🤝｜tech-support

[bright hill]

Just so I can get it out of the way

Incase it starting conflicting with the Browser extension

[vocal burrow]

what did you rename

[karmic crown]

if it is lora it will be in lora folder, if it is lycoris then in locon, unless you check option to treat both as lora

[bright hill]

"fucking why" is the new name

[vocal burrow]

so there's no models/lora folder anymore ?

[bright hill]

I've been aggrivated that it has taken me this long to get Loras to show up in auto1111

Yes

That's where everything online said to put Loras

I used the extension, and it works now

But if I dont find out where the extension placed the Lora I downloaded, it's going to bother me

[vocal burrow]

I don't see why it would put them anywhere else than models/Lora

[karmic crown]

civitai browser plus never put anything in any other folder except the default ones (on any of my machines)

and i have it on two macs, windows and linux

[bright hill]

It's in the webui, but I don't have a lora folder anymore, so I don't know where the extension put it

I've been trying to look for it

[karmic crown]

i always enable this

go to settings

[bright hill]

Found it

[karmic crown]

find civitai browser plus

find this

expand that

and see if you changed anything here

[bright hill]

Nope

All options are "None"

[karmic crown]

than i have no idea how can you have loras in any ohter place except in lora folder

[bright hill]

I'll restart the cmd in a minute

I'm currently generating a pic

[karmic crown]

actully, you can change a folder here

but i never touch that

you can try this

https://github.com/alexandersokol/sd-model-organizer

i use those two together

[bright hill]

I'll try it now

[vocal burrow]

you did refresh that explorer just to make sure that it didn't create back that lora folder ?

[bright hill]

Yep

[vocal burrow]

Also you can just search for Cecilia Alcott.safetensors in stable-diffusion-webui and track where it is installed

[karmic crown]

ctrl + r

[bright hill]

already tried that

[karmic crown]

i hope i didnt forgot the sortcut, since i am using windows like 10 times a year in last few years 🙂

[bright hill]

The only one I have is in the "fucking why" subfolder

[vocal burrow]

I'd huse f5 but I don't even know if ctrl + r works or not ^^"

[vocal burrow]

wait ....

you had that lora already in the folder before renaming it ?

[bright hill]

I forgot the order of events

I had the Lora pre downloaded for quite a while

In modes/Lora

models/Lora

[vocal burrow]

hit the refresh button in lora tab

[bright hill]

New showed up in the webui

[vocal burrow]

see if it's still there after that

[bright hill]

Yeah

Just refreshed

[karmic crown]

maybe it was for total commander only?

[vocal burrow]

I haven't heard that name in ages

damn we're old

[karmic crown]

recently i found my business card from the last century (from 1999) 🙈

and i used to make websites for netscape 3 and 4 🙈

[bright hill]

0.o

I've only heard of those browsers in ancient legends

[karmic crown]

i am so old i used net before google 🤣

[lean ridge]

hey, when running crystaldisk info there's a way to fix the Reallocated Sector Count?

[vocal burrow]

let's try to keep the tech support related to stable diffusion problems.

@bright hill is the lora actually usable ? Does it show up in the list of lora used when generating a picture ?

[bright hill]

Yes

I was playing with it a few minutes ago

Also, a strange error popped up when I tried to make a large picture, said it needed 12 GB and I only had 11GB

Thing is my rig is definatiely able to provide that

Is there some command line AARg thing I have to do, to uncap the memory usage?

[vocal burrow]

ok I wasn't there "a few minutes ago" and because you're not sure of the order of events let's establish a ground floor :

• close your a1111
• move that "fking why" folder somewhere far away (in desktop for instance)
• restart a1111 and see if it still shows up in the Lora listing

Also what's in your webui-user.bat ?

[vocal burrow]

no command line arguments is gonna create more vram out of thin air for you.

what's your gpu ?

[bright hill]

NVIDIA GeForcee RTX 4060 Ti

Default webui-user.bat

[vocal burrow]

so 8gb of vram

use set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram-sdxl

[bright hill]

Okay

dxdiag says I have 16123 MB of Display memory

webui relaunched

"fucking why" has been moved to desktop

Cecilia Lora is still present

[vocal burrow]

oh yeah it's TI my bad, you can remove --medvram-sdxl then

[bright hill]

Mate you scared me

[vocal burrow]

and is it usable ?

[bright hill]

Yes

[vocal burrow]

use it and screenshot the whole page

[vocal burrow]

it's not the whole page

I want to see that part specifically

[bright hill]

What part of the page is that?

I don't recognize that layout of the webui

[bright hill]

I do remember reading that at one point, Lora wasa button, instead of a tab

[vocal burrow]

it's just what you get at the end of the generation

[bright hill]

I had to zoom out

[vocal burrow]

it's still gerenating ???

[karmic crown]

you need to wait to finish

[vocal burrow]

1 minute 30 ?!?

with a 4060ti

[bright hill]

1024x2048 resolution?

[vocal burrow]

and you're upscaling

[bright hill]

That was a default option?

[vocal burrow]

definitely not

[karmic crown]

if hires is not collapsed it is ACTIVE

same applies for refiner

colapse both

[vocal burrow]

asking 1024x2048 native from a SD1.5 is gonna give you broken results no matter what

[bright hill]

Ah

[karmic crown]

with sd1.5 use 512x512

or 768x512

[vocal burrow]

SD1.5 is trained for 512x512, SD2 768x768, SDXL 1024x1024

[karmic crown]

or something like that

[bright hill]

Ah

[vocal burrow]

stick to 512x512 no upscale for testing, we don't care about how good the result look

[bright hill]

Several lessons learned, then

[vocal burrow]

we just don't want to waste our time waiting for it to finish

[karmic crown]

when you need to test something do not use refiner and upscaler (unless you are testhing those)

[bright hill]

I was trying to recreat that one pic as a test

Said it had a clip size of "2", and I wasn't aable to find that option

But I'm skipping over that for now

[vocal burrow]

yes, don't spread over multiple problems at the same time

[bright hill]

Gotcha

Standard generatio isn't a problem

The error I had was when I was trying to replicate

[vocal burrow]

problem now is "where is that lora"

[bright hill]

Oh yeah

[karmic crown]

and for that, use a simple setup, like 10 steps euler a

no refiner, no upscaler

[bright hill]

The Lora is working?

[vocal burrow]

screenshot what I asked for please

[vocal burrow]

that's not what I aksed for

[chrome rampart]

evening. i see new version 1.7 or am i wrong?

[vocal burrow]

generate a image using that lora and screenshot the result with the text under it

[bright hill]

I'm sorry, I'm lost

Ah

[vocal burrow]

yes it released a few hours ago

[chrome rampart]

hop everything worsk since auto update lol. thks areytis

[bright hill]

Right, so I think I figured out my problem

When I had that "not enough memory" problem I was trying to generate a 1024x2048 image

Since I thought that was the resolution of the test image

[kindred tinsel]

Hi! I'm having some issues with my generation times (seconds per iteration exactly). Here is the description of the problem:
My generation times doubled from literally yesterday to today, I've always kept my gen times at 10sec per iterations on my 5500 XT smth like that AMD gpu. But now I am getting double that and sometimes triple that. I am running a promp with:

• No Loras,
• vae 84000 ft pruned
• One 1.99 gb model pruned
• No control NEt
• No hire.s fix

The only thing that I changed and now reverted back to was the location of my pythonn installation from an ssd to a hd. But then I reverted back from this change and pushed it to my c drive and used the 3.10 distribution in the windows store.

Can it be that by uninstalling python I removed some caches for stable diffusion?

[faint shuttle]

Please guys how to edit a torch file, if that's possible !!!

[bright hill]

@karmic crown @vocal burrow Thank you for your help

I have learned a lot about operating SD

[karmic crown]

what is torch file?

[vocal burrow]

go to Settings->SysInfo->Download system info and drop the result file here

[vocal burrow]

You don't want to use microsoft's python as you don't have control about its version. Use the proper and standard python from python.org (any version from 3.10.6 to 3.10.13 will do just fine)

[kindred tinsel]

So like, could that be the problem because that is the only thing that I've changed, I dont want to think I fried my card with the amount of images I have generated

[vocal burrow]

if your gpu is fried it would simply not work anymore, it wouldn't "half work"

and unless you messed with it you're not gonna kill it with SD, just like you're not gonna kill it gaming on it

[kindred tinsel]

Alright, am uninstalling python 3.10.11 which was located I think in the hdd now, and also uninstalled python from windows store. Lets see if that changes this

[chrome rampart]

can i see / read somewere whats new or changed in V1.7?

[karmic crown]

in release notes

[chrome rampart]

tip were to look for that ?

[karmic crown]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/releases/tag/v1.7.0

[faint shuttle]

im facing a problem with an extension, its asking me to change the load from gpu to cpu, cause im running an amd rx580 which doesnt have cuda

[chrome rampart]

thks and wish amazing weekend

[faint shuttle]

Attempting to deserialize object on a CUDA device but torch.cuda.is_available() is False. If you are running on a CPU-only machine, please use torch.load with map_location=torch.device('cpu') to map your storages to the CPU.

[karmic crown]

windows or linux? which software? a1111?

[vocal burrow]

just to be sure that's the C:\Users\Weston\Desktop\stable-diffusion-webui-master\models folder ?

[bright hill]

Yes

C:\Users\Weston\Desktop\stable-diffusion-webui-master\models

The exact filepath I copy/pasted from windows explorer

[vocal burrow]

ok drop your webui-user.bat in there

and how did you initially install that lora ?

[bright hill]

Downloaded manually from Civitai, moved manually into models
Had to create a Lora subfolder manually

Placed it into the Models/Lora

You want me to more web-user.bat into models?

[karmic crown]

no

[vocal burrow]

no I want you to drop your webui-user.bat in discord

[karmic crown]

that looks good for your card

[vocal burrow]

Honestly I have no clue where that damn lora is nor why it's working....

[karmic crown]

what will happen if you close cmd window and run bat again?

[bright hill]

Could Auto11 had changed something?

[vocal burrow]

But on the other side I'm pretty sure you didn't install your stable-diffusion-webui correctly because there is no commit version showing up in the log.... So I'd suggest to reeinstall it (corre'ctly this time, using git https://github.com/AUTOMATIC1111/stable-diffusion-webui/#automatic-installation-on-windows) that way you'll also have the 1.7.0 verion instead of your 1.6.1 you're running right now.

[jaunty sphinx]

Help I require tech support

[vocal burrow]

what are you trying to do ? what's the complete console log ?

[jaunty sphinx]

I am struggling to do stable diffusion web ui open vino with intel iris xe graphics for sdxl turbo

I get this error: TypeError: StableDiffusionPipeline.init() got an unexpected keyword argument 'text_encoder_2'

Is it my torch not update or wot

[bright hill]

I downloaded auto11 like two days ago

[jaunty sphinx]

I only got it to work on stable diffusion 1.5

[bright hill]

I'll try again

[vocal burrow]

yup and it got updated today

[jaunty sphinx]

open vino only works on stable diffusion 1.5

[bright hill]

Oh, okay

[jaunty sphinx]

for me at least

[bright hill]

I was gonna be anoyed with myself if I somehow messed that up

[jaunty sphinx]

albiet the tutorial says it can work on sdxl turbo, sdxl, and sd 2.1

the intel tutorial\

for openvino

or openvino.ai

[vocal burrow]

what tutorial ?

[jaunty sphinx]

here ill copy and paste the list that I use: https://github.com/openvinotoolkit/stable-diffusion-webui/wiki/Installation-on-Intel-Silicon, https://www.intel.com/content/www/us/en/developer/tools/openvino-toolkit/overview.html, https://www.intel.com/content/www/us/en/developer/tools/openvino-toolkit/download.html?VERSION=v_2023_2_0&OP_SYSTEM=WINDOWS&DISTRIBUTION=PIP, https://docs.openvino.ai/2022.3/notebooks-installation.html, https://www.intel.com/content/www/us/en/developer/tools/oneapi/ai-tools-selector.html, https://docs.openvino.ai/2023.0/documentation.html#:~:text=The Intel® Distribution of OpenVINO™ toolkit supports neural,deprecated and will be removed in the future). https://www.youtube.com/watch?v=5X0RmlH6JI4

theres a whole list for u

but its basically intel, open vino

[vocal burrow]

which one did you follow ...

[jaunty sphinx]

the youtube one

the youtube one redirects me to all the links

theres also an article

linked from the youtube video

https://game.intel.com/story/intel-arc-graphics-generative-ai-art/

its the intel youtube guy

youtube and that article he linked which links all the other links i posted

so yea I do not know how to change the stable diffusion models for open vino compatible despite downloading the intel tools to do so.

do I do some commands in cmd or do I use a ui or an extension

[karmic crown]

just a side note: I updated my A1111 guide for Macs 2 times in the last 3 days. Many things constantly happes so that video guides might become obsolete fast

[jaunty sphinx]

dam

because he made it look so easy

just downlaoded a model

and done

[karmic crown]

i am not saying it is, just saying MIGHT

[jaunty sphinx]

he can use realistic vision5 and everythang

just by doing git clone open vino

then me

its like

JALSKDJASLKDJSALDJ

not working

and bob duffy pretty credible too

darn

[vocal burrow]

ok but what's your console log then ? when does the error happen ?

(also please no need to spam)

[kindred tinsel]

I tried reinstalling python and now I'm using the distribution form python,org 3.10.11 version. But I am still seeing dead slow generation times for a pruned model, even with lowvram a pruned model would usually give me something like 10.smth seconds per iterations but now for 20 iterations an image takes upwards of 10 minutes :c.

My gpu isnt bad at all and I've seen faster generations even with literal control net on.

[vocal burrow]

did you delete the venv folder ?

[kindred tinsel]

No......I figured I only needed to update the direction for the python install in the .cfg file

[jaunty sphinx]

oh sorry for spam

the error only happens for other stable diffusion models

[karmic crown]

remove venv, it will be created again

[vocal burrow]

delete it, then share your webui-user.bat here and send us the result for a simple 512x512 SD1.5 generation

[jaunty sphinx]

do you want the whole console log or just the thang

[vocal burrow]

everything

[jaunty sphinx]

i think thats it

[kindred tinsel]

omw omw

[jaunty sphinx]

or this one when i did it with the other stable diffusion 1.5

[karmic crown]

xl model + 512x512

no no

[jaunty sphinx]

uh oh

you scaring me man

oh yeah does reactor or roop work with open vino?

and control net too?

i haven't searched if the extensions worked with open vino

[karmic crown]

you cant use sdxl model to generate 512x512

[jaunty sphinx]

hm

so should i generate 1048x1048?

i thought it works

[karmic crown]

768x768, 768x1024, 1024x1024

[jaunty sphinx]

ok

[karmic crown]

512x512 and 512x768 with 1.5 models

[jaunty sphinx]

interesting

cause i could do 512x512 on fooocus collab b4

thanks viking!

[karmic crown]

sdxl models are trained on 1024x1024

1.5 on 512x512

2.x on 768x768

[vocal burrow]

1/ you're not up-to-date, you're using the old 1.6.0,
2/ your uddetailer extension is broken
==>
update stable-diffusion-webui + your extensions and then try again

[kindred tinsel]

wait but why some models in sd1.5 allow for 578 x 1024 sizes? is that bad ?

[karmic crown]

also every time when you get a memory related error - restart web ui (close cmd and run again)

[vocal burrow]

but yes it might break things because openvivo is ..... kinda hacky .... and a new version of stable-diffusion-webui just released today soooooo

[jaunty sphinx]

oh thanks aryetis! i will try to update stable diffusion, is there a proper link?

yea i downloaded stable diffusion for open vino yesterday

i am testing it on my laptop

oh how do i update uddetailer also

[vocal burrow]

you seem to have installed it properly so just running git pull in stable-diffusion-webui folder should be enough (you'd still have to upgrade extensions in the UI tho)

[vocal burrow]

check for updates, Apply and restart UI

[jaunty sphinx]

oh yeah i forgot about that

[karmic crown]

smaller size than trained is usually a problem

larger is not a problem

[vocal burrow]

probably not, roop is held together by faith and ducktapes. So throwing open-vivo on top of it is asking for troubles

try to get a proper vanilla open vivo install working first

then try to add stuff on top of it later

[jaunty sphinx]

dam i'm trying to copy fooocus faceswap for open vino

oh when i do gitpull it says i'm alraedy up to date in the cmd prompt

[vocal burrow]

hmmm

ok I guess you're using the openvivotoolkit repository then

[jaunty sphinx]

oh yeah, do you know how to make open vino use gpu for switching models, I see when i switch models using the a1111 webui it seems to use cpu and lags the hell out of my computer.

yea I did the tutorial

the gitclone copy thing

[vocal burrow]

So the moment extensions are updated by devs to support anything 1.7.0 specific it's gonna break in your version

[jaunty sphinx]

dam son

[karmic crown]

what git log gives?

[vocal burrow]

did I mention that openvivo is hacky ?

[jaunty sphinx]

yes sir

\stable-diffusion-webui>git pull
Already up to date.

d;

[karmic crown]

git log

not pull

[jaunty sphinx]

commit 44006297e03a07f28505d54d6ba5fd55e0c1292d (HEAD -> master, origin/master, origin/HEAD)
Merge: f2a35b84 ed6be6e7
Author: Qiacheng Li [e-mail]
Date: Thu Nov 16 10:54:40 2023 -0800

simplify installation

Merge dev to master

commit ed6be6e7f0449e228e6d18b5e40aeef8f1959830
Merge: 35e72f24 f2a35b84
Author: Qiacheng [e-mail]
Date: Thu Nov 16 10:52:40 2023 -0800

solve conlifcts

commit 35e72f24cdd56f9a49b245ad326fa27249ff397e
Author: Qiacheng [e-mail]
Date: Thu Nov 16 10:43:06 2023 -0800

remove torch-install.bat

commit 62618d1c91cd82a84eac6e9f7d7ec6d292e82798
Merge: d172a08d 9241c00e
Author: Qiacheng Li [e-mail]
Date: Thu Nov 16 10:39:11 2023 -0800

Merge pull request #77 from openvinotoolkit/simplify_install

sry

[vocal burrow]

yup that's the openvivotoolkit repo

[jaunty sphinx]

nice

should i use a different open tool kit

[vocal burrow]

I don't think there is

[jaunty sphinx]

damn

is this where i must accept defeat

[vocal burrow]

you can try using sd.next instead

supposedly it has openvivo support and it's based on auto1111

https://github.com/vladmandic/automatic

[kindred tinsel]

thats the resuilt 512*512 took close to 4 minutes to finish

[jaunty sphinx]

holey shit 4 minutes

[kindred tinsel]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --enable-insecure-extension-access --precision full --no-half --no-half-vae --lowvram --disable-nan-check --opt-sub-quad-attention --opt-split-attention-v1 --no-gradio-queue --sub-quad-q-chunk-size 128 --sub-quad-kv-chunk-size 128 --sub-quad-chunk-threshold 70 --autolaunch

call webui.bat

[jaunty sphinx]

the longest my render was on cpu was 2 hours ngl

[kindred tinsel]

:0

arc gpu?

[jaunty sphinx]

nah pure cpu

intel core

[kindred tinsel]

DAMN

but..... you aint meant to do that

or like, it will be shitlong

[jaunty sphinx]

yes

i was experimenting

i was watching a whole movie while waiting

[karmic crown]

have you tried to use all options that exists?

[jaunty sphinx]

CPU is last resort to render things in my opinion but its not bad quality

[vocal burrow]

try using sdnext with the --use-openvino launch parameter

[jaunty sphinx]

sdnext is better?

[kindred tinsel]

no no, but that is literally what they recommend for amd gpus.

In the autmatic111 fork for directml they recommend you add those so that it works.

Should I remove them?

[jaunty sphinx]

yolo no balls i'ma try sdnext

[karmic crown]

@kindred tinsel what is your gpu?

[jaunty sphinx]

oh yeah does anyone know how to make stable diffusion run in local disk D: instead of C:

i heard some can run stable diffusion just using a usb also

[vocal burrow]

it has openvivo support

[kindred tinsel]

AMD RX 5500 XT 8GB RAM 2019

[vocal burrow]

you can install stable diffusion wherever you like, it doesn't matter

[jaunty sphinx]

oh

dam

[vocal burrow]

take your sd folder and move it somewhere else (make sure to take the folder and not the files inside, as there's some hidden .git folder that you need to move too)

[kindred tinsel]

its old viking I know, for what you would expect for 2023 its old af ik

[karmic crown]

I guess you can use something like --skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half --medvram but @vocal burrow knows better

[kindred tinsel]

if I use --medvram stable diffusion tends to GO WILD and create grandeur and then fail cus of lack of memory, while it ofc creates stuff much faster

[karmic crown]

i know what people with Amd in Macs need 🙂

[lucid stirrup]

does anyone know what is happening here

it was working perfectly i was messing around with downloading custom cuda for keras and now after few pictures it generates noise or just grey/black images, gpu rtx 3060 12gb

my args:
set COMMANDLINE_ARGS=--no-half --no-half-vae --precision full --disable-nan-check

i tried with this argsset
COMMANDLINE_ARGS= --xformers --no-half-vae
i get this error (i deleted venv)

[kindred tinsel]

also how do i remove the xformer warning while we are at it haha

[jaunty sphinx]

where is the hidden git folder

woa ur computer is faster than mine @lucid stirrup

[lucid stirrup]

bruh it doesnt matter when i have bugged out mz SD

[jaunty sphinx]

are segmentation faults good?: C:\arrow\cpp\src\arrow\filesystem\s3fs.cc:2904: arrow::fs::FinalizeS3 was not called even though S3 was initialized. This could lead to a segmentation fault at exit

lol

[lucid stirrup]

first 10 images are perfectlz normal and next 10 are this mess

[karmic crown]

just use --skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half --medvram for now

[lucid stirrup]

and after a while it starts generating black/grey pictures

[jaunty sphinx]

lol

[karmic crown]

if something should be removed someone will tell you, but that should work for you

[vocal burrow]

set COMMANDLINE ARGS=--medvram --no-half-vae --opt-sub-quad-attention --opt-split-attention-v1 if it complains about lack of vram, change --medvram to --lowvram, for the RX5000 series you might need to replace --no-half-vae with --no-half --precision full
@kindred tinsel

[jaunty sphinx]

dam why does it take a while for apply and restart ui for me

[kindred tinsel]

Okay Okay and what do I do about xformers, I keep getting that warning about no module found

[jaunty sphinx]

already 10 minutes now 😮

yea i strugled with xformers

[vocal burrow]

you can't

[kindred tinsel]

a

[jaunty sphinx]

i tried doing pip install xformers

lol

[karmic crown]

@vocal burrow amd does not need --skip-torch-cuda-test?

[vocal burrow]

I don't know, it should be along the other files

that's why I'm telling you to move the whole folder

[jaunty sphinx]

oh i did do that

thanks

ez cake

[vocal burrow]

xformers is nvidia only

[jaunty sphinx]

yea

i found out da hard way 😦

oh well waiting for comp to load sd.next

[vocal burrow]

what's your gpu ?

[lucid stirrup]

rtx 3060 12gb

[lucid stirrup]

.

[vocal burrow]

I'd say your "custom cuda" exeprimentation has gone wrong and you should revert to the default ones then.

[lucid stirrup]

i already uninstalled cuda and reinstalled new ones and same thing happens

[vocal burrow]

With a 3060 you shouldn't have Nans errors in Unet especially when using SD1.5 based models

[lucid stirrup]

now i have newest verrsion

[kindred tinsel]

Okay the good news are that with the new command line params, thing went down from literal 22 SECONDS per iteration for a 600x 1024

to 12.41 SECONDS per iteration for same size,

Thanks @vocal burrow and @karmic crown appreciate the help with this!

[jaunty sphinx]

dam ati got the nvidia

he speeeeed

good job volv

[lucid stirrup]

what gpu do you have

[lucid stirrup]

i had rx 580 and bruh was that slow

[jaunty sphinx]

has anyone got stable diffusion to work on an android samsung phone

[kindred tinsel]

AMD RX5500 XT

[jaunty sphinx]

i only heard qualcomm got it to work

[kindred tinsel]

why?

[jaunty sphinx]

AMD is fast too

i wish my cellphone could run ai locally ;p

or make a blackberry or raspberry do ai ;p

[vocal burrow]

using --no-half argument should fix those kind of errors but still it shouldn't happen with a 3060 and it's gonna increase your vram usage for nothing....
You can also try to set the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion to true instead.

I'm off, bye

[lucid stirrup]

thanks bye

[kindred tinsel]

hmmmmm two more things sorry,

1. How can I install olive for model optimization in amd, anyone have a guide for that for automatic 1111
2. Anyone know why my instances measures in seconds per iteration, I hear that it should be iterations per second as if I am literaly going the wrong way about it

[jaunty sphinx]

woa what is olive

[karmic crown]

same question 🙂

[vocal burrow]

https://community.amd.com/t5/ai/updated-how-to-running-optimized-automatic1111-stable-diffusion/ba-p/630252
good luck

[karmic crown]

i can answer question number 2

[jaunty sphinx]

i think iterations per second or seconds per iterations is the same thing

not sure d;

[lucid stirrup]

nope seconds per iterations is slower

[kindred tinsel]

oh no the post :c that thing sucks at explaining it :c. It literally forces you to isntall a new conda thingy and shit

[jaunty sphinx]

dam

i must be slow

[karmic crown]

that depends of speed, if it can do more than one iteration per second than it is it/s, if it need a few seconds per iteration that it backame s/it

[jaunty sphinx]

darn how do i get my laptop to do iterations per second

[lucid stirrup]

2 iteration per seconds is 1/2 seconds per iteration

[karmic crown]

buy new one 🙂

[lucid stirrup]

so 2 iteration per seconds is faster than 2 seconds per iteration

[jaunty sphinx]

rip wallet

[kindred tinsel]

damn so I am literally taking close to 13 seconds per single iteration, that do be shit

[jaunty sphinx]

what is a good reasonable priced computer laptop to spam my ai everywhere

[jaunty sphinx]

is that slow or fast

[kindred tinsel]

yeah thats me ^^

[karmic crown]

M1 Mac with SDXL 1024x1024

that is SLOW

[lucid stirrup]

that is good, i have had 30 sec/it on rx 580

[jaunty sphinx]

shit son

dam

i should read those youtuber specs they got their images loading on multiple stable diffusions in like 3 seconds bruv

[kindred tinsel]

One question,

there is a section that says should I do that in the webui.bat file like the one that doesnt even mention user and run it there?

[vocal burrow]

I mean theorically you don't have to create the conda stuff, it should work the same with a usual install (as long as you use the correct directml fork). Just make sure to use webui.bat --onnx --backend directml

[kindred tinsel]

so where oes that go?

[jaunty sphinx]

oooo --onnx i heard thats for open vino models not sure how to use it

[kindred tinsel]

webui-user.bat or webui.bat normal file

Now we are talking, perfect image for a literal 600x1024 in under five minutes

[jaunty sphinx]

lol

i wonder how long does a text to vid takes

[vocal burrow]

put --onnx --backend directml in the command line args of your webui-user.bat

[kindred tinsel]

okay okay Ill do that

[jaunty sphinx]

is comfyui good, I can get that using cpu or gpu using open vino support also

albiet still pretty slow

i usually used cpu on comfyui

oh well still waiting for sd.next to load cause muh wifi bad

[karmic crown]

@kindred tinsel do you want to hate me? check this out 🤣#🏞｜general-with-images message

[jaunty sphinx]

dam thats hot

has anyone did nsfw pictures and models yet

thats the most challenging on my computer tbh

[vocal burrow]

Yes but rule #4 please #✍🏼｜rules-and-tos message

[kindred tinsel]

wait wait wait wait wait i got A BUNCH of errors

@vocal burrow

should I do as they say and update diffusers or will that break my webui stuff?

[karmic crown]

restart

and be sure to check model

[vocal burrow]

it's telling you everything you need to know

[kindred tinsel]

but the diffusers thing? should I do the pip thingy?

[kindred tinsel]

oh yeah woah sdxl!!!! I cant run it to compete

[karmic crown]

i am talking about this error

[jaunty sphinx]

oh i just used google

[vocal burrow]

treat error one by one :

• fix the first one
• re run
• repeat until it works

[kindred tinsel]

oke im reinstalling this little thing called torch!!!

[vocal burrow]

I really have to start my workout now (it's already 11pm here lol)

bye

[karmic crown]

bye

[kindred tinsel]

oh okay bye!

[karmic crown]

good night

[kindred tinsel]

lol every tech support basically signing off :c

[karmic crown]

that is 1.5 model, without upscaling, directly generated as 768x1024, and the image was generated in just 8 steps in a little more than 1 minute on my mac

[kindred tinsel]

how long did it take?

[jaunty sphinx]

bye

[karmic crown]

klick reply on that image, and ask what you want in that other channel

lets get back to techical problems here

but i need to go soon, i am to tired

long day

[kindred tinsel]

oh yeah, I still have one tho, 1) somehow the python installation is not showing up in my cmd, I have it in the environ variables and yet it doesn't show up, andd 2) even if i run it from the python 3.10 command line it doesnt work

[karmic crown]

open cmd, and type pyton --version does it work?

[kindred tinsel]

yeah !

[karmic crown]

what version it is?

[kindred tinsel]

3.10.11

[karmic crown]

perfect

so your python works correctly, and you have recommended version

[kindred tinsel]

Ugh the pip doesn't even show up in files :c nor in the command prompt

[karmic crown]

pip --version

does it work?

[kindred tinsel]

nope it says pip is not recognized as an internal command

[kindred tinsel]

I did that!!!!! and it did install it but now it wont run saying that it needs that other version of diffusers

[vocal burrow]

(I have to stop looking at discord whenever I pass by my computer)

[vocal burrow]

then show us the error log that you have now

[kindred tinsel]

I got it fixed wait a min

[karmic crown]

have you installed this pyhton? https://www.python.org/downloads/release/python-31011/
and checked "add python to path" when installing it?

it is strange that pip doesnt work

[kindred tinsel]

Yeah I redid it now thank you!!!. Im waiting to see if the webui passes the argument to reinstall torch since the error was about diffusers last time

[karmic crown]

just run
webui.bat --reinstall-torch
without changing webui-user.bat

[kindred tinsel]

I did that, it says it installed version 2.0.0
now its trying to launch the webui.

But i get errors in the onxx webui-user.bat file saying that I do not have the correct version of diffuers when I already updated those dependencies in the folder for sd and in general in the pc

[karmic crown]

show me the error

i can stay here max 2 more minutes, my head will explode

[kindred tinsel]

its okeee dw I will try to debug it on my own if you are tired ^^. I need to re run this stuff because it seems like it only updated the webui.bat file and not the webui-user.bat file which I use to access sd, therefore I will replicate everything I did on this other file, the erorr is basically

py torch version 1.13 +cpu instead of 2.0
pip install --upgrade diffusers

Its the same thing

[karmic crown]

friendly advice, run them from cmd, do not click shortcuts

that way you will be sure that you are running right file form a right place

i have a felling that you have two installations

[vocal burrow]

activate the venv and then try to upgrade diffuser

going back to do my reps

[karmic crown]

venv/bin/pip
is a dirty way to launch pip from venv without activating venv 🙂

actually on windows is venv\bin\pip

sorry, mac and linux all day for years 🙂

[kindred tinsel]

yeah here?or like uh pip in the python install foldere

[jaunty sphinx]

is linux op

[karmic crown]

open cmd first

[jaunty sphinx]

use cmd yee

pip --version

[jaunty sphinx]

cmd line op

[kindred tinsel]

i tried running the command they said, pip install --upgrade diffusers but it says requirement already satisfied (1.26.0) which I suppose is the version

[jaunty sphinx]

ye

[karmic crown]

i am out for today, bye people

[jaunty sphinx]

adios viking

may there be another ship

[kindred tinsel]

bye viking!!! Ill try to get this done !

[jaunty sphinx]

yo volv do u know why the stable diffusion sometimes connection errors out bruv

[kindred tinsel]

memory errors are the usual for me tbh

[jaunty sphinx]

dam son

it keeps happenign to me

[kindred tinsel]

but I dont know a lot about it, it usually only happens when i truly create a huge ass image

[kindred tinsel]

is the console open?

[jaunty sphinx]

and also why switching models also use my cpu instead of my gpu d;

ye console op

[kindred tinsel]

I see the issue is coming from this

1. I run the reinstall command for pytorch and it leaves it at 2.0.0 + cpu (which is supposed to be the required version)

2. I remove that command from the webui so it doesnt try to reinstall it every single time I open it and place back --onnx and --backent directml
   (here the thing reverts back to 1.13+cpu as if nothing had happened in stage 1 and gives me errors about torch, and to top it all off it gives me errors aobut diffusers too, when i have them in version 1.26 it says they are 0.16.1)

[jaunty sphinx]

nice

[kindred tinsel]

not nice ma friend not nice :c I am now deep in the venv updating directly in it I hope this works

Nope I killed it, it looked as if it was good but nope, now it says it cant find the entry point for the application, I messed up some dlls or smth idk imma delete this venv and just go back to waiting 12.49 seconds per iteration

[jaunty sphinx]

lol

yea i dont think juggernaut xl works with open vino every time i do it i get an error lol

[ornate elk]

hey, what are you trying to do?

[jaunty sphinx]

he trying to make his amd gpu and his computer go faster

to run dat stable diffusion supa fast

[ornate elk]

i wouldtn recommend any amd user to try the MS Olive version, its super limited in usage

[jaunty sphinx]

yea idk whats olive tbh

but do u know how to make juggernaut xl run on open vino

that shit hard af

or make any model run on open vino not just the basic ones lol

[ornate elk]

have you tried dreamshaperv8?

[jaunty sphinx]

oooo ill try that

[ornate elk]

2gb model should work, if openvino uses normal .safetensor files

[kindred tinsel]

hI I was trying to install basically diffusers version 1.18 because I needed that for stable diffusion to run and when I try to install it it said that the requirement was already satisfied. But when I tried to run the web UI it gave me an error that the thing wasn't installed correctly. And then when I tried to run for instance with reinstall torch because I needed that for basically running this direct ML backend these onxx that AMD released and when I tried to use that it also didn't work because it said that I didn't have call torch installed and that I didn't have the diffusers installed so it couldn't install any models and it basically couldn't run. It kept keeping me errors and I just went back to the normal stable diffusion and now it runs.

[jaunty sphinx]

yeee thats what he wants

[kindred tinsel]

ssupposedly that should make my models run faster with amd

[jaunty sphinx]

i forgot how diffusers work to be honest

[ornate elk]

okay, whats your gpu?
and did you used directml before or the cpu mode?

[kindred tinsel]

I don't understand what you mean by did you use directml before? I basically just forked the webui. I don't even remember who made it, but it was like the stable diffusion web UI direct ML and I just installed that in my computer. I used to run it normally. my GPU is an AMD RX5500XT. It's from 2019. It is not a good card for. say it has eight gigs of VRAM. It's basically not a lot but he should be able in theory to run these models and and right now I'm I'm running some iterations and I don't know without onxx it works.

[ornate elk]

okay, yea the normal directml fork works for AMD cards, you should stick to it or use linux with the default auto1111. because olive is to limited. you would need to use converted onnx models. these are limited to a locked resolution and not every model is convertible. lots of problems and not really good supported in the directml fork.

If you just want a fast generation in 512x512 or 768x768 you can use the Shark webui. it has the same apporach as Olive but is much more stable and usable.

[kindred tinsel]

Ah, I see I didn't know directml had so many problems, and onxx has also these problems. Thanks for the help. I'll stick with directml fork from webui. I mean, it works fine. It's just a little slow, but I figure that's also my card. So yeah, thanks for the help, though!

[ornate elk]

directml is very vram hungry and not optimized for heavy workloads. so thats the problem we have on windows rn.
rocm is the answer on linux and will come to windows soon hopefully. But your gpu might be not supported

[kindred tinsel]

Damn sad its only from 2019 god damn inma need anew one again

[ornate elk]

yea its sad :/ i mean your lucky that AMD FSR3 will support your gpu but your on the edge of support end for the most software features of amd

[kindred tinsel]

Would you recommend changing to Nvidia gpu like 4090 or lower like 30 series? Or sticking with amd since they are slightly cheaper

[jaunty sphinx]

guys guys i found from another discord

they implemented intel gpus for a1111 now

they said they just updated it today d; so i dont have to use open vino

lets goooo

i just have to change the commandline args to uh whatever the dood on unstable diffusion said was --use-ipex

i hope it works lads

[ornate elk]

depends on your use case. if you primarily wanto to use ai stuff then nvidia is better supported, but they have high prices and have low vram. Thats why i switched to AMD.
If you primarily want to game and to play around with ai stuff ocassionely i would go with AMD as they offer more vram for lower prices. Also when Rocm will come to windows you can use SD much better with it than now.

[jaunty sphinx]

and destroy the venv

[kindred tinsel]

oh I don't game that often and was thinking of getting into lora training and model training for sd1.5 to test it out so probably 8 gb of cram won't cut it rn but I think 24 should do the trick with the 4090, I know training isn't supported with amd so yeah thanks for the infot !

[ornate elk]

ohh right, hope it works for you, let us know

[ornate elk]

yep for training go with nvidia, 4090 would be capable of everything ^^

[azure magnet]

With 1.7.0 need i to upgrade torch in xformers?

[vocal burrow]

I don't think it works for your iris xe as it's an integrated chipset.
From what I can read IPEX support is only for Intel Arc GPUs.
I don't have an intel chipset nor an arc gpu to confirm this.
So yeah don't waste too much time on this.

[sharp lichen]

what should my denoising strength be for ESRGAN_4x?

[lucid stirrup]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/14346

any help would be great

[ornate elk]

Below 0.5
The lower the less the image will change

[vocal burrow]

don't use --disable-nan-check it will only hide useful error message instead of fixing them

also --no-half-vae is included within --no-half so there's no point having both

[lucid stirrup]

i have tried with --no-half only and this weird generated images

[sacred lark]

im trying to download automatic1111 but the instant i try to install it this error shows up and when i clicked any key it just closes

[sacred lark]

Help pls

[lucid stirrup]

i have the same error in WSL, you do not have pip installed or pip and python version you have are missmatched

i mean i guess that is the problem i did not solve it yet

[sacred lark]

yup i dont have pip installed i didnt even knew i needed it :/

[ornate elk]

Delete the venv folder and rerun the webui-user.bat

[next tapir]

how can I make stable diffusion use rocm? I installed the rocm drivers, it's still using cpu

[ornate elk]

On Linux or windows?

[next tapir]

Linux

[ornate elk]

Then it should work If you delete the venv folder again

[vocal burrow]

Make sure you have set the permissions and user groups correctly sudo usermod -aG video,render $LOGNAME. And that rocm can actually use your gpu (if it shows up when you type rocminfo you should be good to go)

[next tapir]

Python 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0]
Version: v1.6.1
Commit hash: 4afaaf8a020c1df457bcf7250cb1c7f609699fa7
Launching Web UI with arguments: --precision full --no-half --skip-torch-cuda-test --upcast-sampling --port 10100
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
Loading weights [6ce0161689] from /home/vellu/Diffusion/stable-diffusion-webui/models/Stable-diffusion/v1-5-pruned-emaonly.safetensors
Running on local URL:  http://127.0.0.1:10100

To create a public link, set `share=True` in `launch()`.
Creating model from config: /home/vellu/Diffusion/stable-diffusion-webui/configs/v1-inference.yaml
Startup time: 4.1s (import torch: 1.5s, import gradio: 0.6s, setup paths: 0.7s, other imports: 0.3s, load scripts: 0.3s, create ui: 0.4s, gradio launch: 0.2s).
Applying attention optimization: Doggettx... done.
Authorization required, but no authorization protocol specified
Model loaded in 3.0s (load weights from disk: 0.5s, create model: 0.2s, apply weights to model: 1.8s, calculate empty prompt: 0.4s).
MIOpen(HIP): Error [Compile] 'hiprtcCompileProgram(prog.get(), c_options.size(), c_options.data())' naive_conv.cpp: HIPRTC_ERROR_COMPILATION (6)
MIOpen(HIP): Error [BuildHip] HIPRTC status = HIPRTC_ERROR_COMPILATION (6), source file: naive_conv.cpp
MIOpen(HIP): Warning [BuildHip] /tmp/comgr-95ec5f/input/naive_conv.cpp:39:10: fatal error: 'limits' file not found
#include <limits> // std::numeric_limits
         ^~~~~~~~
1 error generated when compiling for gfx1030.
terminate called after throwing an instance of 'miopen::Exception'
  what():  /long_pathname_so_that_rpms_can_package_the_debug_info/data/driver/MLOpen/src/hipoc/hipoc_program.cpp:304: Code object build failed. Source: naive_conv.cpp
./webui.sh: line 255: 33956 Aborted                 (core dumped) "${python_cmd}" -u "${LAUNCH_SCRIPT}" "$@"

now i get this

installing libstdc++-12-dev fixed it

it does not sow my models tho

oh wait, i made a typo in the model format, nvm

very odd, when i try to switch to the model it says this

torch.cuda.OutOfMemoryError: HIP out of memory. Tried to allocate 58.00 MiB (GPU 0; 11.98 GiB total capacity; 11.06 GiB already allocated; 120.00 MiB free; 11.80 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_HIP_ALLOC_CONF

and my screen briefly turns black. I have 12 gb of vram, cant i run sdxl?

[ornate elk]

You need some commandline args in your webui-user.sh
Uncomment the Commandline_ARGS= line
There add --medvram --opt-sub-quad-attention --opt-split-attention-v1 --upcast-sampling --no-half-vae

[next tapir]

Where do i need to uncomment it?

Nvm i found it

It works!

Thank you

[ornate elk]

No problem 🙂

[ebon kernel]

Hi friends - I'm not sure if I installed stable diffusion correctly on my mac (16GB memory, Mac1), but when I went to make an image in stable diffusion I got the below error. I went to settings and toggled on the setting it indicated. I got the same error. I'm not a technical person at all but was able to get this far with a tutorial. Now i'm stuck. Any help is much appreciated. Thank you.

[karmic crown]

show me the bottom of your webui

this part

[karmic crown]

m1 mac?

[ebon kernel]

yes

m1 pro, it says

[karmic crown]

https://github.com/viking1304/a1111-setup/discussions/2

you can go step by step, or just run my script

whatever you want

https://github.com/viking1304/a1111-setup/releases/tag/0.0.2

[ebon kernel]

Thank you. 👍

[karmic crown]

but

since you have m1

which is same as mine

i would recommend to run my script like this
sh a1111-setup.sh -f all

[tight terrace]

Hi all! I have a question (probably stupid question for most of you). How do I add the editing icon in the img2img picture?

I can't see the pen in the top right corner. Turotial I'm following has this option but not me....

[sweet kraken]

can anyone help me with this problem i am using an imac and google collab

[karmic crown]

I can help with Mac, but not in that specific combination; how does that even work? which guide did you follow?

[sweet kraken]

hi viking1304 — Today at 18:34
i have been making videos with this for a long time but it has now stopped working and when i go to the xformers site there is no help for mac there

[karmic crown]

xformers is for nvidia

not for macs

but I am confused, how do you combine collab with the local machine? where did you find info on how to set up your enviroment?

normally, Google Colab uses Google Cloud resources only, and the only thing that is used on the local machine is the browser

so usually it is not important if you are using a mac, pc or tablet

[sweet kraken]

are you still there viking1304 drunk nieghbours came to my door ,is there any way i can make SD videos with my setup

i get the collab notebook from the internet then run the cells and fill in what i want it to do and it used to work without any problems

[karmic crown]

in that case you are using Google Colab as expected (in Google Cloud) so it is not importnat what device you are using

local mac install is somthing compleatly diffrent

thats why i was confused

i thought someone actully find a way to combine resources

i do not think that anyone here can help you too much with that, since most of use use locally installed a1111 or comfy

[sweet kraken]

can i install a1111 or comfy

[karmic crown]

if someone can help, he will jump in, do not worry

where does that link lead? here?

if it does, wait for someone who knows more about it

can you please post the links from there so I can check them?

[sweet kraken]

yes i think the lead brings me back here

[karmic crown]

send me a github link please

[sweet kraken]

sorry i dont know what that is

[karmic crown]

3rd link from a screenshot

the one saying New issue

[unique cape]

Somehow i get a problem on stable diffusion it says could not allocate ''nonsense amount of vram''
Using a RX 6700 XT and cannot generate images bigger than 512x512
What might be the problem and how may i fix it?

[karmic crown]

your command line params are probably wrong

show us your webui-user.bat

btw. do you run that or webui.bat?

[unique cape]

runnin this

[karmic crown]

please wait for someone who knows more about correct params for your card

but you can try those while we waiting for confirmation or better version

--medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae

[unique cape]

this instead of --medvram right

[karmic crown]

i found those in history as reply for someone else with same card

yes

[unique cape]

thx will try in a couple of mins

[karmic crown]

it has 12GB?

since that was for 12GB version

[ornate elk]

you still need --medvram

[unique cape]

yes it does

[karmic crown]

he asking where to put that

it has medvram in those params

you sent those

hahaha

[ornate elk]

😄

[karmic crown]

i am publishing 0.0.3 of my a1111-setup

🙂

[ornate elk]

ohh nice

[karmic crown]

and it is not bugfix release, just some improvements and minor tweaks 🙂

and yes i published 0.0.2 yesterday 🙈

[unique cape]

guess it worked

got the same error on 1024x1024

[ornate elk]

you cant do 1024x1024 with an AMD gpu on windows for now. Also for 1.5 models you shouldnt even try. (deformed output)
you can use 512x768 or 768x512 or 512x512 and then use hiresfix to upscale the image

[unique cape]

damn

thanks a lot

[karmic crown]

and... done

https://github.com/viking1304/a1111-setup/releases/tag/0.0.3

[tardy pivot]

anyone know how to fix this?
C:\StableDiffusion\ComfyUI\comfy\model_sampling.py:81: UserWarning: The operator 'aten::frac.out' is not currently supported on the DML backend and will fall back to run on the CPU. This may have performance implications. (Triggered internally at D:\a\_work\1\s\pytorch-directml-plugin\torch_directml\csrc\dml\dml_cpu_fallback.cpp:17.) (comfyUI)

[ornate elk]

np, if you want the exact settings for hires fix or upscaling in img2img let me know.

[desert garnet]

@elder turtle @echo hill

[ornate elk]

got it

sry that it took so long

report bot has a technical problem

[desert garnet]

ah

how do you report anyways

[ornate elk]

you can react to messages with this emoji to report ⚠️

[unique cape]

Would be awesome

[karmic crown]

@ornate elk do you know who made this or who can help with this?

[desert garnet]

ok

[ornate elk]

nope, cant help with google collab sry

[ornate elk]

For Upscaling 1.5 models use these settings:
512x768 or 768x512 resolution
30 steps,
cfg 7-12
Hires Fix:
Upscale by 2
Hires Steps: 10
Denois 0.5
Upscaler: Esrgan 4x+

[desert garnet]

how do i do --share in comfyui? and how would i load and switch between workflows on my computer from my shared remote?

[unique cape]

Thx^^

[karmic crown]

https://github.com/comfyanonymous/ComfyUI/issues/608

[karmic crown]

the only advice i can give you is to raise a ticket there

@ornate elk btw, with 1.7.0, torch 2.1.2 and some minor params tweaks i have a faster generation on my m1, 1.44s/it for 1.5 model 512x512

[ornate elk]

oh thats nice

[karmic crown]

sdxl is much faster

in relative terms 🤣

Steps: 10
Sampler: Euler a
CFG scale: 7
Size: 1024x1024
Batch count: 4

about 8 min 30 sec, instead of almost 12 min

[tough bison]

https://github.com/damo-vilab/AnyDoor?tab=readme-ov-file

Hey guys, I'm trying to run a repo like this one on macOS, but I'm running to an issue where it says certain packages aren't available on Mac. Is there anything I can do about that?

[karmic crown]

i am not sure about that one, but you can use https://github.com/AUTOMATIC1111/stable-diffusion-webui without any problem on mac, a1111 has many plugins and supports a lot of features, so maybe you can do something similar

[lucid stirrup]

i solved my problem by DDU-ing my drivers and deleting all nvidia software i had and installing new drivers

i guess i had some nvidia software that was messing up with cuda drivers

[karmic crown]

it requires xformers==0.0.18, which is for nvidia, that is your main problem on mac

[tough bison]

ahhh, appreciate it!
are there anyway around that?

can i host this easily somewhere remote? Just wnana play with the gradio demo.

[karmic crown]

as long as it has nvidia card

[echo ingot]

Hey guys RTX 3060 12gb VS GTX 1080ti Sli 2x11=22 gb. What is better for AI?

[karmic crown]

i am almost sure that the answer is 3060

[jaunty sphinx]

shit

i dont think it works

sunnva boooch

god fookin dammit trying to make intel xe graphics card work on every stable diffusion model

time to run open vino again

agony

the only best i got is open vino running stable diffusion 1.5 d;

[jaunty sphinx]

asjdlfa;kfjlskafjksaf;lasjfaslfasjflajf

so the ipex thing doesnt work only for amu or something

d;

now i must cry on my computer

at least cpu still works

pog champo

[unique cape]

Gives error at %95 when i do 512x768

strange

Can't understand what causes the issue

[karmic crown]

show the error (whole)

[unique cape]

Sure a sec

asap i get it again

[onyx solar]

I've asked SDXL to go ahead and give you a graphical breakdown based on your question:

[unique cape]

any idea @karmic crown ?

[karmic crown]

not ui, cmd window

[unique cape]

the strange part is

its not happening all the time

[jaunty sphinx]

holey crap

what is thato range stuff

[desert garnet]

is there a node i can add to the standard inpainting comfyui workflow that lets me draw/erase a mask to use so i dont have to actually edit the photo? and how do I make it so that only the mask gets inpainted because i think my entire image is being slightly affected?

[heady stone]

Im using the draw things app. the image on left is a model i downloaded called realism sdxl and its all glitched out. then i tried the base XL base 1.0 and the image looks like shit. using the same prompts they used. is there something i'm missing with this or is image generation a lot more complex than i thought

[onyx solar]

Left image needs the SDXL VAE.

[desert garnet]

hey i cant get this working, could you help me?

[karmic crown]

this one? https://civitai.com/models/152525

[unique cape]

@karmic crown sent the cmd from dms

[karmic crown]

pleas send it here, i do not check my dms here, nothing personal

[heady stone]

i see. it says "This version include the VAE to help newcomers with their setup." file is Realism Engine SDXL V20 VAE.safetensors.

[karmic crown]

which model you are using? i do not see that here

[karmic crown]

it is not vae

[unique cape]

ghostmix v12

[karmic crown]

plase show me what setting you used

[karmic crown]

can you try with a different upscaler?

[unique cape]

sure

[heady stone]

Im on mac so im using this app called Draw Things. i tried comfyui but had issues with it. basically i imported the Realism Engine SDXL V20 VAE.safetensors. There is an option for custom VAE

[karmic crown]

try this https://github.com/Coyote-A/ultimate-upscale-for-automatic1111

[unique cape]

How can i add this to my stable diffusion?

[karmic crown]

i can help you to install automatic1111 on mac if you want, than you can do whatever you want
what mac do you have?

[heady stone]

I would like to get comfyui working. right now im just trying to see if i can teven generate images in reasonable time. with comfyui, it took 20 minutes and produced a gray square, so clearly something isnt right. in this app, it at least does it within a minute but there's some other kind of issue. I have M1 macbook

[karmic crown]

and add this url https://github.com/Coyote-A/ultimate-upscale-for-automatic1111.git

[unique cape]

AAAH thanks

[heady stone]

How does automatic compare to these others?

[karmic crown]

automatic is maybe little slower than comfy

[heady stone]

I have to go now, but I'll continue tinkering around later and see

[karmic crown]

https://github.com/viking1304/a1111-setup/releases/tag/0.0.3

option 1, just run my script and wait for a1111 to show up

[unique cape]

guess i did it?

[karmic crown]

option 2, follow my step by step guide https://github.com/viking1304/a1111-setup/discussions/2

[unique cape]

Cannot see it tho

[karmic crown]

reload ui

[ornate elk]

Hey @unique cape if the hires fix failed at 512x768 you need the tiled diffusion extension. There enable tiled vae

Its normal that it can fail with 12gb vram for AMD if many tabs or programms are running in the background

Tiled VAE helps to not get an out of vram error

[unique cape]

how can i add that?

[ornate elk]

In the extensions tab

[unique cape]

did it

[ornate elk]

Click on available. Then on load from. Then search for Tiled

Then install

[karmic crown]

@ornate elk this is another solution for low vra https://github.com/Coyote-A/ultimate-upscale-for-automatic1111

[heady stone]

saving for later

[unique cape]

is the hires fix this one

[ornate elk]

I know, img2img has also a default script that can do the same

[ornate elk]

No

You need Tiled Diffusion

[unique cape]

i've got it and restarted the UI but cannot see it

lemme close it and open it

Still cannot see it

probably my dumbass doin something wrong sorry

[ornate elk]

In txt2img you should now have a tab called Tiled VAE and Tiled Diffusion.
Ignore the Tiled Diffusion tab.

Before testing anything you should know that every time you get an Out of Memory error. You have to restart the whole webui.

So restart

[unique cape]

AAAH

Yeah im stupid...

any spesific setting

[ornate elk]

Yes. Set the encoder to 1024 and the encoder to 128

[unique cape]

gotcha

[ornate elk]

Then you should be able to upscale 512x768 and even 540x960 for FullHD

[unique cape]

I wonder why i cannot see this on upscale tab tho

[ornate elk]

Its in img2img under scripts

[karmic crown]

i managed to install it and access webui on mac, but it is slow as hell, and i had to change half of the code, since cuda is hardcoded everywhere

[unique cape]

aah okay okay

[ornate elk]

Used for further upscaling

[unique cape]

thanks a lot really appreciate that

i probably sound so stupid

testing like this

[ornate elk]

Close the refiner tab.

[unique cape]

Does it affect something?

[leaden shadow]

Anyone notice Kohya ss behaving weird lately?

[ornate elk]

Yes it won't work

[unique cape]

aah

thx

[leaden shadow]

Like the outputted lora files being way too small

[unique cape]

new error

[ornate elk]

What's your model?

[unique cape]

Ghostmix v12

[ornate elk]

With baked VAE ?

[unique cape]

baked vae?

nope

[ornate elk]

Ok strange error then. Do you have --no-half-vae in your webui-user.bat ?

[karmic crown]

he is using old 1.2 version

[ornate elk]

Do you have an other model to try?

[unique cape]

same error

[ornate elk]

Can you paste the full cmd log?

Screenshot

[unique cape]

Sure a sec

[ornate elk]

Is the refiner tab collapsed?

[unique cape]

it is

Waiting for it to finish to see the error

this is the .bat btw

what does no half vae means?

[ornate elk]

Add --upcast-sampling there too

[unique cape]

Does no half vae means i cannot use models that says vae?

[leaden shadow]

Anyone know what might cause Kohya ss training gui to randomly start producing 9000 KB loras when it used to produce 147000 KB loras? Using the same settings.

[ornate elk]

Nope. No-half-vae will use the VAE files in fp32 mode and not in fp16
That helps for compatibility issues

Try add the --upcast-sampling to the webui-user.bat and try again

[unique cape]

this happened after adding --upcast-sampling

[ornate elk]

Remove there too

[unique cape]

...

....

Lets act like it didn't happened

[karmic crown]

this starts to look like radeon cofing for mac 🙂

[ornate elk]

Yea upcast-sampling is like no-half but less vram hungry

[unique cape]

Guess it worked

But the image quality dropped really bad

[ornate elk]

Because you used denois 0.7

Not 0.5 or lower

[unique cape]

and somehow everything is faster now

[ornate elk]

Thats good too

So yea if you use hires fix dont forget to enable tiled VAE with the specific settings

[karmic crown]

bye people, cu

[ornate elk]

Bye

[unique cape]

cya thx viking

what was the resolution for hd?

[ornate elk]

540x960

[unique cape]

thx again!

[ornate elk]

You can't go much larger than that with hires fix. It would run out of vram even with tiled vae

But for higher resolution you have work in img2img after generating the base image in txt2img

[unique cape]

But the image quality is really bad rn

wonder why

[ornate elk]

How bad?

[unique cape]

a sec

Hmm

guess its because of the model that i was trying

now that i changed

everything turned better

[ornate elk]

There are a few models that can have problems with upscaling. Sometimes switching the Sampler can help.
Most models work fine

[unique cape]

what samplers do you recommend?

[ornate elk]

DPM++ 2M SDE Karras, DPM++ SDE, Euler a

[unique cape]

would closing the tiled vae make it like how it used to do

but cause errors again

[ornate elk]

Closing hires fix will make it normal

Tiled VAE doesn't do anything if the resolution isn't set to high

[unique cape]

Gotcha thx

[ornate elk]

But yea most models work without problems

When using hires fix

You can play around with the denois value. Lowering under 0.5 will make it look closer to its default look

I'm off gn

[plucky marsh]

hey guys, how can i fix a face of my girl model if the image is already exported on my pc? I know that i can use inpaint or outpaint but that will not modify if to look like my girl model

[sturdy ginkgo]

use a low denoise value with the inpaint should not affect the input too much .. do multiple passes if needed

[hybrid siren]

Hey guys, I'm using ComfyUI and when the workflow ends with VHS Videocombine I get the error "Exception: An error occured in the ffmpeg subprocess:
Unrecognized option 'crf'" only when the output is a video format, not GIF. I recompiled ffmpeg with apt-get install ffmpeg per a reddit thread I found, but I didn't solve. Anyone experiencing something similar or have a fix? CRF is currently set to 20

[fickle rock]

Would anyone be willing to get into some DMs with me and help me figure out the best way to:
use the rig I have (RT 2060 6GB)
generate fantasy character portraits and landscapes
perhaps take a hand-drawn sketch and turn it into an image
possible models/LORAs/inversions, settings, etc.

Familiar, but still feel newb! (Feel free to DM!)

[arctic zinc]

Anyone use adeatlier for mac? It runs extremely slow and was wondering if it should be set differently for mac users. I know stable diffusion runs slower on macs.

[graceful loom]

How to remove this error @ornate elk

[mystic sequoia]

Hello, I am getting this error for image to image transformation within 512x512 image from stability.ai: "Engine ID stable-diffusion-v1-6 does not support image_to_image at this time"
Which image to image model is available for 512x512 image from stability.ai?

[ornate elk]

Try 1.5, 2.1 or sdxl

[ornate elk]

Some setting in decorum is wrong. But idk which one. Maybe copy some exact settings from a tutorial

[uncut jewel]

Hey, anyone knows what that node refeer to ? (it's pluged into a latent_image on a Ksampler) but don't know how to add this node :/

[onyx solar]

Download Comfy UI Manager and then use its Install Missing Nodes feature.

[torn kernel]

hey guys what is the best way to train a model for interior design, architecture?

[uncut jewel]

Yeah i tried that already, but there'snt any nodes showing

in the install missing node tab

already updated ComfyUI too

[onyx solar]

I took a quick look through and it's not coming up on any channel for any registered add-on through the manager. From some searching, it looks like it's being considered a Primitive node, but extracting and attempting to connect a primitive failed for me. So best guess is that it's deprecated at this point and you should just use a different latent image widget in its place since that's what it seems to be connected to.

[uncut jewel]

Alright, yeah i see the logic. Thanks for all your research to help me man! That was really helpful 🙂

[fair oxide]

Why do I get this error trying to create images any other size than 512x512?

onnxruntime.capi.onnxruntime_pybind11_state.RuntimeException: [ONNXRuntimeError] : 6 : RUNTIME_EXCEPTION : Non-zero status code returned while running Concat node. Name:'/up_blocks.3/Concat' Status Message: D:\a_work\1\s\onnxruntime\core\providers\dml\DmlExecutionProvider\src\MLOperatorAuthorImpl.cpp(2449)\onnxruntime_pybind11_state.pyd!00007FFC4C1BF31F: (caller: 00007FFC4C1C07C4) Exception(3) tid(5a34) 80070057 The parameter is incorrect.

[prisma dome]

I’ve been having an issue when the command window errors out with the issue “modeled failed to load”. Any ideas how to fix this?

[ornate elk]

What's your webui and GPU?

[fair oxide]

Using the directml version for AMD GPUs (https://github.com/lshqqytiger/stable-diffusion-webui-directml), rx 7800xt, should have specified that

[ornate elk]

And you probably followed the AMD guide for MS olive integration.
Thats why you got that error. Using olive In directml is not recommended.

You have two options.
Using the directml version without olive (slower but more compatible and without resolution restrictions)
Or use the Shark webui. Its faster and works like olive but better. But its till missing features.

[fair oxide]

How would I run the directml version without olive? Do I just use models without optimizing them through the olive tab? Should I change my commandline arguments?

set COMMANDLINE_ARGS=--opt-split-attention-v1 --onnx --backend directml

[ornate elk]

Yes you should run it without olive. The needed args are:
--medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling

Remove --onnx --Backend directml
Also delete the venv folder before restarting the webui-user.bat.

Then you can use any model from civitai but start with 1.5 models

[fair oxide]

That's significantly slower compared to the generation of 512x512 images with Olive, is there no way to further optimize this? If not I'd try Shark WebUI, does it support inpainting and img2img as well?

[ornate elk]

I think its capable of img2img, but not inpainting. Also your stuck on 768x768 Max resolution mostly.
If you want speed and Usability you would need to run auto1111 on Linux.
But I would say use directml until Rocm support comes to windows in maybe 1-2 months.

For me directml isn't really slow.
What are your txt2img settings?

[karmic crown]

I saw that you sent me some messages in DM, but if I start accepting DMs from all servers and channels, I will have no life anymore. nothing personal
what are your A1111 and torch versions?
what command parameters do you have in webui-user.sh
what is your mac?
and how long does it take to generate this:
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Size: 512x512 with any 1.5 model, without any hiresfix or refiner

[fair oxide]

Then Shark WebUI is not an option for me either. Should installing automatic1111 on Linux fix the problem? It'd allow no restrictions on resolution, inpainting and everything else?

[fair oxide]

Those are my settings, just the default ones for the purpose of testing the speed compared to using Olive - it's about 2.5 it/s now compared to 13 it/s with Olive

Olive did allow me to create 1024x1024 by the way, I figured it out, it only wouldn't allow resolutions that do not divide by 64

Or resolutions that are too large. Nothing above 1024x1024 worked for me.

Black screen

And then it brought a different error entirely on missing kernel

[ornate elk]

Auto1111 on linux would be the best of speed+usability.
But In speed nothing compares to shark.

Its up to you. If you want more information or help regarding Linux and stable diffusion, I also recommend joining the Shark discord.

For me the usability from directml is more important than the speed.
Upscaling is a huge quality improvement that only auto1111 does good

[fair oxide]

I must have inpainting

So I won't be able to use Shark WebUI as long as there's no inpainting available

[ornate elk]

For inpainting to work in directml you need to add --no-half to the webui-user.bat
But that will increase vram usage in txt2img.

[fair oxide]

Is there no way to get Olive and inpainting to work together?

[ornate elk]

I dont think so.
The creator of the directml fork just implemented the basics for olive.

@fair oxide looks like shark has inpainting capabilitys

[karmic crown]

@ornate elk i need your opinion on something

there is a guy with Intel Mac Pro 2019 (16-Core Intel Xeon W, 576 GB RAM), with an AMD Radeon Pro W5700X 16 GB MPX GPU and an AMD Radeon RX 6900 XT 16 GB GPU (yes, you read everything correctly) 🙂
since he has two GPUs, a1111 decides (somehow) to use RX 6900 XT, but he gets only about 14% of GPU usage
Mac has no cuda, so he can't use that option to select the device by ID or number, right?
he also can't use --no-half-vae, only --no-half is working (is there any other similar case with that GPU, on windows maybe)
what are the recommended params for that card on Windows and Linux?

ok, that's all for now 🙂

[fair oxide]

That's really unfortunate, Olive is extremely fast, I hope a method to utilize inpainting with Olive is discovered. As far as I'm aware, I get this error whenever I run resolutions that do not divide by 64 or try inpainting, so it encourages me to believe that inpainting doesn't work because the resolutions of images sent by inpainting are not divisible by 64 - everything else works perfectly fine:

Non-zero status code returned while running Concat node

[ornate elk]

@fair oxide shark has inpainting it seems. So check that out

[fair oxide]

Will do that straight away

[ornate elk]

Idk how advanced it is.

[fair oxide]

It should be optimized for AMD GPUs as well, yes?

[ornate elk]

Did you selected "only mask" for inpainting method ?

[fair oxide]

No tutorials or guides anywhere, just found the github page

[ornate elk]

Its made for AMD and Shark got acquired by AMD a few weeks ago

[fair oxide]

Tried whole picture, tried only masked, tried everything

[karmic crown]

wiat, let me put all together in one message

[ornate elk]

Okay then it doesn't seem to work with olive.
Join the shark discord for any questions.
It works on windows and Linux nativly

Place the .exe in a new folder before running it

[karmic crown]

#🤝｜tech-support message

[ornate elk]

It still should work with --device-id

A 6000series normaly dont need --no-half but idk how it is for these metal drivers

Would add the common amd args including --upcast-sampling

[karmic crown]

are you sure about device-id?
i can tell him to try, but i think it will not work, see the desciption

btw. default mac params are those (set by a1111 team) export COMMANDLINE_ARGS="--skip-torch-cuda-test --upcast-sampling --no-half-vae --use-cpu interrogate"

[ornate elk]

He should try. --device-id can switch between nvidia and Intel igpu so it should detect amd aswell.
But idk how it behaves on mac

[karmic crown]

what params are those for 6900 on linux and windows?

[ornate elk]

The one from my guide:
--medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling

[karmic crown]

its better for him to use AMD Radeon RX 6900 XT, right?

[ornate elk]

Yes its the better supported card

[karmic crown]

i am writing this from my iMac, which has 580X, and it works with exactly the same params as my M1, and that is the thing that bothers me the most, that guy and the other with eGPU are the only exceptions with macs that have problems that i have never seen 🤷‍♂️

maybe difference starts with cards that have 16GB, i do not have any to test 🤷‍♂️

python + brave = almost 100% on my iMac

i am confused

btw. what would you use with 580x 8GB on linux? so i can try on my mac

i dont understan how he gets only 14%

[ornate elk]

I would start with --medvram and if oom errors appear switch to --lowvram for 8gb

[karmic crown]

this works, but i am not sure if i can make it faster
export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half-vae --medvram-sdxl --use-cpu interrogate"

its the same thing i am using on m1

[ornate elk]

14% is very strange

Normaly it should go to 100%

[karmic crown]

thats how it works for me

i guess i need to disable gpu for brave, to leave more for python

i disabled gpu acceleration for brave and gpu us on 95% as expected, how he gets 14%, i guess only god knows

Steps: 10, Sampler: Euler a, CFG scale: 7, Seed: 4080909017, Size: 1024x1024, Model hash: a88bfb6ffa, Model: hephaistosNextgenxlLCM_v10LCMBF16, Version: v1.7.0

Time taken: 2 min. 50.7 sec.

this is not bad for 580X with 8GB

[spare osprey]

yo g

how can I extend a two second clip on stable video?

[dark owl]

I only get 4% lmao during 150 image batches at 12min each 512x968, 20 steps, openpose, hed, depth, cfg 7, Euler a.

[sturdy ginkgo]

So i am using ComfyUI and for some reason my workflow doesn't load XL models ... someone knows what could be wrong?

[vocal burrow]

I havent used comfy in months but wasn t there specific nodes to load sdxl models ?

you seem to be using the "default" one, maybe they got updated and now support sdxl too.

[sturdy ginkgo]

ok.. will take a look into it

it always fails a the first ksampler node .. that is why i didn't think it could be the loader

[vocal burrow]

hmmmm looking at the error message, the error seems to come from an Impact node, not a "load checkpoint" node

try to remove the controlnet stuff and see if it works

[vagrant quarry]

Hey, iam running SD on my pc (i7-13700k, RX 7900XT, 32GB DDR5).
I habe followed the AMD tutorial to use AMD GPUSs, that works fine so far (the gpu is used to generate) but there are no olive riders in the WebUI to convert the models.
How can i activate them ?

I only have 2,Xnits/s

[vocal burrow]

"the AMD tutorial" which one ?

[vagrant quarry]

One sec

https://community.amd.com/t5/ai/updated-how-to-running-optimized-automatic1111-stable-diffusion/ba-p/630252

https://youtu.be/kw0WT5sDBIY?si=-NkR3UzZ_RwTqdGn

[vocal burrow]

check for any errors in the console log (from startup to generation)

[vagrant quarry]

One sec

[vocal burrow]

(gonna have to afk for a while, gota commute back home from work)

[vagrant quarry]

this is the log:

L:\stable diffusion\stable-diffusion-webui-directml>git pull
Already up to date.
venv "L:\stable diffusion\stable-diffusion-webui-directml\venv\Scripts\Python.exe"
fatal: No names found, cannot describe anything.
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: 1.6.1
Commit hash: 03eec1791be011e087985ae93c1f66315d5a250e
Launching Web UI with arguments: --medvram --backend directml --no-half --precision full --opt-sub-quad-attention --opt-split-attention-v1 --disable-nan-check --theme dark --autolaunch
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
Loading weights [e5f3cbc5f7] from L:\stable diffusion\stable-diffusion-webui-directml\models\Stable-diffusion\realisticVisionV60B1_v60B1VAE.safetensors
Running on local URL: http://127.0.0.1:7860

To create a public link, set share=True in launch().
Creating model from config: L:\stable diffusion\stable-diffusion-webui-directml\configs\v1-inference.yaml
Startup time: 13.5s (prepare environment: 0.6s, import torch: 5.4s, import gradio: 1.7s, setup paths: 2.0s, initialize shared: 2.2s, other imports: 0.4s, load scripts: 0.6s, create ui: 0.4s, gradio launch: 0.2s).
Applying attention optimization: sub-quadratic... done.
Model loaded in 3.0s (load weights from disk: 0.6s, create model: 0.4s, apply weights to model: 1.7s, calculate empty prompt: 0.3s).

this is my UI

and this is how it should look like:

[sturdy ginkgo]

works..thank you

[vagrant quarry]

i think i worked it out

there was a common error i fixed

[kindred tinsel]

Aryetis!! Mind helping me out too with onxx?

[primal apex]

yo im trying to install stable diffusion and i keep getting this error: https://gyazo.com/41040d3a675f0822f1cc7e1d20924f7c

i looked up some solutions one of them were clear "repositories" folder and it was already empty, i also tried to disable firewall

[sturdy ginkgo]

try installing to a folder not linked to onedrive (cloudstorage) .. suggest a folder directly in root like C:\SDWEBUI

desktop/documents etc all sync to onedrive when using a microsoft account to login to windows unless you turn onedrive off

[vagrant quarry]

its working

[vocal burrow]

don't put stable diffusion in any kind of "sync"/cloud folder (eg : OneDrive)

welp, Kuromi said it