#💬｜general-chat

1 messages · Page 51 of 1

fervent thunder
#

Kaarssteun

empty depot
fervent thunder
#

bruh many people come from ntts videos and all that stuff

#

they come here

#

and ping Kaarssteun

#

but he never answers

broken gull
#

You didn't even ping xd

#

Also why would he

#

You're pretty late to the "controversy " which was merely miscommunication from discords side.

young ivy
#

huh?

#

Maybe I'm just keeping myself to generating and talking about generating and rando crap, but I've never heard of any controversy?

fervent thunder
# 
OutOfMemoryError: CUDA out of memory. Tried to allocate 1014.00 MiB (GPU 0; 7.92 GiB total capacity; 4.52 GiB already allocated; 1.02 GiB free; 5.52 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Time taken: 0.63s

Torch active/reserved: 5143/5694 MiB, Sys VRAM: 7111/8114 MiB (87.64%)
``` 200 mb 😭
#

what can I do to release vram

warm junco
fervent thunder
#

thanks, im using linux though

#

so the .sh

fervent thunder
warm junco
fervent thunder
#

Woah

#

what does it do?

warm junco
fervent thunder
#

...damn

warm junco
#

For nvidia on Linux you have to build your xformers by yourself, dont know how to do that

fervent thunder
#

i just need to install this?

fervent thunder
#

Note: Both Windows and linux should no longer need to build. On 1/23/23 webui shifted from a user-built wheel to an official wheel along with other package upgrades, shown in this PR

#

nvm?

warm junco
#

Hmmm

#

You can try to add --xformers and See of its usable when launching

fervent thunder
#

yeah, i am rn

warm junco
fervent thunder
#

seems to have worked?

fervent thunder
warm junco
fervent thunder
#

yep!

warm junco
#

Nice!

fervent thunder
#

oh my god it uses so little vram!!!

#

thank you so much

warm junco
#

It will help a lot for vram usage

#

No Problem

pastel fern
#

Anyone else shit is breaking cos xformers is forcing to upgrade to pytorch2.0

pastel fern
#

by right the latest version doesnt need xformers

#

but its breaking shit left and right

hasty juniper
vast ingot
#

there isn't a dreambot working currently, sorry morpher

surreal perchBOT
#

@hasty juniper

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

chilly crow
#

This server is managed by Stability Ai right?

vast ingot
#

yep

chilly crow
#

Gotcha

#

Considering the timeline has passed for Artists to opt out of training dataset last week
Does this mean training for V3 is just going to start?

vast ingot
#

I'll be honest, I'm just a community moderator, a community member, and I have no official info on it.
I don't think any public info has gone out about this
So my personal speculation would be, it may start soon, there may be some last minutes things to verify to be certain it's good to start, but if the dataset is set and captioned, it would be the time to start.
But then again, they could also do it other ways, like have secured a core dataset that they could already be training on for some time, before adding the opt'd dataset in

#

just personal speculation once again

jovial maple
#

Sup

#

Og here

pastel fern
#

anyone else getting issues with latest version of xformers?

vast ingot
#

Sup Og'

thorn hare
#

on civitai, when it says "copy Generation Data" how do i insert all the parameters into A1111 ?

cedar garden
#

Hey hey

vast ingot
thorn hare
#

thanks

#

btw, is there a way to pick a certain pic from a created batch and tell A1111, "yes, more like this!"

vast ingot
#

you need to click the "Extra" checkbox that is on the right of the Seed thing, for those options to show

broken gull
chilly crow
lament steeple
#

Is anyone using DAZ studio to generate poses for controlnet? I just want to know what content it comes with. I've been collecting DAZ content since 2007 and I am wanting to do a tutorial, but I need to know what is available to everyone

strong marten
#

i was wondering how can i turnoff safety filter in dreamstudio i got tired every picture being blurred

whole garden
#

thx

fervent thunder
#

when i copy the data from a promt on civit i get way different results what could be the cause?

fervent thunder
#

What is the best website to find cool prompts

flint harbor
warm junco
fervent thunder
#

Thanks once again appreciate it

warm junco
neat condor
wise flax
#

what is the best uncensored text ai?

surreal perchBOT
#
FAQ: What is Stability AI?

Our vibrant communities consist of experts, leaders and partners across the globe. They are developing cutting-edge open AI models for Image, Language, Audio, Video, 3D and Biology.. AI by the people, for the people. Learn more here stability

regal snow
#

Something bizarre is happening here... I have essentially the same setup in 2 locations.
In one location, I am able to use SD Ultimate Upscaler with Inpainting to preserve some details (Inpaint not masked).

#

In another location, when I try the same thing the output image is totally F'd

#

I compared my installed Extensions and they are identical. Checked and am using the same settings. Its pretty infuriating

fervent thunder
#

Hi folks. Looking into ControlNet tutorial on YouTube. Seems like he is able to set two Control Model where I have only one. Why is that?

regal snow
#

The SD Upscaler gets totally screwed after I attempt the Inpainting - like I can't even use it again in Img2Img without reloading SD

regal snow
fervent thunder
#

Thanks a alot!

halcyon granite
#

during dreambooth training my unique prompt word for a character is generating lots of electronics, there's a brand similar to my prompt. Should I cancel the training because that word has something associated with it?

regal snow
#

My SD Ultimate upscale is just very whacky at this location... ControlNet isn't working with it either, is working fine with it at my other location

strange aurora
#

Is there a specific way to use a model for character posing in diffusion, like a mannequin model.

regal snow
regal snow
quasi crescent
#

Is there python code available to use stable diffusion by just passing In a text prompt and getting a image out, I don't want a UI around it, I want to add it to a app

buoyant depot
#

hi

vast ingot
#

Hey !

strange aurora
#

Going back to my original idea of using 3d animation software to make 2d animation through diffusion

regal snow
#

so just stick your manaquin into the cnet window and select openpose for prepocessor and openpose model

proud nova
#

I am trying to generate a photo of two people using two LoRAs, but I can't manage to get both of them inside one image. SD just merges them into one

#

Any ideas?

regal snow
#

have you used latent couple + composable lora?

#

It also works a lot better if you have openpose maps available

proud nova
#

I am trying it right now

#

LoRAs completely stopped working

gloomy anvil
#

aw man, I capped out on pickapic. Anybody else getting shut out after hitting a 3k "score"?

vestal dew
#

I've been following all sorts of controlnet tutorials for hands - none of them seem to work

strange aurora
proud nova
#

With composable LoRA, my LoRAs are completely effect-less

#

SD ignores them

#

And generates random characters not even next to each other

radiant talon
#

is the dream studio model trainable in any way ?

vestal dew
strange aurora
#

I mean, if CD can use an I phone to make a rotoscope anime of rock paper scissors…

#

(I actually don’t think it was filmed on a phone, but still, lol)

proud nova
regal snow
vestal dew
#

Im trying it with images. As long as I reset between images, it does a decent job, and then the txt2img person is in a similar pose

#

I am about to try with someone facing away from the camera.

#

as suspected, it doesn't work

astral pivot
#

any idea what they used to make this video?

urban sluice
#

im making some horror facility images

#

anyone got ideas?

orchid wadi
#

I love the fact that pickapic was basically what I been saying for... 4 monthsm what we need.

#

I actually thought about model level system for finetuning based around it. But alas my grad takes too much time.

#

I'm having so much fun with it 😄

#

Good timewaster

#

Just needs an app and my toilet time is sorted

#

If only there was no NSFW filtering. Since I like to explore the fringes and WTF side of the prompts.ö-

#

What we need now is leaderboards to see who has ranked most. And then peer review sampling, where prompts and earlier solutions are compared randomly.

#

Giving it hard to get phrases, like "Young adult", "diaper with daiper pattern". (Daiper is type cloth pattern - and common typo). "soft hard cheese".

#

So on and so forth.

#

Also... It is still generating stock images with regocnisable watermarks and shutterstock ids!

sleek pendant
#

I am having trouble doing inpainting with stable diffusion 2 using Python gRPC API (stable-inpainting-512-v2-0). I am trying to change the background of a product (in this case, a cream) and although the background is changed correctly, the text inside the product ("facial cream") is distorted in the final result, even though it is not selected in the mask (the body of cream does not change! only the text inside).

I have checked the mask and it is correctly configured, and I am using the API and algorithm correctly according to documentation. However, I am still having this problem and have not been able to find a solution.

Could you help me understand what might be causing this problem and how I can solve it? Any help or suggestions would be greatly appreciated.

past edge
sleek osprey
#

@wise stratus since deep floyd is based on imagen can it do video aswell

nocturne citrus
#

Guys, I want to disable training in Stable Diffusion

#

I got model already trained

regal snow
#

You can disable it in the GUI editing webui.py

nocturne citrus
#

codeformer.setup_model(cmd_opts.codeformer_models_path)
gfpgan.setup_model(cmd_opts.gfpgan_models_path)
modules.sd_models.load_model()

#

Commenting out those?

#

Or those?
import modules.hypernetworks.hypernetwork
import modules.textual_inversion.textual_inversion

#

I only want to generate

lean terrace
#

Looking for a way to (in bulk) fill in black bars of a big old 4:3 full length cartoon movie, would stable diffusion be a right fit? I think not because it has no temporal coherence to my knowledge, maybe some of you all know something else that I could use?

thorny hornet
nocturne prairie
#

Hey everyone, hows it going?

thorny hornet
#

Doing great!!

twin nymph
#

guys, is there any anime model that generates mostly sfw?

zenith igloo
candid kernel
#

Let’s gooo stable diffusion new model released

fervent thunder
#

where?

#

prob just another fan made model

candid kernel
#

No

#

It’s nothing big but I love it

fervent thunder
#

thats just 2.1

#

thats old as hell bro

candid kernel
#

No

#

It’s a finetune of it for image variations

fervent thunder
#

oh

candid kernel
#

(The model itself is hella old though)

odd oxide
#

Anyone know a way I can incorporate stable diffusion into a python script with api? Working on a game where it would be nice to generate visuals for new environments in real time. I will pay for help 🙂

stray ibex
#

Hi all! If I am faced with the task of teaching sb to a certain person, body structure - which method should I use better? Hypernetwork, Lora, train or Embinding?

white mountain
#

lot that can be done with a variations model

#

anyone manage to download it yet?

stray ibex
candid kernel
#

I haven’t done it cause I’m on my phone atm

white mountain
#

I'm still downloading the ckpts myself

wheat beacon
#

hello i m trying to instal stable diffusion and i downloaded all the right models and everything but whern i try to double click on the webui-user it shows me an error message, could i gwet help here ?

#

i m new

buoyant jacinth
wheat beacon
#

exit code: 1

stderr:
P:\Users\oscar\Downloads\python-3.10.10-embed-amd64\python.exe: No module named pip

Launch unsuccessful. Exiting.
Appuyez sur une touche pour continuer...

#

here

#

the last line just says too push a key to close

shrewd trail
#

I will train stable diffusion model on faces, style, objects as ckpt if any one is interested

thick matrix
#

Hello 👋

#

Is this Stable Diffusion server?

vast ingot
#

hello

#

yes !

vast ingot
near silo
#

Howdy heckin hello people

vast ingot
#

Heya Sytan :=) how you doin' ?

lunar vigil
#

hi

near silo
#

I'm doing fine, and yourself?

vast ingot
#

Hello azam

lunar vigil
#

just joined a minute ago lol

vast ingot
#

and welcome around 🙂

#

I'm training a model on cats currently

near silo
#

Welcome :>

vast ingot
#

how could I feel better ?

lunar vigil
#

cats is cute

near silo
vast ingot
#

cats are fun, cute, fuzzy

#

and we just had a contest where everyone had to draw cats

#

it's a fun contest 🙂

near silo
#

@vast ingot oh, is this something I could submit to? What are the images being submitted for?

vast ingot
#

well, it's a little late now

#

the submissions ended 2 hours ago

near silo
vast ingot
#

I thought you knew about it tbh

#

it's every week, open for 5 days

near silo
#

I kept seeing it, but I always forgot to check

#

Interesting, interesting

vast ingot
#

now we vote, then the winner is elected, then we start again with a new theme

near silo
#

So what exactly are they? If you wouldn't mind explaining

vast ingot
#

there is the theme for the current week

#

but currently is voting time

#

and I try to do a model each week with the submissions of each and everyone, to continue the fun

thick matrix
#

What website y'all use to use Stable Diffusion? I only get poor results

vast ingot
#

lots use DreamStudio, but on this discord, a lot of people have it installed or use it in the cloud, through AUTOMATIC, InvokeAI, or other UIs that are available around (check #1080946152318443610 )

poor quality, if you compare to MJ or dallE for same prompt, sure.
SD is a lot more generalistic, being able to be finetuned and do mostly anything, where the others are a lot more prepare to do great art, but not a lot more.
So for an experience "out of the box", for sure SD is worse
but if you start to really play with it, there is no comparison

vast ingot
#

84%...

brave matrix
#

Curious what you guys think are the current best models for photorealism?

trail wing
high falcon
#

Is Civitai.com botching the roll out of new content moderation right now? I just got a new modal popup asking me what I'm comfortable with seeing and everything I clicked is what they're actually hiding lol

#

They're trying to give you flexibility in your boundaries and then immediately violating them 😄

jagged karma
#

I just found one of my creations in the 'output' folder, can I somehow get it back through its prompts etc? 🙂 Thanks!

#

Id love to try out some new variations on it

warm junco
jagged karma
warm junco
near silo
#

I find it has better looking faces/proportions/and a little more tonal contrast

fervent thunder
#

Hello! Just joined the server, was wondering what the name of stability’s language model is called so I can do some YouTubing?

#

I want to learn how to use it

#

Or I thought I just joined. Apparently I’ve been here since Jan.

brisk wagon
#

why don't I have permission to chat in dream bot channels?

near silo
#

don't quote me on this, but I believe the bot isn't active anymore

#

or at least I think i heard something about that

astral goblet
#

the info

near silo
#

ah, my bad

#

I read those a longgg time ago when I joined haha

#

The sheer amount of people being blown away by Adobe Firefly shows just how little people know about AI art generation

fierce sleet
#

hello

fervent thunder
shut forge
narrow sluice
#

It's actually not a bad thing tbh, we're essentially specialists lol

near silo
#

I am working on an image to submit to an AI image generator comparison

narrow sluice
#

And I roasted the shit out of it

#

It can't tell vector from AI generated stuff

#

There's a few styles that would be way harder for people to tell that it's AI, which includes traditional styled drawings, animations, heavily stylized, mixed media, 3D renders, architectural drawings and said vector example I just brought up

near silo
#

@narrow sluice oh no, this is just somebody in reddit comparing the same prompt across MJ, Dalle 2, Bing AI, and now I will be submitting one on behalf of SD lol

narrow sluice
#

I've been in the trenches with the Anti-AI crowd, so I apologize for my reaction. I need to work on it.

near silo
#

No worries haha

red panther
#

Hi

near silo
narrow sluice
near silo
#

MJ V5 is pretty damn amazing

red panther
#

Nobody needs me(

near silo
#

But SD can still beat it with work

fervent thunder
#

Hello everyone

red panther
#

Hello

#

Looks like I'm not welcome here(

shut forge
tawdry portal
#

anyone else encounter a 'zip bomb' warning when using stable diffusion in a colab?

shut forge
shut forge
tawdry portal
shut forge
#

I think it may have something to do with the pcis update, when I initialized it it went away, not sure why but that's what I got when I was troubleshooting and I haven't ran into it since, it could totally be linked to something else though bit that's where I'd start

red panther
#

I can't access Instagram

shut forge
#

Oh darn

red panther
#

What?

tawdry portal
#

what are your favorite google colab templates to use for working with SD?

reef holly
#

How to control the characters in img2img mode not to look at the camera?🤔

jade meteor
#

Hay how can i use stable diffusion also other models too loke realistic vision on Android for free!!!

near silo
feral kelp
#

ahhh man I finally get some time to work on a project and thelastben colab isn't working. Maybe it's getting updated or something. Anyone have another favorite colab that has img2img and the option to load different models?

near silo
#

Sadly I am not sure, this community is very much a local installation sort of thing. Best of luck to you! I know there are a lot of people that go through a lot for generations, and I respect the grind for sure

#

if I hear anything, I will make sure to tell you <3

red panther
#

May i ask?

vast ingot
#

Hello, yes 🙂 ask away

#

and welcome around/sorry for those few first experiences that don't seem to match what you were looking for it seems

tawdry portal
feral kelp
tawdry portal
#

ya but can pay what you want. Low as $0, high as you want.

fervent thunder
#

oh no...what happenned to sd after update, model list looks so bad now...

red panther
vast ingot
# red panther Is it possible to use Stable Diffusion to process people's faces in the same sty...

it is possible, but it can require quite some work, depending on the quality and versatility you are looking for
1/ the most basic way would be to use img2img, and use a mask to "protect" the face from any changes. This will limit what the AI is able to produce for sure, but you can still do quite a lot of things with this. Here is an example I did on my old account : #1011634831467221033 message
2/ controlnet can help keep the likeliness of what you are modifying too now, it wasn't out yet when I did that 1/ example https://www.reddit.com/r/StableDiffusion/comments/119o71b/a1111_controlnet_extension_explained_like_youre_5/
3/ training your Stable diffusion model on that specific face, create a new "model" out of it, that knows how to draw this face in lots of context. This is a whole other subject and requires to learn quite some things to get it right. here is a guide on training (generic, not only for faces) https://github.com/Guizmus/sd-training-intro

red panther
#

Thank you)

#

But I have problems with processing power (Ryzen 5 5600G)

vast ingot
red panther
timid spruce
#

Where to download a stand alone version with a nice GUI... Anyone?

vast ingot
#

those 2 are the most appreciated local UI currently, from what I see around

timid spruce
fervent thunder
#

yo wsp

cyan pond
vast ingot
surreal perchBOT
#
FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

thick matrix
#

Hey 👋 what's the best web based Stable diffusion? DreamStudio, Huggingface, Gradio or other?

boreal prairie
#

Hi everyone, I'm an illustrator and I'm wandering if it's possible to apply a style transfer from my illustrations to photos with Stable Diffusion

vast ingot
vast ingot
# boreal prairie Hi everyone, I'm an illustrator and I'm wandering if it's possible to apply a st...

you can, but it currently requires quite some work.
You'll first need to make a new "model", a memory for SD if you like, that will be trained on your style. This is a little harder than just making pictures, but works really well
Once that is done, you can use that model to make anything with your style, so you can also modify pictures in img2Img and apply your style, yes.
#🔧｜finetune is the place where it's most talked about

boreal prairie
#

Thank you very much for your reply. I understand. Do you know any tutorial or guide on how to create a new model to train SD?

vast ingot
#

as a matter of fact yes, I wrote one recently

vast ingot
untold chasm
#

new video is out!!!

#

enjoy

#

tried to spend more time on sounds with this one too

vast ingot
#

aZ, you post them in #🎥｜animation ? I love your animations but they deserve their spot there too

untold chasm
#

oh didnt know that was there - nice one!! 😄

wraith smelt
#

hello im new

untold chasm
#

hello new im dad

wraith smelt
#

why can I do a link?

#

...

vast ingot
#

no discord invites

wraith smelt
#

he did a link

vast ingot
#

links are ok

#

it's an invite to a server you sent

wraith smelt
#

aaaaaaaaaaaaaaaaaa

#

okay

vast ingot
#

just no discord invites

wraith smelt
#

can you do videos any were?

untold chasm
wraith smelt
vast ingot
#

I said except...
And also, still, this isn't a video link you are trying to share, it won't work on any channel, it's a discord invite and we don't allow those

fervent thunder
#

@vast ingot

vast ingot
#

cleaned up

#

and to report, right click on it, applications, report to staff
or react to the message with ⚠️

#

it sends us a notification

stable pawn
#

Is Automatic1111 Web-Ui act strange today? For me the Model list look strange and I can not click on the image to see it tiled.

astral goblet
#

if you have it auto updating with git pull in your bat file, you should know there was a massive bulk merge today that changed a lot of the UI and is likely breaking a lot of people's extensions. i update often as i can but even i don't put git pull in the launch bat

stable pawn
#

Ah... and now I feel stupid. I saw it, clicked past it and forgot it. Darn stroke I had made my brain all dumb. Thank you, I need those small friendly reminders.

fervent thunder
#

yea, this update didn't do good on model list visuals...

#

spacing between movels is nice (but imo there's no reason for that big of a gap), but damn, word wrap is so bad

stable pawn
#

It is work in progress I hope, it had more upscalers.

astral goblet
#

i blame the parents for not having the foresight to teach us all programming at a young age

stable pawn
#

Naaaa, we need to keep our culture and always blame everything on the programmers, bad lay out, bugs, Greta Thunberg; all programmers fault.

fervent thunder
#

contrast on output description isn't great too

astral goblet
#

you're right. greta's scowl bares a lot of responsibility.

fervent thunder
#

They made at least 3 updates today, so prolly gonna fix issues

astral goblet
#

the mac support for webui has pytorch 2 now. we're getting closer to seeing it proper on windows

#

i always like looking through github insights. they're like patch notes lol. each with it's own little discussion thread

slender vault
#

any good tuts on how to set up SD and other models on colab with little python knowledge?

#

by little i mean none lol

digital wraith
vast ingot
digital wraith
#

Just run I guess?

vast ingot
#

I don't know yours exactly so hard to be sure, but yeah

#

it should just be that

#

run each cell one acter the other by clicking on the arrow that appears when you move over the [ ]

slender vault
#

so here i could paste a path of a model i find on civ i like? ah rip no posting perms

digital wraith
#

or ctrl-F9 to run all

lofty scarab
#

I started to use open pose editor with controlnet to get the exact pose I want from a character. The thing is that when I use controlnet, when I select the image I just created nothing happens when I click on the ''send to img2img'' button. It's kind of annoying because I have to save the image then go to img2img and import the image. Is there a fix to this? If anyone could help please.

astral goblet
hazy pulsar
#

What’s the current best upscale model for old photos?

crimson granite
#

Where can I use stable diffusion ai?

dawn gulch
#

Auto1111 would be a common go to but is currently broken and being fixed, afaik.

astral goblet
# hazy pulsar What’s the current best upscale model for old photos?

the one that works. There are a variety of techniques. Old photos especially benefit from a few passes before upscale. The SD ultimate upscaler extension for a1 webui works wonders if you dial it in well enough. i'm not sure of actual upscaling models yet though. All my best results have come from using sd ultimate upscaler, which works by splitting an image into smaller sections and running them through an img2img prompt with all your selected settings.

lofty scarab
astral goblet
fervent thunder
#

hm, buttons work for me

#

well, at least from png info , never tried from output images

astral goblet
#

oh yeah, sometimes they work for me too! i just stopped trusting them

lofty scarab
astral goblet
#

another thing i get is that sometimes the red onion skin that shows how your generation dimensions line up to the image in img2img tab, totally breaks until i restart the whole process. my work around is that the inpainting tab never breaks so i just send it there to get the numbers i need for proper aspects

#

i should just get a util that lets me click an image an see it's aspect ratio, then calculate any width:height off that

proud nova
#

Aw man, A1111 is broken

astral goblet
#

no just updated

lofty scarab
#

do you guy gets blurry image? I mean not as sharp as usual.

astral goblet
#

i dont know why so many people have git pull in their bat file. A youtube tutorial maybe?

#

it doesn't come that way afaik

#

and of course, the tutorial just says "paste this here" instead of explaining what the purpose and caveats of it are. i'm guessing.

fervent thunder
#

I had seperate file before to update a1111, but...at some point I had too many files to launch a1111 with different parameters, so I just put update in all f them

astral goblet
#

i just type it before i launch. i use my quake style command terminal and double clicking the bat file doesn't launch into that, so i always manually type the commands in. it's all muscle memory at this point

#

start + ` .. its my favorite new hotkey

tribal brook
#

Not sure if this is the proper chat, there are some many XD.
I noticed that my local stable diffusion doesn't use all CPU nor GPU when running.
In fact, the python process only uses ~7% of CPU and GPU is almost always 0%, with rare moments of ~40% GPU.
Is that normal? Images takes minutes to run, can't the process use more of the PC?

fervent thunder
#

I don't like to do repetitive things or to have too messy workflow flooded with unnecessary things.
I mean...I have a mess on my PC, but if it's something I use often or work with - it just has to be somewhat clean...

fervent thunder
astral goblet
#

hell, there was even a period when i was doing early web developmental work where i used total commander, the modern big bro of NC

#

shit, that app may actually help me with image sets now that i think of it. great for file managing

tribal brook
fervent thunder
astral goblet
#

speed differences should be very similar

tribal brook
fervent thunder
astral goblet
fervent thunder
#

my eyes hurt looking at it 😄

astral goblet
#

its all we had and we loved it. in all its' 30hz crt glory

fervent thunder
proud nova
#

Switching models in new Auto1111 is not working anymore. Hope they'll fix it

astral goblet
#

works here on the freshest pull

#

model loaded in 4s even. good times!

#

going back to the other model took 20s. ugh these loaaad times

#

thats a 400% increase. gut wrenching

river sparrow
#

i thought of a brilliant idea. if you help enough with pick a pic on https://pickapic.io/ then you get free stable diffusion credits on dream studio

astral goblet
#

brilliant if it actually works as a promotion and does it's job. otherwise it's just being a charity. it's already offering free generations there

drifting rapids
#

hi

keen pewter
#

Is there a go-to guide for lora training on mac?

fervent thunder
#

does anyone ask gpt for creating prompts? im guessing you all do

storm ermine
#

Good afternoon, can someone help me make an image please

astral goblet
#

chat gpt would slow me down a lot in those workflows

fervent thunder
#

interesting

#

i will look all of that up thanks

astral goblet
#

chatgpt often doesn't know how specific models were refined or what keywords work too

#

iit's a good boost when you're lacking inspiration, but its not a be end all prompt generator

fervent thunder
#

yeah that's what I got too from trying it recently

astral goblet
#

youtube hype artists will often be like "use chatgpt, NEVER PROMPT AGAIN" and i just 🙄 at those kind of sensationalistic headlines

fervent thunder
#

I really dislike those people

astral goblet
#

maybe they see some sort of increase in click turnover on their metrics, but what they're really doing is confusing the entire field instead of offering advice

fervent thunder
#

yeah

astral goblet
#

i prefer to read where i can, but so much of teh content these days is all in a youtube video instead of a page

#

"like and subscribe!!" ugh

chilly shoal
#

my computer keeps freezing when i run the user.bat file

vestal dew
#

anyone know what this means when merging checkpoints? "Error merging checkpoints: 'NoneType' object has no attribute 'pop'
"

cold drum
#

Does anyone know whether anything 4.5 is based off of v1.5 or v2? Need to know for lora training

charred mesa
#

1.5 iirc (can't confirm though)

cold drum
#

I couldn't find anything on it being based off 2.0 and above so I think safe to say below that

#

Thanks

astral goblet
proud nova
#

Damnn.. Why is auto1111 so broken. Like literally half of the extensions I use don't work properly - including controlnet

#

Hope it will be fixed soon

harsh dome
#

its for a big proyect very important in my life

#

its for the change of the world

astral goblet
vast ingot
proud nova
#

Can it be used with CoLab as well?

vast ingot
#

I guess by modifying the notebook to add the hash somewhere but not sure where that would be

proud nova
untold plank
#

Hello chat

vast ingot
vast ingot
proud nova
untold plank
#

Are folks having git trouble?

vast ingot
#

so it seems yeah. the latest version of automatic seems to have problems and folks need to use git to revert

untold plank
#

Unfortunate

vast ingot
#

welcome around by the way 🙂 new to SD?

untold plank
#

Thanks! Yeah, I started looking into image generation a couple days ago

vast ingot
#

quite a magical world to explore 🙂

untold plank
#

It's impressive, but I've got to do some reading (and get my old ML tower out of storage)

#

I've got a bit of an odd background

#

I did computer science in college, and I work in a non-ML section of industry

#

But I'm interested in SD for making book covers / illustrations

#

I volunteer for a small press that republishes out-of-print books

vast ingot
#

I worked in IT most of my life, nothing ML related

#

but never could let go since I installed it

#

for sure, you'll be able to produce quite some interesting, professional looking things, once you get the hang of all there is around 🙂

vestal dew
#

ML?

untold plank
#

Machine Learning

#

(An older and more accurate term for this kind of AI)

vestal dew
#

SO some old ML computer will be good for modern apps?

vast ingot
#

still runing around

untold plank
vestal dew
#

this tech is ML on the back end, but not while it's being used

vestal dew
untold plank
#

It's 64GB RAM + 8GB VRAM on the graphics card

#

2016 processor though, so possibly slow

vestal dew
#

if it could learn, then hands wouldn't still be mangled, and feet, even with all the controlnet stuff

vast ingot
#

most important componant for SD is the GC. 8GB VRAM is enough yeah

undone fox
#

hi all 👋

sterile gust
#

Have they already told us SDXL isn't going to realase publicly yet?

vestal dew
#

8 VRAM is kind of low now

undone fox
#

new here, I was wondering about logo generation

vast ingot
#

logo gens, yep, what are you wondering ?

undone fox
#

bit lost with all these channels but my prompt are all coming out real crusty

vast ingot
#

ok, so to give a little tour of the good channels for that, since we are becoming a little crowded

undone fox
#

alrighty, thanks

untold plank
dapper crater
#

sdxl beta is horrendous rip

vestal dew
#

What if there are a dozen vultures in my yard, and they're digging up the cable for Internet to my house? Is there a channel for that?

vestal dew
vast ingot
vestal dew
#

I have a 4070TI with 12

#

still not enough

vast ingot
#

I'm training on 3090TI 24Gb currently

vestal dew
#

24 is a good number

#

not as good as 40 or 80

untold plank
#

Wew

#

Much as I'd like to have a huge battlestation, that's gonna be a tough sell for the org

#

Since we already have to budget for printing equipment, which is surprisingly expensive

vestal dew
#

you can get 24 for 1500 or so

#

like an RTX 4090

#

still a budget consumer GPU

untold plank
#

How are the pretrained models? Servicable?

vestal dew
#

pretty good

#

but people train for custom things like characters

#

you can mix and match famous people too, to get good looks

#

But for a consistent character, you need something trained

untold plank
#

Most of the images we're looking to generate are inanimate objects

vestal dew
#

such as what?

#

for what purpose?

#

how many images? how often? How much work can someone do on an image?

untold plank
#

It'll vary from title to title, but the book I'm working on right now has a some how-to guides involving household objects, and it would be nice to generate pictures of things like drywall knives

#

Typically a title will have a cover image plus 5-20 illustrations, but again, varies from book to book

#

Since this one has relatively little text, but the instructions are dense, we'll probably add 30+ illustrations to make it easier to follow

vestal dew
#

why do you need a printer to write a book?

untold plank
#

Oh we're not writing books, we're republishing out-of-print stuff

#

Primary sources from major historical events mostly

vestal dew
#

Ah, so you print it

untold plank
#

Exactly

#

Get an ISBN number, register with the library of congress, and print copies for history students to use in their studies

#

It's amazing how many things go out of circulation

vestal dew
#

What kind of drywall knife, for example

gilded wedge
#

hey guys-- been learning how to train my own models. anyone know if there's a way to merge models without using automatic1111? i can't get auto1111 to work for some reason lol. all i want to do is merge a couple models together.

vestal dew
#

one reason a1111 11111 often doesn't merge is that one or more of the models isn't really the model file

#

check the sizes. They should be a few giggity gigs

#

drywall saw, utility knife, curved knife?

untold plank
gilded wedge
#

aw sorry, i meant i can't get automatic1111 to even open ( trying to launch it from colab but i keep getting the error 'Stable diffusion model failed to load, exiting' )

#

was just trying to use it to merge models-- didn't know if there was another way.

vestal dew
#

did you upload a model there?

#

SD needs model files

#

you cant merge models without having some models

gilded wedge
#

i haven't uploaded any models to it yet, i'm just trying to get automatic1111 to launch from colab and it refuses to launch the webui at all. so haven't even gotten to the point where i can upload and attempt to merge the models 🤣

#

i'm not technical as you can tell.

#

so not sure if i'm just doing something stupid

vast ingot
#

yeah the current automatic seems to have a problem

#

lots of people reporting it tonight

#

I'm not sure how to merge outside of it tbh sorry

gilded wedge
#

oh!!! okay! that's actually good news because i've got it to work before and it wasnt working tonight so i thought i was doing something dumb. if thats the case i'll just wait for it to get resolved i guess.

vast ingot
#

yeah, there are posts on reddit about it too

gilded wedge
#

kk cool, thanks for letting me know. ill keep an eye out for when its fixed then

vast ingot
#

how is training going ?

untold plank
vast ingot
#

fix inside

#

but not sure how to apply to collab

gilded wedge
#

pretty good! i'm still trying to figure out all of the different technical pieces, but i've managed to create a model that merges the art styles of some of my favorite 2D animated movies. and it's producing results that look somewhat cool ( sometimes a jumbled mess though ). super fun so far

#

oh awesome, thanks guys, i'll take a look

vast ingot
#

using what tool do you train ?

untold plank
#

Definitely learn git if you want to get into anything technical

#

It's a great tool for all kinds of work

#

I use it for code, taking notes, etc

#

Even have a composer friend who keeps his music in there, not sure how that workflow looks tho

vast ingot
#

yeah but we also have lots of people that come here with 0 technical knowledge, so sometimes I just give the git fish without teaching how to git pull

untold plank
#

That's fair

gilded wedge
#

i've been using dreambooth in colab to train. i grabbed a shit ton of HD images from prince of egypt, road to el dorado, and some ghibli movies. here's one of my fav pieces it's generated so far.

#

oh i guess i cant send images on here

vast ingot
gilded wedge
vestal dew
untold plank
#

That way you'll know what each tool does

vast ingot
#

it's a real hydra on its own, SD. each time you think you got it you learn there's more. so managing the energy investment is important

untold plank
#

And you'll have an easier time figuring out where the problem is

vestal dew
#

post in general with images. I'd like to 👁️

warm jasper
#

Is there a ControlNet or something that you upload a image and specify something like "the green part of the image should be this, the red part that, the blue part..."? Because I'm feeling I'm fighting against SD right now

meager nest
#

So what's the best model right now that can do just about anything? I mostly use Hassan, but I been outta these scene, so idk what makes best photo realistic images these days

#

I mostly use Hassan and novel ai

proud nova
meager nest
#

Interesting, is it stable diffusion 2.0? I have only ever used 1.5 and idk if I need to download anything extra besides that model to get it working

#

Idk if I need a base model for 2 in my folder

fervent thunder
#

thinking I didn't like realistic vision, maybe I don't do prompts right with it or something idk...

meager nest
#

I like Hassan cuz it can do nsfw and regular stuff all in one lol

#

So I can make whatever I feel like without switching models

proud nova
#

Ah 💩, here we go again

fervent thunder
meager nest
#

Ai be great for art poses, if I can't get a pose I like because I need anatomy to copy, it's fantastic for portraits

#

I'll try those 🙂

proud nova
vestal dew
#

I find protogen awful

fervent thunder
#

yea I don't like it that much too

proud nova
#

RV Is better imo

vestal dew
#

all the people it makes look the same, have a doughy look to them

fervent thunder
#

It's just most popular models for realism, different people like different things, let him figure it out.

vestal dew
#

I like EldrithsLucidMix for realism

fervent thunder
#

haven't tried it

#

and don't see it on civi, do you have a link?

vestal dew
meager nest
#

Better than Hassan?

#

I have gotten extremely lifelike results with hassan

vestal dew
#

which hassan?

meager nest
#

Whatever one that was out 2-4 months ago was

fervent thunder
vestal dew
#

I use this one: vae-ft-mse-840000-ema-pruned

fervent thunder
#

yea, same

vestal dew
#

there is hassan dreamlike

#

Feb 10

#

Anything Hassan Dec 22

#

I am going to try the dreamlike hassan

fervent thunder
#

oh damn there's new interesting model

#

kinda a bit too far into fantasy or even paintings based on examples, but still interesting results

candid flume
#

I know that i will look like a noob

#

But

vast ingot
#

I think it closed not long ago ?

#

following the AIs doing more and more photorealistic

candid flume
#

What AI ?

fervent thunder
#

SD?

#

I think even mj5 can do somewhat photorealistic now too

#

anyone in adobe firefly

shadow blaze
#

Hi! I have lots of very good img2img images I did looking for good prompts for an animation in Deforum so, I now that you can set an Init image and that´s great! My question is the following: Is there a way to set or keyframe different images as inits every X frames? To tell the proces to "activate" a reference y an specific time?

vast ingot
#

you can change the denoising, there is a timeline for it, based on keyframes or functions, just like all the other params

#

so you can set the denoising very high for a single frame

#

and change the prompt on the same frame

#

that will change the scene completly

#

very high like 0.99

shadow blaze
#

Ho! Sounds like a great hack!

vast ingot
#

other way is to do your animation in multiple steps

#

because the problem of that first technique is

#

you can't choose an "input image" on that new frame

#

you can just change the prompt

shadow blaze
#

I thought about that but I need a sequence

vast ingot
#

so doing in multiple times lets you handpick the init frame

shadow blaze
#

I cant find the denoise option

vast ingot
#

I haven't used deforum in a while, gimme a sec

shadow blaze
#

THANKS!!!

vast ingot
#

strength_schedule

#

put it very low

shadow blaze
#

There is a Anti Blur option. Is that the one?

vast ingot
#

strengh is the contrary to denoise

#

so put it at 0.01 for a frame instead of my previous tip

#

and I'm not sure about what the antiblur is...

shadow blaze
#

HOOO!

#

sure!

#

YHANKS amigo!

vast ingot
#

no problemo, happy to help !

sick wave
#

Hello, sorry for the werid question. I haven't used Stable diffusion fora couple of month and I had it set to update on start. After using it again, I started getting error about half vae and it feel like generating take more resources. Did something change over the past month?

vast ingot
#

hey !

#

there has been some breaking updates to automatic today yeah

#

there are 2 directions you can go

#

the other is to try and update the extensions, and deactivate those who still are breaking

#

bad luck, updating today after months

sick wave
#

It happens :p Thanks for the help 😉

vast ingot
#

no problem 🙂

fervent thunder
#

hey guys i am about to start to rely on stable diffusion completely to make money, is there a chance in the future when a update will start charging me? do i have to turn off my auto update function?

astral goblet
#

should get a lawyer to break down the licenses the tech is under imo. They're pretty good liceses, but it helps to consult a lawyer on these matters especially if it's your livelihood.

#

anyone doing any kind of business that relies on FOSS should get some sort of lecture on what the licenses are all about imo

fervent thunder
#

i see, thank you

#

i think it will change a lot in the upcoming years

astral goblet
#

these versions now will always be free. you can potentially back up what you got now and it'll always be safe to use

abstract ice
#

Wish they wouldn't call people AI artists unless the artist is 100% AI driven

astral goblet
#

people call themselves ai artists. but whatever. lol

hushed python
#

n stable diffusion, what does "batch count" mean?

#

what does it do

astral goblet
#

right now it seems to do the same seed over an over. maybe fixed by now actually. but it does that many batches one after another. so if your batch size is 4 and batch count is 4, that 16 images

willow needle
#

Does anyone have any resources on how to write prompts that hide hands well or mind pointing me towards anything to help a newbie deal with hands? One other thing in particular is perhaps a resource that shows the nuances of prompts that I might learn to make them better. I've got Stable Diffusion, using WaifuDiffusion, NovelAI(Mostly), Doppleganger, and a generic model all in all. I make basic babe images so far, really just trying to understand it as best I can at the moment.

#

I've been pouring over the resources tab here for a while on another account and this is my main one I've joined with.

rich plover
#

In my respurce manager, I see my GPU memory being used, but it doesn't show the GPU processing power being used. I have a laptop 3080 ti, and I expected I would see usage because of the Cuda Cores. I'm using A1111.

rich plover
haughty mountain
#

Hi everyone, what do I use to sign in on Runpod stable diffusion. It’s asking me for a password

fervent thunder
#

Everyone seems to have monster GPUs here lol

#

Stuff's hard to do with my 1650

#

But I make it work, I think

zenith thunder
#

you can make smaller images and then upscale

#

helps a lot

pliant mauve
#

which channel to go for help?

vast ingot
fervent thunder
#

hi

#

seems like they released a new model on their os but i cant login to test it 😭

#

wuwt (whats up with that)

rich plover
#

I thought SD was supposed to be safe for work. I put in a prompt from a prompt reference site and the output was very nude, despite there being no nudity prompt.

fervent thunder
#

embarrassing

#

but I think you need to have a setting checked

#

someone in the technical support maybe knows more precisely where

rich plover
#

A 'nudity' negative prompt fixed it. But still, it was a bit surprising.

fervent thunder
#

I bet haha

#

glad you got it fixed

#

hope your boss wasn't in your office

rich plover
astral goblet
#

i think you're in a good place with your 1650 to learn an develop your understanding. by the time you get onto the 4090ti, holy moly you'll be balls to the wall skills

vestal dew
#

video - depth library lets you easily fix hands.

Reality - nope

#

I have tried multiple controlnet hand fixes - none work

astral goblet
#

i used that without following a video at all just the other day. it was all pretty straight forward and made sense. i found that the scaling of the hands did change the levels of the greys to a point where the depth got a little washed out though, so i adjusted the levels myself and punched it back into ctrlnet again and hazah! while it wasn't a direct feed into it, it helped a lot.

#

overlaying the hands on the image was a big help. adjusted the png it made and then made the depth guidance come in as a second layer hard and only for 30%. just something i try sometimes for no real reason at all.

#

i also got a lot of success doing that same approach, then using it in inpainting, "whole picture" masking onlywhere the depth hand was and brute forcing that

#

so much success in those processes. orrrr you can just "icant" about it

#

sometimes for hands i'll just sketch one lol

weak kelp
#

hi all. hello from late nights here in california 👋

vernal cliff
#

sleep bruh

vestal dew
vast ingot
vestal dew
#

"this is a game changer" turns out to be the same thing every time - if you generate 10,000 images, maybe one will be good

astral goblet
#

like, i just threw that together so easily. i'm not sure why your'e all sulky about not being able to. if you keep saying you can't you're just telling yourself you can't

#

but i litereally took 3min to do that process

vestal dew
#

I just tried it....6 finger hand

astral goblet
#

the key is to stop watching youtube tutorials. those guys ar eusually just giving you their first go around the block settings. they've not tried anything else and have rushed the tutorial to delivery

vestal dew
#

what does "google hand line art" mean?

#

I've been going on this...just trying to fix one hand on one image, for 14.5 hours

astral goblet
#

open up google right? go to images. "hands" then select the tools menu in the results, and change the mode to "lineart" wabamamam.

vestal dew
#

setting things up, letting it generate, coming back, setting up more

astral goblet
#

old googlefu skills suddenly relevant to new hobbies

vestal dew
#

Google is in SD?

astral goblet
#

lol no. you're allowed to use any image you want. it doesnt' have to come from an extension.

vestal dew
#

ok terrific

#

I went to the google web site, searched, gott images of hands

#

\so?

#

if I go to sketch, it makes a whole new image

#

If I put an image of a hand in sketch, it generates an image of a hand

#

that doesn't fix a hand on an existing image

astral goblet
#

you're doing a lot of "this doesn't that doesn't i can't" but ok

#

just select none for preprocessor

#

ez

vestal dew
#

I just tried what you said...take an image of a hand, put it in sketch

#

it generates a hand

astral goblet
#

i'm just giving you loose advice. it's not meant to be a step by step tutorial. deconstruct the steps. intuit them into your mind. reapply them in your workflows

vestal dew
#

I have tried pictures of my hands

#

3d hands

#

line art hands

#

the hands in depth tool

#

creating depth maps in blender, poser, DAZ

#

I've tried drawing hands

astral goblet
#

well okay, but like, the process works for me with only a little effort. usually get it real nailed in very easily. i only tried describing that to you, not start another argument about why SD can't do hands at all

#

have a great night frank. i'm off to sleep

vestal dew
#

thats what the folks who make the videos claim

#

yet their results show 6 finger hands, hidden hands, mangled hands, flat hands, mitten hands, glove hands

astral goblet
#

yeah, it seems sulky to blame them if those steps aren't working for you. there are a million different setting configurations to explore on your own

#

don't sulk frank. be your best self

vestal dew
#

many of the videos show the process, but never a good result

#

"if you keep playing with cfg and otther settings and generate lots of images, it will be terrific"

#

my best self is a destroyer, a bubble burster\

astral goblet
#

did you just quote me, but then add "generate lots of images" bro, that's just lieing...

vestal dew
#

I am a MIGHTY MAN who lives by my name, FRANK...the truth

#

I destroy BS

#

"THIS IS REVOLUTIONARY"...ok, show us\

astral goblet
#

😿

#

you didn't look at what i showed?

vestal dew
#

I was helping my neighbor find an SEO company. One extolled their virtues...getting you to the top of a search

#

their company was not in the top 10 pages of any search

#

I looked at line art images of hands

#

that's not a hand attached to a human

#

that's not fixing a hand in a SD image, nor is it generating an image 100 the fuck percent with good hands

astral goblet
#

i tend to only show my throw away work on discord. anything you upload here, discord owns. they're going to train on it all

#

you do this often. its weird. you demand results then nitpick why they don't fit for hours

vestal dew
#

when I made a shield to fight off vultures, it didn't take 100 attacks for it to to work. It worked every time so I didn't get cut to ribbons by vultures

#

I demand nothing

#

When someone says "fix hands" and "this is revolutionary - no more bad hands in SD" but it comes down to generating 100's of images - then it's not revolutionary

exotic acorn
#

Yo guys
How do i combine face to body using inpainting ?
And is there a better option ?

vestal dew
#

that's how SD works. Generate 100 images, you get 10 or 20 or sometimes more with good hands

#

what do you mean combine face to body?>

#

You have a body with no head, a head with no body?

#

if you are extending an image to add the rest of a head or add legs, use outpainting

#

but it's the same thing - you'll have to try many images before you get something decent

#

ah...I finally got a decent hand after generating for 12+ hours

#

gigs of images\

#

it has a bad thumb, but I will now try to inpaint just the thumb

#

however, the fix I seek is one that makes SD generate good hands every time, so I don't have to fix images

exotic acorn
vestal dew
#

outpaint

#

but you will have to generate many times\

#

and legs get tricky

exotic acorn
#

Arigato

vestal dew
#

they bend in odd ways when you outpaint

#

use an inpainting model

#

I add legs to head and torso, add the top of heads when cut off

exotic acorn
#

Well i made a head of myself

Should i use an random sd to outpainting the body ?

#

Or i should use the sd 1.5 model i made my face with ?

vestal dew
#

use an inpainting model

#

there is a 1.5 inpainting model

astral goblet
vestal dew
#

I am fixated on getting good images\

astral goblet
#

those people make over $100k a year doing videos. They have that cash in their best interest. let them go. do not obsess over their headlines

vestal dew
#

I am tired of a good one in every way but hands or feet

astral goblet
#

you're hung up on headline claims lol

vestal dew
#

I dont even mind extra limbs - it doesn't happen too often

astral goblet
#

acting like a hero, okay we get it. headlines sensationalise. welcome to the cycle

vestal dew
#

Im hung up on getting images with good hands and feet

astral goblet
#

bro

#

dial it back

vestal dew
#

not monstrocities that look like a guy who hit me with a sledge hammer after I finished with him

#

perfect image with 6 fingers on a hand

#

5 fingers but no thumb

#

8 toes

#

no toes

#

then every time I come in here to see if anyone has a new solution, you take over the conversation with "Stop saying what you can't do"

#

today, you made suggestions, which was excellent. That's appreciated

bleak matrix
#

Good morning, everyone! How are we all today?

fervent thunder
#

Sleepy

#

But I still need to attend church

bleak matrix
#

I know the feeling lol

#

Right now I am just trying to keep myself warm enough to move. There's just enough bite in the air where these bones are like, "Ahh!"

fervent thunder
#

I hate being cold when I wake up

#

Worst feeling ever

proud nova
#

I am starting to be really frustrated

#

Nobody seems to know how to rollback the update of webui on CoLab

dry mulch
#

Hello guys I'm new here, How can I generate image please

vast ingot
surreal perchBOT
#
FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no bot on the server that generates images. However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610 for more details! You can also stop by #1025467151206854736 for any issues you experience while using DreamStudio or #🤝｜tech-support for any problems you encounter while installing it locally!

fervent thunder
#

wew

sterile tree
#

hi

#

where did the dream bot tabs go?

fervent thunder
#

Who even uses dreamshaper or dreamlike

cyan pond
#

Stable diffusion is suddenly twice as slow. I' I saw somewhere that this could be because I updated my GPU driver and that I should re-install cuda. I'm using A1111 on windows 10 with a RTX 2060.

Does anyone know if that could be it, and whats the best way to reinstall cuda?

fervent thunder
#

probably launch installer and see what it suggest, there probably should be delete option

meager nest
#

So I discovered what causes the last few steps to ruin an image: if you have a vae in the models folder that's what causes it

#

Old models of automatic 111 you needed the vae in models, now they should all be in vae folder

#

It causes buggy generations that look deep fried

cyan pond
#

On github it's still in models

warm jasper
#

I'm noticing now how the lines are more straight using CPU instead of GPU. Is this related to the launch options?
CPU: --no-half --use-cpu all --skip-torch-cuda-test
GPU: --precision full --no-half --lowvram --opt-split-attention --always-batch-cond-uncond --xformers

fervent thunder
#

idk, I have one vae inside model folder for 2dn and my SD doing just fine
You don't need them inside model folder, but that's a way to use model+it's vae without actually changing vae
If Ignore selected VAE for stable diffusion checkpoints that have their own .vae.pt next to them is on.

It shouldn't cause any issues if naming scheme is correct

proud nova
#

Does anybody have older version of SD webui on CoLab? I need one real quick.

queen granite
#

are you looking for v 1.5

proud nova
#

No. I need the webui itself. The latest version is broken

maiden moat
obsidian meadow
#

Hi all, is this good compared to Automatic1111? Stable Diffusion web UI/UX

vernal cliff
#

what

obsidian meadow
#

@vernal cliff Askin me?

proper ingot
#

hi,👋 Speak Spanish

sick venture
#

RX 7900XTX or RTX 4080 for Stable Diffusion? AI training included.

astral goblet
#

i've got a 4080. i wanted the xtx but couldn't find it (damn paper launches) . For peopel who don't like to wrangle custom libraries and unsupported driver level code, don't use AMD for AI stuff. the XTX has 24gb ? right? so that's good for training, but the 4080 has 16gb which is still so good for training. There's so much room to operate and learn. So while 16gb might not sound as much, for a starting point its so good. And if you need more memory in the future you can just rent a cloud service for that one job you need it for, or trade up because 24gb cards will get cheaper too.

software support, huge optimizations like xformers don't work on AMD yet. Everything's made for CUDA and AMD works with nvidia's CUDA through janky layers like RocM and directML. While they work, they've don't got a decade of polish on them. You need to wrangle that stuff with technical know how.

For ease of use, nvidia. I say this as someone whose had AMD gpus since they were still ATI gpus. It wasn't until SD came along that suddenly i had a need for Nvidia i didn't want to compromise on.

sick venture
#

How big difference in AI training is 16 vs 24GB VRAM? Namely RTX 4080 vs 4090? Does 4090 have better price/performance ratio in this stuff?

alpine bear
# sick venture How big difference in AI training is 16 vs 24GB VRAM? Namely RTX 4080 vs 4090? D...

The higher VRAM on the 4090 allows larger batch sizes while training, meaning it can work on more images at once, reducing the number of total steps needed to achieve the desired result.

As per price to performance, the 4080 is an abysmal value on its own. It is priced so damn high for what it is that the 4090 is a great value by contrast. I had a 4080 for a few weeks and decided to sell it and buy a 4090 because I believe the 4080 will lose its value very quickly, and the 4090 will at least hold on for a while.

little hemlock
#

hello everyone, can some one help me with textual inversion?

alpine bear
little hemlock
#

so im not quite sure if my machine is able to train an art style

alpine bear
#

specs-wise?

little hemlock
#

i have an intergraded rtx 2060

#

yes, ive tried googles collab also, but no success

#

so its like not sure what to do now

alpine bear
#

A laptop 2060 should be able to train 512x512 embeddings at batch size 1. It should also be able to train LoRAs with lowvram settings enabled.

sick venture
#

@astral goblet @alpine bear Thanks for complex answers.

vast ingot
#

The stars stay the stars.

alpine bear
little hemlock
#

or somewhere else

vast ingot
alpine bear
little hemlock
#

okay thanks

alpine bear
vast ingot
astral goblet
#

i wouldn't get a laptop for training reasons. talking about value, laptops are very poor performance to price values. plus they have horrible thermals in general. meaning you'd have to use the "pause every 5 epoch" settings that i see now an then... to give your machine time to rest between runs.

#

if this is a hobby you want to proceed with, a couple thousand is not a steep investment at all. Most hobbies see price tags like that as an entry point.

#

$2g assumes you've got a desktop to drop the gpu into too lol

#

also $2g for a 4080/7900xtx is a rounded off high figure for "hobby entree fee" discussion purposes

little hemlock
astral goblet
little hemlock
#

alright ill try lora then

astral goblet
#

oh your oriignal question though.. --low-vram in the webui-user.bat file

#

at this point i'm not so sure what the og question was now haha. i'm scattered this morning. need a coffee to get some focus. Good morning generators

#

i'm really happy with my 4080 and i honestly suspect it will hold it's value for a few years quite well

#

resell will be very possible and i'm looking serioulsy at boosting to the Titan ADA when it drops

#

48gb of mem

#

i was hesitant on 16gb too but i've had so much leg room since jumping into it. The speed is so great

alpine bear
#

the 16gb of vram isn't a reason to hesitate imo - it is more than sufficient for AI tasks

#

depends on the task, obviously, but for generating images it is solid

astral goblet
#

i was worried i couldn't train on 2.1 with it but i haven't run into many hiccups yet with that since, i'm still building data sets and ram don't hold that back. i do

alpine bear
#

just fine for training

astral goblet
#

similar to what we see now. the SDXL variant model will be the higher requirement model

alpine bear
#

Whatever it is, I'd wager a 4080 will slay it.

#

@vast ingot do you know the system requirements for SDXL or where we can find em?

astral goblet
#

well maybe the 4080 might approach it cautiously and strategically kill it with smarts

alpine bear
#

They might not be out yet

astral goblet
#

wont just leroy jenkins it

alpine bear
astral goblet
#

all we know is the parameter size whihc is BIG, but i forgot it

vast ingot
alpine bear
#

performance, yes

vast ingot
alpine bear
astral goblet
#

types cheat code tell me all your secrets

alpine bear
#

in my opinion, value. But this is debatable.

vast ingot
#

And by di** i mean dirt of course

#

Lots of dirt here

#

Xl dirt

alpine bear
#

I am trying to find an emoji to respond to this egregious abuse of power

vast ingot
#

I would ban myself if I could reverse it or if weren't Sunday evening

alpine bear
#

Guiz:
owoLickLretardedpepeowoLickR

vast ingot
#

That's dirty

#

I should ban you

alpine bear
#

You should, really you should

vast ingot
#

Nsfw line crossed for sure

alpine bear
#

100%

vast ingot
#

Any other bystander to witness ?

#

I wonder if I could catch problems for that lol

alpine bear
little hemlock
#

erm is it weird if i ask someone if they can train 3 images for me (art srtyle) ? 😅

astral goblet
#

can you train things on that low of a set? i've never attempted

#

lowest i've done is 15

vast ingot
#

He asked for it

#

You are all witness

alpine bear
#

@mighty pond @midnight sphinx @long wind this is the end for me

#

It has been an absolute pleasure

vast ingot
#

NO THAT IS ENOUGH TREXDEL

alpine bear
#

@sudden marlin admin abuse incoming

vast ingot
#

Done

little hemlock
mighty pond
midnight sphinx
#

R.I.P

vast ingot
#

Hate speach like that is too much

astral goblet
little hemlock
astral goblet
vast ingot
#

HAHAHA

long wind
#

rip?

vast ingot
#

For real though, i think i killed them irl with the ban

astral goblet
#

you want me to put the hammer down?! ⚡

midnight sphinx
#

oh you actually did it you mad lad

#

lol

vast ingot
#

Yeah he DISRESPECTED ME

#

Hum

midnight sphinx
#

I'm going to go back to playing runescape

vast ingot
#

Ok so end of joke because i don't want to scare anyone either

long wind
#

I'm literally shaking rn

astral goblet
#

48gb on the titan ADA. whose getting one with me?

#

we'll splits and i'll keep it in my pc

alpine bear
#

back from the dead

midnight sphinx
#

lost your roles

alpine bear
#

the only role I need is miku role

long wind
midnight sphinx
#

I need a croissant

alpine bear
astral goblet
#

Roles? more like, bounces. Lowwww - riiiii -dahhhh

#

y'all are fun to watch ❤️

vast ingot
#

Yeah i didn't know about the roles

#

We'll have that fixed fast sorry

#

I though it was a joke without consequence

#

Last time

alpine bear
#

lmfao xD

astral goblet
#

everyone always thinks jokes until someone gets their roles poked out

alpine bear
#

I knew the roles would disappear

vast ingot
#

I have had Kevi explain to me since

alpine bear
#

It is more work for Jae at the end of the day so honestly it was worth

alpine bear
vast ingot
sudden marlin
alpine bear
#

I'm more disappointed that you did the work for Jae

wild steppe
alpine bear
#

Man I wish I could see the mod chat. Guiz probs got slapped on the wrist like a child for that stunt. What an absolute pleasure.

wild steppe
#

Promoting Guiz to admin

sudden marlin
#

He was thrown into the pit

alpine bear
vast ingot
#

🥲

sudden marlin
#

How'd he get out?

alpine bear
vast ingot
alpine bear
minor rain
#

hello chat

astral goblet
#

batmanclimbingoutofthepit.gif

digital wraith
#

OMG, love the new UI

fervent thunder
#

yea...overall design change isn't bad

#

but some details are a bit broken

lavish current
#

When do you guys think txt2video will be as good as txt2img

#

I wanna make my characters move around and show expression

#

Will even settle for a good enough img2video model for now

warm junco
#

should i update to the newest version?

#

heard its pretty bugged

fervent thunder
astral goblet
fervent thunder
#

seems optimistic to me but i have no idea

astral goblet
#

based on progression trends already

#

1.0 dropped in august

fervent thunder
#

noice

#

xl model seems pretty cool

astral goblet
#

i have no other basis other than "well shit i don't know.. we've come this far in 7months!"

fervent thunder
#

haha

proud nova
astral goblet
#

i didn't think we'd have anything close to controlnet by now

fervent thunder
#

I hope this XL model gets public release this week

#

seems pretty damn cool

astral goblet
#

newest pull works fine for me. i had to do a clean build of venv, copy dll's back over, and fix my launch bat file. thats it

proud nova