some of those are more popular than SD in the first place

[pale latch]

SD has a larger userbase than all of those combined. Since anyone can install it and there are many more services around it

[potent spire]

yet it is limiting the accessibility due to specs demand

[pale latch]

It's like how linux servers are dominant and not IBM or Microsoft

[potent spire]

so a lot of people use services like MJ

ease of use, no playing around with complicated installs and customizations etc.

[pale latch]

naw. wait 2 years. hardware will be cheaper

long term, accessibility is fine

[potent spire]

we will see but even that doesnt kill off the closed models

[plucky forge]

Hi,

I am looking for a way to make a professional caricature from my photo, I would like to use AI for this. I see on your channel how you use Stable Diffusion to create graphics, Do you know the tools to make a professional caricature from your photo? caricatures in various styles, www.etsy.com/listing/1557483961/midjourney-prompt-for-funny-and www.fiverr.com/deokeo/draw-political-caricature-and-cartoon Can you tell me how to achieve such effects in Fookus and Stable Diffusion? links to install the necessary tools and tutorials showing such effects from your photo are welcome,

[pale latch]

oh okay. if it comes down to ignoring the past and waitng on the future leaning towards your unlikely prediction, then yeah. i guess we'll see. Moot discussion huh

[potent spire]

well you are purely speculating my friend

Midjourney is currently the most used one even tho its closed model and costs, why?

[hallow sable]

its easy to use thats why lmao

[pale latch]

i got a whole century of computing and technology advancement backing my "speculations" oddly.

Why even discuss the future at all if that'll be your great "nah uh"

like i said, i was never expecting an honest conversation

[potent spire]

you have nothing backing your speculational claims

[pale latch]

no. no its really not. that's just mj marketing catching you. You can't count how many people install SD

[hallow sable]

any idiot can go on their phone, create a discord and say make me this or that .

[potent spire]

or its your FOSS ideology thats catching you? 😄

[potent spire]

this as well

[pale latch]

Not really. Real world facts support my biases. Do yourself a favor and look up how many linux servers vs closed servers there are

[potent spire]

no, they arent

[pale latch]

you're cliche. i'm bored. later

[potent spire]

your claim that closed models are basically about to die is a speculation that has no backed evidence or anything in that regard

[pale latch]

cough strawman. have a great day

[hallow sable]

too be honest I dont know why 3dsmax and maya exist when blender, modo, cinema 4d exist

[potent spire]

man your discussion style is weird lol

[potent spire]

because they are industry standard for a reason

have established themselves over the decades

got strong add ons as well

and companies have built up on top of them

[pale latch]

very typical

[hallow sable]

they are industry standards because they had the most marketing. but more and more studios are switching. its also the reason why you get pig slop like starfailed because of this "industry standard" where they only hire people on degrees and not talent.

[potent spire]

or they are simply the best lol

[pearl ocean]

Imagine running SD in a SD Card

[potent spire]

nobody beats Maya in animation and rigging and 3ds Max is unmatched in modelling

[hallow sable]

they are not the best. I switched from 3ds max 2014 to blender and never looked back

[potent spire]

Blender has a long way to go

but at least its free

which is very good

how do we get to that topic actually? 😄

[hallow sable]

bohemia interactive, embark studio, and a few others have fully switched over to blender... blender and Houdini is becoming the standard

[pale latch]

situations like pantone and draconian regulations lead to this too. Oracle database engineers are a great relic example of proprietary systems being the standard failing.

Adobe is an outlier most def. Alias perhaps. Their time has come to an end though and they are hanging on only through marketing. Their fate was cast when the biggest studios like pixar and weta digital started building their own tools on top of open research instead

[potent spire]

Autodesk dominates several industries, Blender getting used here and there doesnt change that

[pale latch]

oh yes autodesk ated alias

eventually they'll eat themselves. it's the standard business model in these situations. unlimited growth can't sustain

[potent spire]

and yeah people like to bring up stuff like Blender being used in Spider-Man animated series only to ignore that it was used briefly because of grease pencil, in reality Maya was the main animation software

[hallow sable]

autodesk dominates the majority of cad / manufacturing industry. and no there are many films that have been made with blender.

[potent spire]

Adobe lives better than ever lol

[pale latch]

i'd wager that their main tools were proprietary situations tooled for their studio, and that the editors are just front ends for their artists

[potent spire]

yes, and film and game industry as well

and i know Blender was used to make some movies

[hallow sable]

no only pig slope triple AAA and pedowood

[pale latch]

adobe doesn't have nearly anywhere close to the same dominance it did in the industry as 20 years ago.

[potent spire]

with the revenue they have they easily dominate still

[potent spire]

sure buddy 😄

[hallow sable]

there are many talentless hacks that skate by in game studios from coders to artist

[potent spire]

this includes amateur BLender users

[true ermine]

Why not actually? I believe Maya is just like any other DCC

Also hi

[potent spire]

most of them never had any networking with the industries in the first place

[potent spire]

its unmatched in animation and rigging, ofc will people including me use it

[pale latch]

the reason maya is used is because its the same old business model. Autodesk paying for schools to use maya in their courses. Institutions that teach exclusively on opensource tools are blacklisted in the wider industry more or less. It's a huge social engineering situation.

[potent spire]

and if someone comes up with the 3.000 dollar price tag this person proves that he is not informed well

[true ermine]

I mean there's no reason to question its existence, or other software for that matter

[pale latch]

Artists who get a little seasoned with experience soon realize that they don't need to license expensive front ends

[potent spire]

the only people that do that are open source fanatics, thats it

and most of those people have zero touch with the industries

its a FOSS amateur/hobbyist circlejerk most of the time

i dont even know how we came to that topic but yeah

[pale latch]

See. there's just no sincerity to this guy. no honest conversation skills. I spotted it immediately. Very cliche

"ackshually"

[true ermine]

I mainly use Blender but Maya and 3DS Max are surely great tools in their own right

[pale latch]

sigh

Fanatics are people who will martyr themselves. Hyperboles are just dishonest conversation.

[potent spire]

which is a mature way to handle it. Some people actually use 2 of those 3 or even all 3 in their own workflow

[pale latch]

weird that you bring gay imagery into this conversation as a slide against your opponent. Super weird dude.

nothing wrong with gay

[potent spire]

i dont use Blender (anymore) but 3ds Max, Maya, ZBrush and some others but yet i donate to Blender Foundation

[pale latch]

plus, why even go there? whats on your mind?

[hallow sable]

ive seen amateurs use 3ds max and maya too... you can get a student license for 3 years. I know many artist who freelance and in studios that use blender everyday...

[potent spire]

there are indie licenses as well for those tho which i took advantage off

otherwise id had to pay much more

in that case unless certain circumstances i likely would stick to Blender i believe

its UI got much better than back in 2016-2017

[pale latch]

by default. there were always mods to make it behave like maya even back when i was in college in 2003

[potent spire]

that UI got much better with the arrival of 2.80

[trail lion]

That's what makes it great

[potent spire]

hardly disagree there, well technically it would be the community, but fanatical cultists are not something that makes it great and i dont blame Blender devs for that or the rest of the normal part of the community

[hallow sable]

imagine paying for garbage and supporting it

[potent spire]

because those kind of people only bring toxic behaviour to the table and a mass of false informations and more trouble

[hallow sable]

how many times have adope and autoschiet changed their licensing?

[rich kestrel]

THIS

nothing is more pathetic than someone who pays for adobe

except maybe college students/unis who are forced to do it

[hallow sable]

and then saying others are cult like...

[potent spire]

well i could listen to people with experience in the industry and professionals, or to some blind cultists with zero touch with those

[rich kestrel]

adobe is NOT your friend

[potent spire]

thats not the point anyway

[hallow sable]

oh I forgot im not a professional... i make zero money

[potent spire]

i pay them for their tools and service

and use their platforms also for networking as well

and more than that actually

[hallow sable]

so are the many other people that I work with and work for

[potent spire]

apparently, yeah

[hallow sable]

ive also worked as electrician out of high school and ive met many idiots who said I only buy Kleins, Milwaukee, fluke and thats why I'm a professional...

[potent spire]

but it doesnt change your situation here on what you asked and claimed about 3D software

[hallow sable]

I didnt claim anything ive stated the truth. which also exist in other fields like AI art

[potent spire]

you didnt state the truth, you stated lies based on your disinformation

it was foreshadowing from your first sentence already where you asked why people still use 3ds Max and Maya when there are Blender and C4D

[hallow sable]

my disinfo that studios and professionals use X product instead of Y ? while you say that only Y exist

[trim nymph]

discussions that are getting onto an rhetoric clash are a waste of time

[potent spire]

i actually tried to get more on the line but just read above

[outer crescent]

https://www.theguardian.com/uk-news/2024/feb/27/glasgow-willy-wonka-experience-slammed-as-farce-as-tickets-refunded ok this is funny. the promo material is clearly AI generated.

[potent spire]

Just read about that like 30 mins ago

[outer crescent]

aah 🙂

[potent spire]

Kinda a scam lol

I mean the AI generated promo image is the problem...but only because its literally a false promise

[outer crescent]

the reality looks like one of those dog show courses

[potent spire]

What the hell were they thinking

[outer crescent]

$$$ is what

😄

[potent spire]

I mean did they expect people to ignore this bs?

"Yeah they lied to us, this looks nothing like they promised but whatever"

[outer crescent]

its scotland, brits tend to put up with awful things and moan quietly about them.

[potent spire]

[outer crescent]

its a weird mix of stocism and defeatism

[potent spire]

Such a weird case, really

[outer crescent]

LMAO found the website. https://willyschocolateexperience.com/index.html its like a spot the deliberate mistake in every one of these ai images

wonder if i could peel the prompts from them, site looks lazy

aww, no cant get the prompt info.

ill never know what prompt made "imagnation lab"

[potent spire]

Lmao

[outer crescent]

https://willyschocolateexperience.com/index_files/twilighttunnel.png these are amazing

[potent spire]

Are those some scamming AI bros? Lol

Their Twitter is seemingly gone, FB a lot of deleted links and full of people mocking them

[rich kestrel]

fuck fuck fuck fuuuuuuuuuuuuuuuuuuuuuuuuuuck

I updated kohya and now my json doesnt work and its slow as shit

now to figure out which commit I had

goddamit I hate devs

[outer crescent]

more likely they paid like their nephew whos "good at computers" to do it

[potent spire]

Damn lol

And those that paid for it are dumb as well tbh

I mean they dont even provide any images of whats there?

And you still decide to buy tickets based on some AI images

Or did i miss something

[outer crescent]

https://houseofilluminati.com/ this is a rabbithole. these are the "experience creators" its a whole big AI image BS parade

[potent spire]

Yeah, what i meant

[outer crescent]

this is just bad bad AI all the way down. if anyone's an AI content creator, this stuff is a gem!

[potent spire]

Why the hell do people buy tickets from such a place

[outer crescent]

yeah they send out this AI generated promo envisioning, and you get there and its awful

[potent spire]

Well now they have angry parents AND anti AI people amongst all on their neck

[outer crescent]

looked through the source. "/DALL·E-2024-01-30-09.50.54-Imagine-a-scene-where-fantasy-and-reality-merge-seamlessly.-In-the-foreground-a-grand-interactive-gala-is-taking-place-filled-with-elegant-guests-i-1024x585.png" Its ok people, we're safe. It wasn't Stable Diffusion. It was DALLE

[potent spire]

XD

[outer crescent]

this thing would honestly make a good little AI based comedy video, but i dont think most AI content creators want to show AI taking the L 😄

[green raven]

Holy crap. Ethereum classic ETC is rocketing.

[outer crescent]

im not buying your pretend money stop pumping crap here 😄

[fervent thunder]

its all rocketing but don't buy in now, it's over with

bitcoin goes up, so do all other ALTS etc

[karmic cedar]

We’re going to need a new kind of olympics for terrible, redundant AI.

[potent spire]

My work colleague is insane lol

[fervent thunder]

welcome to life

my whole job sucks

everyone in that mf except for the people in other towns are just trash bro

[potent spire]

He tells me since i wanted to 3Dfy and replicate my home village and prepare to implement it in Unreal Engine i should pay for a satellite service to gather data of my village lol

[fervent thunder]

thats some gen z type shit

[potent spire]

He mentions Starlink

Since when does Starlink do that

They provide internet

[fervent thunder]

but u also have to realize like

[potent spire]

Not special satellite data

[fervent thunder]

half of the world is autistic these days

cant expect everyone to be on point all the time

[potent spire]

Lol

Yeah

I mean if he said i should pay for a drone service to do that and gather aerial data and so on

But satellite? Lol

And even the former one is not that easy

[outer crescent]

we should have a challenge to make photorealistic versions of what it SHOULD have looked like. 😄

[karmic cedar]

i like it

[outer crescent]

man i just spent 15 minutes going davinci code on this guy, finding all his other ventures and stuff, and turns out real journalists have been doing so as well. he's just a massive thread you have to pull. i wont name names or anything, but same guy behind this wonka experience has been churning out all kinds of AI get rich quick schemes including AI books on amazon.

anyway, that was a fun little diversion before bed. nn

[trim gazelle]

[pearl ocean]

Me when I generate realistic cat pictures

[light lily]

this server lacks some furry art lul

[hallow sable]

get mental help

also please keep away from children

[trim magnet]

as long as is sfw u can post it here

[light lily]

L M F A O
u are a little misinformed my guy

[trim magnet]

everything is a fetish,even wearing boots

[light lily]

ok

[mortal delta]

what did i just walk in on....

[trim magnet]

dont worry these gen z's wont bother u anymore,wont be long before u hit the ground

[mortal delta]

ok

[shell tendon]

https://ideogram.ai/t/explore

[mortal delta]

i just Frankenstein a pc some time ago.

[shell tendon]

prompt: a hybrid of a shark and a clown holding up a sign that says "SD3 was the king of prompts for less than a week". The shark clown should be standing in army boots on top of a cow which is doing a handstand on top of an alligator. This is happening on the moon.

#🏞｜general-with-images message

and there's the result

[trim magnet]

well u def a hack thats for sure,a top one

[hallow sable]

why dont they make the anime with a real story like a film... most of it is over the top junk or things related to high school...

[full lichen]

Hi