💬｜general-chat

[low moon]

so whatever

[charred mesa]

fr fr

[low moon]

but still

[charred mesa]

SO THIS IS WHY

GIVE US SD3

[low moon]

18 years motherfuckers

i put 18 years into one project

whether u think its stupid or not

you gotta apprceciate that

thats it slonger than some of you bozos ave lived

how old yare you guys

on average

?

[narrow kernel]

you can practice something wrong 10,000 times and become perfect at doing it wrong

[low moon]

like 17?

im 43

loooooool

[narrow kernel]

im 37

[low moon]

so what then h,mmmm?

you know where you will be at 43?:

[narrow kernel]

dead

[low moon]

the way things are going?

yes

well

hopegully not dead

lol

see?

this si sad?

like wtf

people have been reduced to such self loathing animals

apes

by schools

[narrow kernel]

no, i mean statistically, i will likely be dead. high T all my life, was an athlete from 10-30ish and there's something wrong with my heart now

[low moon]

I didnt do it worng actually I am very good at it

[potent spire]

guys, enjoy the creative hobbies you have

[low moon]

Thing is no

[potent spire]

dont mess too much with politics and sh*t

[low moon]

please no you ar enot animals

[potent spire]

it just drains down

[willow crescent]

Ironically it wouldn"t even, for example in Germany the cost of the adminstrative side of social departments could finance any jobless person alone from what I know. And this is merely one example.

[low moon]

I tried all my life to show that to somehow wake people up

and commerce never rewarded me I get that who cares

what do these suiccesful people get anyway a decade of joy if that while the tax man houdns them

the point is there is more to life than this body

I wish Emad sees this

and that moron leading öpen"AI

if they think theyll be rich and powerfula and stuff by puttign things behind paywall they ar emistaking

[narrow kernel]

idk, all i'm hearing is a bunch of envy, jealousy and narcissistic entitlement. for every millionaire you see, there are thousands that didn't get there. for every professional basketball player you see, there are hundreds of thousands of kids that didn't make the cut. each one thought they were the next lebron

[potent spire]

well they are already rich

[low moon]

theyll be miserable fools

and then thyell die

[low moon]

So what ?>

all thsoe thousands how are ehitr dreams and efforts worth less?

[narrow kernel]

if you want to be a top percentile at something, you have to be damn good at it or extremely lucky

[low moon]

Where is the jealousy?

[potent spire]

the funny part is that anti-corporationists and others are at the end of the day still ending up crawling to those companies either openly or with backdoor

[narrow kernel]

no, this falls back to what i said about the dunning kruger effect

[low moon]

I know many people who wokred hardedr and siffered way more and achieved nothing and are not in the papers

how do you explain that

also

the arab oil shjeicks

[narrow kernel]

i'm sorry, but when it comes to making it big, we aren't all cut from the same cloth

[low moon]

who are born into it please explain their meirt

go ahead

explain to me whay some randoim dude in the middle east is worth 10000 more than you

please enlighten me

go ahead

[narrow kernel]

the luck factor

that i mentioned

[low moon]

there is no luck

[narrow kernel]

RNG, nat20

[low moon]

nothign is random

there is 0 luck

[narrow kernel]

they were born there and not here

[lean timber]

Yo, you don't think the people who lived in the beginning of the 20th century, working their asses off in dirty factories without regulations and unions, or thrown into war... you don't think they'd be pretty damn envious of the life you've lived? Be happy with what you've got.

[low moon]

if you think somethign that determines the quality of life of someone is up to lucl you are a fool

[narrow kernel]

so again, life isn't fair and doesn't care about you, me, or anyone. stop comparing yourself to everyone else and set realistic goals and expectations in life

[low moon]

there is no way that middle eastern dude gets billins to play with while you ar ein some trailor park in the use or whever

there is no way that is luck

its not possible

and you know it

deep down

you know it

[tacit tiger]

guys what r u chatting about? like how did u get from stable diffusion to this?

[narrow kernel]

dudes rambling about life

[low moon]

Everything is connected my friend

[potent spire]

compare life in EU or USA in general to life in India

[low moon]

SD3 is very much so connected to a lot of things

🙂

it ruffles a lot fo feathers

Hey EMAD!

listen!

I hope he reads these

theres a lot going on right now

everyoine is in pain

are you in pain?

yes you are

so am i

we all are

things are not right

Papa musk is a madman

but his heart is in the right place

[narrow kernel]

Porn is illegal in India, which is why there are so many SD users that rely on being able to finetune for porn.

[low moon]

haha porn

[lean timber]

Yeah, and compare life in India to life on the frontlines in Ukraine, or continue to compare that to WW1 trenches. There's always FAR worse places to be than we sitting by computers can imagine. We have it pretty damn good.

[low moon]

Yes indeed count your blessings but it can be better for all of us eincluding Uklraine and Israel and everywhere else where WW# (allegedly not yet WW3) is playing out

Israel yes

The center of the world

[potent spire]

yup

[low moon]

the chosen land by God

[potent spire]

[low moon]

Israel is awesome

[potent spire]

Israel is a nice country

lmao this bro

[teal pagoda]

SD3 is not as good if it's not fine-tuned by the community trainers

[low moon]

No one there does that but Hamas

using doctors as human shields

Jews are the chosen people fo God

[potent spire]

is this a nazi over here or what lol

[low moon]

all nations will be vblessed and cursed thrpough Jews

Jews have a covenant with God

[potent spire]

whats the issue with satanism

xD

ok i was right, this is a nazi

lol

[low moon]

Jews were kicked ut of so many places because everyone is jealouse of Jews

because they have a covenant with god

also no one wnats to accept theyoke og heaven hecne they hate jews

because Jews represent God

[potent spire]

my man your system failed finally 1945

[low moon]

and god has certain comandemnts that everyone must abide buy

and most peopel do not wan tto abide by gods commandmnents

this is the simple reaosn why everyone hates hjews

and even now they side with killer animal terrorist who behead kids and gad babies

[potent spire]

you are supporting the left-wing of the NSDAP

[low moon]

its obvious

[potent spire]

obviously

[low moon]

Peaople dont eant God

and Jews represent God

hence why most rebel against God/Jews

no God does not condone slaves whatsoever

[willow crescent]

Even if, what elites want is a whole nother topic, I was referring to stuff that could be done in relation to what rx808 was saying in general about capitalism and it being how the world works

[narrow kernel]

Oh ffs, this convo is going to far now

[low moon]

No one says anyone is the slave of jews

[narrow kernel]

Stop talking about religion and politics

[low moon]

it just says just because Jews are the chosen one s( like a special commando in an army unit) they will bring the rectification of creation and so the nations will benefit from than and they will follow jews in the messianic age to benefit from the grace of God

its not abotu slavery

[potent spire]

from Stable Diffusion to Unstable Confusion

[low moon]

lol

[narrow kernel]

Well house Slytherin is known for being manipulative

[low moon]

those who have eyes will see and who have ears will hear

[narrow kernel]

But house gryphondor is know for having rule breakers

[low moon]

Well I am sorry

but if you think that way then you are going to be dealt with

[narrow kernel]

And the sith only operate in extremes

[teal pagoda]

maybe a hacker who can access some rockets from his PC?

[low moon]

Believe me when God decides your time is up the you are gone

[narrow kernel]

But once we have all 7 dragonballz we can summon Odin

[low moon]

no one will want to be 100 miles near you when judgement is execited upon you

[narrow kernel]

And the Loki can make a world dinner for everyone to dine on

[low moon]

Remember Sodoma my friend

[narrow kernel]

While poissidon brings fish to the feast

[low moon]

To this day nothing grows where Sodom was

dont play with God

[narrow kernel]

Do you see how silly all this sounds?

[low moon]

yeha man

yu have your little dinoisaurs and ape men and all the othe rlittle systems you ahve buoilt up as protection against god

well good luck in the days to come

[narrow kernel]

4000+ gods and religions throughout recorded history and each one swore up and down theirs was right

[low moon]

bla bla bla

you know 0 about the torah

like 0

u have no clue

but yet you badmouth

[craggy crag]

Anyone knows how to use sd webui as an api guys? For some reason the the /docs there is no post route for img2img or any other generation

[olive creek]

Yo, you shoukd take this conversation private. Some of us are really here waiting to get some assistance related to the app

[teal pagoda]

It's not like these rats are playing with weather weapons. Check the "HARP" project if you don't believe me. Even they recognized that they can "control" the weather.

[narrow kernel]

You know nothing about Harry Potter or Lord of the rings, yet you're indirectly saying they aren't right

[low moon]

Insults over insults....

Well go ahead

keep insulting

but when you hear the sounds and you realize deep in your bones the truth

[narrow kernel]

You're almost as atheistic as I am, you denounce every other god/religion out of the 4000+ known, I just denounce one more than you

[low moon]

dont say you did not have the opportunity

See thats the thing

out of those 4000

one must be right

so dotn gamble

because you know which one si right

you know

\it

all of you

muslims and christians

you know the right one

so pleaZe

[teal pagoda]

I studied/researched this for a very long time now. All the conclusions redirect to this simple fact: you as a good human (worker) won't be able in your entire life to be a millionaire, let alone to just enjoy some luxurious moments. All the millionaires/billionaires did something illegal/immoral at least one time in their career to be where they are now and they won't say that secret to anyone.

These are facts

[low moon]

No

the fact is God is in charge

and he determines to the Penny

how much you earn in a year

and if you think otherwise

you are a fool

You are an absolute moron to think Bezos or Zuckerberg or the apple man knew how to make it

a total absolute bafoon

[teal pagoda]

unfortunately true

they are really sick

[low moon]

God determine sot the last penny how much Bezos will make in ayear do you udnerstand what that means?

[teal pagoda]

don't even search for what the Dubai millionaires rats are asking from Instagram models

do not even search :)))))

[low moon]

Yes

Poop fetish

[teal pagoda]

I know this feeling

true :))

hope that something will happen as soon as possible

[craggy crag]

Is stable diffusion automatic 1111 web ui still able to be used as an api? For some reason when I’m hitting the docs route I can’t see the option for inpainting or im2img

@everyon

e

[teal pagoda]

nah, or sign the papers with the d3vil

like 90% of the rappers

[craggy crag]

Is stable diffusion automatic 1111 web ui still able to be used as an api? For some reason when I’m hitting the docs route I can’t see the option for inpainting or im2img

@everybody

[teal pagoda]

exactly

I can agree that this is 100% legit

from all my discoveries

And I'll give just one example

really quick

0nlyF@ns - "girlfriends" material wannabes who just pretend to talk to you (so you feel important and "special" for someone) in exchange for money. In reality, you are talking with some indian chatters who just try to sell you the model's nud3s and that woman never cared about you (not even if you donated 10K$ to her). So this business is pretty simple: the women/girls know that there are so many SIMPs out there nowadays who are too ashamed to get a "real" girlfriend irl, so they planned to play with their emotions to get the money from their pockets. It's a simple social engineering game which has so much success, because 70% of "men" are legit SIMPs rats without courage to talk to a girl irl.

good luck, i'm out

[low moon]

sad;y yes

also the rich people i know are all dead because of stress and heart attacks

they did not get thir wealth right

its sad but true

🤷‍♀️

ome is partalyzed from the waist doen one is heart attack and others are dead

so idk how they got their wealth but they are not enjoying it

it makes you think

the rich guy i know had divorce and always with laywaers meetings

so think about thsi please

before you buy the idea of money will fix all your problems slogan

he worked on bunkers for the elite

he wa sone of the richest people in the country

my mother could have married him butshe saw the truth

pleas

please

do not fall for the propaganda

money will not make you happy

and its not about jews

asshole

antisemite idiot

its the same everywhere

fukcing moron

you know most jews ar enot rich right

you fucking idiot

0.01% of humanity is jews

and 30% of them are nobel prise winners

and you still hate jews?

you fucking disrespectful swine

you dontdeserve to be on the same planet as jews

animal

youd be in the dark ages without jews and this is what we get

thousands of years of killings!

no worries tho

youll see soon what you did

Egypt will pail compared to whats coming

[sage reef]

now tell us how you really feel

[merry helm]

look at his youtube channel plz sub, i just randomly found it

https://www.youtube.com/@PlantForwardMDlife

[astral lotus]

How does one uses the "GetNewSession" api endpoint?

It askes me for a session Id, which seems counter productive since that it what I am trying to get😅

[merry helm]

plz subscribe to that channel, it gave me a lot of good advice

[pine fiber]

a month since sd3 api, id give it another 3 weeks

[severe pendant]

can somone help me with a quick and simple image request, I need to put a shirt over a stock image to use as a model

[trail lion]

and the award for most triggered user today goes to...

[olive creek]

Hello, no matter what I have tried I can't get open webui to interact with a1111. I keep on getting something went wrong string indices must be integers not str. Has anyone seen this before? My openwebui instance runs on Ubuntu with python 3.11 and my a1111 I stance runs on win11 with python 3.10. I can access a1111 from ooenwebui host just fine. Also, running --api --share --listen flags on a1111 side

Somewhere I read something that could be related however I'm unsure.. regarding mistmatch of expected format from a1111 api's output (JSON)?

[merry helm]

https://www.youtube.com/@PlantForwardMDlife

SUBSCRIBE AND I WILL HELP

[sage reef]

help with what?

[olive creek]

Checking the logs from ooenwebui container I see GET /images/API/v1/models HTTP/1.1" 400 Bad Request. So to me is as if a1111 is returning not expected output

[untold latch]

Would you say an RTX 4090 is a good card for training pony sdxl? Or I need something even more powerful? like what if so?

I mostly worried about that 24gb Vram as I have heard its not enough for training sdxl

[fervent thunder]

dayum the requirements are gettin' heavy

[trail lion]

4090 is the standard right now for training local, anything better than that is a comercial card

[fervent thunder]

damn i just have a 4060

[untold latch]

Ikr T-T thats why I wonder if even a 4090 is enough lol

[olive sparrow]

hi

[trail lion]

but you can do almost anything with 24G vram, at least when talking about training a lora or finetuning a checkpoint

and tbh if you actually had a use case for something more than 24g vram, just rent it, that's what I do...get this notion of needing to have the best local hardware out of your head

[untold latch]

I think for training lora you need just like 16gb vram, I guess waiting until 2026 to see what the next gen cards will get is tempting too

[trail lion]

I have a 16g card now, I rent when I train, I use my card for local image creation, it's totally fine

[untold latch]

Yeah, heard 16gb is fine when its about training lora, and its more accessible

[trail lion]

I could use my card to train also, but it would take hours, and I wouldnt have access to it during that time

so I guess if you're deciding to buy a 4090, decide if you need something that will give you the ability to train in a small footprint of time (like maybe 1hr or less) and whether you need the 24g vram, but financially you'll pay the price...it's like the price of a used car

[cursive magnet]

If you're into AI training, you may consider nvidia tesla

Maybe can find some used ones for good price

[trail lion]

that wouldnt boost the vram though right? just the processing, and unknown at that

[cursive magnet]

tbh, im dont know about them much, but i seen used A40/L40 in russia with 48gb vram

[untold latch]

Uh weird, refurbished Tesla with 32GB are pretty cheap than what I would imagine, might actually consider

[trail lion]

dont consider it unless you're sure what it will do

[untold latch]

Yeah refurbished stuff is quite a risk ikr

[trail lion]

not just the refurbished aspect ,but the fact that you can get some of these cards at a fraction of the price, do they actually help with what is needed for stable diffusion? I'm skeptical

[untold latch]

Haven't look into details, but yeah its something to have in mind, thanks

[trail lion]

my plan is just to wait for 3090's to drop in price

[untold latch]

Haha like so many more people lol (like me)

[trail lion]

which they havent, so there's tons of demand still

[cursive magnet]

These cards are made for AI purposes, so they surely do

[trail lion]

meaning what? I can buy a card with what specs to do what? the devil is in the details

if 3090's are still > 1k, then I'm out

stupid to pay that kind of money for a card

[untold latch]

I heard an scandal time ago when nvidia started selling cards cheaper and scalpers just bought them and resold these for higher prices, I don't think prices will drop all that much because of these sort of people

It was something about bots in the web

[trail lion]

some people just need to have the best thing, and companies like apple and nvidia take advantage of that

[cursive magnet]

Idk, i never had such cards so i can't compare their performance to RTX. But i know some people buying used xeon processors, tesla cards from aliexpress and they're happy with LLMs, trainings etc

[pine fiber]

sd3 gonna be dead on release

Check this new meta paper https://arxiv.org/abs/2405.09818

[fleet karma]

Hello people

[pine fiber]

no

the speed of the drive that it’s stored on will matter when comfyui loads it into memory

if you have an nvme or ssd put the models on there if you find it’s slow

yea

just make sure you aren’t maxing out your ram

[fleet karma]

Excuse me, sorry to intrude, but I'm trying to find where to look for someone that is knowledgeable in training AI's for art generation and if I may be guided in the right direction.

[hybrid thistle]

does anyone here use stable diffusion on comfy ui, im trying to find a server specifically for comfy ui

[compact swan]

Which model should I use for a 3060 Ti?

Just to generate images

not training

[peak orchid]

That depends on what sort of training you want to do. Nobody here seems to have the the knowledge for training with 1 million image datasets, but many thousands have the skills necessary to train with 10 to a couple thousand images. You also need to figure out if you want to finetune the main model or train a peripheral one like a Lora

[fleet karma]

Thank you for your response, I'm honestly still familiarizing myself with the concepts (is Lora the character based?) Someone capable of training with around 500 images might be more than enough. The main focus of the project would be of anime art, not sure if that may narrow down the search.

[peak orchid]

Someone still needs to train and release the weights such a model for it to be an SD replacement, and it needs to not require a datacenter gpu to run

[pine fiber]

It’s only 7 and 34b

I’m getting numbers slightly wrong

yeah you will have to train it but this is a step in the right direction

[peak orchid]

I haven't done a ton of lora training, but others here have. UD, the pony discord, and other Discords also have knowledgeable folks

[pine fiber]

they divide the image into input and output tokens. Which is different from diffusers, and it’ll run well

[fleet karma]

Oh so it would indeed be Lora training what I am looking for? And may I trouble you with guidance on how to find the discords you mention? I would very much appreciate it!

[peak orchid]

We also need to be able to test it, as research papers often cherry pick results and don't fully show the issues at play. Example: 4o's image text generation looks great but the rest of the image looks worse than dalle 3

[pine fiber]

yeah u talked about that the other day

those weren’t cherry picked tho imo

[peak orchid]

Go to Pony XL on Civitai to find their discord link, and search for "Unstable Diffusion" to find their discord link

They do a lot of sfw as well

They allow nsfw, but do other stuff as well

Same with pony

[fleet karma]

Thanks so much, I'll look into it!!

[peak orchid]

When i wrote neural-style-pt back in 2018, researchers for years afterwards published papers showing better results from their methods that ended up just being cherry picked, highly specific/situational, or traded quality for speed. So I'm a bit jaded after all these years in the AI art community

[pale latch]

yeah but they were comparing to you for years still

[finite cloak]

when you install swarm it by default installs comfy itself

if you really want to use your existing comfy install as the backend, in the installer select None on the backend choice, and then afterward go to Server -> Backends -> add a ComfyUI Self-Starting instance -> configure the path to the startscript and anything else you prefer -> save and go. Also go to Server Configuration and set the model paths to your comfy model dir

[fervent condor]

yo is there any llama 3 42b uncensored finetune ?

[languid plover]

can we genrate images now ?

[viral rune]

?

?

[raven field]

Not for free

[languid plover]

because of lack of donor or something else

??

[atomic pier]

Hi. I'd like to hear from you guys.
I made a virtual person similar to a real person as a prompt.
We used 0.45 weights as prompts and specific person loras.
I hope the person (called A) likes it and continues to use it.
I put it in seed and recycle it, but the similarity with the desired person A is inferior.
The seed results have a similar facial feel, but when I asked 15 people, the similarity was about 50%.

If I think the result generated by seed is the same face as A, I would like to make more than 10 pictures and use them to continue to generate A characters by learning with Lora.

Is it correct that the way to continue using virtual characters is the way I've described it, or if there are better ways to do it?

[craggy plinth]

[young surge]

Quick question. Whats the promt to use more than 1 LoRA?

[atomic pier]

You can connect and use the LoRA prompt.
Like lora:last:1lora:last_0510:1

[young surge]

So lora:pack1:1Lora:pack2:1 ?

[young surge]

If I'm understanding your answer correct. Coding is not my friend

[atomic pier]

Yes, that’s right.

[young surge]

And i assume adding more than 3 tends to break things?

Now to wait for people to wake up and get into Tech support xD

[sterile flicker]

Hey guys, I'm having trouble generating images when running stable diffusion for the first time. The error message says 'NoneType' object has no attribute 'lowvram'. Has anyone encountered this issue?

@dry escarp

[young surge]

Pardon the ping.
So when people have several resources listed in a post in Civitai, but no apparent lora:pack:1 in the promt window, what, and how they did done it?

Such as on (Hard to find a non-nsfw image as an example xD) https://civitai.com/images/12597091

(Repeated question in Tech)

[rugged mirage]

if you add the lora via e.g. a seperate Comfy node it doesn't add the <> synthax to the prompt

[feral halo]

Hey! I am new to this discord (and sd), what would be a good channel to ask some noob questions in regard to how to start building their own model?

[young surge]

So pack1:1, pack2:1, is sufficient... or?

[potent bane]

怎么用了

[neon inlet]

Hi guys, am I able to access / use SD3 on this server? If so how and where can I do that, thanks 🙂

[fading niche]

哈喽

[vale steeple]

用什么？

[lusty beacon]

SD 3 Weights when?

[loud solar]

if you ask often enough ...

[lusty beacon]

Not sure if they will ever come.

[vale steeple]

My ETA is the end of the month May or early of the month June

seriously just a guess

quite a lot of people getting pessismistic about this

[loud solar]

Only thing we can do is: wait ...

[lusty beacon]

and pray 🙏

[vale steeple]

#🆕｜sd3 message

[lusty beacon]

Does he mean drop like release or like fail

[vale steeple]

the original context is he congratulate Google Veo team and someone ask him will there be more usable in production, user-hardware-friendly model release

[lusty beacon]

ah okay

[opal hedge]

https://x.com/EMostaque/status/1790451196915831018?t=YJuHqJJ-YCivInuOrZ2_Lw&s=33

"Sd3 is due to drop now I don’t think folk will need that much more tbh with right pipelines"

Where sd3

[sage reef]

the pipelines are leaking

[lament tendon]

yes

i am the female

[trim magnet]

i am young asian lady looking for famous football star with at least 3 ballon d'ors so i can marry (pref portuguese) ,i can clean,cook and i can do laundry

[lament tendon]

I too am authentic female human

you mean you are a liar?!?! and I thought we were gonna get married... 😭

as an authentic 100% real woman I decline this request

you hurt my feelings

you lied to me

i wanna divorce

[frozen ridge]

ola

i need help

i found my lillith

[lament gulch]

I logged in VIA my Discord on Stable Diffusion. Now the loggin method isn't there anymore to do the API credits. repost

[strange epoch]

привет

хеллоу гайсы

[olive sparrow]

gm

[shrewd flare]

gm

[shrewd flare]

hi

[fossil hedge]

Good Afternoon, I'm just interested in stable diffusion. Please advice for VGA. Is Rtx 3060 12GB better than 3060Ti 8GB?

[rugged mirage]

more vram means you can go to higher resolutions/more frames/bigger models and is generally more important

the Ti will be faster but more limited if you want to do more complex workflows (also probably wont be enough for the bigger sd3)

[ionic oriole]

hello

[strange heath]

hello,everyone

[hasty arch]

hello

[hasty arch]

你好！

[hushed karma]

gm

[dawn warren]

how to use it

[sand tide]

hello good afternoon

[zenith forge]

how to download

[rotund creek]

ho boy

[astral lotus]

click download

[static cape]

And another week without news...

[crisp iron]

Is topaz the top of the line video upscaler at the moment or is there a better method out there?
I'm looking for a creative upscaler like magnific but then for video

[lavish lake]

topaz video is the easiest one, if you go full custom via comfyUI, there are multiple ways to upscale/interpolate and get great results.

[azure nebula]

SD looks like big debt

Report saying it's collapsing

[static cape]

Which is why they should just release all SD3 Models instead of still trying to appear like they can finance training it longer.

[radiant nacelle]

Hello

[astral goblet]

If you're building something to last, don't fudge the foundation. If that's shit, it'll all be shit.

They should keep training and refining it as long as they can while not facing diminishing returns, and then pass the baton

[covert sand]

are a lot of ppl still using 1.5 or whats popular these days

[shell tendon]

sdxl is the go to now

[narrow kernel]

I will work with even the base sdxl model before I'll touch a 1.5 model these days. I can't stand prompting for 1.5. Once on a blue moon, I'll use it for some multi stage workflow that always ends with sdxl, but pixart sigma took over that role immediately lol

[naive thorn]

so much new ai stuff came out and they cant release this yet

:_:

[covert sand]

is sdxl control net as good as 1.5?

[low moon]

Supir?

[loud solar]

Never tried supir with video ... could take hours ^^

[low moon]

yep :/

but its the best upscaler

I wish fixing hands would be as easy as making a really good "hands" lora...

[loud solar]

😄

[soft iron]

hi, SD3 will never be released

[teal pagoda]

Right, that's kinda not possible (maybe if you have tens of thousands of dollars and you don't have something better to do with them).

[teal pagoda]

:))))))))))

It seems that you bring the bad news every single day

[low moon]

If Emad said repeatedly that it will be released why everyday people say the contrary?

[soft iron]

because they lie 10 times in a row

[teal pagoda]

they are at the 4th lie, but yea, i hate the liars anyway

[teal pagoda]

yep, topaz is the best local upscaler for video, but you need a good GPU (3090+)

or you'll wait days to upscale a 5 mins video

[loud solar]

Topaz is pretty expensive ...

[teal pagoda]

yea, but you pay one time and have it for life

instead of paying monthly subscriptions for ever

[loud solar]

AFAIK only updates for 1 year ...

[teal pagoda]

did they introduce this now?

as far as I know you paid for like a lifetime key

[loud solar]

I think I've seen that a longer time ago ...

but maybe no more updates?

[low moon]

What happened to you buy a software once and it's yours forever?

[loud solar]

2 small for SDXL

[low moon]

use 1024 by 1024

then use

turbo

or lightning

well

SD15 has a different flavor but sdxl at this point is probbaly better

wildcardxXLTURBO_wildcardxXLTURBOV10

in my experience thats the best turbo model

by far

the only reason to use turbo is to get fester gens

they result is a bit blurrier than regular

but considering u do less than 10 steps its pretty decent

[sage reef]

haha this is like your daily message 🤣

[soft iron]

for the last month I wake up every day and sit waiting for SD3, looking at the wall

I don't do anything other than this in a day

[low moon]

That's impressive and sad at the same time.

[sage reef]

[low moon]

[pearl ocean]

[pearl ocean]

Impressively sad

[sage reef]

this is gen chat 🙂

did you have that problem before?

huh, only thing i know fam is that comfy has no such problems 🙂

understandable, but you have to seduce the noodles and nodes and make love to it, then it will work out 🙂

haha

i should be a writer eh?

but yea jokes aside, i have no idea what the hell is going on with that stuff man, what's your gpu?

oh 3060

no but i mean it should be very fine

the world of neural networks sometimes requires a lot of VRAM indeed 🙂

just think of it as a digital brain

which makes you just wonder how the hell does a pure biological brain work eh, complete magic 🙂

i mean no offense to my brain, even tho i can imagine a picture in my head, i sadly cant transfer it on some screen for me to look at... so yea... i guess digital brains are better in that sense...

maybe the future will have some bio tech where you can project what you currently see or imagine in your brain, along with vision of dreams, and thoughts, etc, but that is prob some very far future

[low moon]

Welcome to Noodletown.

Once you go noodle you never go back.

[sage reef]

ramen diffusion

you whipped a noodle eh?

[low moon]

I dont blame you

Noodletown is scary hood at first sight

but ocne you warm you way in it has many secret pleasures to give

[hardy sedge]

What is the best Method for realistic face Swaps?

[low moon]

reactor?

[hardy sedge]

And for Video?

Is there a better way then reactor?

[low moon]

i dont know much about video sadly...

[hardy sedge]

Reactor is good but it can be better.

Right

[foggy halo]

heyyy

guys

[low moon]

and galz!

😄

[foggy halo]

hiii ^^ oopsiee i forgor

how are you

[low moon]

pretty good!

you?

welcome to AI land

[graceful peak]

my webui wont open

[low moon]

We need more info than that

forge? A1111?

fooocus?

[rich bolt]

Not sure where to ask this: ComfyUI - How do I fix hands as a post processing step, or make certain that hands render correctly in the first place?

[graceful peak]

a1111

i posted my error thing in #🤝｜tech-support

[low moon]

welp when it happene dto me and it acted up i just did a new install git cloned it into a brand new folder...

[graceful peak]

ill try that

[graceful peak]

will it mess up any updates etc

[low moon]

not if its a new folder

it wont mess anythign else up then

[covert sand]

do i have to uninstall sd1.5 if i wanna dl and use sdxl

[knotty plover]

Hey

[graceful peak]

i think i figured it out

it was my lack of storage space

[low moon]

ouch yes that can happen

[low moon]

no

[covert sand]

do i have to update A1111 or something or is it as easy as downloading a new checkpoint

[trail lion]

you can have multiple checkpoints and versions, there was a time when you did need to update like when SDXL was just released, but the current version that you likely already have supports all the released versions

[covert sand]

i downloaded it right around when SDXL first came out and haven't updated since

i can't remember if it was right before or right after SDXL dropped

[trail lion]

if you have 1.6 or later you're good

[covert sand]

thank you

[primal current]

Open Source alternative to Leia Pix? (to create 3D animations from 2D images, other than SVD)

[graceful peak]

what does connection errored out mean

for a1111

[opal hedge]

wen sd3

[soft iron]

never

[opal hedge]

Say it ain't so

[low moon]

never mind these doomers. emad is the rightous source and he says it will release.

[charred mesa]

Alex (StableSwarm) and Comfy implemented them already and they are waiting for SD3 to finish training. I wonder why comfy is even waiting to merge if it apparently won't come out.

[opal hedge]

I'm pretty sure it will, though I do wish they gave up updates on how long they expect the training to take

Any sort of communication would be great

[charred mesa]

Even the CTO, whom has given a way longer ETA was wrong with it

4-6 weeks instead of "hihi 😊 soon"

I just hope its by the end of may with controlnet models and training code

[opal hedge]

Yep. Personally, I don't mind delays but I wish they would be open with them so we're not waiting for something that's supposed to happen daily that never comes

Like "I know we said April, but we realized the model needed more training and now we're predicting late May"

I think that would have put a lot of the community at ease

[charred mesa]

I am glad that they are training it more

I bet even with this extra month of training it'll be undertrained

but I still welcome this further pre-training

[opal hedge]

Pretty much this. Meta recently proved that more training is always better with their mid-level llama 3 model

They trained the shit out of it, and it's really good. They said they could train it more and it would continue to learn

[charred mesa]

alex had the same suspcision since its more transformer based

imagine if Meta suddenly bought it out of nowhere kek

and continued training it on their massive server GPU clusters

[opal hedge]

I'd be okay with Meta or HuggingFace buying SD3

Zuck has earned a lot of credit with the open source community as of late

[charred mesa]

especially if meta made it openrail++ instead of noncommercial-openrail or whatever

I don't even profit from AI art and never intend to

but it would still be an upside

[opal hedge]

I imagine that SAI will update SD3's commercial license once it releases

[charred mesa]

I mean they literally opted out a bunch of artists

it feels more appropriate now to even consider stuff like this

[charred mesa]

I mean..

shi maybe

we can all dream

idk what the intentions are

if they released it with generic open-rail++ (so commercial) then the investors would go even more super mad lmao

can't even imagine that

and if stability evaporates for good, what would the non-commercial do if you can't even license it?

does it become abandonware LMAO

[opal hedge]

Yeah, I'm not really sure what the gameplan was

SAI clearly has very passionate researchers and devs, but the leadership needs work

[gaunt arrow]

hello,every one

[proud spindle]

hello

[nimble violet]

gm

[flat blade]

hello

[upper geyser]

Open Source alternative to Leia Pix? (to create 3D animations from 2D images, other than SVD)

[empty patrol]

do i have to update A1111 or something or is it as easy as downloading a new checkpoint

[ornate lily]

gm

[fervent thunder]

sup y'all anybody in chat?

[craggy plinth]

hihi

[versed goblet]

how to

[sage ravine]

any good tutorial for comfy ai ( for beginner ) i use automatic 1111 on my pc already...

[wicked bolt]

what's your name?

[sage reef]

[trim path]

hello,I'm pretty sure it will

[covert sand]

do i have to download a different controlnet from the one i used for sd1.5 to use with sdxl?

[opaque gate]

hello

[narrow kernel]

Yes, they are not interchangeable. Sdxl uses a different set.

[covert sand]

where do you get them?

[narrow kernel]

google is pretty good at finding things like this

don't mean to be that dick, but you know

[covert sand]

its good when you understand things about ui, sd etc

none of this stuff makes sense to me

[narrow kernel]

well google, youtube and patience will come in handy. it's a field that is evolving at an exponential rate, so even the content creators can't keep up

[narrow kernel]

https://github.com/lllyasviel/stable-diffusion-webui-forge/wiki/ControlNet-model-download might get you started. but anyways, i'm more of a fan of teaching a person how to fish, rather than constantly handing them a fish. being able to actually research for things is an important life skill

the https://huggingface.co/lllyasviel/sd_control_collection/tree/main ones are going to be your best bet though

[covert sand]

i know how to research lol

the reason i came here was to expedite that

bc ppl here already know stuff i want to know

[narrow kernel]

well you would have known that there would likely be a large propagation delay in responses vs spending 30 seconds googling or youtubing "sdxl controlnet"

but anyways, in terms of the model versions, you really don't need to go above the mid or 256 ones. a lot of the time, the 128 ones work just as well

(if they have multiple versions of them)

[covert sand]

this isnt something you can figure out in 30 seconds

[narrow kernel]

idk, it kind of was that way for me when i first started out 🤷‍♂️

but if you're going to go the youtube route, make sure to pay attention mostly to guides from this year. lots of things have changed and lots of workflows have been replaced

[dry hull]

/video

[narrow kernel]

there are also some decent written guides as well like https://stable-diffusion-art.com/ which is one of the first search results when you google for things. they have a ton of good tutorials with lots of pictures comparing different settings and what not

[orchid vault]

hello,everyone

[half karma]

gm

[molten sigil]

hello,man

[fossil briar]

gn

[tame rampart]

we can all dream

[static cape]

At this point I don't think it would since they haven't met any projections for SD3 yet.

[jade rivet]

Yeah, I'm not really sure what the gameplan was

[clever hemlock]

can't even imagine that

hello

[echo jacinth]

Prompt:dog Superman

[hushed raptor]

guys i need help can i ask here?

[loud solar]

Try 🙂

[low moon]

AI and commerce as we know it seems incompatible, like fitting a square into a circle.

Let's face it no one cares about money, people want a decent dignified life, if that can be had without money everyone would accept it.

What seems to be happening, world over, is that money is becoming a hindrance isntead of an incentive.

[wicked mauve]

Hello Guys

[low moon]

Hi Jack.

[wicked mauve]

hey Dpdge :]

I've been working on some technology to make images similar to other images using sd

if anyone want to test it out you give me an image and i try to make one like it

[low moon]

Concistency?

consistent characters!?

[wicked mauve]

not really but it will try to make similiar one

[low moon]

is this a comfy workflow

[wicked mauve]

80%

[low moon]

or some new approach

[wicked mauve]

no it's not comfy it's a system i made using AI

[low moon]

id check it out if its an online API

[wicked mauve]

it's still in beta so you need to send me image i try it for u

but we can't send images here i think

[teal pagoda]

yo

any news regarding SD3 never coming out?

[fervent thunder]

not that i know

[graceful sapphire]

666

[rocky island]

How sd generates anime ? Like mj or still cant

Sd3

[humble sierra]

no it's not comfy it's a system i made using AI

[pallid merlin]

hii

[teal pagoda]

it's not really at the level of niji from midjourney

[low moon]

🤣

[azure nebula]

hey what control net to donwload to use the image to sketch?

canny?

[trail lion]

canny or lineart or sketch, they're all sightly different so it's trial and error to see if it does what you want

[last ridge]

How to use drawing

I don't know where it is. Who can tell me?

[trail lion]

it's one of the tabs under inpainting, but drawing with a mouse is difficult, you can also drop a modified image back into img2img or controlnet etc

[fervent thunder]

...

[patent nymph]

Hey! Does anyone know a similar node as cut by mask node from https://github.com/BadCafeCode/masquerade-nodes-comfyui ? I need to remove places marked with a mask from an image with an alpha channel.

[trail lion]

no clue, but there's a comfy channel on this discord

[tender night]

yo

[loud solar]

And if we feel comfy here? 😄

[wanton kraken]

Hello

https://idm-vton.github.io/

Is it possible with stable diffusion?

Please take a look and let me know. Thanks.

[warm junco]

hey there is no extension for auto1111 of it. its an seperate tool you have to install manually or via https://pinokio.computer/

[lavish lake]

shady looking link

[warm junco]

nah its fine

[wanton kraken]

Aha.

[warm junco]

its a clothing swap tool released a few weeks ago

[wanton kraken]

So it has nothing to do with stable diffusion, right?

Can I test it online?

[warm junco]

it uses the diffusion process so it has something to do with SD

[wanton kraken]

Is there any online free src for this?

[warm junco]

idk

[wanton kraken]

I checked this site but no function for this.

How to test it?

[warm junco]

you need to install pinokio and in pinokio you can choose to install vton

[wanton kraken]

Understand.

In that case, we don't need to train any model for this function right?

[warm junco]

i dont think so as it comes with its own model/dataset

[wanton kraken]

SO Vton is kind of model, right?

[warm junco]

i dont think so

[wanton kraken]

What is vton?

[warm junco]

its a programm containing many things

here is more information:
https://github.com/yisol/IDM-VTON

[wanton kraken]

Thank you.

Have you ever tried this?

[warm junco]

nope

maybe someone will do an extension for auto1111 in the future

but i did clothing+faceswap in auto1111 using Controlnet IP-Adapter

[wanton kraken]

Oh.

Controlnet IP-Adapter is Cool.

But Vton would be better than it, what do you think?

[warm junco]

yea i think so too

[wanton kraken]

Thank you.

Hi, @warm junco , Can you help me do it with code without external api?

[warm junco]

do what?

[wanton kraken]

Implement Vton.

[low moon]

Yeah, a very good clothing swapper would be amazing in comfyui.

Ipdapater is janky.

[wanton kraken]

Do I need to do with comfyUI?

Can it be done with model?

[low moon]

I suggest you do it for Comfui.

anyone even remotely serious who cares about dressing their characters a certain way is in comfui.

Casual folks settle for A1111 Forge and Fooocus.

[wanton kraken]

Casual folks settle for A1111 Forge and Fooocus.?
what does it mean?

[trail lion]

I keep reading about the AI chatbot replacing search engines (ie the google), and I was reluctant to believe it, but more and more I'm convinced it's true because people need immediate gratification. My parents spent hours in libraries, and so did I, until search engines, and now it's like...why would I take the time to search to find the answer in 10 mins when someone can walk me through it right now

[low moon]

We need a promt for our promt.

Just type in "Cool image" and get 3 pages of intricate promt description.

[wanton kraken]

?

Hi, @low moon

Only comfyUI can do VTon.

?

I would like to discuss it with you in DM.

Okay?

[low moon]

I am not an expert...

[wanton kraken]

Do you find me an expert please?

[potent spire]

Chatbots are too flawed to replace it

them*

[wanton kraken]

Hi

I 've also tested the Fooocus.

I used "female model, luxury fashion photoshoot, stunning Paris, Summer, Haussmann-style buildings, intricate architecture, tree-lined boulevards, leaves, the Eiffel Tower in the distance, and the blue Seine flows under bridges." for text prompt and set faceswap.

And set the speed as "extreme speed"

BUt the result image quality is bad and lots of background features are missing.

Can you figure it out please?

[trail lion]

status on finding the guy an expert?

this chat is seriously going downhill

[wanton kraken]

Please help me.

[trail lion]

it's like a waystation for entitled people who want instant solutions=

[wanton kraken]

Hi.

What is the guidance scale in stable diffusion?

[wanton kraken]

?

[trail lion]

the what?

[fallow veldt]

anyone good with inpainting? I'm inpainting details in a large resolution image and keep getting bad detail and a halo that doesn't blend with the rest of the image

https://discordapp.com/channels/1002292111942635562/1089974139927920741/1241293682435428352

[velvet slate]

God I am loving learning SD/comfy. It's great to actually feel like a complete noob again. Reminds me of when I first started learning my major.

[shell tendon]

try differential diffusion in comfyui

[narrow kernel]

It's called CFG and it determines how hard it will try to force the image to listen to the prompt. The common range most people work with for most models is 4-8 or so. A cfg of 1(or 0 depending on the app) will completely ignore your negative prompt.

Google for "stable diffusion cfg explained" or something like that to learn more.

[winged birch]

ひうｄｗ

お

https://tenor.com/view/apo-solary-apo-apofps-gif-9924237009492714744

https://tenor.com/view/smash-gif-21365305

https://tenor.com/view/mood-sad-panda-gif-14650463720672021603

https://tenor.com/view/high-five-amy-santiago-rosa-diaz-stephanie-beatriz-melissa-fumero-gif-23124416

https://tenor.com/view/aw-shucks-aww-thank-you-gif-25109804

[stuck sun]

hey uhh been out of the loop a bit. Is SD3 coming out, ever now?

[opaque yarrow]

API: yes, install locally: no

[stuck sun]

never?

[opaque yarrow]

nah they would release the checkpoint

it's just takes time

idk why

u can see stable diffusion 3 api doing weird output

[stuck sun]

idk, like i said been out of loop on this

[stuck sun]

i assume to make it impossible to do gens of things going up butts thru extensive butt redaction testing

[fervent thunder]

bummer

why!? 🙌

someone contact me with with SD managers, I'll take on the job

let's train that goddamn GAN

[fallow veldt]

Comfy oh no, I have to install... might try, ty

[shell tendon]

if you get it set up i can get you my inpainting diff diff workflow

it really does integrate with the image better than anything else i've tried and i've been pretty exhaustive

[fallow veldt]

yes I've saw some videos, interesting. I'm really frustrated as I don't understand why it doesn't work on a simple task. It seems I'm better off trying to fix it in Photoshop.

[shell tendon]

hwat are you trying to do?

it wasn't really clear, just saw a random image

[shell tendon]

100%... it's amazing

there's always so much to learn, it's endlessly exciting

[shell tendon]

the difference is libraries are still good

google is not

[fallow veldt]

I upscaled an SD XL image by 4 with Ultimate SD Upscale, I got seams lines in some areas, I'm trying to make them go away

[stuck sun]

google was good 10 years ago, it would actually let you find the things you were looking for!

[shell tendon]

prolly want to use the checkerboard pattern and all the other stuff on it

[fallow veldt]

I've tried different denosing, blur mask, soft inpainting on/off, it doesn't make the detail good and always leaves some kind of halo

[stuck sun]

have you tried supir

[shell tendon]

tbh i've stopped upscaling using that so i forget the names of the individual options but i'd recognize them from a screencap

are you trying to add new detail?

or are you trying to just blow it up without blurring?

[velvet slate]

It's a nice feeling to see how much better everyone is too 😅
After enough years in the field I stopped feeling that I was trying to catch up.

[fallow veldt]

no, I'm a tired man, tired of installing and troubleshooting things xD

[stuck sun]

sure but supir is genuinely fucking good

[shell tendon]

yeah, it is

it's really, really good if your source image is already good and doesn't need anything fixed during the upscale

[fallow veldt]

mmm I guess I have to try it, let me see, a Saturday of installing stuff

[shell tendon]

(important to point out, i think, as we use upscaling to fix stuff or hallucinate new detail so often)

it's really easy to install lol

comfy... you download it, run like 3 commands and you're off and running

you can install all nodes from the comfyui manager

including supir

type in the name, click install, and then restart

[stuck sun]

you can use it to hallucinate too if you mess with the settings but its best for preserving the original image, better than anything else, ever

[shell tendon]

totally agree with the above and i've tried fn everything

but again, gotta have your image nailed down. if there's any muddy details... watch out especially for hands... it's just going to magnify it

[stuck sun]

I was paying for magnific and cancelled it when supir came out

[shell tendon]

you can also use it to sharpen

upscale with supir, then downscale back to original size

[fallow veldt]

yes I wanted to give it more detail, for the lulz https://discordapp.com/channels/1002292111942635562/1089974139927920741/1241614349315870793

[fallow veldt]

uh models are huge, how much disk space?

[stuck sun]

my install is 68gb i think

i'm using a standalone installer for it and it installs llava as well, which automates prompting from the original image to help guide the upscale

i tried supir on your image in the sdxl channel

[oblique elk]

Yes the difference between both methods is the "creativity" you allow the ai. Supir keeps nealry 100% of the input while controlnet tile changes a lot depending on the prompt and strength.

[verbal quest]

newbie questions about SV3D

what are the limitations?

can it really make a me a 3d model of anything?
with specifics?
and how detailed can it get?

looks impressive from what im seeing

[fallow veldt]

Mm in my Linux disk I don't have that much space

[teal pagoda]

what's the name of that "chatbot"?

[trail lion]

just generically referring to chatgpt, gemini and the like

[teal pagoda]

nah, they aren't really at that lvl yet

[teal pagoda]

it's hard for them to apply advanced filtered search like on google (with commands like "OR", "site:" and so on).

And the problem is that they'll give you only the most popular results

I tried them myself

claude, gemini, gpt, copilot

they all give you the most popular results

this is not a good thing if you're into researching/finding as many things as possible for example

and if you regenerate the answer, they still copy results from the previous answer

example case

you want to make a list with 300 companies selling PC components

and you want them to be all different, not the same stores, markets

the AI will give you max 100 names of the most popular ones

after that, it'll regenerate results from the previous 100

and this won't help you if you wish to make a complete list with the cheapest stores for example

so google advanced search and the web crawlers are the best atm

[hidden dagger]

hello, guys

what do ya do with Diffusion model

[teal pagoda]

:))))))

images?

what else?

[hidden dagger]

so does texts or videoes

but you know images have different forms

[teal pagoda]

you said diffusion model

[hidden dagger]

yh

diffusion model is generateive

so you can generate so many things, not just images

[teal pagoda]

but this is a server for images

[hidden dagger]

IK

but you can use different modality for generating the images

[fair pewter]

?

anything related to images

you can tweak it and get videos out of them though

[hidden dagger]

Ik

but more precisely, everything is related to matrices or vectors

[fair pewter]

both

well you dont see the process if you arent using a framework

stable diffusion models are meant to put you at ease when you generate

i would worry about matrices and etc if i was using pytorch and etc

and theres a shit ton of stuff to learn

if you use a framework that is

[hidden dagger]

well, every deep learning model does as well

the training is always a black-box one

[fair pewter]

yep

i agree as a pytorch user

[hidden dagger]

oh, I forget to mention everything is related to tensor more precisely in deep learning

[hidden dagger]

so do I

[fair pewter]

yeah it is

but why are you dragging this topic in the stable diffusion server if i may ask

[hidden dagger]

what's ur research interest

[hidden dagger]

no idea, just asking

I am wondering if everyone study the model

[fair pewter]

ive built gan models before but now im just working with CNN models

[hidden dagger]

do ya use transformer

[fair pewter]

no?

i train the models from 0

[hidden dagger]

alr

I find baseline then try to modify the baseline

[fair pewter]

yeah u can load light pretrained pytorch models already

only ever used transformers to load styleGAN when i was in my GAN era

[hidden dagger]

Ik

my research topic is abt stable diffusion on medical imaging

[fair pewter]

yeah?

i dont think an image generation model is going to help you with that

you want it to generate medical images or just detect them>

[hidden dagger]

well, using SD model on medical imaging does not mean you can only generate the image

u can use SD model combined with other downstram tasks such as segmentation

so I read paper mostly

I bet you don't read paper much

[fair pewter]

never seen anyone use sd models for that

[hidden dagger]

bcz that's the new topic in SD academic research

but I have read some paper with that kinda topic

[fair pewter]

🤷

also what are you going to research

[hidden dagger]

just what I've mentioned previously cz I haven't determined what kinda downstram task I would do

so I study the models from various downstream tasks

[fair pewter]

i mean ive made a detection model from pytorch before to detect the tay sachs disease

i think its best to use a proper framework for these tasks imo

[hidden dagger]

my response is pretty obvious for ur question

[fair pewter]

going to continue trying to use an sd model for these tasks?

[hidden dagger]

different direction I would say. I am doing academic research, so I have to find out the issues on the current models then modify them

[hidden dagger]

yh, it is possible

[fair pewter]

good luck

[hidden dagger]

and doing academic research means you have to do sth that no one have ever done before

that's why we need academic research

[fair pewter]

theres millios of things u can do that no one has ever done

millions*

ai has just got its spotlight and theres millions of things we could do

[hidden dagger]

trew

but still most of the people gatekeep the history

[covert bridge]

are there any alternatives to SD I can install and run on my PC? I don't want to use midjourey/online packages.

[fair pewter]

what do you mean?

sd models or the interface

[covert bridge]

ai image generators that are not SD, and not online only

[fair pewter]

you can go to huggingface

you will find a bunch of models that arent sd

plus sd models are usually .safetensors so theyre not dedicated to sd only

[low moon]

They have to fix hands first.

[charred mesa]

they can't heh

[nova zodiac]

Pixart

[chrome barn]

Guys my friend wants to do "AI Car photography" to see how well it does what model should he try? and does he need a Lora or TI if so please let me know

[hollow goblet]

https://tenor.com/beb52.gif

[low moon]

Buncha car loras on civitai

[strange prairie]

Hello I don't understand that discord can I generate images anywhere?

[oblique elk]

My french friend you here you can find information and help to install a stable diffusion model local on your hardware and run it to create images or you can create a paid account and use the artisan-bot to create images.

[strange prairie]

Thank you Lord of the weed, I was looking for artisan bot but I though it was free, do I need lot of hardware resource to run local model ?

[oblique elk]

Depends on the model and the different extensions you want to use. Creating short videos using multiple different control nets etc. require a decent amount of video memory (of your graphic card)

The gpu would be the most important thing

[strange prairie]

Yeah I have a very little GPU xD

I will try images not video

A last question is it easier to install the model on Linux or windows ?

[oblique elk]

Depends on your skills, i have it installed on Mac with apple silicon and intel cpu and since a few month on a linux / debian pc

[strange prairie]

I am developer and I work on Linux every day

[oblique elk]

In generel you need python with 3.10.X Version (3.11 does sometimes give hassle with some requirements for python packeges)

Well then it should be easy, just create a virtual environment. Decide for a UI download a model and go for it 🙂

2 UIs are mainly used auto11111 (lately replaced by forge) or the node based UI comfyui. As a developer i would guess comfyui would match

[strange prairie]

Okay thank you very much, I will try tonight 😉

[oblique elk]

Do you have a nvidia card?

[strange prairie]

No

Only a burautic pc

Just for development but not for games

I have a laptop gamer too with old Nvidia

Perhaps i will try on that laptop

Perhaps I will try to develop a game with ai graphisms

Until now, graphism was the problem ^^

[oblique elk]

Would try it with the nvidia laptop. good chances it will be faster then the burautic pc.

[fervent thunder]

Hey everyone! I'm working on a personal project called Isari AI that aims to provide fair compensation and visibility to workers using AI tech. I'm looking for feedback and ideas from people interested in AI and promoting fair work opportunities. Would anyone be interested in discussing this?

[covert bridge]

you dont mean this do you? https://play.google.com/store/apps/details?id=com.grisoft.pixart&hl=en&gl=US

[oblique elk]

No pixart-sigma is another model to create images local

[covert bridge]

phew 😉

[tulip yarrow]

do the "masterpiece" and "best quality" tags really do anything to the image

[covert bridge]

does it have its on UI, or is it only usable via something like comfy?

[oblique elk]

can be used with comfyui for example.

[covert bridge]

so its like a checkpoint/model, not something u set up like a1111?

[oblique elk]

yes it is kind of a checkpoint but in comyui you would need another node to work with these models.

Advantage is you can use prompt understanding from pixart and then use sdxl for refinement of the image

[young surge]

@stark fern sup ?

[hardy sedge]

Why does kohya_ss always need internet access when you want to create a lora model? Is there a way around this? Does anyone know which data he always wants to download? Can you download them yourself and put them in a folder?
Thanks

[low moon]

[fervent thunder]

hahahahaha

[fervent thunder]

This is like an AI joke hahahah

[foggy halo]

SD3 release when

[young surge]

Nevah

[soft iron]

never

[trail lion]

are you running it local? it doesnt need internet unless you are trying to pull something, like a model for instance

[young surge]

😐 that moment your gpu and ram runs out of ... ram, and halts the genning on an image to the point of you havent to end the process 😦

[sage reef]

[young surge]

And the preview window showed such promise

[unkempt hatch]

could you train the model specifically on the concept of cartoon - photo, almost like a slider?

like, train cartoon versions for a bunch of characters, and then some 2.5 semi realistic AI renders of them, and then actual photos of people acting them in live action films or cosplay?

it seems Midjourney can already do that. Like you can prompt "real life bart simpson" and it creates a photo of what he'd look like if he were real.

[narrow kernel]

Yeah with loras. You'd have a lora for each style and then you'd pick one and the amount of intensity you want to apply. You can also mix loras, but if they are too different from each other, they will fight each other. Like a cartoon lora would try to avoid using anything realistic looking and a realistic lora would try to avoid anything cartoon looking. So trying something like a 50:50 of them might make them cancel each other out or it might destroy everything. But if you start with say a model like juggernaut that is mostly geared toward making realistic stuff and then applied a cartoon lora at a lower strength, it might yield something 2.5d looking

Basically, you gotta experiment

[austere steppe]

hi

[fervent thunder]

Hello, anyone knew how to install stable diffusion on computer?

[oblique elk]

check the pinned message #🤝｜tech-support

[fervent thunder]

Thanks!

[young surge]

Anyone got any good wasteland LORA's for PonyXL/SDXL Merge?

Or rather, just any good landscape lora for PonyXL?

[drifting sleet]

hello

please i cant find the text to image generator, that is the stable dreamer, please where can i find it

[hardy sedge]

I always get this message when I don't have internet on. I use Kohya locally under Windows. can I download this url: /api/models/stabilityai/stable-diffusion-xl-base-1.0 file manually? If so, which folder do I have to put it in? So Kohya always needs internet. I thought it would work completely offline? It works with internet because he is still downloading something

Thanks.

requests.exceptions.ConnectionError: (MaxRetryError("HTTPSConnectionPool(host='huggingface.co', port=443): Max retries exceeded with url: /api/models/stabilityai/stable-diffusion-xl-base-1.0 (Caused by NewConnectionError('<urllib3.connection.HTTPSConnection object at 0x0000026DE9F71630>: Failed to establish a new connection: [Errno 11001] getaddrinfo failed'))"), '(Request ID: 6ead4e04-602f-4b37-b662-a6b1f0a3f64d)')

[velvet slate]

Hey anyone here really good with QRmonster? I'm having a hard time getting the exact output I need

[patent nymph]

Hey! Is it possible to solve this error or it is limited my GPU and nothing can be done about it?

Error occurred when executing OneFormer-COCO-SemSegPreprocessor:

Allocation on device 0 would exceed allowed memory. (out of memory)
Currently allocated : 5.88 GiB
Requested : 2.34 GiB
Device limit : 8.00 GiB
Free (according to CUDA): 0 bytes
PyTorch limit (set by user-supplied memory fraction)
: 17179869184.00 GiB

[rare parrot]

Use fp8 and tiled VAE

[narrow kernel]

Don't have internet on? Why do you turn it off? Are you paranoid about the stuff you're training or something?

And that's it checking for updates on the model's HF repo.

[trail lion]

point is you can use an offline model, vs pointing to hugginface

[unkempt hatch]

That's what I've done, but only 2.5d like you said, and not necessarily the best adherence to the original.

Whereas somehow Midjourney can simply prompt "Bart Simpson if he were real" and it does it

[hardy sedge]

Where can i found the offline model,?

[astral goblet]

i just use the one i have in my /models/stablediffusion/ folder. Downloaded from huggingface. If you select one of the presets, it'll try to download it from hugging face. If you select custom and just chose a local file you have saved, that's good too.

NGL. Seems weird to be waist deep in this hobby enough to want to train a model on your local machine, but also not know where to find the model. You might have to return to the basics and learn to crawl before you try to run the marathon.

I'm not sure why your connection to huggingface is being refused. That shouldn't happen. Might be a broken python environment

[hardy sedge]

I'm talking about kohya.ss. and he tries to download the file every time I want to train a lora. I wonder why the file is not saved.👍

[astral goblet]

maybe user folder permissions. seems like a local system problem. mine doesn't do that if i select the presets. it only downloads them if they're not already saved

i suspect you're talking about bmaltais' gui wrapper for kohya scripts too

jumping ot an assumption like "it always needs online then" is bad troubleshooting

[hardy sedge]

No idea, but it doesn't matter now, its probably won't get solved.

[weak kettle]

hi guys is there any ai tool that can make shortcut from 30 videos, like you input some videos and the output is one short clip with cutted things from these 30 videos

[trail lion]

doubt it, that sounds pretty complicated

[weak kettle]

i dont need voice matching, only like cutted contect from multiple videos

and make short video 2-3 minutes from it

[trail lion]

well you can do all that, rip the videos to frames, take the ones you want, recombine them...but there will be manual steps, there's nothing that's going to one-shot that effort

that plus the fact that the whole thing is a waste of time to begin with because really video diffusion is meh to begin with, maybe in 5 years