#💬｜general-chat

1 messages · Page 38 of 1

quick narwhal
#

Trying out some nasty shit first

empty knoll
#

lmao

quick narwhal
#

Horniness is the best motivator for learning new tools and skills

empty knoll
#

hey if its motivating lol

quick narwhal
#

I already know I want a way to segment an image so that one portion is depth based and another is openpose based

#

Fucking around with a picture of a tree would not have helped me realize that lol

empty knoll
#

fair lol, memes motivate me, or turning gifs of michael scott into linus from LTT lol

quick narwhal
#

ctrlnet is way nastier than pix2pix

empty knoll
#

lmao

trail ledge
#

in another couple of weeks, something else will probably come along and blow controlnet out of the water 😄

empty knoll
#

right lol

#

kohya is pretty cool too for training

trail ledge
#

I love the speed of development in SD right now

quick narwhal
#

I'm so down for it

empty knoll
#

smaller Lora files to use with additional networks

#

lora extraction

#

I wanna play around with that, I havent yet

quick narwhal
#

It looks like openpose doesn't wanna believe that certain configurations of limbs are possible

empty knoll
#

might be worth trying canny or depth then

quick narwhal
#

and kinda just gives up

#

I see a lot of people using canny, but I'm not sure what that model is interpreting

#

so far I've only played around with depth and pose

trail ledge
empty knoll
#

make your pose here, then use the screenshot tool on it to make an image, then process that for openpose

quick narwhal
trail ledge
quick narwhal
#

Let me fire up canny for a spell

empty knoll
#

That also has premade scenes you can play with and adjust

astral goblet
#

even without alkl the revolutionary new features coming out every few days, people need to chill out on thinking stuff is dead or slowing down. tech is 6 months old lol

#

i love so much that windows 11 lets me run a terminal quake style

#

"wheres the term again.. OH yeh! start+`

fiery pike
#

Hey all, does anyone if there's been a similar Dreambooth exploration for what are good Face hyperparameters in 2.1? Or is textual inversion / LORA the better way to go for this?

#

I've been trying some dreambooth Faces with the same parameters they have for v1-4, but it does not seem like they carried over...

grim crown
#

hello guys

#

can you help me?

#

i how to find automatic1111 docker image link?

#

please help

#

thanks so much

static snow
#

Ok guys, every tutorial I watch and follow give me the worst results when I try to train a model or anything ... I did Dreambooth, I did LORA, I did Textual Inversion... I've spent hours and hours and NOTHING I do works, the results are horrible. horrible.....

#

I don't really know what else to do...

#

I have 12GB GPU but literally I followed everything step by step and this output looks nothing like me

lament steeple
#

I can get a style to work, but haven't successfully done a character yet

static snow
#

In the video tutorials, they say "follow these steps and get these results" and it's like... no it doesn't look anything like that

#

I guess I'll just keep trying but it trains for so damn long it's like wasting time 😄

lament steeple
#

Did you watch Software Engineering's tutorial?

#

it's over an hour, but he goes really in depth

#

even how to check when to stop training

static snow
#

Do you have the link to that? I'm not sure

#

I'll try anything at this point 😄

shell seal
#

can safetensors model be used the same way as ckpt models?

lament steeple
lament steeple
stable pawn
#

I still have hard time understand training of Embeddings, most tutorials show how to train a face, but what if I have 150 black and white images from a concert and want to train a model for "Ilford Delta 3200 ASA" to get the film look but at the same time train it on "punk concert" so Stable Diffusion know tags like "3200 ASA" and "Ilford", how do I associate tags with my training?

twilit panther
#

How to use the ai?

lament steeple
twilit panther
lament steeple
#

What step are you on?

stable pawn
#

@twilit panther Do you have a plan on how you want to use AI, what you want to do?

shell seal
#

what is ensd in generation settings ?

#

found it nvm,

lament steeple
#

Do you have that checkppoint?

shell seal
#

yeah

#

the model hash is same

lament steeple
#

Unless they use clip skip I have no idea. sorry

shell seal
#

oh they did,

#

what is clip skip

chilly valve
#

Morning ☕️

shell seal
#

morning !

hard osprey
#

Are there any models that can make a political cartoon style?

pastel fern
#

WTH is a political cartoon style

lament steeple
#

you could see if any editorital cartoonists are known to stable diffusion

#

I got tired of searching the db. sorry

hard osprey
dense bloom
#

Ali

#

baba

dark blaze
#

morning

broken loom
#

Hello I haven't used sd since back when it was run in a bunch of channels here. I want to run it on my gpu, what is the name of the ui that people use to run it?

bleak matrix
#

Good morning, everyone! How are we all today?

blissful flame
#

who know the best jok?

lament steeple
#

Last good joke I made was 26 years ago

#

he calls me dad

wide juniper
#

Is there any way to get my SD to start up faster?

empty barn
wide juniper
#

It's in SSD

#

Oh NVMe SSD is a thing

autumn sparrow
#

hey guys, i keep getting an error in web-user.bat? saying i need to update pip but i already updated it.

lament steeple
#

did you update it from within the virtual environment?

autumn sparrow
#

sorry i'm not a programmer, don't know what that is

lament steeple
#

you on windows?

autumn sparrow
#

yes

lament steeple
#

ok we're going to do this step by step. open your windows explorer to and go to where your webui-user.bat is

autumn sparrow
#

yes

lament steeple
#

in the address bar type cmd

#

it will bring up a command window

autumn sparrow
#

yes

lament steeple
#

type .\venv\scripts\activate

autumn sparrow
#

with the dot at the beginning?

lament steeple
#

yes

#

then type

autumn sparrow
#

wait

#

it can't find the path i'm looking for or something

lament steeple
#

what does it say at the beginning of the command prompt

autumn sparrow
#

you mean at the top?

lament steeple
#

at the beginning before the cursor

#

I want to make sure you're in the right folder

autumn sparrow
#

oh wait hold up

#

ok uhm

#

if this is what you mean it says

#

C:\users\username

lament steeple
#

ok that's not the right folder. it should end with stable-diffusion-webui

autumn sparrow
#

ok so what do i do then

sick tulip
#

Have you ever had this problem

#

AttributeError: module 'tensorflow' has no attribute 'Tensor'

lament steeple
autumn sparrow
#

idk osrry

#

sorry*

#

explain what that means pls

lament steeple
#

if you did, it will ay (venv) in front of your prompt. if it doesn't, then type

#

.\venv\scripts\activate

#

it's a virtual environment to separate it from your main python installation

autumn sparrow
#

yeah it says venv at the beginning

lament steeple
#

ok type python -m pip install --upgrade pip

dawn epoch
#

First good morning everyone!

#

When I try to run the lora that I installed locally it gives this error! what is it and how do i solve it?

lament steeple
autumn sparrow
#

sorry for not understanding but can we start from the beginning again

lament steeple
#

ok in your windows explorer window for stable-diffusion-webui, type cmd in the address bar

sick tulip
#

Oh, sorry about the mistake

autumn sparrow
#

i have windows 8.1 so it's a bit different i think

sick tulip
#

is ContorlNet

dawn epoch
#

RuntimeError: mat1 and mat2 shapes cannot be multiplied (154x1024 and 768x128)

lament steeple
#

how i windows 8 still running?

autumn sparrow
#

idk

#

tried upgrading to windows 10 or 11 but it just wouldn't

lament steeple
#

ok the old fashioned way.

autumn sparrow
#

i guess yeah

lament steeple
#

webui on your c drive?

autumn sparrow
#

uuuuh hold on

#

no don't see it

lament steeple
#

what is the path?

#

we will need the path to get you to that directory

dawn epoch
autumn sparrow
#

where do i see the path en for what

lament steeple
sick tulip
#

Never met,I don't know

autumn sparrow
#

i'm so confused

lament steeple
#

what folder were you in when you cloned the webui?

autumn sparrow
#

😣 i don't know

lament steeple
#

you have to activate the environment to upgrade pip in the environment and I don't know how to get you there in windows 8. the good news is that that is a warning not an error

#

I was just trying to help you make it go away

autumn sparrow
#

yeah i know, i just have no experience

#

i'm not smart

lament steeple
#

windows 10+ or linux I could get you there. I'm sorry

autumn sparrow
#

i'll look into it

steep dragon
#

how to open a1111 web ui if you deleted the pin? 🤔

lilac pine
#

hello! I have a question, do LoRA models work with like different models prior to what it was originally trained from? For example, if I trained a Blue Archive LoRA model from redacted v3 and used that on another anime model, would it still generate good images that adapts to that anime model?

flint fiber
leaden mural
#

How does one do a side smirk?

#

I tried, and all that it gave me was a weird creepy smile

fervent thunder
#

sounds like most smirks

thorny finch
#

/ imagine

fervent thunder
#

you cant do gens here

lunar dirge
woeful pelican
#

Does anyone know of a good free writer ai that you don't need to log into or download to use?

crisp maple
#

novelai?

woeful pelican
lilac pine
#

how do i make previews for the extra networks

astral goblet
#

Before i leave for the day, i just want to say that Windows 11's hidden feature, making a terminal window into a quake style terminal that comes up with start+`, is my new favorite thing about windows and has fully won me over. Theres no going back now

gentle otter
#

windows 11 was so bad i went out of my way to reinstall windows 10

astral goblet
#

i was on arch linux the past 2 years so i didn't get any of that first wave windows 11 stuff

#

i was using windows 10 until i got a brand new vega64 and eventually realizied that amd's shitty windows drivers were causing all my isssues and i moved to linux

gentle otter
#

classic

odd geode
#

you try your very best, really give it your all

#

wow i have early member role

#

really is an honor

muted mural
#

https://streamable.com/ticsfq very pleased with the ControlNet+EbSynth combo but can anyone help with fixing the weird glitchy artefacts caused by people moving behind the window?

vestal dew
#

I have it installed. When enabled, it doesn't run. SD acts as if ControlNet isn't there/enabled.

static snow
#

Is anyone able to use their stable diffusion to train and get images of themselves? I swear on the lord I have tried everything...

gentle otter
static snow
#

Last night I literally ran it for 100,000 steps and saved embedding every 50 steps, they all worked like crap, they looked nothing like me 😦

#

(that's 2,000 embeddings that I tried today)

trail ledge
#

you probably overtrained with 100,000 steps

#

did you go back and look at intermediate embeddings to see if there was a sweet spot?

#

also, textual inversion embeddings are just super fancy prompt building. It can only get so close depending on what's in the model. Try different models and see if you get better results.

gentle otter
#

first if you got over 100k steps overnight i feel like you're traning wrong. i did it overnight last night and only got 1000. also did you preprocess your images to generate captions? did you manually fix the captions? what settings are you using on the train tab?

cobalt glen
#

Hey can anyone tell me how can i make an brand logo for my shop herr which commands i have to use or something

lilac reef
#

Anyone out there use Kohya Lora as I am still having a hard time comprehending steps, epochs and was wondering what is this exactly in comparison?

override steps. steps for 16 epochs is / 指定エポックまでのステップ数: 544
running training / 学習開始
num train images * repeats / 学習画像の数×繰り返し回数: 600
num reg images / 正則化画像の数: 0
num batches per epoch / 1epochのバッチ数: 34
num epochs / epoch数: 16
batch size per device / バッチサイズ: 18
total train batch size (with parallel & distributed & accumulation) / 総バッチサイズ（並列学習、勾配合計含む）: 18
gradient accumulation steps / 勾配を合計するステップ数 = 1
total optimization steps / 学習ステップ数: 544

#

I am used to HN/TI/DB and this lora stuff I just can't. I know I need 3k epochs in DB, or 1200 for a TI so easy enough but in LoRA I still have no idea what I am really doing.

#

I see 544 steps but that cannot be the true total since it is over 2h to train it making it the absolutely slowest method behind HN that has no optimizations to it.

jade mesa
#

You can use deforum in the cloud with a webui now.

#

You're welcome.

static snow
magic crest
#

how cab i create air?

#

can

#

ai

#

/create ?

scarlet depot
#

hello, i want to install automatic 1111, but can i install it with amd video card

empty barn
scarlet depot
#

when i try to install it, at the end it is written "torch is not able to use GPU" write --skip-torch-cuda-test...

empty barn
scarlet depot
#

Im with windows 10, video card amd ryzen 7

trail ledge
#

[1:32 PM]Occam: Only works on Linux. Which GPU?

empty barn
scarlet depot
#

thank you

#

!

trail ledge
#

yeah, still "In Testing" though

#

so your mileage may vary

stray delta
#

Hey guys, a quick question

#

Does stable diffusion works with 7000 series AMD gpus?

#

Or only NVIDIA?

empty barn
nova lily
#

typically do you thinks good or negative to add alot information in a singular prompt? Would the ai be confused?

For example
"Dark forest woods' instead of having 2 prompts for "Dark forest" "dark woods"

Same thing with sea/ocean, beautiful/gorgeous/pretty

drowsy arch
#

/prompt

rose sedge
#

Where is inpaint at full resolution now?

#

I don't see it

#

where it used to be

brisk bloom
rose sedge
#

thanks

#

but it gives me error

#

when trying to generate with that option

#

ValueError: Coordinate 'right' is less than 'left'

brisk bloom
#

not sure. never had an issue with it

rose sedge
#

maybe it has something to do with my install, as I can't do a proper git pull because it tells me that my "local changes" to some files would be overwritten

#

I haven't done any change to any files

#

but it allways tells me that

#

so I have to update manually by downloading from github page, But, if trying to git pull just after that it still tells me I have modified files.

brisk bloom
#

it will say the name of the modified file. move it to another folder and git pull

#

it will get it again

rose sedge
#

yeah but its allways the same

#

I have to go delete a bunch of files

#

every time I want to do git pull

#

and then it doesnt work

#

If I do that stable diffusion doesn't work

#

it only works if I use the downloaded files

#

manually

brisk bloom
#

maybe do a fresh install

rose sedge
#

Is there an automatic way to do that?

brisk bloom
#

not automatic but its fairly easy

rose sedge
#

thanks

#

oddly enough, I just removed the files it asked, did a git pull, and stable diffusion worked

#

after many times

#

maybe its because I reinstalled git and some other software the other day

#

and the full resolution worked too

brisk bloom
#

updating git would have helped. yeah

wide juniper
#

Is civitai running on donations alone? I’m thinking of donating 20 bucks to them, but I’m not sure if they’re already profitable or not

fervent thunder
#

Do we have explanation on what's the difference between controlnet preprocessors and models?

waxen pasture
# fervent thunder Do we have explanation on what's the difference between controlnet preprocessors...

Preprocessor will turn the input image into whatever the model needs. For instance, if you want to use the depth controlnet model, and you have a regular image, you would select a depth preprocessor to turn that regular image into a depth map. If you have a depth map already made that you want to use the depth model on, no preprocessor is necessary. At least that's my understanding of it, I just downloaded controlnet earlier today and played around for like 20 minutes.

astral goblet
#

you don't have to use a preprocessor. they're just there to assist in creating the information that controlnet models want. if you get that info out of an image any other way, use it

crystal scroll
#

anybody know how to use the k samplers? or am i just misunderstanding something, as I only see the regular euler and euler a etc. samplers

astral goblet
#

there's so much upgrading that can be done to the preprocessing bit. like parameters and all that. it's gonna get neat

#

somebody already made an extension that lets you pose your own skeleton right in the webapp, and use it as a control

fervent thunder
fervent thunder
#

Hi. When bots welcome back for promt??

crystal scroll
fervent thunder
#

Who know it??

sturdy quail
#

So...ControlNet. I look forward to trying that out, so I might have to wait until if one of the web uis integrate that. As much as I want to use triple one, I don't have computer for it. 🙂

jolly jasper
#

3060 for $200 usd for SD, worth it? I already have a 1080 ti

trail ledge
#

depends on if you get more vram

jolly jasper
#

I get around 3it/s on 1080 ti

trail ledge
#

ok, then why are you asking?

jolly jasper
#

Because im asking whether for $200 its a worthwhile upgrade

#

I have no idea the performance difference between the two

#

Gaming wise, its roughly the same. Im more interested in the inference strength of the 3060

#

Accelerating SD training and inference with RT cores would probably cut my time waiting by 2x if what i heard is correct, but im not completely certain on it

west sparrow
#

Wow; SD has improved a lot since I last checked out some of the better images

#

The #🏆｜winner-gallery has really good image res, art-style consistency within the image, and less artefacts

#

Is that just evolution of SD itself, or other tools ontop?

zenith thunder
#

ive seen lots of people merge models

astral goblet
rich dock
#

Hi! Been testing out stable diffusion, is this a good place to ask questions about prompts or technical issues?

#

I'll take a wild swing and ask basic prompting question and if its the wrong place to ask maybe someone can point me in the right direction.

Question ; what is the syntax for blending between two things, like if i wanted to mix a cat and a dog by percentage, or two different embedded face trainings together, or an embedded face and a lora - whats the terminology to do this precisely

trail ledge
rich dock
#

Second question; is it possible to restrict certain attributes to one object or character, so this prompt doesn't bleed in to other thing, like say you wanted someone on a green chair, occasionally it does stuff like make their hair green, or give them a green shirt or something,

rich dock
trail ledge
#

with embedding it's [embedding1: embedding2: 0.25]

#

that'll do the first 25% of steps with embedding1, then the rest with embedding2

rich dock
#

ok cool thanks

trail ledge
#

if you want to restrict something, try inpainting

rich dock
#

and is there a good place to ask qustions about training embeddings?

trail ledge
#

then you can just mask out the subject and prompt specifically for what you want in the mask

#

regarding training, you can probably ask anywhere.

rich dock
trail ledge
#

make sure you have an inpainting model

rich dock
#

yes, even then, i have tested multiple inpainting models

trail ledge
#

and you'll need to tweak your mask blur and paddings to fit how big or small your mask is

#

also, select "only masked" for inpainting area

#

that way you get more resolution

#

if you select "whole picture" you'll lose resolution on parts of the picture you're not painting

rich dock
#

appreciate the help though

lament steeple
#

ok who drank all my coffee

distant oar
#

hello

surreal frigate
#

@mild badge you can't block me from here, just tell me what I did to you so magnifically that upset you so much?

chilly valve
#

morning! Coffee

surreal frigate
#

kind of early bird you are @chilly valve

#

good morning to you and your deeds.

fervent thunder
surreal frigate
#

Euler a, dont like it?

quick bluff
#

Im not you bruh

proud dawn
#

Hi, where can i find the available fine tuned models?

chilly valve
charred wren
#

is it just me or is civitai running like crap rn

#

took me like a good 30 seconds to refresh the front page

forest apex
#

@bleak matrix

bleak matrix
#

oh my

fervent thunder
#

waow didnt have time to check his account

bleak matrix
#

Not an active member, from what I saw 🙂

warm junco
#

Hey sunsmile

bleak matrix
#

But if they are (and got hacked), they can always let us know

#

Hello!

forest apex
#

morning 🙂

bleak matrix
#

mornin!

#

how are ya?

#

(Thanks for the ping lol)

forest apex
#

been better, been worse, lol, yourself? chipper as usual?

#

still messing with mixture of diffusers, it's been getting some updates recently. my UI coding ability is poor but it's nicer than using the CLI.

wide juniper
#

I want my SD to generate 100 really bad sushis. I plan to use chatGPT to generate the ideas in bullet points. Is there a way to input those bullet points and batch generate them?

waxen pasture
#

There's probably an extension for inputting a list of prompts and doing them sequentially, as well as maybe the command line interface can do it? I am not sure

wide juniper
#

I have wild cards, so I think I might use it.

#

guess I'll get duplicates

fervent thunder
#

Do we have a non-anime model , which can do magic?
Particles, effects , auras , e.t.c?

#

Just not anime , anything , game-ish looking is fine

deft bronze
fervent thunder
#

But there isn't alot of magic...at least not in a game , only in a movie

#

sent few screenshots in the channel with images.

waxen pasture
inland geyser
#

gm guys, greeting from China

stable pawn
#

How much of the popular prompt tags are just myth or BS? I tested things like 4K, 8K and so and they tend to work worse than 2K, and "detailed" just zoom in on a detail or add objects, I often try "add sharpness" to get sharper images or I try things like f18, long dof, sharp focus.

#

I often try to take prompts from https://civitai.com/ and then try to remove as much as possible to keep similar result and sometime I can remove more than half of the prompt.

deft bronze
#

yeah, that makes sense. There's a lot of cargo-cult prompting

stable pawn
#

As an example, I do not write "Two girls are hugging" but "girls hugging" then it imply it has to be another girl and the AI often understund that.

#

And the extra sharpness in the first prompt, I often fix that in upscaling and in post work.

fervent thunder
peak orchid
fervent thunder
#

True

stable pawn
#

Question is if some one want to risk that in China just to say "hi".

peak orchid
stable pawn
#

My best tips to avoid deformed hands, "wearing mittens", but it may look fun if you try to create a sexy lady pose on a beach 😄

fervent thunder
#

lol

#

You just reminded me what I wanted to do when I first got SD...

stable pawn
#

Mitten kittens?

fervent thunder
#

almost 😄

stable pawn
#

Prompting is 30 % trying to avoid bad hands, 15 % bad anatomy, 25 % getting rid of signatures, watermarks and so, then 12 % porn.

fervent thunder
#

I don't usually try to avoid bad hands...when I get them, but like that gen , I just inpaint...oever and over again, there are other details I'd want to outpaint and fix anyway

stable pawn
#

And 44 % finding synonyms for shaved cats.

fervent thunder
#

okaaaay...

stable pawn
#

I was talking about Sphynx felines, nothing else.

#

... I gave me a embedding idea.

round harness
#

hi, is it possible to have a metadata reader in the app of this discord ?

forest aspen
#

For everyone who missed the live session I did for AI/CC, here is the recording of the entire event. Where we talked about Environment concept art, Character design, 2D Game props & UI Icons for games and many other things. Using AI Tools with a professional mindset and goal. Hope what I've showed there will help many of you! https://www.youtube.com/watch?v=8wS4O9EojMM

final river
#

Hey people, anyone going to NFT paris? I'm going Saturday and will send comments and contents (as much as possible) on my server.
( I want to know if i can meet anybody randomly there 😄 )

opal crescent
#

anyone knows a good guide to install chatgpt?

empty barn
#

Then hack OpenAI and download the model.

#

Oh, and make sure to buy the 80GB VRAM variant.

forest apex
#

step 4: profit

deft bronze
#

it's like 30 usd a month

astral goblet
opal crescent
#

hmm i thought it can be downloaded like stable diffusion

#

webui

#

nevermind then

astral goblet
#

it seems counter intuitive, but text language models seem harder to train then image denoising models

#

it takes a ton more computer to run chat gpt tech

#

i think you might be able to run just base gpt 3 on a home pc, but i'm not sure. it's kind of jank and not really as cool as SD

#

sorry i'm wrong. there are variants of gpt3 that have been made which can run on home hardware

#

they run more like gpt2 though

empty barn
#

GPT3 hasn't been published, has it?

astral goblet
#

the code to train your own model has

empty barn
#

I thought OpenAI's decision to keep that locked was what prompted so many open source alternatives to crop up, like GPT-Neo and GPT-J

astral goblet
#

stuff that runs locally on a single gpu uses gpt3 level tech but performs more like a gpt2 model due to not using all the possible features that are responsible for driving memory use up

empty barn
astral goblet
#

source code

#

my bad they didn't

#

nothing in there, just research data

#

but,, they did release all their research as a paper. i suppose that's what has been reimplemented

empty barn
#

No, others have developed alternative networks.

#

Don't give OpenAI credit for that. 😄

astral goblet
#

i'm not sure this is the argument you want to have. needlessly being argumentative is one of my giant pet peeves and it's too early on my weekend for that

empty barn
#

It's Monday. 🤔

astral goblet
#

yeah like many full employed people in a job at somewhere that's open on the weekends too, my weekend tends to be irregular

#

again, another needless argument

empty barn
#

Sure, sure. But why are you engaging in it then?

astral goblet
#

do me a favor and try to react to a comment of mine

empty barn
#

What did that do?

astral goblet
#

nothing. ahhh. bliss.

plush forge
#

Hi there! Did someone manage to train a lora with google colab?

silk wren
#

I can't use imagine now?

fervent thunder
#

You didn't let me respond waow

plush forge
fervent thunder
plush forge
#

I'm trying to use the fast-kohya-trainer, have you used it?

fervent thunder
#

nope, don't think I should try training on 1050ti lol

plush forge
#

it's a google colab haha

plush forge
fervent thunder
#

free services going downhill

#

Just a week or two ago I could use collab for 6 hours , now 3 at best
Kaggle added afk check once per 40mins or something

plush forge
#

Welp, at least you could use them hahaa

#

i'm still battling errors

fervent thunder
#

paperspace...just sucks lol, too slow...

fervent thunder
plush forge
#

I mean they all generally suck lol

#

I'd totally pay for a service where I could rent out a gpu to train lora/ti's

#

but if it's the horrible experience that is using google or kaggle then

#

no need

astral goblet
#

renting someone else's hardware vs using your own hardware. the shit that gets beat on by everybody that uses it is probably going to suck more

#

the rental companies are probably overselling capacity too, since why wouldn't you? supporting peak volume all the time is costly and usually a pointless endeavor

plush forge
#

no matter how beaten up it is

fervent thunder
#

I think after a while more renting companies will popup...hopefully

#

rn it's just all google or owned my google

plush forge
fervent thunder
#

we have lots of hosting services , maybe they'll get the catch with GPU's someday too...

astral goblet
#

i've heard azure offers super competitive pricing and they provide serious good infrastructure, but there's nothing like a community building tons of free tools to launch on azure services right now

fervent thunder
#

hm...hosting a111 shouldn't be hard

astral goblet
#

https://tenstorrent.com/ i'm waiting to see what jim keller's company does next to make any predictions about where infrastructure is going

astral goblet
plush forge
#

still it's useless to rent gpu's for training if in order to do the training you need a masters in computer science

#

lol

#

With good GUI, then everyone will jump right on

plush forge
astral goblet
#

even the beefiest grayskull e300 from tenstorrent only offers 16gb though. probably meant to be used in clusters. i'm excited to see where he's going next. likely they'll license the designs they field test thorugh tenstorrent to all major manufacturers. keller is kind of a savant for that shit

astral goblet
plush forge
astral goblet
# plush forge no, just 4 a1111

right. it should run on azure cloud, and should be easy enough for anyone to follow instructions and set it up if they know how to navigate those services. but you still hear more buzz from individual home hobbyists about these other services instead. word of mouth marketing is something that is a double edged sword for microsoft

dawn plume
#

Hey guys! What’s the most interesting thing you find about stable diffusion? Have you tried other similar tools?

astral goblet
#

actually changing answer. How rapidly it's developing is awesome, but trying to understand where it's going is more interesting. It's a denoising algorithm right? that's not just pixel noise either. it's noise, that's it. How soon till software defined radios are using this tech on signal processing, to remove noise from analog data and usher in a new wave of analog data processing

visual zephyr
#

stable balls

astral goblet
#

"new wave of analog" maybe a little bit of a pun. not sure if it works but i like it

gentle otter
#

hosted services just suck, SD's flexibility, extensibility and ability to try unlimited prompts are what make it great.

dawn plume
#

hmmm interesting

astral goblet
#

also controlnet is really neat

#

do that with dall-e

dawn plume
#

nicee

#

so you would say you prefer SD overall?

astral goblet
#

i prefer open tools, but SD is the only one in this category so it succeeds by default. the best kind of success

gentle otter
#

ive gotten better results from others but yeah SD is free and open so I'd rather get good at and support SD

astral goblet
wintry stream
#

As I'm sure you all saw in the #📣｜announcements channel angeryping

We have a short survey we'd really, really appreciate everyone taking the time to fill out to make sure we're keeping up with you on how you like to user this server & what you'd like to see in the future!

https://docs.google.com/forms/d/e/1FAIpQLSdpaL4lcXvdexbLEnAnP-G3w5e56nXuq_0-CfWDFKpmpAlSww/viewform

I'll be around for the next little while so you're most certainly welcome to let me know if you have any questions or run into any issues filling out the survey itself ❤️

(If you'd prefer to remain anonymous on the form, simply leave the first (optional) question blank)

fervent nymph
#

My biggest feedback was to bring back Dreambot, as it was by far the best part of the server

wintry stream
fervent nymph
#

Yeah, filled it out thanks. Would be good if others who miss Dreambot could give their feedback on it too

waxen pasture
#

I have filled it out as well :)

wintry stream
wintry stream
meager acorn
#

I was using ChatGPT for the lols earlier and got hit with the "too many requests in 1 hour" ratelimit error. Did 5 seconds of research and found out that even ChatGPT Plus subscribers will see this error. This is one of many reasons why I hate OpenAI

gentle otter
#

I'm amazed their website loads at all with how fast they've grown. cant really complain about getting their service for free. its far more forgiving than dall-e

open wedge
#

hi can someone with a running version on mac give me some advice please? Im asking for a class of several students

past edge
#

TIL people run this on mac.

vernal iris
#

Hi. I'm... new here.

wild steppe
vernal iris
humble panther
#

someone has shared to me before a link that is just a repository of ai made images i think from SD, anyone know it?

hexed hatch
#

there are websites like lexica .art, publicprompts. art, prompthero .com, and openart .ai that have repositories of prompts of SD generated images, maybe theres a more specific site but thats all i know

astral goblet
# past edge TIL people run this on mac.

Yeah mac got machine learning cores working in tandem with arm processors. they're doing insane things over there and i'd consider buying a modern macbook for this kind of work

umbral flame
#

how use this AI?

light folio
wintry stream
fervent thunder
#

I'd like more 2.1 models, but idk about generalist, I'd prefer to keep different styles in different models, so I'm not getting styles I don't want in a general prompt and don't have to fill my negative or basic prompt with tons of specifications avoiding it 😦

glad oasis
#

how can i generate images help

wintry stream
fervent thunder
stable parrot
#

whats the best model for pixel art sprites?

fervent thunder
stable parrot
#

thanks

plush forge
#

@fervent thunder

#

who did you talk to when you had google colab errors?

fervent thunder
plush forge
dire sun
#

The latest automatic 1111 isnt' switching models automatically, even if this feature is enabled. Anyone know how to correc this?

noble pumice
#

Are there any Mac users here? I had SD automatic 1111 up and running locally and it was working fine but then I closed Terminal and now I am getting a 'Cant connect to server error' so I have two questions: Do I need to have Terminal permanently open in order to run SD, and secondly, how do I get it up and running again as I cannot find anything on this via a Google search? Thanks

fervent thunder
#

no mac users here, just schway individuals

trail wing
earnest shadow
#

ро

#

how old is this neural network

lost marsh
#

i need a 3090 to use now....6900xt workarounds are very inconvenient

obsidian meadow
#

Hi all!

#

I have updated automatic1111, and was hoping that auto-load model was working again...but it doesn't for me? Can someone confirm this? I tried to enable it first and restart, and then disable and restart, but that doesn't help. I am referring to this option in the settings of automatic1111

#

When reading generation parameters from text into UI (from PNG info or pasted text), do not change the selected model/checkpoint.

#

You can find it under: User Interface

magic citrus
past edge
#

Normalize deepfakes lets goooo

pastel fern
#

I want to try and read up and learn what ai/stable diffusion actually does

#

Does anyone know a good place to start?

#

Everything I've read is just so full of technical jargon idk wtf I'm reading

past edge
#

ask chatgpt broh

#

it just learns how to turn x image into noise and back

#

then learns the patterns and remembers how to do it. add some magic to make images congruent and voila. magic art in seconds

naive mauve
#

how do i generate a picture

fervent thunder
#

a rare 2.0 model nice lol

#

oh looks like i was behind, was referring to artificialguybr new model

dire sun
trail wing
dire sun
stiff plaza
#

is this an Open source Ai server

trail wing
#

@dire sun the most annoying thing right now is that in that some changes were made to the generations, and if you talk about it every single simpleton comes in the door with the "you sure that's not transformers?", it's like an eternal spring of idiots, but guess that comes with wider adoption

dire sun
trail wing
dire sun
trail wing
dire sun
trail wing
#

I don't refresh it, it opens, 6 or 7 seconds after everything populates and it's off to the races

dire sun
#

although it does say loading

dire sun
trail wing
#

some stuff was messed with in the way it processes, then some stuff was put back

#

and at least in two time periods, because of interactions, mostly with what info it recorded at the time

#

and some interactions with the pseudo random generator

#

those won't be able to reproduce

#

just wish they stopped effing with it, or at least provide all the compat options

#

but hey

#

move fast, break stuff

dire sun
# trail wing some stuff was messed with in the way it processes, then some stuff was put back

this is why I stay with a version for months on end. The one I have 'use to' make images flawless, but something happpened. It happened after I deleted my venv, however I can get them back, and it's a load of fish to do it. I have to delete the venv file, the config, and use the original files, setup, then let it make a new venv and hope it works. That's a lot of tooling.

#

yeah, automatic shouldn't mess with stuff so much. We don't need a rolling update every 3 days

trail wing
#

dude, scunet pops in and out of existance and I have no clue what is going on

#

sometimes it's there, sometimes it ain't, it's like the tileable button, sometimes it works, sometimes it's radio silent

dire sun
trail wing
#

I have it installed, once in a blue moon it gets detected, 90% of the time it's like it doesn't exist

#

but hey... control net

#

moving on

dire sun
trail wing
#

don't think so, even the extension stuff was diff

dire sun
#

because if I could get it to install, I wouldn't change.

#

see, there's the kicker, we can't move forward

trail wing
#

well, you can try

dire sun
#

it appears that the most stable version of automatic is early Jan, which is just about where I am

#

Dec 31 was the last time I found a stable version

trail wing
#

I stayed with something from around that time that didn't have the override and the broke load model on png info

#

but had to move on in the end

#

annoying af but hey

dire sun
trail wing
#

with the amount of ckpt/safetensors and control net

#

getting better and better

dire sun
#

but you can't recreate any seed?
Neither can I

trail wing
#

yep

dire sun
#

someone needs to figure out why this happens

trail wing
#

we'll always have 1424987 in October of 2022

#

it was not meant to be

#

there's a partial fix for one deviation, at least was, holded up in the pull requests

#

but it won't fix everything

#

there were at least two or three moments where things went berzerk

dire sun
#

not sure what you mean. Isn't that a certain seed that made everything look good

trail wing
#

that was a joke, homage to Humphrey Bogart

dire sun
trail wing
#

it was a movie star from the start of the "talkies"

#

anyway, wish I had frozen the system couple of times along the way

dire sun
trail wing
#

I watched all the old movies in my 10's

#

maybe that explains a lot

#

but I digress

dire sun
#

nod

trail wing
#

at least we get to see all this in the garage phase

#

before the ms's and the big tuna comes in and idiotifies it

dire sun
trail wing
#

was lucky to get on the personal home computer wagon back then

#

so the ride is pretty wild, guess this will boil down to a personal AI friend or something that will provide for every command

#

maybe through a cell or something

#

maybe even much before any sort of general AI is achieved

dire sun
trail wing
#

I need a 4070/4080

#

for heftier stuff

#

even if it's mobile

#

the vram is limiting and any txt2vid will be even worse

dire sun
#

running 1650 super 😄

trail wing
#

I bow to your patience

dire sun
#

if you can wait 2 mins per photo, sometimes longer, you have it 😄

trail wing
#

I know someone running diffusion bee on mac that has to wait a whole day for about 8 pics

dire sun
#

it's exciting, the waiting.

trail wing
#

they just go out, work day, come back to see what came out

dire sun
#

hoping the images turn out right

#

I can't get SD to install in linux, just windows

trail wing
#

@lilac reef could be...

trail wing
#

it's a little faster actually

lilac reef
#

I bet I know it is me having to delete venv

#

it must have a new xformers or some other thing

trail wing
#

I'm doing grids now to explore ckpt/safetensors

dire sun
trail wing
#

it's that bad

dire sun
#

grids?

trail wing
trail wing
#

saves all the manual switching, that is now even more annoying af because of the override thing

dire sun
trail wing
#

let them be turned off, if they'll enjoy it more when it is a commodity on their cell phone, they get what they deserve

dire sun
trail wing
#

lots of karens waiting to post their fake vacations and lots of chuds waiting to post their non existant gfs

#

it's going to be hilarious

dire sun
#

the 2015 reborn 😐

quick bloom
#

hey guys! Currently APi is open source. Will it be opposite any time soon?

elder cradle
#

No

#

The api should always be oss

fierce garden
#

new to stable diffusion and AI art in general and i need a bit of help.

quick bloom
dire sun
#

python.exe -m pip install --upgrade fastapi==0.90.1 does not stop the middleware issue, on local install of automatic 1111

warm junco
dire sun
#

I will try again

warm junco
dire sun
#

well of course it says this :stderr: ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'requirements_versions.txt

warm junco
#

Okay then bring back the file (restore from bin)

#

Add fastapi==0.90.1 to the bottom of the file

#

But i think your git pull isnt working correctly

dire sun
#

Then it started to tell me to upgrade torch and xfomers

#

so it was stuck

warm junco
#

Okay you can upgrade torch and xformers with --reinstall-torch --reinstall-xformers

dire sun
#

I've seen this before, and eventually it panned out

#

but not this time

warm junco
#

Broken?

dire sun
#

yeah, it doesn't load right, buttons aren't in the right places, brokne

#

broken

#

but not able to take a photo of the image to show you

warm junco
#

You can make Screenshots with Win+Shift+S

dire sun
dire sun
warm junco
#

Wow wtf xD

dire sun
minor thistle
#

I'm trying to generate 4 images of 768 * 768 on A10G gpu that has 24gb vram. But running out of memory. any solution for this?

grim crown
#

@warm junco

#

can you help me please?

dire sun
warm junco
#

Yea now you can add git pull back into the webui-user.bat and also --xformers --autolaunch

vale creek
#

I have a question. I'm thinking of buying a 3060ti or a 3070 for gaming.
Is resolution such as 1920x1080 difficult in the SD in local?

warm junco
#

With 8gb vram yes but with upscale its possible

dire sun
#

Is there a way to use controlnet and models in different drives?

#

Since I don't have room for the control net models, they will be stored on another drive

brisk bloom
dire sun
#

I might have room for 2 gb worth

stable pawn
#

A question: I have installed Controllnet and OpenPose through Git, and I use Git pull when I start Automatic1111, when I do so, will it only update Automatin1111 or do other things like OpenPose update at same time?

warm junco
brisk bloom
#

only the webui will update with git pull. extensions like contronet will only update if you manually request an update

warm junco
stable pawn
#

Ty.

dire sun
#

shortcuts work with VSTs, for music

brisk bloom
stable pawn
#

I hope it in the future get a update check script that can say if there is new updates at start, even for models and so. Just now I just forgot I can do so through Extention tab.

dire sun
#

I have space on another drive as well

warm junco
dire sun
#

thanks, I'll dig around

warm junco
#

You could also organize your drives and delete stuff or move it to an other drive

dire sun
#

but I"m in the process of backing up other stuff

#

I heard just two of these controlnet models are all you need, is that true

stable pawn
#

I need to start my day, can you come up with simple and fun prompt ideas or interesting words to use in the category "fun and happy" not "Porn and spank".

warm junco
dire sun
#

thanks

warm junco
storm river
#

Can someone help me, I wanted to start doing stable diffusion and wanted to install a model merger and it already gave good combinations of models that I should try but it says I need AnythingV3.0 huggingface pruned
[2700c435]「543bcbc21294831c6245cd74c8a7707761e28812c690f946cb81fef930d54b5e」but I dont know how I can find the exact model that is needed for that, any ideas how to get that?

#

Or is it even needed

warm junco
storm river
warm junco
#

In the webui you can merge models

#

The anything v3 is an known Anime model

storm river
#

yeah but how can I get the exact version that is there listed? is there a way to get it or just searching?

warm junco
#

Choose the pruned fp32 and the .vae.pt file

storm river
#

So it doesnt matter what exact model I take? because this one is SHA256: 67a115286b56c086b36e323cfef32d7e3afbe20c750c4386a238a11feb6872f7 and the one that was recommended started with 543 so it doesnt really matter?

warm junco
#

The original got taken down so it wont matter

#

Or maybe its from civitai

#

But its the same model so just use the pruned-fp32

storm river
#

Okay thanks then for the quick help I appreciate it :D

warm junco
#

No problem 🙂

storm river
#

doesnt look like there is right one in there I guess it shouldnt change much anyways :D

warm junco
#

Yea its all the same xD

glacial oyster
#

Hi all.... Do you know what model that the result is looks like a real human photo? Generally this model used in pixiv or twitter that the result is cosplayer photo or something like this. Lately i looking for it in Hugging Face, but found nothing.

warm junco
glacial oyster
#

ok, thanks ...

warm junco
#

Yea was a typo sry

spice tulip
#

#roles image

#

I'm so dumb

#

I dunno how to use this

rich dock
#

what causes images in SD to sometimes become overly sharpened looking

wide juniper
#

Why is controlnet in my txt2img tab? what's the use case of it?

brisk bloom
rich dock
#

is controlnet included by default now?

brisk bloom
urban tusk
chilly valve
#

morningggg Coffee

wide juniper
#

Is there ever a need for cfg to be greater than 15?

#

Never gives good results

brisk bloom
wide juniper
#

No real need to even touch CFG right

brisk bloom
#

I like to use 6.5 a lot, but 7 will give you good results. For 2.1 you want to go down to 4 or 5

wide juniper
#

How did you ever figure out you need to go lower? Just ugly highly-saturated results?

brisk bloom
bleak matrix
#

Good morning, everyone sunsmile

brisk bloom
#

morning. its evening here

lament steeple
#

It's morning here, but I'm nocturnal. almost bed time

bleak matrix
#

Yay for nocturnal!

onyx python
blissful flame
onyx python
bleak matrix
#

It's probably just Discord.

onyx python
golden magnet
#

it seems that civit often suffer from internet issue...

bleak matrix
#

Restart Discord, if you're on mobile, restart the app

lilac pine
#

Hey guys, I am getting a HeaderTooLarge error for ControlNet. Any idea how to fix it? Thanks!

onyx python
bleak matrix
#

(if on mobile)

#

Discord is just finicky. But you can always file a ticket with support (Discord)

#

np

onyx python
bleak matrix
#

Hope it works!

restive dawn
#

GM. New here. Just downloaded stable diffusion. Is there a guide to certain settings for realistic images of people?

regal snow
#

I have a 4070ti so I don't think I need x-formers for most operations... but I'll be trying out Dreambooth soon enough, x-formers seems to be required with GPU < 16gb

lilac pine
simple ore
#

Hey all, is anyone running a 3090GS KFA2? I was wondering how the card preforms, thanks. 🙂

carmine shuttle
#

What was the chat where I could create drawings?

regal snow
simple ore
#

Sweet thanks i was more hoping anyone had it installed can could tell me temps also. I'm thinking 2 step over from a 3080 Strix 2 that card and want to be sure it's properly cooled. 🙂

viscid ravine
#

How can I use stable diffusion paint feature.

gentle otter
#

in painting?

proven plinth
#

Wait where can I generate pictures

gentle otter
#

not here

proven plinth
#

I feel lost in this server

gentle otter
#

they dont generate images here like a few other ai bots (which is a weird way to generate images to be honest). this is just a support server

wild steppe
polar summit
#

HI sorry for bothering everyone where do it put a .pt file

trail wing
astral goblet
# wide juniper Is there ever a need for cfg to be greater than 15?

if you do an xy grid of increasing cfg with increasing steps, you'll find a zone of quality on the grid as steps get higher for higher cfgs. its a neat artifact.

Not much need for it, but you can get good results out of cfgs all the way to 30, according to the last of these generation grids i spotted. haven't ever made my own though

gentle otter
#

its good to test ,but yeah in my recent test from 1-20, everything on the 20 side got ugly

fervent thunder
#

It depends from model and what you're doing, on most model there's usually no reason to go above 15.
I'd say for most generations 6-12 is fine

#

From my experience going higher mostly ensures that everything you prompted will be on the image and will be in focus, everything else - probably cut off.

#

if something not on the image, SD probably either have no idea what that is or prompt is filled with too many things...

trail ledge
#

also, if the prompt is long, then high CFG gives it indigestions as it tries to fit all the prompts in

fervent thunder
trail ledge
#

I find the longer my prompt, the lower my CFG should be (9-12 maximum)

vestal dew
#

RTX 3090 with 24 GB VRAM or 4070 TI with 12 GB VRAM - benchmarks show 4070 TI is faster with SD, but 3090 has the 24 GB. Which is better for SD?

fervent thunder
#

I usually go for low prompt, but often do higher cfg is I want to get exactly what I prompted, was doing up to 15 cfg on protogen

vestal dew
#

I would love to find a prompt that gives a clear, full body image, so I can have someone walking or running, while seeing who it is, with it not changing the character to some weird looking, lower res character

fervent thunder
trail ledge
#

try hires.fix and inpainting the face

#

SD needs resolution to work with facial details

vestal dew
fervent thunder
#

idk , depends on model and output.
For anime model should be enough , for realistic - probably will need even higher

vestal dew
#

Ive4 never done hires fix, not sure my crap card can handle it

trail ledge
#

then do inpainting with just the face and "only masked"

vestal dew
#

which is why Im looking for a new computer GPU, but can't decide between 3090, likely used, and 4070

#

it doesn't just low detail the face - it's everything

fervent thunder
#

Another idea would be is to make upper body and then outpaint

vestal dew
#

if I could paint a figure, I wouldn't be using SD

#

when it does do full body, everything loses detail

#

generally, the figure also becomes skinnier, sometimes almost stick figure like

astral goblet
fervent thunder
astral goblet
#

i loaded up atomic heart the new videogame last night and noticed it had supprot for dlss3 on my 4080. i'm like "sweet if i need frames then i got free frames here with frame synthesis tech!" but then the 4080 runs this brand new cutting edge unreal engine game at 120fps 1440p maximum "atomic" settings with ease anyways

#

so the new high tech feature wasn't needed. lovelace is just that good and software quality standards are rising thanks to engineering work that epic has done on unreal engine

vestal dew
trail ledge
#

yes

vestal dew
trail ledge
#

just outpaint downward, and it should add in more details

vestal dew
trail ledge
#

given your prompt

vestal dew
#

what is outpaint downward?

#

I don't know what that means

astral goblet
trail ledge
#

outpainting is to add more to your current image

#

so you start with face and torso

vestal dew
#

Are you talking about in Photoshop or something?

trail ledge
#

then you send to img2img (or use another tool with open canvas like InvokeAI)

#

and you extend the picture by asking SD to "outpaint" downward

vestal dew
#

Ive tried taking a portrait, torso only, in img2img

trail ledge
#

no, it's done in SD

astral goblet
vestal dew
#

it's almost impossible to get it full body than, and that's when the low detail is the worst

trail ledge
#

I don't think you're understanding what we're saying

vestal dew
#

I use A1111

#

there is no outpainting option there

meager acorn
# gentle otter I'm amazed their website loads at all with how fast they've grown. cant really c...

I somewhat agree with this, however if I was paying $20 a month for ChatGPT+, I would certainly expect not to get rate-limited.
It would also be a lot better if it actually told you how how much longer you are being ratelimited for (it always says "too many requests in 1 hour" even if you only need to wait 5 more minutes of the 60m afaik).
It also doesn't actually tell you anywhere how many requests is "too many", you just eventually hit the limit. Additionally, by adding a rate limit like that, they also add an incentive to creating alternative accounts because it makes it so that having more accounts effectively means that you don't get rate-limited.

trail ledge
#

from A1111

#

it's in the img2img scripts

fervent thunder
fervent thunder
trail ledge
#

there are many ways to outpaint in SD, depending on what kind of interface you like

vestal dew
#

I dont like cloud

fervent thunder
#

invoke ai pretty cheap for VRAM on outpaint, it's also pretty quick...but not as comfortable to work with 😦

vestal dew
#

I see, so I can outpaint down, and it will add legs...

fervent thunder
vestal dew
#

Ill see if this works with the crap card I have

trail ledge
#

you may want to outpaint slowly though, over several iterations

vestal dew
#

do I put the original prompt in then, or just about the legs?

fervent thunder
#

If you will be limited on vram - just do it in invoke ai, I can outpaint here even with 4gb vram, just giving it less info to work with, reducing chosen area

vestal dew
#

I have never used invoke

#

is that run locally?

fervent thunder
#

yea

vestal dew
#

Do I write the original prompt, or just prompt about what's being added?

fervent thunder
trail ledge
#

you can use the original if it's a continuation

#

but if your original says "upper body", you'll obviously need to correct for it

fervent thunder
#

yea

trail ledge
#

or if you want to add something else, like "fence in front"

vestal dew
#

OK, I tried adding 128 pixels down

trail ledge
#

need a model that supports inpainting

#

btw

fervent thunder
#

I was doing it with original model usually, same as I used for gen

vestal dew
#

dont they all support inpainting?

trail ledge
#

nope

#

but you can make your own

#

if you have a particular favorite, but it doesn't come with an inpainting version

vestal dew
#

well, it worked with protogen anime

trail ledge
#

if you get weird output like another head or something, then you likely need inpainting version

vestal dew
#

now trying with the model that made the original image

trail ledge
#

if it works, then all good

vestal dew
#

it worked fine, added part of the legs

#

with the original m,odel, vintevivid, it shows him standing in a concrete block

#

instead of adding legs, it just added the concrete block

#

Im trying with more iterations. The script suggests 80-100

#

sampling steps, that is. 10 minutes to generate

#

78%]

trail ledge
#

I ignore the script recommendations, works fine with the same settings I originally used

potent fox
#

Hi guys, i have the problem with controlnet, that with every rendered frame the model gets more and more nude. Very strange.

trail ledge
#

also, sometimes it will be creative and add things like concrete blocks or pillows or whatever, just do a batch

vestal dew
#

ah, now he's behind the concrete block, nmot in it

potent fox
#

Using img2img batch processing. Single frames are ok, but in the batch process it drifts to nude.

trail ledge
potent fox
#

I am using the RealisticVisionV13 model. Maybe it is the reason.

vestal dew
#

Now it made his hips longer, so they go down further like some kind of freak alien

trail ledge
bitter marsh
#

I am newbie, can someone help me the link for stable diffusion for prompting. thank you

young patio
#

Say, what's the name of that AI program which can simulate an inputted voice?

astral goblet
#

there's a cool editor called graphite, it's all made in RUST so it's very "stable" heh... an alpha preview right now, all node based. it can connect to a stable diffusion server but since i got automatic1111 running, it doesn't connect to that wrapped set of stable diffusion. i need to figure out what it wants still. runs in a browser for now while it's being made. entirely open source so i think you can download it and build it yourself.

it's a very node based editor and preview software, so only use it if you're a tinkerer who likes to custom tailor tools to your purposes. https://editor.graphite.rs/

#

the "imagination" tool is the node that SD would connect to

bitter marsh
#

I am a newbie, can someone help me the link for stable diffusion for prompting. thank you

trail ledge
young patio
#

@bitter marsh If you like learning by discovery, this website: https://lexica.art/ is like Google images for Stable Diffusion.

bitter marsh
young patio
#

Type the sort of thing you want, they click on the images you like, and the prompts used to create them will be shown. 🙂

bitter marsh
#

since, right now I am prompting using my image at midjourney suddenly there is a limit, thats sad

young patio
#

Best to download a program onto your computer and use your own GPU, if you have the power.

bitter marsh
bitter marsh
young patio
#

There are a few. I have one downloaded, though I haven't used it much lately. Give me a minute.

trail ledge
bitter marsh
trail ledge
#

using (nude, naked, nsfw:1.3) in my negative prompt

young patio
#

kek

flint copper
#

I want to generate results of a specific face everytime

#

the face isn't in stable diffusion as it isnt a popular person or character

#

so should I finetune it to generate the same face everytime ?

trail ledge
#

is a specific face someone you have photos of? If so, yes, textual inversion or lora probably your best bet

potent fox
#

Maybe i have the same issue. It is a specific face, some models are more trained on it and some less. I am using a lora already.

fervent thunder
#

hello! vedal9Excite

fathom cedar
#

Is it possible to seperate the main image and a masked area output in Auto1111?
for example
Cat with red eyes
-inpaint over eyes - prompt "blue eyes"
-2 images result, one is full image with blue eyes and another is seperated blue eyes based on mask

fervent thunder
#

Hello

vestal dew
#

was it me you're looking for?

potent fox
#

What if i get only greyscale images using controlnet (i am using a simple grey dummy for the poses)

vestal dew
#

Well...outpaint experiment was a failure. It extended the image, I did it 128 at a time. Made the legs look like cylinders with all sorts of weird seams where each down extension was done. When it gets to it, the feet have 12 toes each

gentle otter
potent fox
#

I found out why my images where grayscale. I have enabled the "apply color correction to img2img results" option. Since my pose images are greyscale...😂

trail ledge
proper karma
#

Just putting this out there: please stable staff, don't let krikey pull y'all into crypto/NFT/MLM/pyramid schemes.

astral goblet
#

Have anyone seen this "Self Attention Guidance" stuff? i think it's similar to what controlnet is doing, just using the generation process to get the control information each loop back, maybe? i need to read more about it still and also, be smarter

gaunt peak
#

Hello, has anyone tried the new "Use PNG alpha channel as loss weigth" feature in embedding training?

fervent thunder
gaunt peak
#

@fervent thunder Is this about the new feature, or just generally about hypernetwork training?

fervent thunder
#

It's my attempt at helping beginners who don't really know what to do to get their first training working

#

I remember when I started, I was lost for a while

gaunt peak
#

Thats quite nice.

Did you try training with the "Use PNG alpha channel as loss weigth" ?; It got recently added to automatic1111 and is supposed to make training on specific features easier.

fervent thunder
#

ooh, I'll add that to my list of things to research, thanks

#

I don't know what that is, but it sounds like de-emphasising the transparent areas of the images, which makes total sense

gaunt peak
#

Yes, I would actually like to use it, for my next embedding, but are stuck with my Photoshop knowledge :x

woven swallow
#

i cant use the program, can someon help me with a request?

#

i have an image macro and need background generated for it

astral goblet
#

woah. woahwoahwoah. nobody told me that the smaller controlnet models load near instantly. i just started using them and woah!

dim lily
#

hello guys, i just joined out of curiosity

#

i have 0 idea where to start but I'm reading faqs and stuff

waxen pasture
dim lily
#

I'll try Arki's Installation guide then

warm junco
dim lily
#

oh thanks

twin wharf
#

Any one know how a art style lora dataset looks like?

#

Can only found character/person lora dataset example, but I want to make a art style one.

astral goblet
#

why didn't anyone tell me about this

#

week now i been using the big models cause i thought it made little difference, but, this is instantaneous loads

brisk bloom
astral goblet
#

i'm disappointed iin all of you, allowing me to be this stupid

#

controlnet models

brisk bloom
#

ahh. I posted them a few times in different channels

astral goblet
#

oh i've known about them, the smaller models. i just had already gotten the larger ones and i didn't think one of the biggest benefits was instant load times

#

as opposed to the longest 1:05 ever

waxen pasture
astral goblet
#

life changing really

brisk bloom
#

and the HDD space saving

astral goblet
#

i got tons of room to stretch out for storage and i plan to build a big honkin NAS too

#

you all been getting instant loads on controlent models while i been here loadin a minute like a chump?? this whole time?

waxen pasture
#

yep :)

astral goblet
#

damnit all

#

everytime it loads fast now, i'm going to be chapped

brisk bloom
#

finally Im seeing attachments as a grid now on discord. I heard others saying it changed but only today it did for me

#

much better than long long posts

grave copper
#

how do people make +18 content with stable diffusion? (like ones on pixiv)

dim lily
#

they probably install stable diffusion directly into their computers because websites have ToS against displaying such

#

just my guess

gaunt peak
grave copper
gaunt peak
#

Did you use Stable Diffusion before?

grave copper
#

no

fervent thunder
#

hi, I have stable diffusion and its awesome. Heard something about openjourney. Its working? I need to update stable? Where find more data?

gaunt peak
grave copper
#

only midjourney

gaunt peak
# grave copper only midjourney

Okay , so you know about prompting and settings a little bit.

Stable Diffusion is free to use and open source. You can run it on a google collab or locally on your PC if you have a good setup.

Stable Diffusion has more than one model you can use and they are all free to download. Some are better for photorealistic stuff and some are better for illustration , anime etc. and some are specifically trained on adult themes. ; While every stable diffusion 1.5 derivative should be able to do nudity. Some do it better.

grave copper
#

Since im not very 'smart' about installing and what not, is this ok?

#

cuz with the 'normal' installation im supposed to install 30 different stuff as pre requesites and what not

#

and apparently this is 1-click option

dim lily
#

currently downloading a Model tho

gaunt peak
grave copper
#

apparently what im installing is a local installation

potent fox
vestal dew
#

for outpaint down, for example, is there a way to go more than 256? If I type in a bigger number, it goes back to 256. Then I have to do one more outpaint to get all the way to feet, which then can look a little off, not exactly even with the leg above, so there's a seam

gaunt peak
vestal dew
#

i am...how do I get openoutpaint?

gaunt peak
#

Extensions tab ; Then load the standard extension url ; and in the list there should be openOutpaint somewhere. Install and restart your webui.

vestal dew
#

standard extensions url?

random fulcrum
#

hey

#

where is the bot?

trail ledge
#

bot's been retired

gaunt peak
vestal dew
#

I found it

gaunt peak
#

👍

random fulcrum
#

oh

#

ok

#

i will do another thing

vestal dew
#

trying to install it caused SD to crash

#

restaring

#

eh...openoutpaint not working. It doesn't populate the model window

#

had to enable api

gaunt peak
vestal dew
#

When I use the regular outpaint, I add more pixels. It extends the image.

#

With openoutpaint, I cant start with an image like img2img. I generate an image, generate below it...the images are related, but the top image isn't extended

#

original is portrait of a man...extended is his lower body, but not of the same guy, not connected to the top

#

It'

#

It's doing something, but I don't know what....trying to extend an existing image to add what's below, like the legs

#

in many cases, it just gives me another image below the first when I add the box below

gaunt peak
#

Since openOutpaint is integrated now you can just do your normal txt2img first and then use "send it to openOutpaint" from there the model and the settings and the prompt are added to their text fields. And from there you can either change the model into an inpainting model for better results. Or just start by using the txt2img tool for adding and the img2img tool in combination with the masking brush for changes. ; Masking brush and txt2img tool can also change things, but more from creating something new instead of changing something existent. The Txt2img tool adds over the "edges" and is necessary for outpainting. Txt2img tool doesn't change unmasked things, so not adding too much at a time is better for models that don't have a good inpainting function. If there are seams you can fix them afterwards with masking and the img2img tool. ( Just to avoid confusion everything after the first sentence are steps you can do with openOutpaint)

vestal dew
#

yes

#

but when I make an img in txt2img, send it to outpaint

#

then extend the canvas...what's drawn in the new part isn't the original extended

#

maybe Im doing too much

gaunt peak
#

With models that are not specifically for inpainting, just adding a square beneath is most likely too much. You can take smaller parts by extending only over the "edge" it will just add the nonexistent part instead of changing the existent square, unless you decided to mask it. If it's masked it can be changed.

vestal dew
#

when I try to extend barely over the edge, it wont add

#

I am trying now the minimum it allows

gaunt peak
#

What do you mean with "it won't add"? Could you provide a screenshot?

vestal dew
#

I am generating an image so I can show

#

well...now it's generating

#

if I make an image 512 x 512, then I have to extend at least one square below or above or right or left

#

It wasn't starting a new image when I did it only one square

gaunt peak
gaunt peak
vestal dew
#

cant watch the vid right now

#

for some reason, Discords changes my audio device, and then I cant get audio back inside a browser

gaunt peak
#

I hope that this answers your questions. I have to log off for now. If you have any questions you can DM me and I will answer as soon as I come back again.

gaunt peak
vestal dew
#

then I wont be able to follow it

gaunt peak
brisk bloom
#

all the video is showing is the inpaint box extending out 2 rows of checkers, the rest of the box is overlapping the original image so the inpaint knows what its generating

vestal dew
#

I just posted with general with images...notice what's extended, what's in the extension, is not a continuation of the original

verbal edge
#

/setting

gaunt peak
#

You need to apply your first prompt