#💬｜general-chat

1 messages · Page 23 of 1

cobalt otter
#

Oh?! What's that?

neat condor
#

google it, it's the first result can't link here i think

cobalt otter
cobalt otter
muted temple
#

Does anybody know how far away we are from SD for videos? Like textovideo?

neat condor
fervent thunder
#

maybe the big announcement sd has is text 2 video. i thought emad said we'd have text 2 video this year

earnest echo
#

hi guys

#

I need help to draw a digital painting because I dont know how to do it but I can use most of the drawing software but I don't know how to draw Can you suggest a course for me?

cobalt otter
digital ginkgo
naive aspen
naive aspen
fervent thunder
# cobalt otter Really?!

yeah im almost positive he said it would be in 2023. i wouldnt expect anything too great in terms of fidelity to start but i bet we'll see it soon

naive aspen
fervent thunder
#

probably. but you gotta start with baby steps i suppose

#

give 5 or 6 years and we'll be making our own movies and premium cable series lol

naive aspen
#

its mainly about getting enough fps i think

#

You need to render 24 frames per second to get movie fps

#

For us to start living in a real time rendered VR world we need to get to atleast 120fps

fervent thunder
#

the midjourney guy said he thought we'd be able to generate our own aaa level games in 8 years

#

just his guess of course, but who knows how it'll all shake out. going to be exciting to watch however it plays

naive aspen
#

That would be like real time rendered VR worlds i guess

#

Compute power is the limit

fervent thunder
#

an eventual integration of quantum computing and ai will probably be the real peak. but it sounds we're a long ways from quantum computers being a thing let alone a commercial thing

naive aspen
#

So we need some ai to figure out gpu's and infinite energy

#

With infinite energy we can just brute force everything

knotty wasp
#

what does 3 stages of diffusion mean? img2img upscaling?

dense pine
#

What's a quantum computer???

fervent thunder
#

next level computing. google it up bro

#

seems like emads hinting a new model soon idk

wise stratus
cobalt otter
fervent thunder
wise stratus
#

A few

fervent thunder
wise stratus
slender field
cobalt otter
fervent thunder
#

sounds good to me

neat condor
#

so basically what i've seen is carla ortiz speaking about money and emad being excited about progress, basically whole this conflict in a nutshell, good work CBS

waxen swallow
#

Will we see distilled sd or text to video soon?

sinful talon
#

hello what is the best way to inpaint with stable diffusion 2.1 ? is there a model of 2.1 inpainting ?

low sun
#

People are tryna sue stable diffusion they’re gonna lose

peak orchid
hidden bridge
#

Hey guys,

#

Question trying to create NFL logos to be like a mystic beast wondering if you would like to help me out?

fervent thunder
peak orchid
# wise stratus A few

Are they already uploaded to HuggingFace, awaiting being made public when the time comes? Or is it not that soon yet

fervent thunder
#

If it exists, it exists.

knotty wasp
stone panther
#

Can someone explain this discord to me like im a 5 year old?

astral goblet
#

yeah i confirmed it. i was able to run 11 hours of generations with ease, using --no-half for full precision. but it's so slow. when i remove that tag it works but then randomly the sampling.py process will cause a cuda crash and reset my gpu driver.

people keep telling me it's an OOM error, ignoring that I've got a 4080 and the error message is something about bad inputs to sampling.py, not anything about memory.

peak orchid
#

Guys, I just realized that the T5 v1.1 XXL model used by DeepFloyd is 44.5GB in file size according other T5 V1.1 XXL models

wise stratus
astral goblet
peak orchid
#

Look at the file size lol

astral goblet
#

maybe i'll switch back to amd gpu if they put a 64gb card out first

peak orchid
knotty wasp
#

you only need the encoder part, and you can do half precision, then you are down to like 9gb

astral goblet
#

my older brother's mining rig has over 96gb of memory last i checked. i might buy that off him since he's not using it anymore.

brittle fractal
#

what you guys using as all in one tool? i cant use krita plugin coz it doesnt do outpainting, openoutpaint doesnt have denoising function, inkoke AI doesnt work with 2.0 models, photoshop plugins not working with automatic1111 backends....

astral goblet
# brittle fractal what you guys using as all in one tool? i cant use krita plugin coz it doesnt do...

there isn't really an all in one tool. many try, but like you said, there are compromises all over. i use a1111 generally. i wanted to start using invoke but their installation process is needlessly complex and even though i was born into the command prompt, that still annoys me. Also, i found out that there's zero support for 2x and they don't intend to add it anytime soon. I think that's irresponsible and is very much against the RAIL-m license that the models they rely on declares.

#

development of this field is so bleeding edge that nothing is version 1 yet

brittle fractal
astral goblet
#

i think not enough people are realizing that the model license, rail-m, requires people to make an effort to use the newest version of the code/model at all times. Invoke are spitting on the license by refusing to add 2x support

#

their github says they've got 2.0 and 2.1 support now. maybe a lawyer smartened them up

wise stratus
#

Some models will use less memory some a whole lot more

fervent thunder
storm shard
peak orchid
astral goblet
wise stratus
wise stratus
fervent thunder
astral goblet
#

i was on twitter the other night and saw you finished elden ring. congrats! good to know the leader of this voyage is a gamer

fervent thunder
#

'when theyre done' could be a few days or a year, i mean

storm shard
astral goblet
storm shard
astral goblet
#

do you see a duocorn or a tricorn? no quacorns or quincorns either. just uni

#

i'm a man of simple pleasures

fervent thunder
#

Hello everyone.

#

Any news?

storm shard
fervent thunder
#

What?

storm shard
storm shard
astral goblet
# storm shard Don't you think an embedding could fix that ?

the embedding would over train the result though and i wouldn't be able to get unicorns in various contexts like "tattoo" still. A better text encoder is what i suspect lead to the result we see there. It's simple but it speaks volumes. All the deepfloyd images are examples of the improved encoding really

storm shard
astral goblet
#

once i solve my technical issues, i do intend to train embeddings for unicorns though

#

i'm talking like, ridley scott's legend style embedding. all those gift store paintings of unicorns that have sun shafts in a forest meadow

sage sky
#

can someone tell me if i need this box checked? " move face restoration model from vram into ram after processing" ?

#

😮

astral goblet
ebon pewter
#

any change in auto1111 recently? VAE not being used even it has same name with model

storm shard
astral goblet
#

If i understand it right from the limited information there is, they use the t5 encoder for it. It's not a focus on text so much that their system is powerful enough achieve those results

cold siren
ebon pewter
#

tq

forest sable
#

Is there a way to see which model was used in something generated locally using a WebUI? Like in the metadata?

rotund umbra
#

Hi everyone , how to install abything vae?

#

Does it work like other models?

#

Asking cause Vae is really strange type

warm junco
rotund umbra
#

Thanks)

#

BTW, can videodrive pull models mixing?

warm junco
#

videodrive ?

rotund umbra
#

My Videodrive with 6GB memory

#

Oh, my bad

#

I will rephrase

warm junco
#

Graphics Card ?

rotund umbra
#

Yea

warm junco
#

yea you can merge models with it

hazy wind
rotund umbra
#

Cause author recomended it

warm junco
#

you cant merge with a vae, you have to select the vae for the model

rotund umbra
#

Ah no btw, anything vae has .pt type

#

I have read it incorrectly

warm junco
#

.vae.pt

rotund umbra
#

yea

warm junco
brave matrix
#

hello, i am hearing a lot about this new SD ultimate upsacler? How do I use it? I can't find it in auto111

vale spindle
#

I believe that's an extension mazzy

fervent nymph
#

Is it possible to add dreambot to other discord servers?

ebon pewter
sacred ice
slender oyster
#

I'm using the automatic1111 ui and haven't updated in a while, so just pulled. Did the interface to generating like 100 images move? It seems I can only do 1 at a time, unless I'm blind that was removed? or moved somewhere I can't find

slender oyster
cobalt cliff
#

what do commas in a prompt actually do?

astral goblet
signal nest
#

i'm hoping they'd help with grouping aswell. I'd assume the more punctutation you use, the higher the chance it's going to interpret correctly

#

"a man standing in a street wearing a jacket" "a man, standing in a street, wearing a jacket". we know obviously that the man would be wearing a jacket, not the street, but a statistical langauge engine wouldn't. it's going to have a better chance of seperating the concepts out when its punctuated

peak orchid
#

Has anyone dug into the Joseph Saveri Law Firm that they are using to sue Stable Diffusion yet? They to have multiple lawsuits against companies going on right now, and they were the ones that the GME meme stock crowd used to sue Robinhood for blocking them from buying the stock. They were also the ones who sued Lenovo over the Superfish stuff.
They don't seem like amateurs, but I can't seem to find company history going back past 2018
https://www.saverilawfirm.com/our-cases/

waxen pasture
#

I finally got TI working on my 3070 after hours of oom errors! But does anyone know why Task Manager says I am only using 40% of my memory and 25% of my GPU? I am getting about 2.3s/it and it will supposedly take an hour.

#

btw, the way I fixed my errors was removing no-half and precision-full from my command line arguments, is that what is causing the low usage?

cedar kayak
#

Good morning, why are 90% of the models only blends/merges ? Is there a source somewhere where I can download models that have been trained from scratch? 🤷🏻‍♂️

waxen pasture
cedar kayak
#

And when I want to start training my own model, everything on YouTube shows up as training yourself, not a set made of 10000 images of planes, cars, different paintings, like more broad instead of fixated to 1 style or 1 person or 1 object… is it impossible for normal user pc’s to train with a lot of images ?

waxen pasture
cedar kayak
#

So that shouldn’t be a problem

waxen pasture
#

Well then, start making your dataset! Captioning will take forever with that many images

cedar kayak
#

So I can utilize my machine to train , but I not want only on my own face or someone else lol

waxen pasture
#

Yeah you can do styles and such

#

I guess you could use BLIP to caption for you

cedar kayak
#

Or at least, that’s what I assume people did when scraping millions of low res preview images of Getty and so on

#

🤷🏻‍♂️

waxen pasture
cedar kayak
#

By saving the files using tags of the image on that site lol

cedar kayak
#

I’ll give it a try, not sure where to start though lol

waxen pasture
#

It is a large feat. Maybe do start with a smaller concept, like a basic style or person.

#

I always lioke training styles. I currently have an embedding for LEGO training on my pc.

cedar kayak
#

Yes and slowly once I understand it better increase the volume

cedar kayak
waxen pasture
cedar kayak
#

Cheers !

waxen pasture
fervent thunder
#

has dalle ever had any big updates to their model? seems like they were the shit for a while now i never hear about them

cedar kayak
#

Ah other question. I still like my auto1111 using sd1.5, is there a dreambooth2 model for it that works also with older sd?
I even still use an outdated auto1111, as I am so used to it 🙂

iron elm
waxen pasture
latent flower
#

Id wait for safetensors still

quick bobcat
#

I've got a pokemon model that didn't reliably produce "species" results until about 250,000 steps because I trained it against 1200 unique names/samples.

fervent thunder
#

whats your guys favorite model for photo real atm? analog diffusion is mine from what ive surveyed at least

quick bobcat
#

Shared classes such as "FireType" and "Mega" came through at about 70,000 though. Was able to apply "firetype pokemon" to random shit like a waterbottle.

next birch
#

Does Stable Diffusino have a logo? I want to put a splash screen in my game with the logo on it.

cedar kayak
#

assuming using 4x A6000, as thats what I can fit most in 1 machine right now, any idea ?

quick bobcat
#

Mine is just one case. Total of 1200 unique identities. You probably have more samples and probably fewer identities you're trying to retain. Over 3 sessions probably 18 hours total?

cedar kayak
#

so basically, the amount of time doesnt matter so much 😉 the manually tagging/prompting, yes problematric 😉

quick bobcat
#

I spent way more time writing descriptions. True.

#

So much excel.

cedar kayak
#

yes, manual labor 😢

#

and they said AI gonna do all the work

#

lol

quick bobcat
#

BLIPing the samples actually did give a good starting point. Still had to touch every single one though.

sharp robin
#

I've been out of the loop for like a month or two, any news for a supposedly very fast approach that makes possible multiple gens a second ? I think Emad hyped it

#

The distilled model approach maybe?

#

OK I seached and if I think it is what I was remembering, Emad said it was difficult to train and pushed back. It was overhyped, w&s

dusky laurel
#

IT WAS TOO EASY TO GET IN

#

NOW IT IS DEMON TIME

#

FIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUvvFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUvv

#

FIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUvv

#

FIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUvv

#

FIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUvv

#

FIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUvv

#

FIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUvv

#

IS ANYBODY HERE

#

FIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUvv

#

FIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUvv

#

FIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUvv

#

FIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUvv

#

FIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUFIUHAWIYFGADHUYFGADSKUvv

#

BRUH

#

I ONLY HAVE 1 MIN TO GET BANNED+

#

ILL SWITCH SEVERS

uneven wasp
#

yes, I concur that is an intelligent and well thought out comment

#

really changed the I look at the world

#

empahsis on the "FIUHAWIY" part

winter pike
#

bro got ignored 💀

quick basalt
#

My brother in Christ, what the fuck was that!! lmao

leaden solstice
#

Hey yall. Just curious, what's the big difference between the various protogen models? I see at least 8 models on the account on HF, 6 on Civ

fervent thunder
#

absolute demon move

#

i used to think bad asses didnt exist anymore, then anoosh appeared. wow.

digital folio
#

I created an ai game show using stable diffusion called "the real deal" where I give my friends 3 images and 2 of them are ai generated and I have them guess which one is the real deal, hence the name

next birch
#

5000 steps? LOL. I am trying the opposite. Trying to tweak it to get the best quality image in as few steps as possible. Smallest I got so far is 4 steps.

ebon pewter
#

what prompt should i put if i want some wording on printed on a shirt?

#

e.g: my name omelette on a t-shirt

outer crescent
#

currently doesn't do text well at all

fervent thunder
#

what model is the dream bot using?

#

2.1?

olive scroll
#

Ey, just found out about the big ol lawsuit. Thoughts?

fervent thunder
#

All that would do is just handicap US in Ai race

olive scroll
#

Are copyrighted images public data? Legit question, not being sarcastic

fervent thunder
#

Is inspiration controllable is the main question

fervent thunder
#

No one can stop you from taking inspiration or teaching Ai how it's done

#

it's unenforcable

#

Copyright is gone if you edit the original art to the point where it is no longer comparable

#

And that's what Ai does

#

It never spews the original

olive scroll
#

True. But they are using copyrighted images for profit without consent. Just having it in the database alone, they don't have rights for commercial use?

fervent thunder
#

They're not selling anything copyrighted

signal nest
# olive scroll Are copyrighted images public data? Legit question, not being sarcastic

training AI requires "information bottleneck": the neural network must not be big enough to just store the input data. Stable Diffusion is trained on >2billion images, and it's a 4gb neural net: there would only be 2 bytes available per image, it's impossible for it to copy. Image files require many thousands of bytes each. Creating this "information bottleneck" is part of how training forces an AI to actually learn general rules.

fervent thunder
#

I personally wouldn't even bother paying attention to the lawsuit

#

It's probably some internal political stuff

olive scroll
#

It's more just amusing than anything. I don't think it'll actually stop SD or MJ

signal nest
fervent thunder
#

Gotta get your name out there I guess then

#

And get access to closed stuff

signal nest
#

I am interested in crowdsourcing deliberate training data for generative AI. years ago i saw this chaps project in reddit and helped him out little and often :https://imagemonkey.io/ . I put a few thousand of my own photos up there and did some annotations.

fervent thunder
#

That's super interesting

signal nest
#

if everyone in this community (100,000 people in the discord) did the same we'd have millions of photos. the effort needed to make the data is so small, per person.

#

I am hoping that now that people have tasted opensource image generators , more people would be interested in contributing.

signal nest
fervent thunder
signal nest
#

there's plenty of ways to make data availble of course. I thought this was interesting r.e. encouraging crowdsourcing specifically. I figure he can make the data available to whatever format is eventually most popular for this purpose

outer crescent
grand glen
#

I see mid journey has a very good ability to generate icons and even UIs . Does anyone know of any other ai image generator that can do that because I don't wanna use mj

outer crescent
#

how about this tars? make a little extension that makes it a little game or whatever while people are rendering prompts in automatic?

fervent thunder
#

What I learned that if you just do the thing

signal nest
fervent thunder
#

it works more often than you can expect

unreal eagle
#

any news on the 99.8% speed increaser yet?

outer crescent
signal nest
#

he's got some variations on there like a quiz mode. The main thing I was interested in though was polygonal annotation (i.e. setting it up for segmentation). I've heard that the chinese trained an image generator on fewer images, but with more annotation (i.e. explicitely teaching the inner layers of a net which parts of an image relate to which words, you're giving it a head start on figuring the rest out)

outer crescent
#

well the esp game was brilliant for describing an image in as many words as possible other than the obvious ones.

#

i spent hours playing that way back

signal nest
outer crescent
#

well, salt to taste but yeah

#

ive a feeling depthmap stuff might be better for explaining them, but i think using recaptcha type stuff to have users decide if a hand is real or not would be great

#

training a model on thousands of not-quite-hands might be useful

signal nest
fervent thunder
#

messaged 3 youtubers, let's see what happens

signal nest
# fervent thunder messaged 3 youtubers, let's see what happens

he's already got something in there for automatically re-training a vision net (VGC something) on a selection of labels (he does that through a commandline tool); Imagine integrating the ability to fine-tune a diffusion model this way (a convenient UI to grab a selection of image classes)

#

he used to have a GPU instance for doing that aswell but stopped (he couldn't keep that going)

queen pawn
#

any1 know how i can actvate the gradient accumulation steps? my automatic1111 not showing

odd ore
#

guys I just bought collab pro(not the plus one) and I guess I can use a100 gpu s right now, So whats the benefits? and I was away from sd for 2 months, I m still using ShivamShrirao s collab what do you recommend?

fervent thunder
#

same as graphic designer not being a designer.

odd ore
#

exactly

signal nest
#

IMO - depends on the quality of work produced and how non-trivial it is to create. Someone doing their own finetunes to make unique works, or using 'img2img' could be an artist, whilst most people prompting are not.

#

photography can be art , and indeed not anyone can take a good photo, it takes skill, and artistic sense, but I still dont like calling a photographer an artist. Good prompters might be in a similar bracket?

vagrant rover
#

Who asked

signal nest
#

I dont remember the name , i was too busy replying lol

vivid spear
warm junco
serene hazel
#

Anyone experimented with KLMC2 animation by katherine crowson? Im having trouble with the video showing on animations larger then 1k frames

glass flint
#

code

cobalt onyx
#

hi

severe field
lofty parcel
#

A man ate gas station sushi generated by SD 2.1. This is how his organs shut down.

charred hornet
#

hello ! can we make a AI generated image including a png file hosted in our PC ?

sleek osprey
echo tulip
#

@warm junco Thank you so much for helping me last time, so far SD seems to be running good. I have a question tho is there a method to increase the batch count to over 100 without changing the batch size?

warm junco
winter pike
#

you need to download the stable diffusion .ckpt file from huggingface too

winter pike
young ivy
#

Am I the only one that's got UI problems with auto lately?

charred hornet
valid river
#

Why does midjourney create so much better results than stable diffusion?

winter pike
#

they have a lot of stylization behind doors, i can literally put "fsjdhfkjdshfkjhfsd" and MJ will give me a pretty image

#

not a fan of it, i prefer the freedom of SD

foggy rivet
#

Hi guys

#

I want to know, what does vanilla stable diffusion means? I am pretty new and I was seeing this term many times, thats why.

fervent thunder
worn forum
signal nest
worn forum
cobalt sundial
#

hello guys, did you have a link to get best prompt for stable diffusion v2 ?

signal nest
# worn forum The problem: I add a label and click **Add**. It then adds the placeholder text ...

thanks. Honestly putting an issue on GH showing some interest is great. Personally I exclusively his his 'Unified mode' where you can switch labels and see whats already annotated (and you can still add labels without actually annotating). I know that it only lets you add any text if you signed up. One other thing is if you have any photos lying around it would be awesome to donate them there

worn forum
#

Sure, I'll do. Thanks

foggy rivet
arctic osprey
#

How come my prompts are not turning out good for Stable diffusion offline?
Online it was better
is it gpu power?
I have a 3080ti

river maple
arctic osprey
#

I dont remember which, it was SD 2.1 though

waxen pasture
silent kestrel
#

So how do you utilize seeds again?

#

And change prompts for them

crude radish
arctic osprey
#

It was the model

#

Are there better models for NSFW images?

waxen pasture
arctic osprey
#

What about violence? I am making an adult comic.

waxen pasture
#

Ahh, for making a comic you might want to train a custom model to keep a consistent style and characters

arctic osprey
#

Im not understanding how to make a good prompt. Do I describe the scene first then the action or the other way around?

#

Ive seen some huge prompts on reddit and I dont see the process being used to compile them

#

Is there an order?

warm junco
main sparrow
#

Has anyone using auto1111 noticed that if you do large batchsize with a batch count of like 6 while using embeddings, it creates unreproducible images?

arctic osprey
#

Ok why are my images now turning to black screen right after its done?

#

I downloaded and used a new model

winter pike
arctic osprey
#

how do i disable

winter pike
#

it can be deactivated

#

youre using auto1111? it doesnt have nsfw filter no more so i guess thats not the problem

#

are you using sd 2.x? maybe its the yaml file

arctic osprey
#

Im using arteros

#

aerosneavo

#

hmm, it fixed itself

#

Weird

junior badger
#

posted a script to convert .safetensors to .ckpt in #🤝｜tech-support for those who may need/want it

vernal ore
#

I have an egg microphone

#

It looks like an egg

astral goblet
empty oar
#

did anyone else lose the ability to preview generated images in A1111?

warm junco
arctic osprey
#

how difficult is it to make SD models/

crude radish
mint mirage
#

hello everyone

#

i cant seem to find out a way to either download or use stable difusion

#

i was wondering if anyone could help

astral goblet
mint mirage
#

ah no wories i figured it out

empty oar
#

I seem to have a ton of bugs with A1111 today after accidentally upgrading.

When i use 'highres fix' A1111 crashes. The image gets generated, but no previews. I also need to re-start the cell that re-launches the webui

#

anyone else having a lots of buggy issues?

astral goblet
empty oar
#

sigh

#

thanks

astral goblet
#

alpha software thats at the bleeding edge. pff. am i right? or am i right?

empty oar
#

which version do i have to revert to, to get rid of these changes?

astral goblet
#

might as well just get used to them. it'll be fine in the long run

#

you don't wanna be stuck with old versions when new stuff comes out every week

empty oar
#

i can't re-start webui each time i do a highres fix, can i?

astral goblet
#

you just need to use highres differently. before, it automatically used to figure out what the initial small size was. now you do a small size generation, and when you tick it, select how many times you want to scale it up by. default 2. i usually go around 1.4 or 1.5. you can get the output res you want with a little fidlling

#

i dont like the change, i prefer that check box automatically doing the small resolution calculation itself, but times are changing and this new hires option provides many new tweaks

empty oar
#

also the progress bar is gone?

empty oar
#

appreciate your help!

astral goblet
#

rtx5000?

#

but yeah, turning down the multiplier helps

empty oar
#

yup

astral goblet
#

i set mine to 700ms. 0.7 seconds. not so fast that it red lines the polling, but fast enough that it updates often enough

pearl steeple
#

Hey guys, hope all is well. - I just installed Stable Diffusion 2.0 and Dreambooth - But I am running into this error every time I try to train using Dreambooth -ERROR: Exception training model: 'No executable batch size found, reached zero.'. - I have --xformers installed, I'm using 8bit Adam and fp16 - I'll appreciate any help to fix this - thanks

fervent thunder
#

EVs are not better for the planet when you are Digging up the planet to get the minerals metals and everything else for the battery's and when the battery's die where do they go a waste land?

#

@storm veldt

storm veldt
#

wot

fervent thunder
#

Alex Jones Is RIght

storm veldt
#

i have no idea what you’re talking about

warm junco
fervent thunder
warm junco
#

@fervent thunder if you make anime pictures i can recommend sd-webui-tagcomplete (for booru tags),
for using Lora you need sd-webui-additional-networks

mossy fulcrum
#

1.4 on huggingface is broken

rich sand
#

i like how my computer is extremely slow to generate images using stable-diffusion

#

and i'm not joking

#

i really like it

#

when i start generating the images, i just keep waiting to see what will come out

#

it's exciting, lol!

waxen pasture
#

Yeah but how long are you waiting?

rich sand
#

uhm

#

40 mins

waxen pasture
#

For one image?

rich sand
#

yes..

#

🥹

waxen pasture
#

What resolution?...

rich sand
#

400x400

waxen pasture
#

...

#

what graphics card do you have

rich sand
#

integrated graphics card

#

and i'm not using it

waxen pasture
#

CPU?

rich sand
#

yes..

waxen pasture
#

Hmm

rich sand
#

4ghz

waxen pasture
#

See

#

That's quite slow

rich sand
#

ikr 🥹

waxen pasture
#

I don't think I would be patient enough

#

I wonder how long training would take...

#

The thing is, if you mess up or your image sucks that

rich sand
#

well, at least it's fun, i watch every image slowly being formed and i just keep trying to think how it will be

waxen pasture
#

is 40 minutes wasted

rich sand
#

yeah that's the fun

waxen pasture
rich sand
waxen pasture
#

What's the step interval?

rich sand
#

5

waxen pasture
#

Do you have it displaying at full resolution?

rich sand
#

i forgot the config, but it's the poorest one

ebon pewter
#

13 yrs old me 20 years ago had to wait 15 min for a game to load~ 😅

arctic osprey
#

how do I keep a character and make him/her do other things?

rich sand
waxen pasture
arctic osprey
#

how do I reload the character?

#

do you mean a new model?

waxen pasture
#

What do you mean by reload?

arctic osprey
#

I mean the ai needs to know what the character is

ebon pewter
arctic osprey
#

Embeddings

waxen pasture
# arctic osprey do you mean a new model?

Doesn't have to be a model, embeddings are pretty good as well. I am currently working on an embedding of a character from a video game that is coming along well.

rich sand
arctic osprey
#

How do you do an embedding? Can someone give me a link to a yt video

rich sand
#

i heard abt that old computers had a button to "underclock" the CPU so it could run games and programs on a specific FPS

waxen pasture
rich sand
waxen pasture
rich sand
#

the turbo button

waxen pasture
#

My laptop is the Acer Helios

#

It has a turbo button to slightly overclock it

rich sand
#

is it old?

waxen pasture
#

No

rich sand
#

oh

waxen pasture
rich sand
#

i've never seen computers from 2010+ with turbo button

waxen pasture
#

Well now you have (if you looked at the pictures)

#

The button is on the top left above the keyboard

rich sand
#

very interesting 🤔

#

if i get enough money i'll get one of these

waxen pasture
#

It runs sd pretty well, but ever since I upgraded my PC I have been using that instead

#

Same GPU but one is for PC and one is for laptop, so the PC one is better.

#

I upgraded from 1070 to 3070

rich sand
#

Well yeah technically PCs have more space to fit stuff

waxen pasture
#

Yeah everything is so much faster too, like loading times when starting up the program and switching models

rich sand
#

but also laptops are very portable, but i don't think someone would need an super powerful laptop all the day and keep moving it, lol

#

(btw)

rich sand
waxen pasture
spark kite
thin bough
#

Heyo, I have some troubles with installing, where should I ask?

peak orchid
latent flower
thin bough
#

Alrighty

thin bough
latent flower
thin bough
#

Still feels rude to just throw my problem in there

latent flower
#

no problem, most ppl do that

thin bough
#

ah...

#

Then I'll just do that

peak orchid
#

How do I opt in to helping LAION with Stable Horde? I don't see the option in the online UIs I've checked out

violet garnet
#

does Stable Diffusion have a Chat like function as well?

thin bough
peak orchid
#

Really wish Discord had an exact search option, rather than the terrible autocorrect / similar content search that it currently does

brazen obsidian
#

hola

#

soy nuevo en el servidor

#

😄

#

espero y llevarme bien con ustedes

#

y conocer muchos amigos

winter pike
#

buenas bienvenido

latent flower
brazen obsidian
#

gracias por tan bonito resivimiento

#

❤️

latent flower
brazen obsidian
#

espero y llevarme bien con todos ustedes amigos

#

👍 🥰

junior badger
#

so when is the server admin going to take an art class and see the hypocrisy of censoring art

peak orchid
#

How far along are Stable Diffusion VR projects? I have an education related project that could benefit from such a thing

peak orchid
junior badger
#

I can't even post a picture of a back, yes a back, are backs getting ppl all re-tard strong and sex crazed?

#

that's not true

#

i have a server

#

nice cop-out though... I guess @peak orchid

#

Parenting what ppl see online only ends up in disaster, my opinion
and is also no ones job but the parent

#

but, i guess i'm barking up the wrong tree

#

I just get tired of living in a bubble wrapped fake society of whiners and complainers that ruine life for everyone

#

imo, ppl need to be offended sometimes

#

it's healthy

#

"We're gong nowhere, somebody help me, somebody help me, yeah"

#

"it's ok, you can look the other way"

#

these dudes understand

#

lol

fervent thunder
#

my yaml file is missing where can i download it?

junior badger
#

which one?

#

i can post it here

#

i have a few diff yaml configs i use

fervent thunder
#

for 2.1?

#

i forget how it works

junior badger
#

would you like the one i am using?

fervent thunder
#

that would be great ty

junior badger
#

it works by putting the yaml file in the configs dir

#

or specifing the yaml file at launch time

#

--config=blablabl.yaml

fervent thunder
#

im using automatic1111

keen prism
#

hi everyone!

fervent thunder
#

are yamls generated or downloaded on install?

#

no idea what it is

keen prism
junior badger
#

i'm using invoke, it shouldn't matter, if it does and they fail to work, just delete them

#

you can also get the file by vising the github repo lol

dawn yarrow
#

after school? har har har har har

junior badger
old marten
keen prism
#

bro

#

smh

#

here im not even kiddin

junior badger
#

i posted a tool here already to check your ckpt's for viruses

#

ya

dawn yarrow
#

im 42 after school is like a long time ago hahahaha

keen prism
#

why cant i paste photos

junior badger
#

im fkn cool like that 😏

keen prism
#

after school or work

#

idk it wont paste

dawn yarrow
#

work? nope im retired

old marten
keen prism
#

oh

junior badger
#

no u can't

old marten
#

show me how u relax if u know what I mean 😏

junior badger
#

try

keen prism
#

well i cant i realize now aha

#

well then

dawn yarrow
#

wheres my pot....

keen prism
#

do it if u can

#

if not

#

theres the link

junior badger
#

Hi fives

dawn yarrow
#

ok tiana i think its cool

#

go gfy!

keen prism
#

yo thank you

#

btu like

#

i dont even wana do that shit

#

its for a school project

#

i am so fucking behind

#

so i would apprecieate it

junior badger
dawn yarrow
#

they have you spamming links on discord for school well that sucks

keen prism
#

if i could get some info

#

or data

junior badger
#

"with images"

keen prism
#

please

keen prism
#

and

#

i assumed this would be easier

#

than emailing my entire school

#

or individually going

#

need an age range anyways

dawn yarrow
#

ok ok calnm down its ok

#

must be due tomoorow

#

lol

keen prism
#

ahah

#

day after acutlaly

junior badger
#

uhhhhhh i posted the yaml files

#

☝️

night marten
#

hi, did you found a solution ? If not can you tell me where is the cache folder you are used to clear ?

tidal otter
#

does anyone have any recommendations for managing large image collections? I'm trying out Hydrus, but it seems to not currently support searching/tagging images based on exif information from stable diffusion

#

specifically, I want to be able to pull each word out of a prompt as a unique tag, along with the other data (seed, cfg, sampling method, model, etc)

junior badger
#

@tidal otter have you tried asking this same question to ChatGPT?

formal spruce
#

hey

wraith sky
#

I use it on my laptop and have over 30,000 of my images generated from Midjourney and Stable Diffusion plus 40,000 reference images I've collected over time. Great for doing my own art I can pull up the exact references I want.

neat condor
#

--reinstall-xformers doesnt work some reason, it says ansformers were discontinued and can't be reinstalled, what's with that?

#

Launching Web UI with arguments: --api
E:\AI\SD11282022V2\stable-diffusion-webui\venv\lib\site-packages\pkg_resources_init_.py:123: PkgResourcesDeprecationWarning: ansformers is an invalid version and will not be supported in a future release
warnings.warn(
No module 'xformers'. Proceeding without it.

steady tapir
formal spruce
#

Why can’t I save any image from the web UI

#

It’s all broken

cobalt cliff
#

is this it for stable diffusion? surely the lawsuit has put a lid on further releases? we won't see updates for at least a few years, maybe never?

neat condor
marble plaza
#

is there any good model generating aesthetic human image based on SD 2.X?

neat condor
formal spruce
#

hey

#

ht_load_location)
File "C:\Users\Zayd\Documents\stable-diffusion-webui\modules\safe.py", line 106

#

anyone know how this can be solved?

# 
RuntimeError: [enforce fail at ..\c10\core\impl\alloc_cpu.cpp:81] data. DefaultCPUAllocator: not enough memory: you tried to allocate 117964800 bytes.```
real crag
#

Guys where should i paste lowvram string? And how exactly it should look like?

#

--lowram?

warm junco
# real crag --lowram?

Yes in the webui-user.bat after the Command_ARG=
You need --lowvram
but if you habe 3-4gb use --medvram

real crag
warm junco
real crag
#

Arg=--lowvram

junior pollen
#

I’m new to this so appreciate any tips…
How do I upload an image to act as part of a prompt?
Thanks

warm junco
junior pollen
#

Can I upload an image here in the server in a dreambot channel?

real crag
#

@warm junco do you know any other option with low vram? :)?

#

That i can paste there

warm junco
#

--xformers for nvidia cards

real crag
warm junco
#

Also --autolaunch to Start automaticly

real crag
#

I god nvideo

warm junco
real crag
#

So it should look like this AGR=--lowvram--xformers

warm junco
#

Seperate them with a space

real crag
#

?

#

Oh ok

warm junco
#

How much vram do you have ?

real crag
#

Unfortunately inly 2 :

#

Got gtx 950, i5 cpu

warm junco
#

Ah okay yea then lowvram and xformers is your only hope

real crag
#

I generated some decent images 🙂

#

Already

warm junco
#

Ahh okay ! Nice that it works

real crag
#

You know what works great? For example set pose in blender (or maybe any other 3d modeling program) then use inpaint or img2img

#

Also opera browser support bigger images than Google chrome

warm junco
real crag
#

I get crash at +512 in google and can support 800+ in opera

warm junco
#

Wouldnt use opera as they been bought by a chinese company

peak trench
#

Hi All, just discovered Stable Diffusion and i can see the insane ( maybe scary?!) potential! Happy to join the Discord

real crag
warm junco
real crag
#

Im draftsman my self but have 0! anxiety about ai art, its great helping tool

peak trench
#

I'm a musician and im not scared of technology i see it as evolution ( i just released a movie on the subject concerning music which has obviously drastically changed with tech and samplers too) Still it's a tsunami in what art is gonna be like in the near future

neat condor
#

i'm in this industry for a while now and i lost half my regular clientelle as AI became a thing and yet i was one of the first who adapted it and created new income streams asap.

#

I think textual embeddings will sell like art pieces on itself in the future, basically removing the need for you to reiterate your work and nail what client wants, leaving that part to them.

real crag
#

How ever there are many things that ai cant do atm

neat condor
real crag
#

Like group pictures

#

Or something specific and exact

peak trench
real crag
#

You can have variations of your ideas generated but not exact pixture from your immagination like in drawing

peak trench
neat condor
#

I think at some point it will achieve those things however i don't think it is of any importance at all, when u can photobash in free image editor softwares.

real crag
#

Better example would be actions figure or 3d model that you can pose however you want, in ai art you cant do that atm

#

Not enough controll

#

But, of course i hope this will be fixed

#

In future 🤠

neat condor
# real crag You can have variations of your ideas generated but not exact pixture from your ...

again: photobash, i think every AI artist who wants to push the boundaries must learn how to photobash+ai generate together instead of only prompt-crafting and i think photobashing and personal tune-up touch will always be needed regardless of how good AI will become. even now automatic1111 added sections like (sketch and sketch inpaint) which kinda requires you to have proper drawing tablet. so it's not really "free art on the click of a button", atleast if u want to push for perfection.

formal spruce
#

i cant fix it

#

ive been trying al day

real crag
formal spruce
#

I cant even get the webui to boot

#

its rediculous

neat condor
real crag
#

Use opera browser

formal spruce
#

i have it on linux and the photos dont save

peak trench
#

as a musician allow me to share my experience: with samplers and keyboards in the beginning i was thinking: its great because with a keyboard i can play any instruments except strings / guitars and voice ( and im a guitar player and a singer so happy like this). Today i often use a guitar sample instead of playing it because the quality is greater than what i could record in my studio plus i save a lot of time. Srings...I can play an philarmonic orchestra with 3 fingers and my headset at 3 am in the morning :-). Voices? lead vocals we're not there "atm" but virtual choirs are stunning ! So again I am bullish on tech but clearly it's going to change the way we create big time

formal spruce
#

the photos dont even render properly

#

its all a buggy mess

real crag
formal spruce
#

this is nuts

neat condor
# peak trench as a musician allow me to share my experience: with samplers and keyboards in th...

I'm musician aswell and yeah, i agree. at the end of the day it all boils down to - what looks good, what sounds good, people who appreciate music or piece of art - don't care where it came from, people who appreciate creation story behind the art or music - pay for that story, not the piece of art or music itself. I was quite clear with myself on that regard that i want to sell music and art pieces, not the stories of my struggles whilst creating them. However i can understand why someone would want human touch and be a naturalist.

formal spruce
#

imma go switch to linux and see if its broken

real crag
neat condor
formal spruce
#

its working on linux

#

before i had plasma 5 and it was broken

#

switched to xfce4 seems good

cobalt cliff
neat condor
warm junco
#

Someone know if there is a OpenSource ChatGPT alternative in the Works ?

tidal otter
formal spruce
#

what are the good anime models

warm junco
formal spruce
#

i dont have a browser

#

I know that sounds ridiculous

outer crescent
#

o_O

#

ok ill bite, why do you not have a browser

formal spruce
#

because I broke my pacman config

#

and i cant install anything

#

ill fix it later

#

ok that was far to quick of a fix

#

browser is installing

cobalt cliff
#

which layer of the matrix do you live in

formal spruce
#

43

plucky gate
#

Recently, the opinion has crystallized among some artists, that AI such as stable diffusion is a form of plagiarism. Also that it takes jobs out of their already struggling hands. I am not interested in discussing that here, but would like to ask you where I can find a good place to follow the developments in the public discussion about this? Any specific forums or channels that I can subscribe to?

outer crescent
#

honestly, depends which echochamber you want to deafen yourself in. neither are good places for debate and the other side will be rabidly shouted down. you might find some well balanced videos on youtube, probably one of the better places right now to get both sides any way evenly told. but you're unlikely to find reasoned mature debate in a discord

neat condor
outer crescent
#

yeah 🙂

formal spruce
#

what models do you guys use

#

the models i have are bad

outer crescent
#

or your prompting is weak?

formal spruce
#

what models do you people use

neat condor
#

too many to list here, what do u want to create?

formal spruce
#

I want to create things like my pfp

#

and i want to create realistic things

neat condor
#

well then protogen is good start

formal spruce
#

alr

fringe hinge
#

I have problems accessing huggingface is it working for you guys?

formal spruce
#

link?

formal spruce
neat condor
#

check civitai for it

fringe hinge
neat condor
formal spruce
#

um there is no protogen thing

#

it doesnt exist

main sparrow
# formal spruce the models i have are bad

Start with the 1.5 model and getting interesting prompts from sites like lexica.art, then move on to 2.1 and learn your negative prompts since it looks like that's where things are heading

fringe hinge
formal spruce
#

whats a negative prompt

rotund umbra
#

Hi everyone, I am frightened cause of lawsuit against AI art

#

Any Ideas how to archive all models till they are deleted ?

neat condor
outer crescent
#

lol, hiding models? wouldn't that make them.....Rogue AI? 😄

rotund umbra
#

Xd)

outer crescent
#

the lolsuit is good, it'll mean stuff is clarified moving forward. the AI police arent going to put your computer in jail just yet

neat condor
#

lolsuit - is great name for it

rotund umbra
#

But It can do so in near future ((

elder cradle
rotund umbra
#

Or It can make to ban a lot of models

neat condor
rotund umbra
#

that are great

rotund umbra
#

Closing cvitAI

formal spruce
#

why would ai art be banned

rotund umbra
#

cause most of models are trained with pictures of some artists

neat condor
rotund umbra
neat condor
#

not really, if civit ai closes other guys with the same values will open it outside of US

#

result will be the same

#

i don't even know whether civit ai is in the us or not

rotund umbra
#

But lots of models will be lost (((

neat condor
#

well best thing i can advice, save textual embeddings

#

as for models, if u have $$$ to buy around 50TB hard drive space, sure, knock yourself out.

#

that would be enough to save best models.

rotund umbra
#

But I have no money for hard drive or large cloud to save lots of models

#

(((

neat condor
#

yeah then textual embeddings is the way.

rotund umbra
#

Well, I will try load out all memory on my pc with best models

elder cradle
#

Or make multiple GDrive accounts

neat condor
#

imo it's just waste of your time

#

i think in 1 year we might not even be using ckpt's or safetensors but something much better and faster

#

with realtime generation speeds

rotund umbra
#

Well, my pc is not powerful enough tho

neat condor
#

if past statistics is any indicator of how AI will evolve in the future, i think 90% of what we use today will become useless. so i wouldn't worry too much about it.

rotund umbra
#

I don't think I will be able to take advantage of such techs

neat condor
#

one more thing u can do is create small datasets of images, like generate 100 image per model, so in the future when u can afford it, u can train and revive models based on those datasets.

rotund umbra
#

plus I really doubt that If the artist pool will be much shortened it will be as good as it is now, like, anime style will take great hit

#

cause a lot of artists against AI are in anime style

#

So It will shorten pool of art for training anime style a lot ((

neat condor
#

that's in case they can push the narrative that AI models infringe copyright which is again - pointless panic.

#

because if they can prove that "stealing a style" is a thing - then there will be thousands of lawsuits after that - where artists will sue artists and AI will be the least of their problems.

fervent thunder
neat condor
#

we're basically watching someone aiming a gun at their foot very slowly and playing russian roulette with this lawsuit - i don't understand how can you really think that it will result in banning AI models. 😄

rotund umbra
#

Well, I will still secure myself , and download best models from cvitai any hugginface

#

to be safe

neat condor
#

okay, if those terrible terrible things u predicted happen, i know where to ask for models 😄

rotund umbra
#

( and yea, maybe I am over exaggerating , but I living in stressful informational space . I am Ukrainian, and furthermore OGL scandal is going )

neat condor
#

OGL?

rotund umbra
#

Did you not hear about WOTC trying to revoke OGL 1.0a license ?

#

That is pretty loud rgn

neat condor
#

ah that

fervent thunder
#

lol

neat condor
#

yeah i kinda overheard it but i never created anything DnD so never rlly bothered to dig deeper.

fervent thunder
#

I like how she said like it's common knowledge

rotund umbra
#

Well, the thing is it may cause to stop pathfinder from releasing and developing

neat condor
neat condor
#

i kinda enjoyed first two games, idk about tabletops.

rotund umbra
#

mutant roleplay ( I don't really about it, but still it is concerning one of my friends , and it is still sad )

rotund umbra
fervent thunder
#

Gosh I wish my concerns were related to dnd

neat condor
#

can't decide which one's better, however back to our conversation, since u're ukrainian, u should be even more desensitized towards the drama of this lawsuit. karla ortiz, person who has more money than us, more standing and us etc. etc. acting as starving artist is LEAST of what should be worrying you. I'd just take a deep breath and relax, idk, i'm just chill like that.

rotund umbra
#

And yea. I really like kingmaker and wotc

neat condor
#

that's cool, but everything else is not. you can't live your life in fear, i kinda vaguely remember when i was like you though. i was saving every photoshop brush in case they became rarity which they did.

#

had around 2TB folder for them, however AI came out and now all i use is 3-4 photoshop brush pack i really like, deleted everything else. 😄

#

but well, just for your peace of mind, u can download all the textual inversions there is, they're no less good than models.

rocky beacon
#

what is the ''VAE-approx'' and how is it different from the normal VAE file?

plucky gate
plucky gate
plucky gate
neat condor
plucky gate
#

Ah, ok! Thanks for your opinion and the pointer.

plucky gate
neat condor
#

reddit is reddit afterall.

outer crescent
#

everyone in the peanut galleries on this issue is insufferable

#

entitled nerds versus deluded deviant art types

#

wow, thats a new record for me getting tired of a gimmick.

elder cradle
#

Looking at ur pics, I think you should have a bunch of pics enough to make a decent Dreambooth model

past robin
upbeat tangle
#

another lawsuit against stability

#

do these people just wanna drain stability from the legal fees or something cuz they sure arent winning it

rocky beacon
#

so, I upraded the auto1111, to the last version and now it seems like I'm getting a lot worse results with the exactly same seed, model, vae, etc, any idea why?

neat condor
rocky beacon
vagrant rover
#

Why is there so much ai porn :(((((

cold siren
#

heh why do you think there woudln't be

abstract pivot
#

Are there any articles on Distilled diffusion, how it works and what to expect?

#

Will it work with older models etc

fervent thunder
#

Can I download Stable Diffusion with Intel Integrated graphic

cold siren
#

iirc, the whole idea of distillation is that you take an already trained model and optimize the shit out of it... which also might mean that you won't be able to finetune it later

lilac reef
abstract pivot
fervent thunder
lilac reef
winter pike
fervent thunder
#

I feel like a model or something will release today

#

Im getting those kind of vibes, maaaan

winter pike
#

i hope youre right

old marten
#

Is there someone with a 3090 or with 24GB VRAM who can do a quick test for a brother in need?? 🥺

wide juniper
#

Is there a way to make stable diffusion draw on alpha yet?

plucky gate
fervent thunder
#

How do you get text to render coherently?

#

you dont

#

maybe in the new model

#

Fingers crossed

#

I think I saw some preview for one of the new models on twitter that did render text

#

so maybe it will be a reality in the next few days/weeks

winter pike
#

the deepfloyd model

#

it gives me dalle 2 vibes

#

in a good sense

fervent thunder
#

Yeah I think it was the Deepfloyd

wise flax
#

#nsfw why no nsfw channels here?

#

atrocious.

abstract pivot
wise flax
#

i am there

abstract pivot
#

Ok

brittle fractal
#

holly, i didnt even notice that my SD diffusion was hardly underperforming. i followed a git issue and went from 4-5 it/s to 20 it/s on my 4090
its just night and day

fervent thunder
#

heaven forbid theres a singular place on the internet that doesnt have nsfw

fervent thunder
wise flax
brittle fractal
astral pivot
#

can someone help me with stable diffusion

winter pike
#

from 30 seconds per it to 1:40

neat condor
#

to me too, clean install helped

true dawn
true dawn
#

The 4090s will be underperforming for the next year

#

Not everything is optimized yet, and will continue to do so past a year

#

These changes also affect the 3000s cards

#

Makes them a lot faster, use less vram

#

Be careful though, it might break your training

fervent thunder
#

yeah the cudnn made a big difference for me on my 4090

pale blaze
#

Did anyone using stable diffusion for generating Logo?

brittle fractal
keen narwhal
#

So it is an issue that generations now take twice the time than it was before

#

Didn't use A1111 10 days after the hires fix and CUDA errors. And now this

#

I have a lowvram card but had good performance 31-th of December and then, every new update everything was darker darker for me 🙂

brittle fractal
#

still some people say they get 30 it/s

lofty moss
#

I know this must be the most asked question here, but could someone tell me how I download stable diffusion?

brittle fractal
#

my rig is pretty decent, wonder how the hell they have 50% perf increase like that

lofty moss
keen narwhal
#

I used to make 1 image at 30-40 seconds. 512x512, no hires fix, 30 steps. Now it takes 3 minutes

fervent thunder
#

ouch

keen narwhal
#

i use --api --xformers --lowvram, dunno if I should add anything

fervent thunder
#

fresh install might help but idk

peak orchid
#

Getty Images is now apparently trying to sue Stability AI. They are the same assholes who forced Google drop the "view image" feature in their search. Really hoping that Stability doesn't bend over backwards for Getty as well

fervent thunder
#

they bent over for everyone else it seems with 2.0

#

Getty has enough clout to make them tremble further Im sure, theyre not just some low-rent photo site obviously

peak orchid
#

Getty has enough money to drain their opponents resources on bullshit lawsuits

sudden isle
true dawn
brittle fractal
true dawn
#

Are you on a xx90?

brittle fractal
#

i9 10900 32gb rtx4090

true dawn
#

You should be able to render batches of 25 now pretty easily if you are on 24gb

#

makes working so much smoother

#

also upscaling takes like no time now

sudden isle
#

I looked it up and Getty Images stock went down like 80% over the winter

magic widget
#

Getty can go to hell. They abuse the legal system to send people shakedowns.

#

They're also hypocrites who regularly blatantly violate copyright

#

I hope Stability AI came into this prepared for a fight.

#

I also have to laugh that they're portraying a violation of their EULA (by scraping content) as a matter of law when it has already been well established that scraping is legal and EULAs are bullshit.

#

Oh gosh, I didn't know "copyfraud" was a thing. These bastards send people cease & desist letters over images they falsely claim copyright for.

#

Please put Getty where they belong, Stability AI? 🥺

sudden isle
#

I'm looking at their financials right now and it's not looking good for them brev, Getty's up to their eyeballs in debt

dawn yarrow
#

copyright schmoppyright, you didnt draw it

#

if i draw bart simpson or garfield should i get sued as well?

#

copyright is kinda pointless if im not profiting from it

timber dock
#

Whats good y’all

dawn yarrow
#

whattup

timber dock
#

Just found out there was a discord for stable diffusion

#

I was legit just scrolling through Reddit this entire time

dawn yarrow
#

oh yeah this is the place man, i hear there is an "unstable diffusion discord" as well

timber dock
#

Is that more home run instances?

dawn yarrow
#

idk i havent gone that deep down the rabbit hole yet 😉

#

something something about getting around nsfw filters? not sure

primal terrace
#

Hi everyone 👋
Has there been any news on a release for 2.2, or something like that?

sleek osprey
#

does fp16 give worse images than fp32

timber dock
#

Oh mb my reading comprehension is trash

fervent thunder
#

they need it, 2.0-2.1 are seen as duds by most people

dawn yarrow
timber dock
dawn yarrow
#

ok cool

timber dock
#

I find the idea of subscriptions insulting so I immediately went for that

foggy atlas
#

What are the best sources to stay informed about Stable Diffusion?

#

I tend to come here from the beginning, but lately the announcements are just contests and in the other channels the information is drowned out with uninformative stuff.

fervent thunder
#

probably the reddit i guess

dawn yarrow
#

well i mean what do you need to know

foggy atlas
fervent thunder
#

yeah part of the problem is, theres not much to talk about until theres a new model, and then its everywhere

primal terrace
foggy atlas
fervent thunder
#

if you ever see emad in here and ask him when a model might release, he'll often say 'itll be done when its done' which is fair, and i just mention it to say, they dont really give us progress updates so theres not much to do but speculate until you get an actual model

#

unless its something like they got more people to invest in them

sleek osprey
sudden isle
fervent thunder
#

i think its because they self-neutered so much of it

#

people can get some cool things with 2.1 and embeds, but for me 1.5 photo models still give more pleasing results (and quicker)

peak orchid
fervent thunder
#

to whom it may concern,

I am wondering who i can speak with that's involved with the stability staff.

I have some questions about the lawsuit. I am helping my firm see if we are liable for any damages.

dawn yarrow
#

SUREEEEEE

fervent thunder
#

I'd look like an idiot

dawn yarrow
#

well theres that.

#

tag some mods

peak orchid
fervent thunder
#

we run SD for work

dawn yarrow
#

idk why you so scared

fervent thunder
#

my boss

dawn yarrow
#

let him worry about it then

peak orchid
#

The lawsuits are likely to take years before anything substantial happens

dawn yarrow
#

they cant stop it

#

SD with impunity

peak orchid
#

Getty apparently hasn't even officially filed their lawsuit either

fervent thunder
#

I mean

dawn yarrow
#

i said go

fervent thunder
dawn yarrow
#

dont you have some images youre supposed to be generating as we type?

fervent thunder
#

we just want to make sure our bases our covered

dawn yarrow
#

i say its paranoia

#

you cant stop real tech from emerging

fervent thunder
#

I agree with you lol

#

either way

#

my job is to make sure purplesmart won't get shallacked

dawn yarrow
#

this is only recent "threatening" news so id say they can run away in 3 years scott free

#

you know this red tape legalities take so long

fervent thunder
#

we'll probably be fine

#

but I am playing it safe

#

and I am being PAID to be paranoid

peak orchid
#

On an unrelated note, what do we have for VR related Stable Diffusion projects right now?

#

I have an academic institution that's really interested in that sort of thing right now

mortal hornet
#

has anyone gotten Textural Inversion to work? I'm trying and failing 😄

elder vault
magic widget
#

It's also going to be very jurisdictional, bet on politicians trying to make all kinds of points.

#

Now is the time to rattle everybody's cages about how intellectual property is bullshit.

#

Something tells me Getty's angle is just to move in on Stability AI's revenue. They're copyright bullies, they threaten people into giving them money.

proud nova
#

Can someone explain me what inpaint means?

astral goblet
#

a lawyer would be able to give guidance. current copyright law has a lot of room for fair use. a lot of ai generation would fall under transformative art, since intent is so different. lets say for discussion that midjourney training data is revealed in discovery and they are infringing on high quality copyright materiial they had no license to and they get shut down. that wouldn't mean all ai art is copyright infringement.

Ianal but i have a lot of experience with ip law and studying it so that i can cover my ass. as any professional creative person should

dry trellis
proud nova
#

Sounds good. Thanks for explanation

keen narwhal
#

can anybody help me with "git revert" thing? I try to undo A1111 version with it but I always get an error that I don't understand. 🙂

#

I already had some tries and now there are even more lines than there were before

#

C:\SUPER SD 2.0\stable-diffusion-webui>git revert 9cfd10c
Auto-merging webui.bat
CONFLICT (content): Merge conflict in webui.bat
error: could not revert 9cfd10c... repair #6612
hint: After resolving the conflicts, mark them with
hint: "git add/rm <pathspec>", then run
hint: "git revert --continue".
hint: You can instead skip this commit with "git revert --skip".
hint: To abort and get back to the state before "git revert",
hint: run "git revert --abort".

#

I get this when I try to revert

magic widget
#

Follow the instructions in the message you pasted

#

You could probably also git stash, do whatever with the repo, then git stash pop to move your changes back

keen narwhal
#

In first line it asks to resolve the conflict. In second it asks something I don't even understand. 😄

#

"pathspec" is something specific or I just need to type it?

magic widget
#

This isn't really a git chat

#

or an a1111 chat

keen narwhal
#

I thought it's everything related to AI and generating. Sorry.

gritty ivy
#

https://stablediffusionlitigation.com/ Rather, they are anti-stable diffusion more than all image generators, I think the open-source aspect bothers them. They also have some very foolish explanations for how AI works.

sudden isle
#

I have a feeling the legal defense is gonna use ChatGPT or something for these lawsuits

astral goblet
#

i dont think getty's lawsuit is "anti ai" because they're building their own

#

luckily we live in a world where reasonable minds generally prevail, and people who rush to their own expert conclusions that they did their own research for, don't matter

astral goblet
#

some people squirm anytime a1111's is mentioned

tawny mauve
#

do models not trained with stabilityai/sd-vae-ft-mse benefit from it?

vapid tangle
vagrant rover
wraith briar
#

Is stable diffusion the same as automatic1111

astral goblet
wraith briar
#

I am currently using the stable diffusion webui

#

Would it be wise to delete it, update python, and just use automatic1111

#

because my current python version is 3.10

tawny mauve
#

automatic1111 made the sd web ui

wraith briar
#

and the tutorial im watching is telling me that automatic1111 uses python 3.10.6

#

oh

tawny mauve
#

automatic1111 is a developer who started a project called "stable-diffusion-webui"

#

people refer to them as the same thing though

wraith briar
#

huh i didn't know that