🤝｜tech-support

[ornate elk]

Let it load 🙂

It can take some time

[storm sail]

idk if i can use zluda tho? 😛

[ornate elk]

Yep but it only works with newer GPUs. Not with yours sadly

[fair oxide]

what do i do here 😭

[storm sail]

probably my stuff too old/weak, idk

[ornate elk]

You have a typo in the webui-user.bat

[fair oxide]

oh, how do I fix it?

[ornate elk]

edit the webui-user.bat and screenshot it

also whats your gpu?

[fair oxide]

its an nvidia one im p sure

[ornate elk]

yea but what model? how much vram?

[marble frigate]

hey man i dont have the cmd logs since its happening randomly sometimes it works well and i can create like 5 images and most of the time its just stuck at 100% or 0% with nothing happening nor showing up in the cmd
my gpu is rtx 4070

[fair oxide]

i have 0 idea

its a newer one

[ornate elk]

make sure you have --xformers --no-half-vae in your webui-user.bat
if it happens with sdxl models then add --medvram-sdxl there too

[ornate elk]

you need
--xformers --no-half-vae --medvram

[fair oxide]

alr

[marble frigate]

lemme check

[storm sail]

anyhow, bout ready to install this repo on my dealie, i added --use-directml --medvram --opt-sub-quad-attention --opt-split-attention --no-half-vae --upcast-sampling to commandline args= but i wonder, as the directx dealie says my amd only has 4gigs vram should i change --medvram to --lowvram and idk what is "RX 5xxx cards additional need: --no-half" is that something i should care for? my amd again was radeon r7 m440

[ornate elk]

and if it happens again safe a screenshot of the cmd log

[ornate elk]

try first without --no-half

you can add it later if you got an half error

[storm sail]

righto!

ty 🙂

[ornate elk]

np, make sure you add --lowvram

instead of --medvram

[storm sail]

is --lowvram, gotcha

[ornate elk]

@cosmic mulch the Dev fixed the webui, so you can later git pull again

[storm sail]

i worry bout --nohalf later but still need --nohalf-vae?

[ornate elk]

yes

--no-half-vae

[storm sail]

idk differnce, k, ty

ok, here goes nothing :P, i let this thing go and i be back in while to tell u how it went

[ornate elk]

alright cya!

[storm sail]

man, not caring for --lowvram, i tried another way eaarlier but with same arguement, i try to use stable boy on gimp with this and i get out of memory error for that arg, i wanna try --medvram instead cuz my understanding is --lowvram only use like 40-60% of all vram i have available (weak), i wanna try --medvram (thought i should have more vram) and hope directx thing is wrong (or i read it wrong) 😛

[cosmic mulch]

it loaded. i'll do gitpull again and then run it. thank you so much

[ornate elk]

perfect. also remember the first image generation Will take 15-40 minutes

[cosmic mulch]

none of these are errors i should be worrying about, right?

[ornate elk]

yep thats all fine

[cosmic mulch]

alright, thank you so much! is there any channel in this discord server where i can learn more about the actual models themselves (how to properly use them, the settings to use, stuff like that)

[storm sail]

but i try --lowvram first, again, anyway, first time out here and see if i can use stable boy/gimp with, see how it runs, how long it take to run inference etc, maybe try changing from --lowvram to --medvram later to see difference, etc, try different ways and things 😛

[ornate elk]

we have a channel for questions about image generation.
#📝｜prompting-help

[cosmic mulch]

thanks a lot man, have a nice day. i think that's all the tech support i'll need for a while lol

[ornate elk]

sure try out what works for you, its situational and model and resolution dependend

[ornate elk]

no problem 🙂 yea and if you have any question feel free to ask here again!

[cosmic mulch]

oops im back LOL

i pasted the model in the models tab but i don't see it in the ui for the checkpoint page

and i don't see the lora either

[ornate elk]

whats the model?

make sure its a model and not a lora

[cosmic mulch]

t-ponai3

a version of pony

[ornate elk]

models are
1.5 (2-4gb)
SDXL (6gb)
Pony (6gb)

the model goes into models/stable-diffusion folder

[cosmic mulch]

oh wait.. i think that's a checkpoint

not a model itself

[ornate elk]

model = checkpoint

[cosmic mulch]

oh. welp

[ornate elk]

make sure you first generate one image with default settings

for zluda to work

[cosmic mulch]

oh

i tinkered with the settings a bit

i just changed the sampling steps to 35. i'll change it back to 20 and do a test generation

[ornate elk]

is the 1.5 ema pruned selectet as checkpoint?

[cosmic mulch]

generating with sdxl 1.0 is fine right? these settings

[ornate elk]

yep

now its loading 15-40 minutes

[cosmic mulch]

alright, i'll do some of my work in the meantime

thanks brother

[ornate elk]

no problem

[onyx solar]

For SDXL, you'd not want 512x512.

[cosmic mulch]

oh and, the loras. i pasted them in the lora folder, but i don't se them appearing

[onyx solar]

It should be 1 megapixel or variants of 1024x1024.

[cosmic mulch]

oh i see, i'll change it. thanks

[ornate elk]

you have to check if they are made for pony

[ornate elk]

change it after the first image is finished ^^

dont touch it now xD

[cosmic mulch]

oh yeah about that

i already clicked interrupt

[ornate elk]

:/

[cosmic mulch]

this is the terminal

[ornate elk]

okay looks fine

[cosmic mulch]

it's been stuck on 0% for a while

but im letting it do it's thing

although i did click interrupt

[ornate elk]

it will roughly stay at that for 15-40 mins

without changing

[cosmic mulch]

so me clicking interrupt is fine right? or am i cooked

[ornate elk]

no if you pressed generate now and its grayed out its fine

[cosmic mulch]

it is greyed out and the interrupting... thing is going on, but nothing happened yet. still at 0%, i won't touch it anymore and let it do its thing

[ornate elk]

oh so its still interrupting?

[storm sail]

i guess im typical macho dumb cuz i hate asking for helps usually, lol, i google tons and watch/read tutorials but that windows having issues with torch.compile dealie was confusing me and couldnt find any clear answer to that problem, i reached out and glad i did especially if this directml build works better than openvino, lol

*installing requirements . . . lol

[cosmic mulch]

this is my screen rn

[ornate elk]

you can either let it load and the first image will fail after 40 mins (but after that it will work) or you can close the cmd, relaunch and run a new image generation (to be safe)

[storm sail]

"the Webui will autolaunch in your Default Browser" . . . i wont have to add --autolaunch, hmm?

[ornate elk]

nope

[cosmic mulch]

ill close it, re run, this time don't touch anything at all.

1024 resolution on sdxl, 20 sampling steps

[storm sail]

currently no errors so far, installing reuirements is taking long time (i tap enter few times, hoping it aint stuck) but . . . yup

wondering if i ought to upgrade pip as it says or just leave it, maybe unimportant . . .

[cosmic mulch]

this is the terminal now. the old 100% bar is gone

ill let you guys know in like 40 mins whats up lol

[storm sail]

k, finally moved past "installing requirements" and just installed, woop, now reckon its loading default model

[ornate elk]

pip can be ignored

[storm sail]

and i can use dreamshaper and lcm with this guy no issues?

ok, ignore pip message, check

[ornate elk]

you should be able to use it

but if not then try the lcm lora with it

[storm sail]

reckon, just as my last build, will have to move dreamshaper model to stable diffusion models folder and then edit the couple .py files to install lcm

probably

[ornate elk]

nope lcm sampler is already available by default

you dont need to touch any .py file

[storm sail]

k, i got webui up, guess this thing done installing everything and alls ready to test/play with, gonna screw around with this and let u know later how its working

oh, lcm by default, nice

saves me work screwing with files when im dumb and can mess up, lol, awesome 🙂

then i just move/copy my dreamshaper model to correct folder here and play with this thing some to see how it is

muchos gracias 🙂 ur the best 🙂

ill be back later on to give report how things are going and/or if i gots more questions, meanwhile be blessed 🙂

[ornate elk]

you can also add --ckpt-dir "C:\Path\toMyOtherWebui\models\stable-diffusion"
to the webui-user.bat and it will use the models from there

[cosmic mulch]

oh. the image loaded, barely took any tim wth

like 10 mins or something

i can go on and generate whatever i want now right?

[ornate elk]

yep!

[storm sail]

"you can also add --ckpt-dir "C:\Path\toMyOtherWebui\models\stable-diffusion"
to the webui-user.bat and it will use the models from there" . . . sweet! nice trick to save hard drive space 🙂

[ornate elk]

yep, for loras use --lora-dir "C: path..."

[cosmic mulch]

uhm, it went crazy after the image was at 99%

and then it shut itself down

[storm sail]

ok, i know what lora is kinda, dreamshaper is lora something (ok, fine, idk wtf lora is :P) what is ckpt (ahh, checkpoint?), tbh, i know next to nothing, ok, nothing, bout any of this stable diffusion lol

[ornate elk]

what was the resolution?

[cosmic mulch]

512

[ornate elk]

do you have the full cmd log?

[storm sail]

this directlml does seem faster than openvino/cpu, idk how much faster but . . .

[cosmic mulch]

unfortunately not

[storm sail]

lol, b*tch automatically means a woman to ai, not a female dog, lol

that did seem faster, image was kinda crap illustration but models/inferences/other settings

[ornate elk]

make sure you dont run anything heavy in the background like games or wallpaer engine (or epic games store (buggy))

then relaunch the webui-user.bat and test again

[cosmic mulch]

sorry

how do i send everything here without it taking up the discord msg limit

[ornate elk]

send it as txt file

[cosmic mulch]

basically it only seems to happen when im using the other checkpoint

t-ponynai3

[ornate elk]

the pony ?

ah that was my tought

for sdxl and pony you need to do two things

[storm sail]

i try dreamshaper, better prompt, etc, etc, etc . . . i try gimp/stable boy plugin with this, etc, etc . . . whoa, u just send me info about zluda CS1o?

[cosmic mulch]

yeah?

[storm sail]

@ornate elk i got a flash of something looked like info on zluda then it vanish

[ornate elk]

first download this sdxl vae file:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main
and put it into models/vae

Then in the webui go to
Settings/ UserInterface/ Quicksettings
There add sd_vae
then hit apply and reload ui
Then select the sdxl VAE in txt2img at the top

[cosmic mulch]

which one 😭

the .vae or _vae

[ornate elk]

the .vae

[ornate elk]

top one

[cosmic mulch]

alrighty downloading

[ornate elk]

okay idk where?

[cosmic mulch]

oh and.. aren't most images that you see of ai upscaled or refined in some way? is there any way i can do this automatically? like after i generate an img, it gets auto refined or upscaled to fix the flaws

[ornate elk]

yes thats easy doable with hires fix in txt2img and the adetailer extension

[storm sail]

@ornate elk u did message me something info about zluda only a minute ago?

[cosmic mulch]

is there any yt video or something?

[ornate elk]

nope

[storm sail]

oh, idk, maybe my discord jumped on the page a bit to conversation we was having earlier, ok, anyhow

ok, gonna go screw with this directml and see yall later tell u how things is, meanwhile, be well

[ornate elk]

sure have fun!

[ornate elk]

i dont think so, bu i can screenshot you the hires fix example

[cosmic mulch]

userinterface has no quicksettings page

[ornate elk]

you have to click on User Interface below again

[cosmic mulch]

got it. thanks

in the sd vae on the general page, i select sdxl.vae?

or pick automatic?

[ornate elk]

you select the sdxl one only when you use sdxl or pony models

[cosmic mulch]

pony models work with the sdxl vae?

[ornate elk]

yep

pony is based on sdxl

[cosmic mulch]

dang.. what about sd 1.5? how can i find what vae i need

[ornate elk]

1.5 can work with kl-f8-anime2, clearvae, 84000-mse vae

https://civitai.com/models/276082/vae-ft-mse-840000-ema-pruned-or-840000-or-840k-sd15-vae

put that into models/vae

then you can select it for 1.5 🙂

[cosmic mulch]

on the first error i just used the vae. on the second error i tried the hire's fix and refiner

[cosmic mulch]

okay thanks ^^

i keep getting the 2nd error

from the second error.txt

it crashes without a trace now.. interesting

[storm sail]

thinking maybe that was what flashed in front of me little while ago, some stuff bout zluda, guess walker using zluda or something related, lol

[ornate elk]

You need to set hires steps to 10

[fair oxide]

just a question. what are the best models for using a reference photo?

not sure if its the right question to ask

but i want to give it a few photos to output a drawing with a prompt

[cosmic mulch]

alright let me try it. now im not getting an error when i generate an image, but after it goes to 99%, it just says connection lost on the website and says press any key to continue

are these settings fine?

[cosmic mulch]

even then, sometimes mid generation, it just randomly stops

[storm sail]

is it ok if i sit in this room until im done testing this directml deal and then report how r things?

[ornate elk]

Which model do you use?

[storm sail]

idk if just staying here meanwhile is against some rule . . . oh, ok, cool, ty 🙂

[ornate elk]

If you want to convert images into drawings then you need controlnet with lineart or canny

[cosmic mulch]

t-ponynai3

even with sdxl, i get the same error. it happens randomly

sometimes it generates, sometimes it doesn't

[storm sail]

run outta memory error in webui, gonna try --medvram see if i get same thing . . .

[ornate elk]

You have to check the RAM usage.

Maybe you need to adjust the Pagefile again. But this time increase it

[cosmic mulch]

what should i change it to?

[ornate elk]

Hmm nope that's already correct

[cosmic mulch]

ihave 16 gigs of ram, 12 gigs of vram, and just discord and firefox running. nothing else

[ornate elk]

Ohhhhh wait

[cosmic mulch]

--use-zluda --medvram-sdxl --update-check --skip-ort these are the settings

[ornate elk]

You have refiner selected!

Never do that

[cosmic mulch]

oh

how do i refine the image then?

[ornate elk]

Its not needed

[cosmic mulch]

without the refiner the outputs are pretty.. bad

[ornate elk]

Don't use the basic sdxl model

[cosmic mulch]

so i just use the hires.fix, and that's it?

[storm sail]

*cuz image im tryna do is too highres for vram i guess, probably, --medvram might not change nothing much, might make things worse, but, idk . . . lol

[ornate elk]

Yes

[cosmic mulch]

what upscaler? latent?

[ornate elk]

ESRGAN 4x or latent.
Esrgan4x anime6b for anime or community made ones

Also for pony and sdxl use a 1024x1024 resolution or near that

Like 768x768

[cosmic mulch]

ill try this then

i was doing 512 before

[ornate elk]

512 will look really bad when using sdxl

[storm sail]

this directml just as well if not seems maybe better/faster then openvino tho so . . .

[ornate elk]

What settings did you used?

[cosmic mulch]

this one?

again same error. connection timed out when it went to 99%

no trace of any errors

[ornate elk]

Screenshot the settings please

[storm sail]

512x768 with highres fix, dreamshaper, lcm, etc, what else settings idk all that influence

[ornate elk]

Make sure to set hires steps to 10.
Then you also need the tiled VAE extension to not run out of Vram

[storm sail]

image size mostly i think is what takes the vram? also inference steps 30? clip to 4

[ornate elk]

And when running out of vram you always need to restart the webui-user.bat

[storm sail]

ok, idk nada bout what im doing but i will look at setting like u say

[ornate elk]

In the extensions tab. Click on available, then on "load from".
Then search for tiled
Then install the "tiled diffusion & tiled vae" extension.
After a restart, enable tiled vae.

[storm sail]

what is this "Then you also need the tiled VAE extension"? how do i do that?

ahh, okie dokie

what does highres steps do?

[cosmic mulch]

still got the error. let me open again and send it to you

[storm sail]

what does tiled vae do (besides keep me from running outta vram?)?

[ornate elk]

It tiles the VAE. Because if not that uses to much of vram

[storm sail]

idk what vae even is, lol, cept it makes pics look better, lol

depending on vae i reckon

[ornate elk]

Hmm that should work.
Also make sure the webui is whitelisted in ublock origin.
Also make sure you always launch the webui-user.bat and not the webui.bat

[storm sail]

and model and how use together, lol

[cosmic mulch]

yes

[ornate elk]

And set denois to 0.5

[cosmic mulch]

still get the error, suddenly stops

[cosmic mulch]

alright. and upscale by 2?

[ornate elk]

Try restarting the PC once

And then try again

So its clean

Dont enable refiner

[fair oxide]

stderr:
Python was not found; run without arguments to install from the Microsoft Store, or disable this shortcut from Settings > Manage App Execution Aliases.

Launch unsuccessful. Exiting.
Press any key to continue . . .

but i already installed python?

[ornate elk]

You didn't added it to path. Or didnt deleted the venv.
Make sure you follow the install guide from the pinned messages of this channel

[storm sail]

when i install this tiled vae, only have to press install button one time, i dont have to set it up anywhere or nothing?

k, thanks 🙂

[ornate elk]

@cosmic mulch also try out without hires fix first if that works normaly

[cosmic mulch]

okay

[ornate elk]

Ah okay

[ornate elk]

You also need the tiled vae extension

Thats a normal error when using to much vram

[cosmic mulch]

how do i get it

[storm sail]

what happens if i change highres steps higher or lower exactly?

[ornate elk]

In the extensions tab. Click on available, then on "load from".
Then search for tiled
Then install the "tiled diffusion & tiled vae" extension.
After a restart, enable tiled vae.

[cosmic mulch]

also.. this isn't too bad for my gaming pc right? not any worse than playing AAA video games where i get like barely 70-80 fps

[cosmic mulch]

in the UI itself?

[ornate elk]

It uses more vram.
Hires steps are the upscaling steps (how often the image will be improved)

[fair oxide]

i put the python download location in my E drive which isnt the main computer drive

is that wrong?

[ornate elk]

Better reinstall it on C

And uninstall it from E

[fair oxide]

will that fix the problem i jus got?

[cosmic mulch]

is this fine?

[storm sail]

hmm, more steps more upscaling? i had it set to 0 but now u say set to 10? 10 should take more vram then 0?

[ornate elk]

Yes but make sure to delete the venv folder before relaunching the webui-user.bat

And that you checked "add python to path"

[ornate elk]

Yep

[cosmic mulch]

no titled diffusion or demofusion ?

[ornate elk]

0=it will use the same amount as steps.
Never do that

[ornate elk]

Nope not needed

[cosmic mulch]

still silently stops :(

[storm sail]

otherwise, i have denoise strength at 0.68 (think it has to do with image clarity or something, idk), i have upscale to 2 and resize options at 0 (this all just image size stuff i think cept upscale is clarity/pixels?), upscaler is "latent" (no clue what it does really)

[fair oxide]

did it right no?

[ornate elk]

Thats not for nvidia

[fair oxide]

ik i got a amd gpu

[ornate elk]

Then yes thats right

[paper pilot]

Hello, I been having issues with my stable diffusion since it has been updated to 24.1.1. I have remove my venv file, reinstalled with the webui-user.bat. and installed python 3.10.7 since I gave part of the code to copilot (bing AI) and the stable diffusion won't open. I have the text file of the per say issue if it's ok to send as a txt file

[storm sail]

so, always want highres steps lower value then inference steps

[ornate elk]

What's your GPU?

[fair oxide]

rx 580 8gb ver

[ornate elk]

That GPU currently won't work with Zluda.
You need to use my DirectML guide

[fair oxide]

oh

bet

[fair oxide]

but im not sure where in the guide it says to set the path of python

could u show me if its not too much

[storm sail]

*i say again, i know nothing about all this stable diffusion, lol, all this settings and stuff, so technical, fun but, lol

[ornate elk]

Its a checkbox that appears when installing python for the first time.

[fair oxide]

this is where its set

[ornate elk]

You can run the installer again and click
Modify
Next
"Add python to environment variables"

[fair oxide]

done

[storm sail]

*hardly know anything about computers at all or any image softwares despite how much time i spend screwing round and despite i draw/paint irl (but on paper/canvas mostly, not on computer much, lol)

[ornate elk]

Hey, what's your GPU and what's inside your webui-user.bat?

[cosmic mulch]

it's stuck at this

[ornate elk]

Check the taskmanager if anything is at Max usage

[cosmic mulch]

nope

[ornate elk]

It shouldn't be stuck

[cosmic mulch]

oh. i clicked on it and pressed enter. it got fixed.. weird

[paper pilot]

My gpu is 0 NVIDIA GeForce RTX 3050 and I will send in as txt for what I added ti rhe webui-user.bat

[storm sail]

that tiled vae shows as "hypertile" now in my list installed extensions, ya?

[ornate elk]

Oh you paused the cmd then xD

[fair oxide]

should i keep it to medvram since i got 8 or just go with low? i dont mind utilizing my vram

[ornate elk]

Try --medvram first

[cosmic mulch]

oh also, on the "sampling steps" option, should i keep it at 20 or inc it to 35? i've heard 35 is a good spot

[ornate elk]

You can switch later if needed

[fair oxide]

bet tysm

[cosmic mulch]

oh nvm i got the biiiiiig error again bro

why do i keep getting this aaaaa

[ornate elk]

25-35 is a good spot. I mostly use 30

[storm sail]

that tiled vae shows as "hypertile" now in my list installed extensions, ya? (i think that's the one)

[cosmic mulch]

xception Code: 0xC0000005

[ornate elk]

Would like to see the whole text

[fair oxide]

also just a question. installing stable diffusion a year ago took waayyy longer than this. is it cuz i got a much stronger cpu (old was 2 core) or was it somehow more modified and faster

[cosmic mulch]

oh, i clicked off

[ornate elk]

No its not hypertile

[ornate elk]

Remove set LAUNCH=True
Remove --reinstall-xformers
And add --medvram-sdxl
Then save.
Then delete the venv folder.
Then relaunch the webui-user.bat

[storm sail]

how can i view/be sure that tiled vae installed correct?

[ornate elk]

Go into the extensions tab and check what's there

[cosmic mulch]

again, randomly stops..

[ornate elk]

Thats so strange

[cosmic mulch]

whats up with me and getting the same error over and over again LOL

it's either xception Code: 0xC0000005 or it's randomly stopping

[ornate elk]

What you can try is to set the pagefile to system managed for C then restart the PC and try again

[storm sail]

maybe guess i musta done wrong or somehing cuz i dont see any other tile thingy on list of installed extensions

[paper pilot]

Ok, will try that

[ornate elk]

You mostly installed the wrong extension

The right one is named tiled diffusion & tiled vae

[storm sail]

mostly? lol

[cosmic mulch]

alright

[storm sail]

swear thats one i did but, nevertheless, i try again, lol . . . what is this "hypertile"?

[fair oxide]

should i do this

[ornate elk]

Code changed and got improved

Thats why its faster now. But yea CPU and Ram helps too

[ornate elk]

Can be ignored

[fair oxide]

uhh it said press any key to continue and it closed without giving me the ip stuff

[ornate elk]

Thats an extension which can speed up generation by tiling the image like tiled VAE does with vae

[fair oxide]

do i need to re run after install?

[storm sail]

this what i tried to install - TiledDiffusion with Tiled VAE
manipulation

[ornate elk]

Yea but it shouldn't close itself.
What's in your webui-user.bat?

[ornate elk]

You need the one without manipulation at the end

[storm sail]

its the only tiled one shows when i use search

[ornate elk]

Have you just type in tiled ?

[fair oxide]

  ..\meson.build:1:0: ERROR: Unknown compiler(s): [['icl'], ['cl'], ['cc'], ['gcc'], ['clang'], ['clang-cl'], ['pgcc']]
  The following exception(s) were encountered:
  Running `icl ""` gave "[WinError 2] The system cannot find the file specified"
  Running `cl /?` gave "[WinError 2] The system cannot find the file specified"
  Running `cc --version` gave "[WinError 2] The system cannot find the file specified"
  Running `gcc --version` gave "[WinError 2] The system cannot find the file specified"
  Running `clang --version` gave "[WinError 2] The system cannot find the file specified"
  Running `clang-cl /?` gave "[WinError 2] The system cannot find the file specified"
  Running `pgcc --version` gave "[WinError 2] The system cannot find the file specified"

  A full log can be found at C:\Users\Nori\AppData\Local\Temp\pip-install-hpmr5nxe\scikit-image_26cfff95568841fba332436204980935\.mesonpy-di8862vw\meson-logs\meson-log.txt
  [end of output]

  note: This error originates from a subprocess, and is likely not a problem with pip.
error: metadata-generation-failed

Encountered error while generating package metadata.

See above for output.

note: This is an issue with the package mentioned above, not pip.
hint: See above for details.

[notice] A new release of pip is available: 24.0 -> 24.1.1
[notice] To update, run: E:\Ai\stable-diffusion-webui-amdgpu\venv\Scripts\python.exe -m pip install --upgrade pip

Press any key to continue . . .

[ornate elk]

Why visual studio lol

[storm sail]

@ornate elk ya

[fair oxide]

ion know

[ornate elk]

Can you screenshot it?

[ornate elk]

Do you have the full log?

[fair oxide]

\

i think its becus i didnt purge cache

[ornate elk]

Ahh then do that.
And then delete the venv folder again

[storm sail]

maybe i need to uncheck boxes, lol

[ornate elk]

There it is, thats the right one

[fair oxide]

https://tenor.com/view/pepe-meme-white-eyes-gif-7499091

[storm sail]

didn't make difference

[fair oxide]

white mode

[lost crescent]

I put https://127.0.0.1 into Trusted Sites at Chrome Browser and get this error - [Errno 10048] error while attempting to bind on address ('127.0.0.1', 8188): only one usage of each socket address (protocol/network address/port) is normally permitted
PS Retrying to start the server... (2/5) There is no AdBlocker.

[storm sail]

@ornate elk thats one i just did, thought hypertile but u say no, i try again

nada, but i know something why now at least, lol - i am getting AssertionError: extension access disabled because of command line flags

[ornate elk]

You need to remove --listen and --share to install extensions

You can add them back afterwards

[storm sail]

tried to get lcm extension earlier and this happened, ahhhh, a little weird that flag would mess it up but okie dokie, why listen bad in this case i wonder

need remove from commandline args u mean?

ya, my bad

[fair oxide]

the error

again

[lost crescent]

Blank screen ComfyUI anybody? 🙂

Whitelist browser failed, sad to say

[storm sail]

why does --listen flag not good when install extensions?

[fair oxide]

@ornate elk

[storm sail]

tiled vae called multidiffusion-upscaler-for-automatic1111 on installed extensions

guess ==listen bad for installing extensions cuz something todo with privacy or something . . . .

[ornate elk]

You need to go in your system environment settings and there at the PATH entry edit and make sure that Python is above visual studio

[ornate elk]

Exactly, somebody with access could install extensions from remote

[paper pilot]

I have relauch my webui-user.bat with the log you mention, it's giving me a different message

[storm sail]

then, before i run inference, blah, i need to be sure to have tiled vae checked there ya? and there couple other things (tiled diffusion, demofusion) and settings for each thing, idk what any of this stuff means but i just put check in tiled vae box and leave all rest at default ok?

[fair oxide]

uhh theres no visual studio but onedrive?

[ornate elk]

Select and click edit

[fair oxide]

which one one drive?

[ornate elk]

No the Path

[fair oxide]

'

[ornate elk]

There is no Python 3.10.11

Only 3.12 :/

[fair oxide]

am i supposed to download the older version?

[ornate elk]

Yea like stated in the guide

[fair oxide]

https://tenor.com/view/oh-my-goodness-honey-mustard-cant-take-it-anymore-gif-27459470

igh we good

[ornate elk]

Better uninstall all Python versions

Then delete the venv folder again

And install 3.10.11 on C

[fair oxide]

should i delete pip cache aswell or nah?

[ornate elk]

Nope

[ornate elk]

Only enable tiled vae

[storm sail]

@ornate elk should i set up this tiled vae extension in any such way u think or just leave all at default? especially i see this notice in tiled vae settings "Recommended to set tile sizes as large as possible before got CUDA error: out of memory."

ahh, ok

i leave all alone then after i checked box tiled vae

i leave all at default

be back later, gonna try a batch and take dogs for walk

take care yall til later

[ornate elk]

Cya

[ornate elk]

What's the message?

[ornate elk]

Thats not the full log

[paper pilot]

Ok, let me get the full log then

[ornate elk]

Your webui is outdated and git pull isnt working. Also your python version is old.
Best would be you make a clean install by following my Install guide from the pinned messages

[paper pilot]

Dang, what's crazy is I had this working before the update 24.1.1

[ornate elk]

What update ?

[paper pilot]

Update 24.1.1. Before my version was on version 23 something, I forgotten the full version name. But I had it working then. After when I updated to version 24.1.1, it doesn't work

[ornate elk]

there is no 24.1.1

the webui version you use is 1.3.2

the latest version is 1.9.4
your more than a year outdated

[fair oxide]

it seems to stop here?

[ornate elk]

@paper pilot
Follow my guide from here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides#preparation

then you can copy over your models

[paper pilot]

Dang, I blamed copilot for that, I search for a stable diffusion and it gave me the old version I'm on

[ornate elk]

oh you installed it recently? but 1.3.2 because of copilot?

[paper pilot]

Yeah, I told copliot to find me a stable diffusion that would work with my gpu and it gave me that version

[ornate elk]

relaunch please then paste the full cmd log

[ornate elk]

whats your GPU?

the 3050?

[paper pilot]

The NVIDIA GeForce RTX 3050

[paper pilot]

Yep

[ornate elk]

ah alright, yea seems like copilot is a year behind with his stable info xD

if you follow my guide and install python 3.10.11 64bit your ready

[paper pilot]

Yep, pretty much

[paper pilot]

Yes, I will do that

[fair oxide]

uhh my models are not loading

[cosmic mulch]

so i found out that the model im using (t-ponynai3) already has a built in VAE, so you don't need to use an external vae. maybe that's why i kept randomly crashing, as i was using the sdxl vae.

however, if i don't use a vae, i run into this error

if i do use a vae, then it just crashes randomly

[ornate elk]

Make sure to enable tiled vae

[cosmic mulch]

i did. the model im using apparently doesn't need any vae at all, as it has one built in.

if i use no vae as the model suggests, i get the cuda out of memory error.

if i do use a vae, then it straight up crashes. this crash happens not only on this model, but also on sdxl.

the crashing is very frequent, im getting more crashes than actually successful images. it seems to be completely at random too, and im unsure how to fix it

i've tried setting it to system managed

[fair oxide]

uhh anyone?

[ornate elk]

Tiled VAE is not sdxl vae

You still need to enable tiled vae

Or you will run out of vram

[ornate elk]

How much RAM do you have?

[fair oxide]

8

[ornate elk]

Thats to less for sdxl

Thats why it won't load

[fair oxide]

dang

any way to make it work?

[ornate elk]

You need to increase your windows pagefile

[fair oxide]

oh

say less

imma do 50

heh

[ornate elk]

xD 16000 min and 24000 max should work

For C drive

And disabled for any other drive

[fair oxide]

sd is installed on a different drive though

does it have to be C even though sd is installed on a different drive?

[ornate elk]

That doesn't matter

[fair oxide]

oh

ight

[ornate elk]

C is mostly the fastest SSD on your system

So thats the preferred spot for the Pagefile

[cosmic mulch]

i'll try running just tiled vae, but selecting sd vae to none

[fair oxide]

where do i set the weight of the lora?

@ornate elk

[karmic crown]

in the prompt

[fair oxide]

do i jus say "lora weight 0.7" or somethign like that

[karmic crown]

lora:nameofthelora:strength

like

lora:somelora:0.8

nameofthelora = filename

without extension

[fair oxide]

bet

lora:Jamie_Hewlett_Style_Pony (1):0.8

so ig like this

[karmic crown]

<lora:Jamie_Hewlett_Style_Pony (1):0.8>
more like this

[cosmic mulch]

i used the system chosen option, i still get it to work sometimes, other times to not. do you think it's because of my ram? i saw that it was exceeding 100% and it manually turned off some apps i had in the bg that were using ram

[karmic crown]

but i would renam the file

remove the spaces and brackets

[storm sail]

RuntimeError: Not enough memory, use lower resolution (max approx. 896x896). Need: 0.8GB free, Have:0.4GB free, in my greed i ask with little expectation of a positive answer, "anyone gotta hack for no matter how i tweak image settings, etc, i will never get another error? lol, it could take years to render whatever but . . . ?"

[fair oxide]

do i need to restart stable diffusion for that

or can i just rename and gneerate

[karmic crown]

you can try, but i would restart

[fair oxide]

would u know whats the best width and height for ponydiffusion?

[karmic crown]

i have never used pony

[fair oxide]

iwell its basically sdxl

what would i use for that

[karmic crown]

for xl i use 1024x1024, 768x1024 or 1024x768

for turbo and 1.5 512x512, 768x512 and 512x768

[storm sail]

or, i wonder, how i would increase vram, i know it aint the same as ram but, hmm, is it like paging file size or???

[karmic crown]

to increase vram you need to buy a new graphics card

[storm sail]

that would still leave me limited to certain settings and resolutions etc, so i would rather have a fix to not be limited at all tho it would take years to render whatever but

anyone got any tweaks for crap laptops with only 4gigs vram?

[karmic crown]

unless you are on Mac. then, you can use a command to change the amount of unified RAM that can be used for graphics. on windows - only a new graphics card can help

[storm sail]

ahh

[karmic crown]

but this only applies to new ARM macs, for old intel ones - same as windows

[ornate elk]

Use --lowvram
Use models with a size of 2gb
Don't use sdxl or pony because they are 6gb
And Enable tiled vae

[storm sail]

i see

[karmic crown]

you can even run it without GPU (i run it on my linux server without GPU) and it work, but it is slow as hell

but, as CS1o said, you can use tiled vae, small models and --lowvram setting

baically use sd1.5 models

[storm sail]

@ornate elk gotcha, believe im doing all this already, think dreamshaper is 2gb, and the rest with ur help. got any other tricks?

[karmic crown]

you can try with https://civitai.com/models/171159/sdxlrender
that might help with the quality
i like it

[storm sail]

that is nice model

[ornate elk]

Close every other programm in the background, games, Wallpaper engine, browser tabs

[karmic crown]

it is 1.5 lora that help increase image quality

[storm sail]

but i like style of dreamshaper

[karmic crown]

you can SDXLrender this with dreamshaper to increase image quality

it is lora, not a model

[storm sail]

@ornate elk gotcha, that's that, hmm?

[karmic crown]

try not to have hunderds of tabs open in browser 🙂

[storm sail]

@karmic crown ooohh, ok, i might give that a shot then, tyvm 🙂

i put that sdxl in lora folder then how i enable via webui? idk nothing bout any of this, lol

[karmic crown]

left 1girl,sweater,white background,
right 1girl,sweater,white background,lora:SDXLrender_v2.0:1.2

most images on civitai have this at bottom right

click that to see the details

you can find what they used for that specific image

[storm sail]

so, lora is like render method or, lol, i am seriously stupid at all this stuff 😛

[karmic crown]

lora is like "helper" to a model

[storm sail]

ya, i dig the crazy astronaut lady pic on the site for that sdxlrender too

[karmic crown]

it can be for a specific face, part of cloths... and many other things

[storm sail]

a "helper", cool

i could use all the 'help' i can get, lol

[karmic crown]

i like the other version (with asian girl) on Chinese ai website much more 🙂

[storm sail]

🙂

[ornate elk]

Also enabled taesd VAE decoding in settings
And then upscale by 1.5 or 1.3

Maybe that works

[storm sail]

i gotta lotta learning tthis stuff to do, lol, a lot to stable diffusion, i am just happy to be getting anything out of it for now that i know nothing, lol

taesd

[karmic crown]

this one 🙂

[storm sail]

@karmic crown ya, on civitai there a pic just next to that is crazy 🙂

[cosmic mulch]

do you think i should try a lighter model? because i believe i keep crashing because i run out of memory

i run out of ram before vram , probably

[karmic crown]

@cosmic mulch how big is that model that you are using?

[storm sail]

gonna try see if i can find this = taesd VAE decoding in settings

[ornate elk]

Pony needs more than sdxl and both need more than 1.5 models
You can try set upscale by to 1.5

[storm sail]

see if that gets me more

[karmic crown]

https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111

you need extension

[cosmic mulch]

maybe i should try 1.5

[ornate elk]

Your GPU should work fine with sdxl

But Try upscale by 1.5

[cosmic mulch]

also, i found that upscaling manually helps for some reason, rather than upscaling together

[cosmic mulch]

thats what im sayin... but it keeps randomly crashing

is it because of my 16 gigs of ram?

[storm sail]

@karmic crown already doing that thanks to CS1o

but taesd VAE decoding is different?

[ornate elk]

You could try set the pagefile to
24-32000 and then restart pc

[cosmic mulch]

RhE?

[storm sail]

and they said recommend that i keep upscale to 1.5 or less

[ornate elk]

Yes

Uses less vram

[storm sail]

taesd in vae section of settings, i only do the bottom option (VAE type for decode) leave all else alone?

i wonder whats the trade off for using things like tiled vae and taesd

not as great quality i will reckon

[ornate elk]

Shouldn't change much in the output

[storm sail]

suppose that where the lora like sdxlrenderer will come in

oh, ok

whats difference between 32 bit float and 16 bit float, i thought openvino was using 16 and thought 16 be less demands

[karmic crown]

just use smallest model you find, fp16 pruned in most cases

[storm sail]

idk what any this means, i dont know what 32/16 bit float is for vae, something about models being fp16 too, dreamshaper is fp16, idk if that has anything todo with the bit float stuff, maybe unrelated, fp16, 16 bit float, idk

[ornate elk]

the

[storm sail]

wonder whats difference between vae on cpu or moved to gpu and how that would effect vram usage

[karmic crown]

that's why i said" just use the smallest model you find"

🙂

[storm sail]

lol

[ornate elk]

move to gpu uses more vram so maybe uncheck that

[storm sail]

for sure, gonna do that then straight away, ty 🙂

[karmic crown]

like this

sometimes download can have more files

like 2 or 3

[storm sail]

word

oh, snap

maybe i got wrong dreamshaper model then

[karmic crown]

nope, it has only one version

[storm sail]

ok

[karmic crown]

but you have to watch for those

[ornate elk]

make sure its Dreamshaper v8

[ornate elk]

it has an sdxl version too

[karmic crown]

yes

[storm sail]

cuz i download from that button on top rather then i see another link there but ya cool

is v8

[karmic crown]

if it has more version that button should show dropdown

so you should be able to choose there too

but notice that each of those can have more than 1 file

[storm sail]

believing i got right one but all that horizontal menu above is all different models? i just got one where on horizontal menu create button was highlighted, rather it just says 8, oh ya, i see, on diffusers tab across there, whatever that is, that download is like 6 gb rather

[karmic crown]

think of it as
horizontal = diffrent models
files = diffrent sizes

[storm sail]

i see, these also are versions, etc

[karmic crown]

you will get what is what eventually, but for now, this similified aproach is enough

[storm sail]

the first few tabs is related to version 8 but different bits like (idk if i need inpainting bit if i already have main model, idk what is diffusers, etc) so . . . suppose they for specific tasks to get best results for said task?

[cosmic mulch]

oh okay. btw, how can i use upscaling for img2img?

[ornate elk]

therefore you need to use the sd upscale script at the bottom,
set the resolution to a sqaure
set the denois below 0.3
and enable the script and select an upscaler

[cosmic mulch]

is there any way i can just upscale any image? like i upload the image and then upscale it

[ornate elk]

that can be done in img2img or in the extras tab

[cosmic mulch]

got it. thanks

[storm sail]

does text to image prompts usage have effect on how much vram needed? i was getting out of memory error before but was saying i could make image up to around 800x800, now, i enabled taesd decoding and unchecked move vae to gpu to see if i use less ram, i start getting out of memory worse (says i can make image up to only 700x700 now) so i put them back to vae decoding full and rechecked move vae to gpu (i did these one thing at a time to see which one might be making things worse, only thing is if prompt is having effect or i need restart my computer or, idk

i gots sooooooooo many questions 😛

looking at task manager, everything there looks ok, im only right now using, well, maybe python using quite a lot of ram, but everything else looks ok . . . weird, idk

[ornate elk]

More tokens need more vram

Stay below 150 and your fine

Loras need vram

Higher resolution needs more vram

Also use Euler a or DPM++ 2M as sampler

[storm sail]

musta been prompt then tho i dont know what tokens is

but guess my prompt was too long

that all it could be cuz went from can create an 800x800 image to added "BadDream, (UnrealisticDream:1.3)" to my negative prompt and itt start saying can create only up to 700x700 image

[ornate elk]

Embeddings also increase token count

[storm sail]

about that, i get normal language, idk how to do text prompts too well tho, like surrounding words with ( and ) emphasizes that part of prompt, ya? idk if ( and ) are for words only or can be used with whole phrase . . . anyway, also, what is the numbers after words (ie, UnrealisticDream: 1.3)???

that is embedding?

what it do exactly i wonder and how to know how to use?

what is fast encoder color fix under tiled vae i wonder

running stable diffusion on cpu was still using vram? probably. tempted to go back to openvino as i wasnt getting so many out of memory errors but then again, i wasnt trying to use highres fix when i was doing openvino

[onyx solar]

I assume you're using SD 1.5 or SDXL. Parentheses to set the strength of tokens (the prompts you're writing) can be used for single words or for multiple words. For example, all of these are fine:
(big) car
(((big))) car
(big car:1.2)
big (car:1.1)
((big car))

[karmic crown]

you have https://discordapp.com/channels/1002292111942635562/1011743094309396631 for question about usage 🙂

[storm sail]

took longer to render too but i wonder if there ways to speed up openvino thing

@karmic crown ty, more curious about other stuff right now but good to know

@soul ty, good to know

[onyx solar]

But also, you should know that prompts for SD 1.5 are not really best written with natural language in most cases. You can still do it and you'll get output that works a lot of the time, but it's not the most ideal.

SDXL is a bit more tolerant of it and produces good output for natural language a bit better, but it also does well with just keywords.

[storm sail]

was openvino/running on cpu still using vram?

or is it completely different then using gpu?

guess i test . . . i will try to render large very detailed image with openvino acceleration and see if i get same memory error

oh, but, there was question about windows not able to torch.compile or whatever my original issue was

alrighty all, be blessed

[storm sail]

alrighty well, no offense @ornate elk but i think i gonna stick with stable diffusion and openvino acceleration cpu rather then the directml. probably my amd just crappy/old/weak so why i get so many out of memory errors when trying directml, works fine enough for gaming idk why it gets screwed when comes to sd. sd openvino on my cpu tho so far seems darn near unbreakable even when i use 2x upscale, etc (thinking it uses vram differently somehow), tho its slower then molasses it has yet to give me many errors in comparison to directml version. in fact only error i seen from this openvino version is one i came here for originally . . . . namely, "RuntimeError: Windows not yet supported for torch.compile", if anyone can help me with this issue (and i read earlier there is a way to bypass the whole torch.compile issue) that would be helpful.

[ornate elk]

openvino can use your RAM and is not tied to vram thats why its working "better" than the limited 4gb vram old gpu.
its fine, use what works the best for you

[storm sail]

someone says, "Bypass this on Windows by adding --compile=False to the end of your run statement. Ex:

python train.py config/finetune_conversation.py --compile=False works, whereas
python train.py config/finetune_conversation.py will raise the error" only idk what that means

[fair oxide]

is 1.5 it/s normal for a 2080ti when generating 1024x1024 with an SDXL model?

[storm sail]

i tried adding --compile=false to commandline arguements for webui-user.bat but that didnt work out well, lol

@ornate elk righto! i thought it maybe using ram instead, ya.

[ornate elk]

are you using --xformers --medvram-sdxl --no-half-vae ?

[storm sail]

far as openvino accelerated cpu version of sd, i think (fingers crossed) there be some ways to speed it up too, i seen videos such as https://www.youtube.com/watch?v=KJDCGyZ2fPw tho i havent went into it yet, idk

[fair oxide]

idk... I'm using automatic1111. what are those settings?

[ornate elk]

check whats inside your webui-user.bat

[fair oxide]

ctrl+f "xformers" found nothing

[ornate elk]

not webui.bat

[fair oxide]

oh derp sorry

no, that is not in my commandline args

should I add it?

[ornate elk]

yes

then it should be much faster

also always launch the webui with webui-user.bat

[fair oxide]

Will do, thanks. Any thread I can read on that?

to know more about those settings

[ornate elk]

sure:
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings

[fair oxide]

tyvm!

[storm sail]

about, RuntimeError: Windows not yet supported for torch.compile stable diffusion, another person says "Run accelerate config in the command prompt, and when asked, answer the question "Do you wish to optimize your script with Torch Dynamo" with "No."

only idk what that means either

i tell u i am dumb af with this stuff 😛

[ornate elk]

cant help much with openvino

[storm sail]

something called triton??? anyway, ok. np. can u direct me anywhere i might get more help with this version/way of doing stable diffusion?

[fair oxide]

still getting 1.5-2.5it/s

well closer to 2.0

[ornate elk]

yea thats normal

[fair oxide]

okay cool

thanks man

[ornate elk]

no problem

[fair oxide]

still a nice bump in perf

like 30% more

[ornate elk]

it lowers the vram usage too so you can upscale larger

[fair oxide]

I'm not using hires upscale tho?

[ornate elk]

yea but you can

[fair oxide]

wow really? It would always time out on me when I tried 2x on 1024

--medvram "Enable Stable Diffusion model optimizations for sacrificing a some performance for low VRAM usage." interesting

wow that's crazy it's actually able to do 2x now

that's awesome

[ornate elk]

yea its great ^^

[storm sail]

@ornate elk any idea another discord or anything where i might find someone more knowledgable about stable diffusion on cpu/openvino, etc? who might be able to help me more with RuntimeError: Windows not yet supported for torch.compile stable diffusion???

[ornate elk]

nope sry, also which webui do you use for opnevino? this one?
https://github.com/openvinotoolkit/stable-diffusion-webui

[storm sail]

https://github.com/mrkoykang/stable-diffusion-webui-openvino

[ornate elk]

--medvram-sdxl is better than --medvram for your gpu

[ornate elk]

oh that one is not as updatet as the on i send

maybe try that

[storm sail]

oh, hmm, ill take a looksee, tyvm 🙂

ty for all ur help today 🙂

[ornate elk]

no problem 🙂

[storm sail]

dig that repo u linked, same thing only more recent, nice 🙂

k, well, if i dont see yall again, be well blessed, stay amazing 🙂

omg, lmao, i dont remember how logout discord, lol, tired and lamebrained 😛

[ornate elk]

haha cya around

[fair oxide]

what about for a 3090ti? I have one coming next week

[ornate elk]

that would only need --xformers

[fair oxide]

sorry to keep bombarding with questions, but what kind of performance can I expect out of a 3090ti? I tried doing some research online beforehand and it seems like I could expect maybe a 50% bump in perf at most

[ornate elk]

hmm idk how much better it performs but it will be really good

[midnight grove]

I know this is likely off-topic, but I don't really know where else to go and I feel like someone with more knowledge may be able to help.
I'm running InstantMesh locally. It's a model for turning images into meshes. I'm having issues similar to the ones I had when I started using SD. Namely:
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 15.00 GiB. GPU 0 has a total capacty of 10.75 GiB of which 2.69 GiB is free. Including non-PyTorch memory, this process has 7.70 GiB memory in use. Of the allocated memory 7.22 GiB is allocated by PyTorch, and 284.19 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
I'm guessing that the application is just being aggressive with its allocation when it grabs 15 gigs. Does anyone have any idea how I might be able to track down what's happening and try to squeeze it into my ~11 GBs?

[ornate elk]

if that tool has a venv you should install xformers into it

that helps using less vram and speeds up the gen time

[midnight grove]

The install instructions included
pip install xformers==0.0.22.post7
so I believe that's already on there

and yeah, this is with a conda venv

(I think I'm using that right but I don't know haha)

[ornate elk]

hmm

[midnight grove]

I'm trying to find where in code the actual allocation happens to see if someone just tossed a "15.0" in there but I haven't had luck yet

[ornate elk]

your on a 1080ti ?

[midnight grove]

2080ti

[ornate elk]

ah okay

i guess this tool just needs to much resources right now

[midnight grove]

I'm just wondering if I can change that somehow because the models themselves are quite small

[storm sail]

im back already, lol

so, @ornate elk posted for me a link to a newer build of stable diffusion with openvino acceleration than the one ive had. i was curious to see if i could find a repo even more updated than that and i found https://github.com/JT-Gresham/Auto1111-IntelArc-ArchLinux, granted its only a teensy bit newer (if i am judging correctly). anyway, i can see a big difference already just looking at the webui-user.bat - where many lines of stuff was missing from my old version (set PYTORCH_TRACING_MODE=TORCHFX and set USE_OPENVINO=1) this one has these things set up. however, my issue . . . when i go to set this newer repo up, i get this error

any idea how i fix?

[ornate elk]

why not using the one i linked?

yours is just a fork of it (also its name has arch in it, so maybe only for linux)

[storm sail]

ya, the one u linked probably more legit or whatnot

and only a month or two olde

so

ya

i agree, lol, i try ur idea

idk if ill run into same problem but . . . guess i try and letya know

was this one, ya?

https://github.com/openvinotoolkit/stable-diffusion-webui

i think so

[ornate elk]

yes

[storm sail]

and, in which case, my bad cuz looking again the one i just tried and this one here like exactly same age, lol

*wasting time, its what i do best 😛

curiosity, in case i face same issue here, wonder what i do

i googled but got different results, like web ai says i can try to install cpu only version torch, blah, only idk what to trust so least i ask someone who gots the info and the brains 😛

this what my webui-user.bat looks like btw @echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--skip-torch-cuda-test --precision full --no-half
set PYTORCH_TRACING_MODE=TORCHFX
set USE_OPENVINO=1

call webui.bat

something bout this line set PYTORCH_TRACING_MODE=TORCHFX

idk, least this .bat looks more complete then that older build i had tho, i think should work better once i get it installed

reasons to think ima run into same error tho, cuz this https://github.com/openvinotoolkit/stable-diffusion-webui looks exactly same as https://github.com/JT-Gresham/Auto1111-IntelArc-ArchLinux

CUDA Variant on Non-GPU System:
If you’re using a CUDA variant of PyTorch on a system without an NVIDIA GPU, consider installing the CPU-only package:
Using Conda: conda install pytorch torchvision cpuonly -c pytorch
Using Pip: pip install torch==1.5.0+cpu torchvision==0.6.0+cpu -f https://download.pytorch.org/whl/torch_stable.html1 ??? hmm . . . .

nm, that one u linked me @ornate elk went off without a hitch, guess i needa learn take more care where i pull repo, lol

ty for ur help again 🙂

[fair oxide]

following up on this to see if it's possible

basically want every batch to go into its own folder

I think I figured it out

using these instructions: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Custom-Images-Filename-Name-and-Subdirectory

[low basin]

My quick napkin math says 64% increase based on CUDA core count and boost clock freq. Might vary ~5% or so from there but should be almost exact. Not to mention things like Tensor RT engines would be significantly faster on the 3090ti

[fair oxide]

awesome, thanks so much man

[low basin]

No problem! It's hard finding exact benchmarks for non-gaming and FPS benchmarks do not correlate to generative AI loads. It's just strictly cores and speed haha

[fair oxide]

yeah, it's been tough finding good info

there are some charts that show generic it/s but I'm not sure of the details of those stats

[low basin]

The comparison charts are usually in the ballpark but as new engines and optimizers are created it may change speeds exponentially based on core counts and other things. Either way you're going to see a remarkable speed increase

[fair oxide]

im excited tbh

And another person helped me boost my 2080ti performance today which allows me to finally do 2x upscale on 1k images

can't imagine what I'll be able to do on a 3090ti

[low basin]

Is the 2080ti the 11gb card?

[fair oxide]

yee

so it's a huge jump in VRAM for me

I have some other GPU intensive parts of my workflow, so it'll be nice to have the extra space

[fair oxide]

it doesnt generate any images

and the log looks like this

[lost crescent]

Good morning! I'm stuck with "blank screen" ComfyUI? 😄

Console (via F12)

[solemn moon]

i think u need xformers

[fair oxide]

How do I get those

i turned on lowvram and i got this now when trying to generate

[solemn moon]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Xformers

[fair oxide]

The Xformers library provides an optional method to accelerate image generation. This enhancement is exclusively available for NVIDIA GPUs, optimizing image generation and reducing VRAM usage. Older versions below 0.0.20 will produce non-deterministic results.

i have an amd

[solemn moon]

damn

my bad

[fair oxide]

u good lol

[lost crescent]

Git pull 
Temporarily clears the ComfyUI blank-screen dilemma - but must be done each time you re-open ComfyUI```

[solemn moon]

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

from the pinned messages here

[fair oxide]

yeah this is the guide i used

Auto1111 with DirectML

imma jus use a different tutorial than CS1o's cuz it seems to have a lot problems for me

[solemn moon]

kudos to who ever did this

[sleek lily]

Hello everyone! I have a question and wonder if anybody know how to do. I am an architect and want to fine tune a model for a specific case study. Such as a tower ruin that designer require to create a new design within it, like adding new structures on it (adaptive reuse). I have the photos of tower ruin and 300 images of design alternative renders for the tower (there was a competition and I took images from there). So the question is: can I fine tune the stable diffusion that it can generate new design solution by keeping the site and ruins but adding new structures on it?

[ornate elk]

What was your GPU again?

[fair oxide]

rx 580 8gb

[ornate elk]

Can you try an other model?

Not pony or sdxl

[fair oxide]

like the stock ones that comes with stable diffusion?

[ornate elk]

Yes

[fair oxide]

```To create a public link, set share=True in launch().
Creating model from config: E:\Ai\stable-diffusion-webui-amdgpu\configs\v1-inference.yaml
Startup time: 51.3s (prepare environment: 77.8s, initialize shared: 6.1s, list SD models: 0.4s, load scripts: 0.6s, initialize extra networks: 0.3s, create ui: 1.3s, gradio launch: 2.8s).
E:\Ai\stable-diffusion-webui-amdgpu\venv\lib\site-packages\huggingface_hub\file_download.py:1132: FutureWarning: resume_download is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use force_download=True.
warnings.warn(

it still like

not doing anything

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker

[fair oxide]

should i jus turn off my ublock origin?

[ornate elk]

No, hold Ctrl and click on ublocks blue button

Then its whitelisted

Also how much RAM do you have?

[fair oxide]

8

i put 24000 on the max page file size

now it says in queue

but

oh wait

nvm it started working

cmd was not doing anything for a bit

yeah uhh

@ornate elk my pc crashed mid generation

restarted my whole pc on its own

even though my prompt was "snow mountain"

man sd really doesnt like amd gpus huh

[ornate elk]

Make sure you close all other programms, browser tabs, games and Wallpaper engine

[fair oxide]

already did

still doesnt work..

[urban hearth]

Newbie to this here, any reason as to why mine is stuck in queue? I got 64gb of ram and 8 gb of vram

[ornate elk]

What's in your webui-user.bat?

[ornate elk]

Still crashes the whole PC?

[ornate elk]

I mean in the file

Close the cmd, and right click the webui-user.bat to edit it

[fair oxide]

yeah either that or it doesnt generate anything even with whitelisted adblocker

[ornate elk]

At the line commandline_args=
You add: --xformers --medvram-sdxl --no-half-vae
Then save and relaunch it

[urban hearth]

Gotcha, what do those arguments do?

[ornate elk]

With default settings?

[fair oxide]

yup

[ornate elk]

Can you try Downloading Dreamshaper v8 and try with that?

[urban hearth]

@ornate elk is there documentation that explains what all of this means and how to use it?

[ornate elk]

You can hover over most of the buttons and settings.
But there is no GUI explanation

[urban hearth]

ohh I see, that's cool

[storm sail]

anyone here with knowledge of openvino stable diffusion build?

@ornate elk u still here? no rest for the wicked, hmm? 😛

[fair oxide]

@ornate elk i havent started generating dreamshaper. but something i noticed is when i select the checkpoint it doesnt load for a few minutes. it immediately loads

[storm sail]

i think im doing openvino build wrong, i wondering if openvino is thing i suppose to have installed on my pc in advance of build

[urban hearth]

Alright, what's the best way to go about generating hot anime waifus

[storm sail]

i think maybe i need to pip install optimum intel before i can have build accelerate my inference only idk

and i hate doing crap idk what im doing on my pc cuz i just make messes

[ornate elk]

Using good models like pony based ones or aam xl

Using extensions like
Boorutag autocomplete and adetailer

[storm sail]

only, if im supposed to have optimum intel installed first before i can benefit from https://github.com/openvinotoolkit/stable-diffusion-webui idk why its not mentioned

[urban hearth]

How does one go about getting extensions? Also is there a comprehensive guide to this that way I don't have to keep annoying you lol

[ornate elk]

Extensions can be installed via the extensions tab.
Click on available, then on load from to get a list

[storm sail]

i google stuff but i dont trust im looking at things right etc

[fair oxide]

yup just happened again with dreamshaper. prompt was snow mountain. was generating fast. but when it reached 80 my pc restarted

[ornate elk]

Try this webui with openvino support.
Maybe that works for you:
https://github.com/vladmandic/automatic/wiki/OpenVINO

[ornate elk]

So your GPU draws to much power it seems

Maybe the PSU is to small?

[fair oxide]

i got a 550w

idk

maybe cuz i got a second monitor connected?

lemme try with my second monitor off

[urban hearth]

gotcha thx

[ornate elk]

There is no guide right now for the basics sadly

[storm sail]

i already got the build u linked to me earlier @ornate elk i got https://github.com/openvinotoolkit/stable-diffusion-webui only, to take advantage of the openvino support/acceleration offered in such builds, idk do i need to have first installed optimum intel?

https://huggingface.co/docs/optimum/intel/index or does this come with them builds already?

[urban hearth]

Ah gotcha, cause even as a software developer I had a hard time trying to figure out how stuff worked lol

[ornate elk]

Best way is to play around and try out different settings

And I mean txt2img settings not settings settings

[storm sail]

see that openvino link u just sent me too, it says "Install the drivers for your device" as first step, i reckon this means optimum intel install or is these drivers it means just correct drivers for my cpu/gpu like regular necessities stuff? i get confused . . . sorry

[fair oxide]

yup restarted even with my second monitor off

[storm sail]

like when i think of "drivers" i just think that i need amd driver for my amd card etc

[fair oxide]

we are cooked with this one

[storm sail]

but in this case, it means i need optimum intel software from hugging face it says that is "drivers"?

ahh, fuck it all, ima make a mess, gonna just pip install that optimum intel shit and see what happens

[fair oxide]

do u guys know which prompt should i use for a full body shot. like this?

[storm sail]

anyhow, ty for helping much as u could

@fair oxide mention head details (hair whatever) and feet details (shoes, whatever) of person usually gets me full body in there

maybe say camera looking up at person, etc

basically, i just describe what i see in my minds eye, like and it gets me somewhere in vicinity, then after start simple, refine

hope that may help

oh, and mention hands to get them in there, so have covered all etremities

u could say instead of camera looking up at person maybe person looking down at camera

etc

anyway, yall take care and be well

[real citrus]

Hi guys! I got a problem... My Automatic1111 stopped working even though it was set for no updates... And now its giving me this error:

"A module that was compiled using NumPy 1.x cannot be run in
NumPy 2.0.0 as it may crash. To support both 1.x and 2.x
versions of NumPy, modules must be compiled with NumPy 2.0.
Some module may need to rebuild instead e.g. with 'pybind11>=2.12'.

If you are a user of the module, the easiest solution will be to
downgrade to 'numpy<2' or try to upgrade the affected module.
We expect that some modules will need time to support NumPy 2."

Im not installing the newest Automatic1111 because two extensions i use dont work in the newest Automatic1111. Ans i was just fine using the older version... Until this shit happened... How can i fix it?

[ornate elk]

Hey, What extensions dont work on the new version?

[real citrus]

SD-CN Animation and Roop

[ornate elk]

Roop got replaced by Reactor

[real citrus]

yeah but reactor is crappy

[ornate elk]

There is also roop-unleashed

[real citrus]

but the more important one is SD-CN Animation

[ornate elk]

Ah that is really outdated

[real citrus]

it shows on the list of extensions to install it but when you install it it shows as installed but its not showing a tab

yeah hence i was using older Automatic

and i was just fine using the older one, i hate to be forced to update... 😡

[ornate elk]

You then need to downgrade numpy manually

[real citrus]

how do i do that? I can follow tutorials on how to install automatic1111, but i dont enjoy digging into that stuff much

[ornate elk]

You need to activate the venv then pip uninstall numpy
Then pip install numpy==1.5

Something like that

[real citrus]

yeah but how do i do those things? 😄

[ornate elk]

When I'm at the PC I can give you the instructions

[real citrus]

Thank you!!! 🙂

[storm sail]

@ornate elk are u the only tech support person ever available here?

i know u said u dont know much about running stable diffusion on cpu or openvino, wonder there ever will be anyone here might help me with questions related

[brave reef]

I just ran a git pull in automatic 1111, and now when I try a simple txt2img prompt on default settings I get this error - RuntimeError: expected scalar type Half but found Float , img2img seems fine. I re-loaded the venv already, but still does it. Any ideas anyone?

[ornate elk]

I know how to run it on CPU only mode and nope there are some other people here who know also a lot of SD tech stuff

[ornate elk]

Can you share the full cmd log?
And the txt2img settings?

[storm sail]

@ornate elk tell me how to know i am using openvino already? how to know i am getting promised openvino acceleration? etc? running cpu mode is easy command arg but . . . well, like here https://github.com/vladmandic/automatic/wiki/OpenVINO what does this instruction step "Install the drivers for your device" mean?

i think i havent even got openvino on my pc, i dont think its installed via any these builds by default so even tho it may say "accelerated by openvino" i am not being accelerated cuz i have no openvino, etc

i think i gotta get openvino on my pc first, then i get the benefit cuz these builds only optimized to be used by it

but idk how to get it

[brave reef]

Nevermind. I just removed 'easynegative' from my negative prompt and it worked fine, very strange. I have not done a git pull in a minute, so I guess I just needed to experiment more...sorry.

[crude trellis]

does stable diffusion performs better with amd cards in linux than windows?

[storm sail]

i think i am only running stable diffusion with my intel and no help cuz i havent got any openvino despite these builds are only optimized for usage with

i tried open virtual environment in python and pip install optimum intel like hugging face says but that virtual environment is self contained and so openvino still not really on my pc, like python is on my pc, throughout, so these builds still not using openvino but if openvino was on my pc in right way i would get the acceleration promised

does any of what im saying make any sense?

if i can't be helped here, i already trying hugging face discord too, any idea where i might take my confusion? lol

freaking confused and apologies if i offend

ima retard i only know how to install videogames, all this is brand new to me, i barely can pip install with python, lol

still, i think im onto something believing i am not been using openvino cuz is not even really on my machine

only wish i knew how to tell for sure, i mean, if i got vanilla stable diffusion right and i now have "accelerated with openvino" and i run them both with exact same way, there no difference in how long it take for inference

then i try and pip install openvino, still no difference

cuz im stupid and not knowing how to set up correct

[ornate elk]

Best performance + usage is by using Automatic1111 with ZLUDA Backend.
But yes rocm on Linux performs slitghlty faster

Checkout my guides from the pinned messages

[storm sail]

only, if i could be pointed to exe says openvino runtime like and it would install openvino throughout with "Install the drivers for your device" like this https://github.com/vladmandic/automatic/wiki/OpenVINO says to do but gives no other hint, like how python or git is installed throughout my machine, then i would be setup correct finally - uh, with exception i might need to install specific openvino fit models, somewhere, etc . . . but everything i look at is cconfusing af expecting i know much more bout running script/programming than normal fuckup i am

[crude trellis]

so the best possible scenario for amd cards would be using zluda on linux?

[storm sail]

until then, im running raw, without any help from openvino or anything, just straight cpu naked

[crude trellis]

sorry if im asking smt dumb tho

[ornate elk]

nope, zluda is a cuda emulation backend which allows AMD cards to use cuda code by making a bridge to rocm

so the best performance is ROCm on Linux
the best usage and good performance is Zluda on Windows

[crude trellis]

and what would "usage" be?

I mean, what does it changes?

[ornate elk]

extension support, vram usage optmizations mainly

[storm sail]

lol, is everyone these days computer guru/programmer and i only one in utter dark? lol

[crude trellis]

so overall zluda on windows is better

[ornate elk]

yep

[crude trellis]

but rocm would be a little faster

is that it?

[storm sail]

zluda, rocm, etc, man im dumb a

[ornate elk]

yes

[crude trellis]

ok thank you

I mean it

[ornate elk]

but it depends on your knowledge too, if you a linux guy it might be easy for you to setup and test.
also Zluda doesnt support very old AMD cards (same goes for rocm)

[rotund lava]

only thing i know is that SDNext (another sd webui ) supports openvino directly so you could try to download that ui then test image speed vs the speed of your current directml installation

[storm sail]

even vram, idk wtf that really is, i play videogames 😛

@rotund lava again, i hear "supports" but how am i to benefit since i dont even have openvino on my pc

[crude trellis]

im not a linux guy, is just that I really want to be able to use SD the best way possible having a rx6750xt

[ornate elk]

@crude trellis also since a few days WSL2 is out for windows so you can have the same speed as when dualbooting on linux. but the vram optimisation is bad.

[crude trellis]

honestly zluda seems hard to setup on windows for me

[ornate elk]

okay then its easy, go for my Automatic1111 with ZLUDA guide:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

[crude trellis]

I was thinking that maybe dual booting and testing on linux would be easier

[ornate elk]

nah its way more complicated xD

[crude trellis]

lol

[rotund lava]

just download SDnext then install it and then you can enable openvino mode and test speed

[ornate elk]

checkout my Zluda guide, and for any questions feel free to ask, i helped a lot of AMD Users here get it running

[storm sail]

i hear "accelerated with" "supported" etc, but i dont have openvino and idk how to get it so i dont benefit from "support"

[crude trellis]

thank you, I tried that guide some months ago but ended up giving up

it was too much difficult for me

but ill try it again

[storm sail]

i have tested speeds, no difference between openvino builds and vanilla

[ornate elk]

yea try it 🙂 and let me know if any step isnt clear

im happy to help

[storm sail]

i currently have https://github.com/openvinotoolkit/stable-diffusion-webui, no better than vanilla, i see "Install the drivers for your device" here https://github.com/vladmandic/automatic/wiki/OpenVINO (i think that "drivers" means openvino only idk how to get it)

[ornate elk]

with drivers it means just the intel drivers. download intel support assistant and install its drivers

[storm sail]

ive tried 100 builds vanilla, openvino, directml, yada, i see no difference (cept directml bad cuz mass out of memory errors) only my intel cpu/gpu is stable but slow cuz i not getting promised openvino "support" or "acceleration" cuz i have no openvino when install should come with this stuff

why would say install machine drivers when that oughtta be done long ago, even windows updates installs them drivers tho i get mine via dell

believing still i havent got openvino installed

[ornate elk]

maybe your CPU just doesnt support openvino

[rotund lava]

if you see no difference and 20mins per img is the average speed you get then thats your perfomance you wont get any faster on cpu mode

[storm sail]

how is it i can get app off microsoft to run faster etc then i can get stable diffusion going? i tell u i am dumb af and doing something wrong, im missing obvious shit

[rotund lava]

what app of microsoft?

[storm sail]

i can use just about any other ai art deal and have it run better here, only id prefer to have stable diffusion cuz freedom and can setup offline use (idk how exactly that yet but), no, mmy computer capable of better, i see folk rip out renders in seconds and one pic takes me 20 minutes? no, im not setup correctly

@rotund lava name one

[ornate elk]

no you cant make it any way faster, you only have a old cpu and an gpu with 4gb vram

[crude trellis]

btw

[storm sail]

director? but im not here to sell products, just sying, sd better only im dumb af and not knowing how to setup right

[rotund lava]

theres none,ppl who make imgs in seconds have an nvidia gpu (a good one),you only have an outdated cpu

[crude trellis]

my full setup is i5-11400f, 32gb ram ddr4 3200hz, rx6750xt 12 gb vram

is it enough for running sd?

[rotund lava]

cpu only mode with directml or openvino will only get u one img per 20mins

[storm sail]

is stable diffusion really so demanding and specific that all these other apps/programs ouutdoing it in way of performance (system requireents, speed, etc) ????

[crude trellis]

what is WSL2 btw? would I need to follow another tutorial on how to install it or something?

[ornate elk]

WSL2 is Windows Subsystem for Linux (Running Linux in a VM on Windows)
Its in beta. i will make a guide when its stable or better than zluda

[crude trellis]

hmmm ok

[crude trellis]

would that be enough for running SD? it would suck to follow the entire guide just so I can found out my pc cant run it well

[ornate elk]

that would run good 👍

[crude trellis]

ok

ty

[rotund lava]

yes,SD needs a decent nvidia gpu (at least a gtx 980/1060) to run at decent speeds,even amd and intel arc gpus struggle with it,if you dont even have a dedicated gpu you will only experience pain and dissapointment

[storm sail]

i can get apps off google play on my phone able to do near what sd does .without half the system requirements . . i dont buy it, i thought stable diffusion supposed to be contender

[ornate elk]

these other programms (cloud based ai stuff) are running in cloud on servers with hundres of gpus and cpus

[storm sail]

them apps do nigh on everything sd does but only i want sd cuz the freedom

so, adobe fireflys. and that be my only other option u say, i need to be online

[rotund lava]

they are not using your own cpu/gpu 😉

[storm sail]

how do i get online with sd so to take advntage of ccloud, ettc? any free sites, etc?

then i can deal with slow renders when offline (rare) and desperate

[rotund lava]

not for free,only paid but you can use sites like seaart / glif / tensor.art they give u a limited ammount of free imgs every day

[storm sail]

im sad 😦