🤝｜tech-support

[vocal burrow]

let's go for it again now

yup

[kindred tinsel]

Wait so why would that be an issue, and if it was an issue, should I run the purge command every time?

[sturdy ginkgo]

only when issues arrise .. python/pip loves to stack version upon version upon version in the cache for no reason

[kindred tinsel]

hello!

also okay

[karmic crown]

it will be be faster, since you will not download model again

[kindred tinsel]

Weird, this is a fresh windows install, i wouldn't know about any other versions of any packages in my system

[vocal burrow]

definitely no, I'm just making sure you're not using corrupted / outdated packages

[kindred tinsel]

okay then, it already installed pacakges I am seeing that it is installing requirements for most libraries, clip, open_clip, Codeformer, will update once it launches the webui

@vocal burrow New result

[karmic crown]

looks fine

[vocal burrow]

why use --enable-insecure-extension-access ?

[kindred tinsel]

Always had it there, not sure if it caused any problems

I can remove it if you want

[vocal burrow]

it's removing security features for pretty much no good reasons at all.

so yup remove it, restart, install controlnet and see how it goes

[karmic crown]

i only found one use for it, but i am sure there is some better solution

i had to add it on my sons pc to be able to use infinite image browser over network, but as i sad there is probably better way to do that

[kindred tinsel]

Could it be that backend feature --use-directml

I have never used it before and Idk why the guide asks for it, removing it on my previous isntalls didn't affect performace at all

[vocal burrow]

without it you're not gonna use your gpu at all

[kindred tinsel]

wdym, before I used to see gpu utilization at 100% :c

in the same fork

[vocal burrow]

it's somewhat recent

[kindred tinsel]

okay then

[wet surge]

I don't want to go through this again, I deleted the 3.12 and this is the error

[kindred tinsel]

isntalling controlnet again ;c

Access error

once

again

[vocal burrow]

full log once again :p

we're narrowing the target

[kindred tinsel]

Okay then

[karmic crown]

Command: "C:\StableDiffusionIDES\stable-diffusion-webui-directml\venv\Scripts\python.exe" -m pip install mediapipe --prefer-binary
Error code: 1

again

[kindred tinsel]

ITS THAT C2V fricking file again, jeeeeeeezzzzzzzzz that thing is supposed to be a python dll file or smth god

[karmic crown]

that is not a first error

this one is the problem

[kindred tinsel]

I can't really touch that, its a command created by the program :c

[karmic crown]

try to run this yourself

"C:\StableDiffusionIDES\stable-diffusion-webui-directml\venv\Scripts\python.exe" -m pip install mediapipe --prefer-binary

copy paste

does it fail again?

[kindred tinsel]

yes it does

[steady thistle]

I'm trying to use ReActor with 1.5 and I keep getting weird colors especially around the lips, but it works perfectly with XL models. Any ideas how to get this working on both types?

[kindred tinsel]

Its always that little .pyd file idk what it is but it always dies

[vocal burrow]

probably just using wrong settings (VAE,cfg,etc) then

[vocal burrow]

like we've said, that's not the problem, the real error is mediapipe refusing to install

[steady thistle]

Ah, good thought. I haven't used VAE much. Let me see what I have checked.

[kindred tinsel]

yeah but mediapipe depends on that file,or else it wouldn't even touch it right?

[karmic crown]

try this

"C:\StableDiffusionIDES\stable-diffusion-webui-directml\venv\Scripts\python.exe" -m pip install mediapipe --prefer-binary --user

[kindred tinsel]

isn't that cv2 library smth about open cv

[karmic crown]

this has --user

[kindred tinsel]

This fixed it

removing the fricking space from the name

AND reomving user

[karmic crown]

ok, try to run it again

[kindred tinsel]

try to run what again?

[karmic crown]

webui-user

sorry

🙂

[kindred tinsel]

Also it didn;'t finish installing anything like openmesh that it also needs so like hbow do i get those to reinstall?

[karmic crown]

just run webui-user again, and see if you have any errors left

we will fix one by one with --user

[kindred tinsel]

well no errors arose during installation

[karmic crown]

since that worked

[kindred tinsel]

i mean

[vocal burrow]

the fact that venv can't access user package is weird

[kindred tinsel]

opening that thing

[karmic crown]

at least we managed to install that package that couse the first issue in log

[kindred tinsel]

Could it be that since I am in the root C drive like outside of the user folders it is causing errors?

[vocal burrow]

but it will very probably cause issues later on

[kindred tinsel]

Lets test segment anyting which started this

[vocal burrow]

no

[kindred tinsel]

haha

no

it didnt work

[karmic crown]

new log, please

[kindred tinsel]

that is all it threw

again it is the cv2.pyd file

This is the permissions for that file

it sdays check permissions but idk if this is what it means by that

[karmic crown]

run this one to install supervision properly

"C:\StableDiffusionIDES\stable-diffusion-webui-directml\venv\Scripts\python.exe" -m pip install supervision --prefer-binary --user

[vocal burrow]

wait a sec please

[kindred tinsel]

okay Ill wait :)

[karmic crown]

he sad that to me

🙂

[kindred tinsel]

figured it might apply to both lmao

[vocal burrow]

I'm reading a bunch of stuff

[karmic crown]

installing problematic pacakges with --user will help, but there is no garanty that error will not arise again later as you said

[vocal burrow]

I don't have time to deep dive into it right now.... so go on

[karmic crown]

ok, run this one now

"C:\StableDiffusionIDES\stable-diffusion-webui-directml\venv\Scripts\python.exe" -m pip install supervision --prefer-binary --user

[vocal burrow]

install problematic package one by one using --user but it's weird and kinda ugly, it shouldn't affect your system anyways

[karmic crown]

https://stackoverflow.com/questions/51912999/could-not-install-packages-due-to-an-environmenterror-winerror-5-access-is-de

[vocal burrow]

that's weird....

[kindred tinsel]

The issue is that those are manual isntalls, I can't go around installing manually the dozens of dependencies for each extension :c. This would pretty much render the webui useless for installing directly.

[karmic crown]

you can try this

from that page

[vocal burrow]

problem might be the permisssions on python install folder

lemme check something

[kindred tinsel]

I wouldn't know about that sice for the folder it doesn't show up, that is an install for all users like i had said before :C

[vocal burrow]

go to your python install folder and check permissions for the cv2.pyd file in there Python310\Lib\site-packages\cv2

[kindred tinsel]

rn I did the classic isntall for only my user

Not there

[karmic crown]

it the same, just find where is your installation

[vocal burrow]

run where pythonit should give your your python install folder

[kindred tinsel]

I am there

I am in the one I isntalled,but mycrosoft says there is another thing called python.exe in the market when I haven't installed anyting from the webstore

[vocal burrow]

ok then that's probably the origin of the problem

[vocal burrow]

Yes that's normal, typical microsoft trying to force you to use microsoft store

WindowsApps\python.exe is just a shortcut to microsoft store's python page

Ok so from where I am that's your problem

[kindred tinsel]

So I shouldn't worry about it?

ooop i should

[vocal burrow]

you should have a cv2.pyd in there set like so

hypothesis : because it's installed for "all users" some files are installed somewhere else with different (and incorrect for our use) permissions, permissions that are cloned into your stable-diffusion's venv, so when you're installing stuff in your venv it explodes.

[kindred tinsel]

Where in the original python install?

This is my site pacakges thing

empty af

[kindred tinsel]

That would make sense, but in the isntall I did when we swtarted to work on this issue I installed python only formy user, only adding the python that islocated in the programs folder to the path by checking the Add to path box

[vocal burrow]

potential fix, uninstall everything, including python. Reinstall python ("only for me" not "for all users") and in a different folder to be extra sure that it will use a new folder and therefore permissions, reinstall auto111, reinstall controlnet

[kindred tinsel]

Oh man, okay lets see, for the sd webui should I only remove the venv?

[vocal burrow]

python permissions errors are nightmare ish

[vocal burrow]

it should be enough

remove controlnet extension too while you're at it

[kindred tinsel]

should I purge cache?

[vocal burrow]

also yes just to be sure

have fun

[kindred tinsel]

fun?

meanie

[vocal burrow]

hope it will work

[kindred tinsel]

Installed it here

just for me but there

idk

Okay new fresh install it is in path

One question, in the pyenv.cfg file, it doesn't use the local site-pacakges since that options is false bydefault, is that something I should chagne or nah?

[vocal burrow]

no otherwise it will access your system's python

that's the option I was deliberating on when I told you to wait

[kindred tinsel]

ah i see

well its installing everything once more

[sonic ferry]

Is there some way to search Discord and ignoring a 'user' StableDreamer Bot keeps showing up when I'm trying to look through help/support to see if something has already been asked.

[vocal burrow]

not that I know of, mabe if you block the stabledreamer bot it will work...

[kindred tinsel]

its so weird how none of the initial dependencies cause any errors for the installation, but the fricking cv2 causes so much pain for a little controlnet extension

@vocal burrow well installing again did not work

Im going to something drastic and change the location of my webui install to the user folder

[vocal burrow]

I don't see how that would fix anything

[kindred tinsel]

the c drive doesn't have write permissions on my user for some reason so maybe that idk

[vocal burrow]

however you can try to allow "modification" to every files and folders in your venv

[vocal burrow]

what ?

wtf

[kindred tinsel]

like you heard

[vocal burrow]

hmmm

what's your antivirus ?

[kindred tinsel]

this one is better

[kindred tinsel]

windows defender :0

[vocal burrow]

ok

[kindred tinsel]

Maybe I am missing some software updates?

[vocal burrow]

no clue, there are too many things that could explain why you don't have permissions to write at c: (none good in my book)

[kindred tinsel]

@ornate elk do you know why I might not have persmission to write or modify in the c drive?

[vocal burrow]

try (re)installing sdwebui somewhere you have permissions set-up properly and make sure it's not a sync/cloud folder as it might mess things up too (so no desktop/onedrive/etc)

[sonic ferry]

I have my ComfyUI in Downloads/ComfyUI_Portable/, can't imagine it'd be a root access permission error like it'd probably be in Linux, but just to be safe I put my stuff in Downloads. LOL

[kindred tinsel]

But if the entire ssd, literally the fastest drive that would work best for webui is blocked, this would mean shitty performance and read/write times to the hdds

Literally what I wanted to avoid by buying an ssd big enough to handle os and webui

[sonic ferry]

It's not the drive that's the problem...maybe...I think it's the location, as I said, in Linux, that location would probably give permission issues. I know this is Windows, but maybe move it to Documents, Downloads, somewhere where files would often be saved so less likely to give permission issues.

[vocal burrow]

to save some time you might want to test if a folder will do by simply creating an empty venv and installing one of the default package in there

// go to some folder 
python -m venv ./venv                 // create venv 
venv\Scripts\activate.bat             // activate/enter it
pip install mediapipe --prefer-binary // install mediapipe in it

[sonic ferry]

Actually, why not try and run it as administrator?

[vocal burrow]

you can always install sdwebui on your hard drive and store model on your ssd to get ssd perfs.

hard drive perf is negligeable except when you're loading models and stuff like that

[kindred tinsel]

test venv worked fine

[vocal burrow]

should be good then, let's hope so

[kindred tinsel]

Then probably is that I am installing in local c drive where my permissions are lowered so I dont delete for isntance the windows folder

on other drives the base folder is open for anything since its just a drive

[vocal burrow]

that's not windows's default behavior tho

[kindred tinsel]

Maybe If i install on somewhere that is just a folder inside my user it will work

[vocal burrow]

worth a try

[sonic ferry]

That is where Documents, Downloads, Pictures, typical places to place and save files as I said.

[vocal burrow]

Not a good idea to "run as administrator" anything, especially when you don't know what you're running (some ckpt models can include malicious codes, extensions are not always checked by others, if you're sharing your sessions you could be giving admin privileges to anyone accessing it, etc)

Sure 90% of the time it will work.... until it won't

and then if you've been attacked, good luck. And even you haven't been your whole permissions tree is probably a mess.

[sonic ferry]

Didn't think that'd be something to worry about in Stable Diffusion, imagine someone putting in a crypto miner knowing basically EVERY user in SD has the hardware to make something off of it. LOL

[half island]

Hoi, is there a way to disable pagefile for SD? As i rather it error out with "not enough memory" than hog down my ssd's with large pagefiles that messes up stability/indexing and i have to reboot the pc.

[kindred tinsel]

Okay I reinstalled SD, and rn its running the install for most of the files, clip and the like

[vocal burrow]

not that I know of, but you can manually set your pagefile to be a certain size.

[eternal fox]

Hello

Guys i have rocm intalled and stable diffusion on ubuntu

But everything is choppy i think my 6900XT is not being used

Even a video on youtube was fully lagging

[vocal burrow]

you can monitor gpu usage with rocm-smi

[sturdy ginkgo]

follow all steps from this guide: https://www.reddit.com/r/StableDiffusion/comments/10zfnlj/novice_guide_how_to_fully_setup_linux_to_run/

[kindred tinsel]

@vocal burrow I got the same error on a literal documents folder :c

And its the literal same file

cv2.pyd

[sturdy ginkgo]

try this: https://www.elevenforum.com/t/add-take-ownership-to-context-menu-in-windows-11.1230/

[eternal fox]

it says command not found

[eternal fox]

i will try

[vocal burrow]

< well then try to allow "modify" in permissions for all files / folders / sub files / sub folders of venv

[vocal burrow]

then you probably didn't install rocm correctly, follow the tutorial

I'll be off for now, bye

[kindred tinsel]

nuuuu :c

byeee :c

[vocal burrow]

if nothing works; try opening a ticket on https://github.com/d8ahazard/sd_dreambooth_extension/issues

make sure to precise you're using the directml fork, give them full log.
Name it something like
[Bug] Can't install extension, mediapipe requirements refuses to install => [WinError 5] Access is denied [...]\venv\\Lib\\site-packages\\cv2\\cv2.pyd

[kindred tinsel]

So it didn't work, not even with raised permissions as mentioned in some of the repos for the extensionss, I am not sure what to do now since all of my workflows depend on these stuff. Furthermore, comfy would fail in the same way

[grim bolt]

can someone teach me how to use stable diffusion and avoid google colab from closing all the time? please send me a private message

[glossy edge]

why does it says multiple pictures of nsfw are made with stable diffusion?
does it that work tho

[sturdy ginkgo]

the website has automated nsfw moderation .. so avoid words that could be interpretted to nsfw images .. so it doesn't generate newds etc

[karmic crown]

i might have a solution for you, but you might not like it 🙃

[kindred tinsel]

why?

[sturdy ginkgo]

want nsfw stuf .. use A1111 locally or ComfyUI to run it all local

[kindred tinsel]

why?

[sturdy ginkgo]

VMWare virtual machine with windows with only a admin account and A1111/comfyUI on it 🙂

[karmic crown]

https://helpdeskgeek.com/windows-11/how-to-completely-uninstall-python-on-your-windows-pc/

remove all python installations that you have

[sturdy ginkgo]

use Repo uninstaller so it also cleans the registry and junk folders that the regular uninstaller misses

[kindred tinsel]

That deep? I mean its just literally python :c

@karmic crown Okay I didall of that

everything that had a python on it is gone, direct installer, windows thingy and packages too

[karmic crown]

nice

now install python using installer you downloaded

and install it like list time (with add to path)

and try where python again

[kindred tinsel]

okay?

should I tick tyhat admin priviliges thingy?

[karmic crown]

yes

when you finish, try slightly modifed test of the one you did before

python -m venv ./venv
venv\Scripts\activate.bat
pip install mediapipe --prefer-binary
pip install opencv-python

[kindred tinsel]

I remember I saw something about installing the extension manually and using something about recursive modules or smth like that to install it in the webui

do i need this?

I think I did it last install on my old windows install

[karmic crown]

i would use that

[kindred tinsel]

okay Im going to test that venv thing

this is weird

[karmic crown]

if you manage to install opencv-python without errors, you are fine

that file is not from hell, it is from opencv 🙂

[kindred tinsel]

Both got installed

[karmic crown]

perfect

remove that folder

[karmic crown]

go to your installation

and remove venv

and run webui-user.bat

hopefully it will work this time

since you managed to install two packages that made a mess earlier

[kindred tinsel]

It is installing some requirements now again so we'll have to wait

[sturdy ginkgo]

someone know what this means:

lora key not loaded lora_te2_text_model_encoder_layers_0_mlp_fc1.alpha lora key not loaded lora_te2_text_model_encoder_layers_0_mlp_fc1.lora_down.weight lora key not loaded lora_te2_text_model_encoder_layers_0_mlp_fc1.lora_up.weight lora key not loaded lora_te2_text_model_encoder_layers_0_mlp_fc2.alpha lora key not loaded lora_te2_text_model_encoder_layers_0_mlp_fc2.lora_down.weight lora key not loaded lora_te2_text_model_encoder_layers_0_mlp_fc2.lora_up.weight lora key not loaded lora_te2_text_model_encoder_layers_0_self_attn_k_proj.alpha lora key not loaded lora_te2_text_model_encoder_layers_0_self_attn_k_proj.lora_down.weight lora key not loaded lora_te2_text_model_encoder_layers_0_self_attn_k_proj.lora_up.weight lora key not loaded lora_te2_text_model_encoder_layers_0_self_attn_out_proj.alpha

ERROR diffusion_model.output_blocks.4.1.transformer_blocks.0.attn2.to_q.weight shape '[1280, 1280]' is invalid for input of size 409600 ERROR diffusion_model.output_blocks.5.1.proj_in.weight shape '[1280, 1280, 1, 1]' is invalid for input of size 409600 ERROR diffusion_model.output_blocks.5.1.proj_out.weight shape '[1280, 1280, 1, 1]' is invalid for input of size 409600 ERROR diffusion_model.output_blocks.5.1.transformer_blocks.0.attn1.to_k.weight shape '[1280, 1280]' is invalid for input of size 409600 ERROR diffusion_model.output_blocks.5.1.transformer_blocks.0.attn1.to_q.weight shape '[1280, 1280]' is invalid for input of size 409600 ERROR diffusion_model.output_blocks.5.1.transformer_blocks.0.attn2.to_v.weight shape '[1280, 768]' is invalid for input of size 1310720 ERROR diffusion_model.output_blocks.5.1.transformer_blocks.0.ff.net.2.weight shape '[1280, 5120]' is invalid for input of size 1638400 ERROR diffusion_model.output_blocks.5.1.transformer_blocks.0.attn1.to_v.weight shape '[1280, 1280]' is invalid for input

[sonic ferry]

CPU gen here so gonna be a while but why am I seemingly getting a lot of green after changing my prompt box color???

[kindred tinsel]

@karmic crown should I test by isntalling control net again?

[sturdy ginkgo]

just how the AI works .. it creates blobs of RGB (Red,Green,Blue) and slowly reiterates till it has the shapes of what you asked for, shrinking it blobs down each pass

[karmic crown]

everything else worked correctly so far? no errors in log?

[kindred tinsel]

nope, it started and its running, openpose editor 3d openpose editor both isntalled correctly from the available tab

Now im installing segment anything to see if this will work out sinceit also had errors before

[karmic crown]

if you had no errors so far, try to install extension

[kindred tinsel]

wait wait, im seeing that it is isntalling segment anything requirements, lama cleaner requirements and ultralycits requirements

[sonic ferry]

This is my third try, there is a LOT of green in my last 3 outputs where there wasn't before I started changing my prompt, loaders, and LoRA box colors. 25minutes left, we'll see what I get.

[kindred tinsel]

Im hoping I dont get any errors

[karmic crown]

i hope we finally solved your problem

[sturdy ginkgo]

might just be a CPU thing as it has to render those colors on screen so to save cpu cycles it uses those more, if you don't have a GPU

[sonic ferry]

Oops, CFG bumped up to 8 for some reason when I have originally been using it at 7. Could that be part of it?

[sturdy ginkgo]

https://media.discordapp.net/stickers/967873404181053460.png?size=160&name=shrugdunnoGodfallCharlie

never ran on CPU so

[kindred tinsel]

@karmic crown it got upto here and hasn't thrown me any errors, idk if it died silently or somethying but its there thinking ig

[karmic crown]

give it some time

[vocal burrow]

looks good (still afk) so in the end it was indeed your python and you didn't fully uninstalled before

[kindred tinsel]

I think it was the windows thing, somehow it was probably overwritting something in the installation

fricking windows giving me those executables

[sturdy ginkgo]

go take a walk i would say 😄

[kindred tinsel]

ultralytics is that heavy?

[sturdy ginkgo]

stuff goes slower when you keep looking at it

[kindred tinsel]

HAHAHHAHA

true

fine I will go

https://tenor.com/view/oppenheimer-cillian-murphy-cillian-murphy-peaky-blinders-gif-1787947313354313976

this me, trying to not think about the pc

[karmic crown]

so if i see a truck coming at me, i should just keep looking? 🤣

[vocal burrow]

yes my deer

[kindred tinsel]

if its a volvo it will slow down automatically

if not

ded

[karmic crown]

@vocal burrow save this somewhere, if someone needs to clean everything properly, so you do not need to waste time https://helpdeskgeek.com/windows-11/how-to-completely-uninstall-python-on-your-windows-pc/

[vocal burrow]

already saved :p

[karmic crown]

i like that mac thing 🤣

[vocal burrow]

I wonder why

[karmic crown]

because i like comments like this one 🙃
#🤝｜tech-support message

[kindred tinsel]

well I am scared

it hasn't moved at all

Not sure if the webui is dead or smth but it is stuck there

[karmic crown]

what you have in ui?

[karmic crown]

this looks fine

[kindred tinsel]

the check for updates command went through with no issue

[half island]

Because youtube is using hardware acceleration. You need to disable that in the browser to use cpu for youtube instead. Anything that used GPU will lag while SD is using every inch of performance

[karmic crown]

aplly and restart

[kindred tinsel]

This transformed into

[karmic crown]

we all waithing to see if it work

[karmic crown]

ok, have some patience than

[kindred tinsel]

lmao literally a test case for how to fix this random cv2.pyd file

[karmic crown]

sell that, buy mac, use my script, wait 2 minute and you are done 🤣

[kindred tinsel]

No NEVER

[vocal burrow]

As long as you're patient and willing to learn we're usually ok banging our heads on the wall with you. Your suffering will help others.

[kindred tinsel]

quick tell me a model to download

like onw of the segment anything ones

This is a new error but I think its not related to the install of any library, rather to whatever settings i have

[vocal burrow]

yup probably wrong settings

[karmic crown]

exactly

[kindred tinsel]

yeah but which

[karmic crown]

as long as you do not have permissions erros like you did, it is fine

[kindred tinsel]

But okay this is something else, lets try control net as last thing THAT WAS FAILING hehe

there it goes, its fetching files from github rn,

[vocal burrow]

I haven't used sam in a while, can't help you off the top of my head (I'm still busy)

[kindred tinsel]

its okay (busy on a saturday? damn hard worker)

I regret to inform you that the error is still there @karmic crown

Mediapipe is still the tumor of this extension

[vocal burrow]

time to open a ticket, let the upper mind know

[kindred tinsel]

yeah idk what the hell is wrong with this extension but a ton of people have the same issue and there isn't an straightforward solution

[vocal burrow]

https://github.com/Mikubill/sd-webui-controlnet/issues/2297

(still reading through it)

[kindred tinsel]

There is a commend at the bottom which seems like a good solution, installing in the venv the pacakges direclty and then installing controlnet again

[karmic crown]

and let me guess use --user?

[vocal burrow]

if you can install without it, it would be better

[kindred tinsel]

no it doesn't use user, it just says, install using the venv

then reinstall through repository url and that it

[karmic crown]

good

[kindred tinsel]

For what i read yesterday it is an error with most if not all extensions that affect this library and make changes on the fly during windows installtions. Dreambooth users have the same issue

[vocal burrow]

dreambooth users have many problems.... Save yourself some troubles and do your training outside of auto1111 if possible

[kindred tinsel]

haha I cant even train 8gb of vram aint that much formodel, for loras possibly but idk any other places to train

through command line venv

Or this might be the error, How can I have the requirement already satisfied if the install failed? Therefore the webui already installed that thing somewhere during its own installation. And the thing was trying to change a dll file which I don't think you can on the fly without admin privileges or smth so that might be it

[karmic crown]

try to run web ui now

[kindred tinsel]

Web ui is in Installing from github

It works.

Okay so here is what we did so far:

1. Clean, thorough,of python, be it user installed or the windows default links for python 3 that come with windows 11.

2. Create web ui stuff and normally work, (the webui works fine even if you get those access errors, those only happen with control net)

3. Once you install every other extension (if they give errors the most likely solution is to reinstall them using the following method), and some give you errors you do this:

3.1) Check in their github or in your command line what required libraries failed to install and save them somewhere, along with (if needed) the versions of them. Once you have done so head over to your venv

venv->Scripts

And run activate.bat in a cmd window

3.2) Write line by line each pip install command with the required parameters,
3.3) Then you can try to eithee resintall through extension list or by link

3.4) If everything is done like this I can say it might work ^^

Thought I might put it out there so you guys don't have to manually write all of that and pin it, if you want to use it ofc ^^

***on another note, now that I know how to do it im about to do it allll again to move this install to the c drive where it belongs

[vocal burrow]

Not pinning it because we're trying to keep the top pins clean.
But I'm making a note somewhere about it that controlnet deps versions might be broken atm for some users.

[kindred tinsel]

yeye ^^, as long as our painstaking half a day of troubleshooting helps someone else we did smth right

[vocal burrow]

basically installing every requirements like you did can work.... until some lib makes some drastic change from version n to version n+1 that the extension couldh'nt have expected.

that's why some extensions put a specific version of some libs in their requirements

And that's also why some extensions break everything by installing a specific version of a lib when everyone else is expecting a different version. I'm looking at you dreambooth

But hey. It works 🎉

[sturdy ginkgo]

TL;DR don't use the latest for everything 🙂

[kindred tinsel]

But the good thing is that you can update it through the list since it shows as being up to date, while I dont think drastic dependency changes will happen anytime soon. The entire thing is being held by four libraries, adding or removing would break a fuck ton of stuff

Unless the code is pretty flexible and the like but yeah my method is not as aafe as everyone would want

[kindred tinsel]

So which version of controlnet would you have installed? like how can you even specify that

[sturdy ginkgo]

i use comfy so the manager monitors that and pretty sure will not install on conflicted versions

[kindred tinsel]

aaaaah well comfy does do that but a1111 is pretty bleak with that

[eternal fox]

I have followed that guide till I reboot, but it still says rock not found when I type rocm-smi, why does it do that?

[sturdy ginkgo]

https://media.discordapp.net/stickers/967873404181053460.png?size=160&name=shrugdunnoGodfallCharlie

[karmic crown]

what you get with rocminfo

[sturdy ginkgo]

try running commands with sudo

haven't used linux in a decade

[karmic crown]

do not run things with sudo UNLESS it is really necessary

[karmic crown]

i am sure you do not run everythin as admin on windows, it is same with sudo

[karmic crown]

if you type
rocminfo

what you get?

[eternal fox]

Ill try in a bit

[sturdy ginkgo]

well we need driver level access so i bet Sudo is needed for RoCM installation and usage

Linux is very closed off on that regard

[karmic crown]

i was talking about a1111 and venv

do not run those things with sudo

of course you need sudo to install drivers

my linux server does not have GPU, and this is the only sudo command i used
sudo apt install python3.10 python3.10-venv

i do not count sudo -H -u www-data as use of sudo, since in normal installation that is not need i wanted to use a user with not login for installation, so i had to use some things that normal people not do 🙂

[karmic crown]

i copied the wrong line from history

[eternal fox]

Guys one question

If you install rocm you don't have to install normal drivers

Right?

[kindred tinsel]

Quick question, is there a way to get the schduler methods like sgm_uniform, normal, exponential and the like that comfy has in the webui for testing?

[eternal fox]

I mean the Radeon center ones

[vocal burrow]

I'm not sure it's in the depencies. So I'd say yes you do need to install AMD official drivers.

[hallow latch]

Hey

[eternal fox]

Oh is there someone if knows he's right?

[hallow latch]

Can I remove these somehow?

Like, so that I only got like 3 options?

[eternal fox]

@vocal burrow @karmic crown

this is what i get

[shadow willow]

Help.

[ornate elk]

seems like a connection issue

[shadow willow]

So how do I fix it?

I'm trying to download stable diffusion... Everyone told that it will go easy.

[ornate elk]

using ethernet, no vpn, no third party antivirus

[eternal fox]

try using dmz and no firewall for a moment, and see if it works

dmz is in your router

firewall from windows

[shadow willow]

So I'm suppose to turn off all my antiviruses, disable ethernet cable?

[vocal burrow]

yup looks like you don't have amd drivers

[ornate elk]

no, you should use an ethernet cable

[karmic crown]

looks like drivers are not correctly installed

[eternal fox]

but which one i have to install?

[shadow willow]

Well I am using it

[eternal fox]

i only have to install only rocm, or normal amd too?

[ornate elk]

then make a new folder and try the git clone again

[karmic crown]

see this step in that guide:

6. go to Terminal and type

sudo amdgpu-install --usecase=rocm --no-dkms          

this installs only the machine learning package and keeps the built in AMD gpu drivers

"built in AMD gpu drivers"

[shadow willow]

That stuff is really loading slow

[eternal fox]

they are already installed

[hallow latch]

This is my set COMMANDLINE_ARGS=--xformers --no-half-vae --listen

It takes me like 1 min and 40 sec to generate a image with 20 steps? I used to be like below 10 seconds. ??

[shadow willow]

Please stay with me @ornate elk

[karmic crown]

7. REBOOT your computer <- did you do this one?

[shadow willow]

https://tenor.com/view/wow-weird-skeptical-worried-disgusted-gif-4990489

[eternal fox]

Yes sir

[karmic crown]

and rocminfo reports no drivers installed?!

[eternal fox]

yeah i dont know how that's possible

but if i do rocm show lib -a it says 6.0

[karmic crown]

my server has no GPU, so i cant help too much

byt that thing needs to work in order to continue

[eternal fox]

How is this possible that it doesnt work

[karmic crown]

https://github.com/ROCm/ROCm/issues/1555

[vocal burrow]

sudo lshw -c video

[karmic crown]

try this
/opt/rocm/bin/rocminfo

maybe it is just a path issue

[shadow willow]

I don't really think I have any working Anti virus, as well as VPN so I'm pretty sure it's the firewall

[karmic crown]

https://github.com/ROCm/ROCm/issues/1850

[shadow willow]

@ornate elk Yo how do I disable firewall???

[vocal burrow]

I don't see any driver info in this, try glxinfo

[eternal fox]

do i have to downgrade?

[ornate elk]

are you from asia?

[karmic crown]

is that ubuntu 23 maybe?

[shadow willow]

Nah? More like Middle of Europe

[eternal fox]

no 22.04

[shadow willow]

Central Europe

[ornate elk]

okay good

[shadow willow]

Why you asking me If I'm Asian

[karmic crown]

https://askubuntu.com/questions/1460861/amd-driver-error-on-ubuntu-22-04

[shadow willow]

Okay nvm

Let's forget about it

[eternal fox]

i have kernel 6 @karmic crown

maybe i have to downgrade to kernel 5

[ornate elk]

you can try Restart your PC or Router

[karmic crown]

this one said he solved that

[shadow willow]

Alright.

[vocal burrow]

so what did that glxinfo said ?

[eternal fox]

It showed a long wall of text without any good info

[karmic crown]

this issue is still opened

https://github.com/ROCm/ROCm/issues/1864

[eternal fox]

I'm trying something with the kernel now

Thanks guys I'm trying and I'll tell youc

[karmic crown]

this is your case i am afraid

[vocal burrow]

I love long walll of text without any good info, show it

[eternal fox]

Hahah in a sec

[shadow willow]

@ornate elk BRO IT WORKED

I changed the name of the folder to small letters

[ornate elk]

perfect, make sure you follow my install guide in the pinned message of this channel

[shadow willow]

I'm watching a tutorial on youtube

[vocal burrow]

again, looks like no AMD gpu driver are installed

[shadow willow]

Damn I'm so hyped. I've waited like half year for this, I was always scared of this coding but It doesn't look scary at all

Or that I'm gonna have to download some large files

[vocal burrow]

rocm 5.7.1 should have support for 7900 XT / XTX btw iirc

[shadow willow]

@ornate elk Yo what does "checkpoint" mean? And do I have to download this?

What does the "weights" mean? Do I have to lift something?

[vocal burrow]

"checkpoints" are "stable diffusion models" both are somewhat vague terms.
No you don't need to download any of these, you should already have the SD 1.5 model by default

[shadow willow]

Oh okay. I would like to download a model. Can I just post the url somewhere? Is the download necessary...

[vocal burrow]

What model version to pick ? aka "what does those subtitles and weird letters in the title mean" ? (And yes they can stack up)

• fp16 : model is using float16/16 bits to store its values. Making it lighter, easier on vram, will very slightly reduce the output "quality".
• fp32 : model is using the "default" float32/32 bits to store its values.
• inpainting : model is trained for inpainting specifically (will be worse than other versions when it comes to txt2img/img2im)
• no vae : the model does not include a VAE => read description to see if you are required to use an additional VAE to get best results
• baked vae : the model does include a VAE => do not use an additional VAE with it
• pruned : model has been optimized to have the less impacting part of it removed, will slightly impact the output but not necessarily in a bad/good way.
  ==> tldr : if you've got a gpu with less than 8gb, download fp16 or pruned fp16 models.

[shadow willow]

Yeah the fp16???

[vocal burrow]

yes you need to download whatever model you want to use and put it in the stable-diffusion-webui\models\Stable-diffusion folder

[shadow willow]

How do I download "LoRa" tho?

[shadow willow]

So I have to put the safetensor file in those folders? That's it?

[ornate elk]

whats your GPU?
in my guide are optimization steps for the best performance

[karmic crown]

you might add LCM to this, since many models now has LCM lora merged inside

[vocal burrow]

yes

[shadow willow]

Intel core i7-9700F

Not a bad one

[vocal burrow]

LoRa/Lycoris quick guide :
1/ Where do I put my LoRa : stable-diffusion-webui\models\Lora, Loras are NOT stable-diffusion model but rather patches for said models.
2/ How do I use my LoRa, you can either :

• Invoke it directly in your prompt by using the following syntax <lora:filename:weight> (or <lora:alias:weight> depending of the setting in setting->extra network). eg : <lora:animeTarotCardArtStyleLora_v20Offset:1> for using the Lora animeTarotCardArtStyleLora_v20Offset.safetensors
• Use the UI "Extra Networks" / card system by doing as described in my screenshot. Clicking an associated lora card should automatically insert the lora prompt liked mentioned above. https://cdn.discordapp.com/attachments/1002602742667280404/1191779742534348870/loraHowTo17.jpg
  3/ Be careful, LoRa are trained specifically for some SD version. Lora trained for SD1.5 (and derivatives) models will not load on SD2 models and vice versa (same goes for SDXL). Also even though Lora trained on SD1.5 models should work for every SD1.5 derivatives, they'll probably work best with the model that has been used for its training.
  4/ LoRa, LoHa, LoCon, what are these ? They're all commonly refered as "Lora" but differ slightly technically :
• LoHa : Lora with hadamard decomposition
• LoCon : Lora extended to redisual blocks.
• More explanations : https://stable-diffusion-art.com/lora/, https://www.sdcompendium.com/doku.php?id=lora_0014, https://www.sdcompendium.com/doku.php?id=locon_training_0071, etc

[shadow willow]

Wait. Can I use multiple loras at once?

Also thanks for the guide

[vocal burrow]

yes but the more you stack the trickier it is to get a proper result.

[shadow willow]

I want that model btw. https://civitai.com/models/4468?modelVersionId=57618

So Uh... I need to install these 3gb.

?

[vocal burrow]

yes

[shadow willow]

Crap biscuits.

[vocal burrow]

3.95gb to be more precise

yeah if you're living with 1998 adsl connection you're not gonna have a good time installing stable diffusion

[shadow willow]

Do I have to install stable diffusion too? Or just these models/loras only

[hallow latch]

My stable diffusion is really slow, never been this slow before? What do I need to do? Just a info is that i have re-installed windows and have not downloaded a lot .. So is it something that I may miss?

[vocal burrow]

did you reinstall your gpu drivers ? what does the log says ?

[vocal burrow]

what do you mean "install stable diffusion too" ? you already did didn't you

[hallow latch]

I have installed my drivers

Log as in the console?

[shadow willow]

Yeah?

[vocal burrow]

yes

[hallow latch]

Just showing the progress?

[vocal burrow]

before that

copy the whole log in a .txt and dump it in there

[ornate elk]

i meant GPU not CPU

[shadow willow]

Ohhh you mean the graphics card right?

[vocal burrow]

yes

[shadow willow]

This week Imma be upgrading my specs but for now I'm using 2060 rtx

I'm thinking of 4060-90

[vocal burrow]

check the pinned thread, step 5 #🤝｜tech-support message

make sure to set the correct command line to get the most out of your gpu

[shadow willow]

Uhh I already did the 4 steps

[vocal burrow]

ok ? but I'm pointing you at step 5 tho.

[shadow willow]

Could guide me through?

Where do I find the user bat

[vocal burrow]

though what ? I can copy paste here if you can't click the link nor the pins for some reasons .

[vocal burrow]

in your stable-diffusion-webui folder

[shadow willow]

I meant like tell me the steps

Which?

[vocal burrow]

your explorer doesn't show file extensions by default

[shadow willow]

Ohhh So It's a "batch" and not a "bat"

I clicked on it

[vocal burrow]

it's a .bat

but your explorer is hiding it

like I said, it's hiding file extensions by default

it's installing

[shadow willow]

2gb???

I'm already installing the model

[vocal burrow]

and if it's installing just now you definitely didn't follow any guide / readme / tutorial / pinned message.

[vocal burrow]

you mean downloading not installing.

[vocal burrow]

yup and it's only the first of many downloads.

you'll be there for quiiiiite a while

see you tomorrow

[shadow willow]

...

https://tenor.com/view/one-punch-man-saitama-xmxmx-otterpaka-stare-gif-25655187

[vocal burrow]

and before you ask, no there's no way to skip it.

[shadow willow]

I think I'm quitting.

[vocal burrow]

not really representative of a default install

[karmic crown]

default will have about 3-4GB

[shadow willow]

Bro probably installed like 20 models

[shadow willow]

Nah I'm not doing this coding shit

[vocal burrow]

connection is going kaput

if you're using wifi switch to a cable connection for more stability

[shadow willow]

For frank sake. Before I was using this colab link on google and everything was fine, until they've did an update so that free users can't use it anymore.

https://colab.research.google.com/drive/1wEa-tS10h4LlDykd87TF5zzpXIIQoCmq

Aren't there other google colab links existing?

Can't life be that easy?

[vocal burrow]

don't know, I don't use colab. (partially because of it's unreliability)

[shadow willow]

I didn't had to install anything

Just copy and paste model url and that's it

[sturdy ginkgo]

you had to install linux/windows and a browser 😄

[shadow willow]

The fuck does Linux even mean?

https://tenor.com/view/parks-and-rec-ron-swanson-computer-trash-garbage-gif-21878872

[vocal burrow]

Yup and now it's broken because the lord and might google decided to turn off the valve. People were definitely abusing it.

There's no such things as "free gpu"

[shadow willow]

BRUH

[sturdy ginkgo]

if you want single install SD get fooocus..no need to code ..just run the exe from desktop

you just get limited on what you can do

[vocal burrow]

they'll still have to download a ton of stuff, problem is the same

[shadow willow]

Lady what the hell do you want me to do? Run Stable Diffusion exe? That's not real

Maybe If things were like that, Colab would still be real, but It's not. Because of yall incels abusing it on making hentai

[sturdy ginkgo]

keep it civil before you get banned

[shadow willow]

Excuse me? But who the fuck are you to lecture me?

I'm serious, who are you

Did you like finish harvard?

Or hogwart for that matter?

That's what I Thought

You're a nobody.

[vocal burrow]

ok and now I am telling it to you too.

We're all here volunteering free time to help each others.

So yes let's keep it civil or you'll get bonked for sure. Let's not escalate this any further.

[sturdy ginkgo]

btw @vocal burrow you know why maybe this might happen?

#🤝｜tech-support message

usually shows when using more than 3 lora’s or batch generating

[vocal burrow]

I'd say you're using some models / lora incorrectly

would need to see what settings you're using and full log

[karmic crown]

check this out, but i am not sure if that is the same problem https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/5116#issuecomment-1330941281

[vocal burrow]

blind shot, it could be sd2 model with incorrect yaml, mixing lora with incompatible sd model, etc

[karmic crown]

also this

[sturdy ginkgo]

ah..perhaps i had a sdxl lora instead of all sd15 in the stack

[vocal burrow]

yup

[sturdy ginkgo]

still figuring out how to setup more than 3 lora’s in comfy proper

[vocal burrow]

Theoretically you can mix and match some lora between different sd version.....
A subet of 1.5 loras do in fact also work on XL, those that trained the text encoder more than the unet should work with some limitations.
But in practice.... stick to SD15 / SD21 / SDXL across your whole pipeline and it should be fine.

[hallow latch]

https://pastes.dev/hbUUGiacg9

Like this?

[karmic crown]

btw, do you have any idea how a1111 know when to hide "incompatible" lora? i see some loras in both cases, but some others are only visible depending on model

[sturdy ginkgo]

metadata

basemodel is inside the meta data

[karmic crown]

make sense, but what those "universal" have then there?

since i see a few loras with all models

for example

i always see those two, even they are trained on 1.5

i guess i should compare what they have with others

[sturdy ginkgo]

believe metadata can be set universal upon training if the dataset is on the lower resolution side so 512x512

[vocal burrow]

yes, metadata https://github.com/AUTOMATIC1111/stable-diffusion-webui/commit/699108bfbb05c2a7d2ee4a2c7abcfaa0a244d8ea

[vocal burrow]

sure it works too

[hallow latch]

1 image with DPM++ 2M Karras and 30 steps took Time taken: 1 min. 21.4 sec.

That was with some launch settings

I remove them and now same settings Time taken: 7.9 sec.

[vocal burrow]

what's your gpu ?

[hallow latch]

Something with --xformers --no-half-vae --listen slowed it down very much

[hallow latch]

2080TI 11 GB

[vocal burrow]

and can you screenshot your webpage showing all the settings you've used in your txt2img tab

[hallow latch]

This?

4 Images with same settings now
Time taken: 31.0 sec.

Maybe something with --listen?

[vocal burrow]

--xformers --no-half-vae --medvram-sdxl should be enough

[vocal burrow]

you don't need --listen unless you plan to use SD throught another computer on your local network

[hallow latch]

I do it while at work sometimes.

[sturdy ginkgo]

4 images in batch or after each other?

[hallow latch]

Batch

[sturdy ginkgo]

in batch is slower as it is generating 4 at once

[vocal burrow]

seems ok to me

[hallow latch]

Yeah but before

1 image took 1 min and 30 sec

[vocal burrow]

hmmm do you want me to make it go slower again ? :p

jokes aside

[hallow latch]

Well, it was faster after I removed those launch things ..

[sturdy ginkgo]

1920x1920 16-bit colorspace 😉

[hallow latch]

But maybe with --listen made it super slow?

[vocal burrow]

30 secs for 30 steps of DPM++ on 2080 hardware seems good to me

[hallow latch]

Yeah now it's good

But I don't know why it was so slow before

[vocal burrow]

easy to test, put it back in and see how it goes

[sturdy ginkgo]

it also depends if the model used is still in VRAM or RAM so it doesn’t have to load it in again

[hallow latch]

How do I know that?

[sturdy ginkgo]

keep an eye on the console window when you press generate and it should say loading ….safetensor etc

or just go to the loading bar

[vocal burrow]

by default it only keeps one model in vram at a time.

so whenever you switch model, you'll have to wait for it to load

[hallow latch]

So strange.. I added --listen to it now and work as before

I have no idea what happen

It was super slow yesterday to

and I did relaunch it couple times

``Time taken: 25.2 sec.

A: 3.25 GB, R: 3.72 GB, Sys: 5.7/11 GB (52.0%)``

[vocal burrow]

"as before" as in "slow" ?

[sturdy ginkgo]

GPU cold maybe

[hallow latch]

Ohh.. I mean it works better

[sturdy ginkgo]

my 3060 takes a good 10 minutes of genning before it can go into it/s when i boot my pc after being off all night

[hallow latch]

First time it was this slow actually :/

[vocal burrow]

no idea... could be a lot of things but without proper reproduction steps I wouldn't waste any time trying to chase a ghost.

[hallow latch]

True

[vocal burrow]

that's not normal ^^"""", gpu is not a motor engine.

you might have some bad caps on your gpu

[sturdy ginkgo]

pretty sure it is a win11 thing

[vocal burrow]

I don't have it and I've never heard of anyone having it around here

[sturdy ginkgo]

gpu is at about 23°C on when i turn my pc on in the morning

when genning it goes to 58°C after 10 minutes or so and stays at around 38°C on idle

[vocal burrow]

it's cold but I don't think it's "too cold" for it to malfunction.

[sturdy ginkgo]

curves are wack i think

[vocal burrow]

too cold would be "getting close to condensation" kind of temperature. So under 10°c.

[sturdy ginkgo]

gpu tuning took a good 90 minutes after i installed this card 😄

no biggie ..it generates just fine

can run all my games and do VR at 120hz easily

[vocal burrow]

ok then, just letting you know "no that's not normal"

[sturdy ginkgo]

my pc is never normal ..hahaha was running elden ring on my 1070 at 100+ fps ultra settings with only some hiccups when entering a new zone at 1080p

[hallow latch]

Uhm, anything I can do?

I tried to run a Prompt with Interrogator

Console just says

[lavish meadow]

fellas question, how can i get the best i want from both prompts?

example i like counterfeit but also like olympus

@ornate elk

[kindred tinsel]

Does anyone know how to fix this, I am running into our of memory errors with 8gb vram. For instance I send in a series of 10 images and it fails on the eight image out of memory. I am using the standard webui user bar arguments that are in the pinned section but I was wondering what I could do (be it extension or command line arguments) to avoid these errors but without going down to low ram which would kill my generation time

[ornate elk]

You need to use tiled vae extension

[kindred tinsel]

Wait how?

Tiled diffusion for generation or tiled vae for decoding?

Also doesn't tiled vae lower image quality cus of the way it works?

[lavish meadow]

thats why you use high res fix

and when i used to use it i didnt notice any bad side effects

[kindred tinsel]

Dude I’m here asking because that doesn’t work on 8gb of vram unless you want to be opening and closing the webui (with its launch time) every single time.

[ornate elk]

Nope

Only use tiled vae

When you upscale

Its needed for AMD to not get out of vram that fast when using hires fix

With 8gb vram you should use --lowvram

Also dont go over 225 tokens

[kindred tinsel]

I literally send in 36 tokens haha but I will try.

So if I use that should I turn it on while also using high res fix?

[kindred tinsel]

But isn’t like —low ram meant for like 2-4 go vram

[ornate elk]

yea for nvidia xD

for amd its everytthing below 8 on windows

[kindred tinsel]

Ah man, if I do that I’m going to be left at generations of minutes instead of seconds. God….. well I guess I will try that out, see if the memory assistant also helps but thanks for the info

[ornate elk]

you can still use --medvram too instead of --lowvram

you maybe only able to upscale by 1.5

for your gpu i would recommend upscaling img2img to not get out of vram

[kindred tinsel]

Yeah, the only thing that worries me is that I sometimes leave batches of 30 images to then upscale the best, but rn its failing after eight or sometimes after 20, like randomly, I might have to keep to webui-user.bat files, one for batch with lowvram and one for faster single image generation

[simple tartan]

yea, I'm definitely going to add an a4000 to my pc, doing SD on the AMD and LLM on the CPU is killing me... I can generate 8 images before I get a text response....

[simple tartan]

Just leave the file open in Code and comment out the lines as you need

[kindred tinsel]

but would that even work?

Oh wait I got what you mean

keep it open edit and then relaunch, I thought you mean change as I need during execution lmao

[simple tartan]

nah, change and save, then run, which you can do all from inside Code

[kindred tinsel]

Okay I overclocked my vram lets see if that does something

and also I will use that technique yeah, thank you ^^

[ornate elk]

Do you run oobabooga on Linux with amd?

I think it works with rocm

[simple tartan]

i'm running it in windows on CPU

but if i get an a4000, i can use taht for ooba, and the 7900XTX for SD

[ornate elk]

Ahh okay

[kindred tinsel]

you guys know what could';ve happened in here, literally jumped from 2.82s to 77s per iteration

[umbral compass]

Do I need to download the Controlnet Lora separately for Comyfui? All of my checkpoints from Auto1111 is there in Comfyui when I linked it by editing the extra_model_path.yaml, but the Controlnet doesn't seem to show any LoRa or is hard to use.

Also, so Auto1111 Mikubill Controlnet can't be used for Comfyui. I had to install another form of Controlnet preprocessor.

[ornate elk]

Make sure you dont have Wallpaper Engine or Epic Games Launcher running the background

[kindred tinsel]

there is the steam launcher, but i think its something about the console since clicking on it updates the render cycle and it shows the new speed and all that

[simple tartan]

plus, its the last step from making this machine the ultimate beast, well... unless I fill all 4 M.2 slots with 4TB drives 😄

[simple tartan]

dude, lol, i completely forgot i have WE running....no wonder my SD's were slower than yours

[kindred tinsel]

Okay so far no fails, 10 generations out of 30

[hallow latch]

Is it possible to have Interrogator clip models stored on the PC to avoid having to load them in everytime?

[lavish meadow]

I keep getting two women in my thing when I have in negative multiple

How do I stop that?

[hallow latch]

Are you generating 512x512?

[ornate elk]

yea WE makes a huge impact an can also cause driver crash xD

[lavish meadow]

768x640

[simple tartan]

it also explains why my SD crashed on a generation, immediately after running 100% same parameters

[hallow latch]

Bigger size can make ai generate two.. but you can try.. Solo, 1 female, 1 girl as prompt to try

[lavish meadow]

I'll try that but I've never had this problem with any other model I'm using olympus version 3

[hallow latch]

I have an thinker for you 😉

If you got time :3

[hallow latch]

Some model can also make two characters more often

[kindred tinsel]

Huh @ornate elk do you know why a gpu could just kinda stutter like it gets stuck out of nowhere on a generation step say 7/20 and without intervention from me by touching the screen or keyboard it gets stuck there

[hallow latch]

So .. I have noticed that my prompt are slow sometimes like 1 image is 1 min and 30 sec... I don't know why but I just noticed .. This only happens after I have loaded in a Interrogator clip model!

Only 1 specific one as far as I know right now..

If I use the Clip Model ViT-bigG-14/laion2b_s39b_b160k and use it to get prompt my 1 single runs takes 1min and 30 sec~ isch..

If I restart my stable diffusion and don't run it.. it takes about 20-30 sec for 4!! images!!

[ornate elk]

thats a random issue i also encountered before. nothing much you can do. close all background stuff, minimize the browser, if it still hangs, restarting the webui can fix it

[kindred tinsel]

Also this happened when I changed the viewing size of the webui cmd

maybe this could be? Im not entire sure but rn that it is in full screen its running without issue lmao

[lavish meadow]

ill try that thanks for the help

[hallow latch]

Happy to help 🙂

[kindred tinsel]

I aslo have amd adrenaline pro to monitor gpu stats better so maybe that

[ornate elk]

i would say the clip model doesnt get removed of your vram after you switched to generating. so it uses vram that could be used for faster generating.

[kindred tinsel]

basically this, im not so sure if all of that would cause stutters

[ornate elk]

no if they are all minimized its okay

[hallow latch]

Ohh damn :/ I found that clip model to generate good prompts. Need to avoid this one

[rotund lava]

you are using a1111 with an amd gpu on windows?

[kindred tinsel]

I defy the norms

[rotund lava]

my condolences 🙏

[kindred tinsel]

oh haha then that was the issue then most of them were open

[kindred tinsel]

its actually running quite fast, only thing I cant get is ipadater afaik

Cs1o's guide here helped a lot when setting it so it was actually was fast and not dog water

[ornate elk]

do other controlnets work?

[kindred tinsel]

Yeah all of them, canny, all of the controlnet preprocessors, depth, even that iweird thing that puts all of the colors in high contrast for mapping, line art and anime thingies too

Im only surprised that ipadapter doesn't

Maybe Ill run some other tests on it later and see if it actually just needs a little tweaking

[ornate elk]

do you get an error with ip-adapter?

[kindred tinsel]

It doesn't do anything rather than throw an error

in failed diffusions I sent an image of what the output of one generationw ith ipadapter was

[ornate elk]

the latest?

[kindred tinsel]

yeah, I installed in the morning through the install from url section

[ornate elk]

wait lol

you dont install it trough extensions xD

its a controlnet model

[kindred tinsel]

I have these in preprocessors but models I downloaded all of the ones in the page but nothing came out

I might be making a mistaqke so mind linking me to those models then?

[ornate elk]

you need these: (safetensors)
https://huggingface.co/h94/IP-Adapter/tree/main/models

[kindred tinsel]

Thank you! Ill put them in the controlnet models folder and comment on the results ^^

[ornate elk]

and delete whatever you installed for it in extensions xD

[kindred tinsel]

lmao I will

[lavish meadow]

when you merge a model, what does primary model, secondary model and etionary mean and ow do they work?

Cause i like the pop of this

but the other model dosent do detail aswell as this one its a pastelmix and counterfeit mix

the one in the img above

gening one made in olympus now

but i also like the look and vibrance of this

but it misses alot of the detail

how do i get the vibrance from that to the as i call is pastelcounter model?

[fair oxide]

Hjelo, i could use some help since im a bit of a novice and there arent much i understand. I tried downloading the EasyFluff model, but everything ended up being really "deepfried" and had heavy contrast. I looked at the installation bit and reading the tutorial, it got me hella confused, so i could use some assistance.

when responding, could yall reply to my message so it doesn't get lost

[hallow latch]

I am not sure if this is really Tech Support.. But how can I save Size, steps and sampling settings so that I always start with it?

I try to create one but i can only get it to work with prompts ..

[lost crescent]

Make a photo with these settings - drop that photo into ComfyUI/A1111 and the settings will install?

[hallow latch]

But can't you use this ?

[lost crescent]

Dunno, mine seems to have stopped recording my prompts! And it used to do it so well!

[willow zodiac]

error using ControlNet_aux and the MiDas Depth Map, can't find path specified in C:\Users\$%^&@#\Documents\StableDiffusionAI\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\custom_nodes\comfyui_controlnet_aux\ckpts\lllyasviel/Annotators\cache\models--lllyasviel--Annotators\snapshots\982e7edaec38759d914a963c48c4726685de7d96'. I uninstalled and reinstalled and still get the same error. Any help is much appreciated.

[unborn vigil]

Hey all, I can't seem to get xformers to work in dreambooth. I have the CL Args set as --xformers, and I can see the xformers 0.0.20 version at the bottom of the webui. But I can only select "default" under Memory Attenion... version: 1.7.0  •  python: 3.10.6  •  torch: 2.0.1+cu118

[cobalt lark]

whats the latest version of python sd runs on?

i remember that 3.11 didn't work yet when i used it

[ornate elk]

On windows its 3.10.11

[cobalt lark]

aight thank you!

[ornate elk]

Np, check the pinned message in this Channel for an install Guide

[karmic crown]

those are "styles" they only have positive and negative prompts, and they are stored in a file styles.csv

[karmic crown]

in the root folder of webui is a file name ui-config.json. that file stores all the default settings, which are loaded when you start the webui.

open the file ui-config.json with a text editor

replace

"txt2img/Sampling method/value": "DPM++ 2M Karras"

with (for example)

"img2img/Sampling method/value": "DPM++ 2M",

save & restart(!) (not reload) the whole webui

"txt2img/Sampling steps/value": 20

with

"txt2img/Sampling steps/value": 24

or whatever you want

[hallow latch]

Damn, I was thinking more of the size of image, steps etc

[karmic crown]

you can change whatever you want

there as a million settings there

[hallow latch]

Well, some models I wanna use certain size or steps etc:D

[karmic crown]

"txt2img/Width/value": 512,

[hallow latch]

Like to start with, I may change some stuff

But thats for all?

[karmic crown]

this will change the defaults

(when you just open webui)

[hallow latch]

Yeah

But no way to have several settings?

[karmic crown]

actually there should be an extension for what you want

https://github.com/Zyin055/Config-Presets

i guess you want this

[rugged compass]

anyone know why my upscale by option is perma blurred out?

[jagged flicker]

what should i do now?

[karmic crown]

go to extensions tab, then available and click load from button
type the name of extension and click on Install
when finish, go back to installed tab and click on apply and restart ui button

[jagged flicker]

ty i will try to find controlnet now

raise ValueError("Reference at %r does not exist" % ref_path) it says

[chrome rampart]

question. is it bad that when i look at stat GPU i see peak above 100% for for TDP? More powersupply is 650W which should be ok for 3060 card

[karmic crown]

hmmm, this is strange

[jagged flicker]

cant download from here too

GitCommandError: Cmd('git') failed due to: exit code(128) cmdline: git clone -v --filter=blob:none

where should i download sd? is there a specific location?

[karmic crown]

how did you install webui? did you just download zip maybe?

have you installed git as one of the steps?

[jagged flicker]

i used automatic 1111

[karmic crown]

automatic1111 = webui

[jagged flicker]

i did it

[karmic crown]

downloading zip is not a proper way to install it

what GPU do you have?

[jagged flicker]

my files name has ö letter so

[karmic crown]

which OS?

[jagged flicker]

now its installed

ty for support

3060 12gb and windows btw

[karmic crown]

do not foget to download controlnet models

[jagged flicker]

ty for support again

[karmic crown]

from here here and here

i hope i didnt missed any (1.5, sdxl and ipadapter)

[ornate elk]

Try denois at 0.45
15 hires steps
And upscale by 2
Dont use custom resolution

[rugged compass]

the thing is i literally cant swap back to it

whenever i try adjusting the scale bar it is still greyed out

i dont want to use custom resolution

[simple tartan]

so when i leave another gradia service running and start SD, it opens on port 7861, which is great, but NOTHING seems to work, perhaps I should force the port?

or i just use comfy alongside the other service, since comfy is on some other 8000 port range

[ornate elk]

Restart the webui then, it will set everything to default

[ornate elk]

In auto1111 you can change and force the port with --port

It needs to be a port in the high range

[simple tartan]

yea, just wondering if i should even bother, from what i see, once you master A1111, you move to comfyUI anyway

[rugged compass]

it is still fucked after restarting

[ornate elk]

Not for me xD I never switch fully to comfyui. I dont like that you can waste so much time for making the perfect workflow when I just want to generate some images.
Also everyone has different workflows in comfyui so you never know if yours is even good or what others do better.

But i make images just for fun.
If your on the professional site using comfyui giving you the full Control of everything is perfect.

[rugged compass]

guess i'll just do a fresh installation

[ornate elk]

If you do. Then pls checkout my install guide.
Its in this channel under the Pinned Messages

[rugged compass]

👍

[simple tartan]

Makes sense, I need to train my character Lora first, then decide which I like better

[ornate elk]

Also noteworthy if your on AMD Windows, Tiled Diffusion doesn't work on Comfyui.

Its not compatible but they may rewrite the whole extension

[simple tartan]

I haven't got that working, so I've been playing with sdxl's

[rugged compass]

i fixed it

for some reason in the ui config json, the value is randomly set to false instead of true

wtf nvm it broke again

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker

[rugged compass]

good point if it happesn after the reinstall i'll try it

[strange thunder]

Is there a way to make adetailer "skip" when you use the "skip" button? It is always, for example, filling in faces on a seed I want to skip and its just time and space wasted.

Also, is there a way to get a filename to say something like "after adetailer" or "adetailer used" like images saved before it goes into effect have the wording "before adetailer"?

[willow zodiac]

I am having issues with my ControlNet nodes. It is not finding path in the Illyasviel folders. It is happening with HED Lines node and MiDaS Depth Map at least. Anyone else have this problem or an idea of what I need to do to fix it?

[warm tusk]

I can't seem to get automatic1111 to work remotely today. After I log in it loads for about 30 seconds and then says connection errored out. Upon reloading it says there is no instance running right now. Local (via splashtop) works but is very clunky and hard to use.

[ornate elk]

do you use --share or --listen?

[ornate elk]

what webui are you using?

[warm tusk]

share is the one with the problem, listen works fine. 1.7.0 and i just did a git pull, no help.

[ornate elk]

could be a gradio bug then, would recommend useing --listen + port forwarding in router. Or using Remote Desktop tool like yours, or RustDesk (works good for me but quality isnt that good)

[warm tusk]

Can you set a password with listen?

[karmic crown]

try with --no-gradio-queue

[ornate elk]

nope :/

[karmic crown]

it use http requests instead of websockets, but i have no idea if that works on windows

[karmic crown]

listen works with local network only, while share use globaly accessable link

if you use listen behind reverse proxy, you can add authentification on apache/nginx

[ornate elk]

not if used with port-forwarding

[karmic crown]

that is absolutely bad practice

[warm tusk]

No effect. It was working fine two days ago.

[ornate elk]

make sure you have whitelisted the webui in any browser adblocker

[karmic crown]

oh yes, that might be the issue

[viscid flax]

im trying to set up stable diffusion but it gives me an error which says couldn't clone stable diffusion xl

can anyone help?

[karmic crown]

what GPU and OS do you have? what UI you are trying to setup? which guide you are following?

[viscid flax]

rtx 2050ti, windows

[ornate elk]

hey, if your on Windows you should use the Install Guide at the Pinned Messages in this channel

[viscid flax]

3050**

[viscid flax]

ok ill go do that

[warm tusk]

It does almost the same thing on chrome which doesn't support adblocking of any kind (short of using something like nextdns). No error message but it doesn't work and on a reload says no instance is running.

Perhaps gradio is just being cranky today?

[ornate elk]

chrome still supports adblocking if you using a browser extension for it

[ornate elk]

my guess too, if it not happened the last days

[karmic crown]

let me check

i am starting my linux instance with share

give me a second to try

[warm tusk]

Not on mobile.

[ornate elk]

true, but we didnt know your tried on mobile. are you at work, at home or where?

some networks can be very restrictive

[karmic crown]

for some it currently install everything, like i never run it before, maybe i removed venv for some testing and forgot

[ornate elk]

if you have any questions feel free to ask 🙂

[karmic crown]

Downloading: "https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned-emaonly.safetensors" 🙈

[viscid flax]

it says python 3.12 doesn't work

thats the one i had

i just downloaded 3.10.11

[karmic crown]

you need 3.10

[viscid flax]

now im trying it again

[karmic crown]

my advice is to remove 3.12 unless you really need it

since it will make things more complicated if you leave it

[viscid flax]

remove it how?

k i deleted it

[warm tusk]

At home for my desktop, on mobile via verizon 5g. Using a vpn for one or even both didn't help either.

[karmic crown]

it works fine for me

it is extreamly slow, since there is no GPU, but it works fine

[karmic crown]

may you please open cmd, type where python press enter, and show me the result?

[karmic crown]

1girl
Negative prompt: ugly, nsfw, child
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 191597071, Size: 512x512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly, Version: v1.7.0

Time taken: 2 min. 53.4 sec.

[viscid flax]

i tried it again

but it still gives the same error

[viscid flax]

i went to the file location and deleted 3.12

[viscid flax]

RuntimeError: Couldn't clone Stable Diffusion XL.
Command: "git" clone "https://github.com/Stability-AI/generative-models.git" "C:\Users\cemtu\OneDrive\Belgeler\sd.webui\webui\repositories\generative-models"
Error code: 128

[karmic crown]

that one is not from the guide CS1o made

#🤝｜tech-support message <- this guide

[vocal burrow]

you've installed it in a OneDrive directory + there's non ascii characters in the path ( . )

[karmic crown]

i would still like to see the result of where python to see what your windows see

[karmic crown]

oh... you always notice that first 🙂

do not install inside onedrive as he said, onedrive sync will make a total mess

[viscid flax]

C:\Users\cemtu\AppData\Local\Microsoft\WindowsApps\python.exe

[viscid flax]

ohhh

can i download it anywhere else?

[karmic crown]

as long as it not inside onedrive, google drive, dropbox or similar

[vocal burrow]

if that's the only thing it shows then you either : didn't install python (correctly) or didin't tick the "add to PATH" checkbox during installation

[willow zodiac]

ComfyUI

[viscid flax]

ohh lemme reinstall it

i dont remember it gave me that option

[viscid flax]

yep

did it

[karmic crown]

open cmd window
and check if those works

python --version
pip --version
git --version

[viscid flax]

C:\Users\cemtu>python --version
Python 3.10.11

C:\Users\cemtu>pip --version
pip 23.0.1 from C:\Users\cemtu\AppData\Local\Programs\Python\Python310\lib\site-packages\pip (python 3.10)

C:\Users\cemtu>git --version
git version 2.43.0.windows.

[karmic crown]

perfect

[viscid flax]

RuntimeError: Couldn't clone Stable Diffusion XL.
Command: "git" clone "https://github.com/Stability-AI/generative-models.git" "C:\Program Files\sd.webui\webui\repositories\generative-models"
Error code: 128

i moved it to program files

[warm tusk]

Disabling all extra extensions seems to have fixed it. Odd that it would break share but not listen.

[karmic crown]

that should not be the case

i had problem with infinite image browser that does not work over network (even local one) but ui works fine

[karmic crown]

if you are not a developer and just want to create images, i advice you to install automatic1111 instead, using CS1o's guide. automatic1111 is easy to use web based user interface and you can use SDXL models traiend by users which have better quality than default one (but you can use that one if you want too)

this is the UI https://github.com/AUTOMATIC1111/stable-diffusion-webui/

[viscid flax]

i think thats the one i downloaded

[karmic crown]

but use [this guide](#🤝｜tech-support message) to install it

since your python and git works, skip those steps

start from this step

Make a new Folder on your Drive (not on Desktop, Downloads, Documents, Programms, Onedrive) and name it Ai for example:
C:\Ai\

[viscid flax]

oh ok

[karmic crown]

as Aryetis said yesterday, as long as you are willing to learn, and to listen, we will give our best to help you

[viscid flax]

i followed another guide i found on youtube

and the guy said download the sd.webui file

[karmic crown]

CS1o made a guide that is really simple and easy to understand

[viscid flax]

and run the bat file

should i delete it?

[karmic crown]

and you will be able to update your webui whenever you want

[karmic crown]

leave it for now, as long is not interfere with the guide i told you to follow

you can rename the folder

but if you havent downloaded any models, loras... just delete it

[ornate elk]

Delete your stuff and follow my guide

Then you won't have any problems in the future

[viscid flax]

yeah i deleted it

C:\Program Files\ai>git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
fatal: could not create work tree dir 'stable-diffusion-webui': Permission denied

[karmic crown]

"Make a new Folder on your Drive (not on Desktop, Downloads, Documents, Programms, Onedrive) "

so not in program files (programms)

[willow zodiac]

@ornate elk I may have made my file path too long. I moved it up much closer to the C: drive and they worked.

https://tenor.com/view/facepalm-gif-23730687

[ornate elk]

oh okay 😄 good that it works now

idk if it would helped but in python installer there is also an option to disable file path limit for windows

[ivory bramble]

Anyone know what this means? It's in my webui-user.bat cmd terminal thing Traceback (most recent call last): File "C:\a1111\stable-diffusion-webui\modules\textual_inversion\textual_inversion.py", line 203, in load_from_dir self.load_from_file(fullfn, fn) File "C:\a1111\stable-diffusion-webui\modules\textual_inversion\textual_inversion.py", line 184, in load_from_file embedding = create_embedding_from_data(data, name, filename=filename, filepath=path) File "C:\a1111\stable-diffusion-webui\modules\textual_inversion\textual_inversion.py", line 284, in create_embedding_from_data if 'string_to_param' in data: # textual inversion embeddings TypeError: argument of type 'NoneType' is not iterable

[willow zodiac]

Does that have to be done during installation or can it be done from the CMD prompt?

[ornate elk]

maybe with powershell, but idk, you could just rerun the python installer and select the checkbox at the end

[ornate elk]

are you trying to use an embedding or do you try to train one?

[karmic crown]

based on his questions on promting help, to use

[willow zodiac]

I tried rerunning the python installer and it did not give me that option. But when I uninstalled and reinstalled, I did get the option. Thanks.

[ivory bramble]

Embedding

[carmine sand]

does anyone know where I can get models optimised for olive?

[willow zodiac]

Does git install LFS have to be installed in each folder or does installing once act globally? If not global, does it have to be done each pull/clone...?

[ornate elk]

If your on AMD i wouldnt recommend to use Olive. As its not very good supported in any way and converted models to onnx are very rare.
Better to use the Directml Webui without Olive. Then you can use everything you see on Civitai.com.
For installing follow my guide from here: #🤝｜tech-support message
im using AMD too if you have any questions.

[carmine sand]

thanks

[ornate elk]

Olive is faster but without it you have more usability.

[white thorn]

Hello there, I have an issue with Lora training. Im using Kohya for training and A1111 to test them. (win11 RTX4080 16Gb vram) . everything seems to work when using Network rank and Network alpha at 8 but I get a lora file of 8 to 10Mb .

I saw many people recomand 128 in both but when I use 128 I get a "NansException: A tensor with all NaNs was produced in Unet"
I tried to disable nan check , set it to float 32 in settings and no half but nothing fix it and this is only with my loras at 128 network rank and network alpha not other loras or any other image I create with my a1111.

So if anyone have advice regarding this .. I take it ! Ty !

[ornate elk]

can you show me the txt2img settings when getting this error?

[ivory bramble]

Hi all, wondering if my system is sufficient enough to be running A111 smoothly:

EVGA GeForce GTX 1080 Ti 11GB GDDR5X
G.SKILL TridentZ 3000 Mhz 16GB (2 x 8GB)```
I built this in 2018 so it's a bit dated

and also with the 1080ti, do I need to use low-vram parameter?

[karmic crown]

since you have 11GB, i guess --medvram-sdxl would be fine

[ivory bramble]

What is the sdxl?

[karmic crown]

chek CS1o's guide for nvidia in pinned messages of this channel

[ivory bramble]

I'm bookmarking so many guides today lol

[ornate elk]

yes, i have the same cpu and a 1080 and everything works okay, follow my install guide 🙂

[karmic crown]

this basicly says = use medvram only for sdxl models, but not for 1.5 based models

[ivory bramble]

I found your guide. Awesome!

Ty

I might have to re-install if I have the wrong version of python hmm

[carmine sand]

to install models, can I just use the .safetensors file from civitai?

[ivory bramble]

"Python 3.10.11, 64bit: (any python above or 3.10.6) Not 3.11, 3.12 " how do I find out which version I have?

[ornate elk]

yes exactly, but only if you use the Non olive version of directml

[delicate hornet]

Como creo una imagen?

[carmine sand]

ok thanks for the help

[karmic crown]

open cmd and type python --version

[eager ingot]

Hello I am trying to get to terms of service but everything is blacked out?

[ivory bramble]

Oh it says "Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]"

[ornate elk]

that will work

[ivory bramble]

Ok awesome

[karmic crown]

i am not sure if anyone can help if you do not write in English

[ivory bramble]

Any tips or tricks to make generations a bit faster?

[karmic crown]

use parameters from CS1o's guide