🤝｜tech-support

[sacred sierra]

where is config?

[ornate elk]

Version? Following my guide will install the latest directml version. Then its up to you what models you download 🙂

[sacred sierra]

Hello?! Can you please respond?! WHERE is the config file?!?!

[lean laurel]

oh yeah sorry sorry, i just didnt know what is the link for actual the stable diffusion program

[buoyant pawn]

same, i might need a full reinstall,

[ornate elk]

Hey, no problem at all. The actual programm will be downloaded with the git clone command

Running the webui-user.bat will install it

[vocal burrow]

Tolerant but not patient... I'm installing the whole foocus as we speak. I am not your robot, thank you.
Regardless I don't see why you're looking for it in appdata

[lean laurel]

ohh i see, thanks a lot!! i didnt know that. really thanks for helping me ❤️

[sacred sierra]

I am looking EVERYWHERE because you did not TELL where it is?!!

[vocal burrow]

exactly where I told you it was

as did the documentation

[sacred sierra]

I have done a Search for CONFIG.TXT and returned NOTHING. I know you are not my robot but you told me to check a file that does not seem to exist, so you left me hanging. Sorry if my sitting here watching you respond to others and ignore me did not go over well with you. I do appreciate the help but just sitting here staring and getting no respone is not exactly helpful, ya know?

[ornate elk]

No problem, BTW what's your GPU?

[sacred sierra]

No You did NOT tell me WHERE it was.. NOT at all.

That is not helpful if I dont know wtf FOOOCUS is

[vocal burrow]

Wherever you installed it, that I cannot guess.

[sacred sierra]

"Somewhere on my PC" is a file.. Oh fucking helpful

I installed it using the stable package that this channel is for. I would think that would be a pretty standard install location.

[ornate elk]

Dude calm down, if you dont know what Fooocus is, then you shouldnt have tried to install it in the first place

[sacred sierra]

I know what it is.

I am saying there is an error and they told me to look in a file but wont tell me wtf the file is.

Telling me I SHOULD KNOW?!

[lean laurel]

let me check it, i m bad with pc may be AMD Radeon? or Ryzen?

[ornate elk]

He told you its the config.txt inside the Fooocus folder, wherever YOU installed it

[vocal burrow]

"the stable package" ?

[sacred sierra]

THERE IS NO "FOOOCUS" FOLDER

And there are folders and more folders and more linked folders.. and if search won't find a Config.txt .. ?!

[vocal burrow]

you can check with dxdiag or by looking at the GPU tab in task manager

[ornate elk]

Radeon is GPU Ryzen is cpu

[sacred sierra]

SO yea I am frustrated.

[lean laurel]

may it be, AMD Radeon (TM) vega 3 graphics?

[sacred sierra]

nevermind .. this is not worth it.

[vocal burrow]

Again, what's that "stable package" you're talking about ?

[ornate elk]

Check in the task manager under performance, if you have multiple gpus

[vocal burrow]

that would be it.

Also yes, check if you have multiple gpu like cs1o said

[lean laurel]

oh i didnt know that i can see that in task manager

is my gpu low?

[vocal burrow]

it should be somewhere there
(there can be GPU 0 and GPU 1 and so on if you're running on some exotic hardware)

[ornate elk]

Its an onboard GPU chip with low vram amount.
It can still work for SD but not the best performance and some limitations

[lean laurel]

vram is ram?? my pc have 16 ram

[vocal burrow]

no, vram is the gpu's ram.

it's much much faster than your "system ram" which sit outside of your gpu

[ornate elk]

Okay, first reinstall using my guide, then try it again

[lean laurel]

thanks a lot u r teaching me a lot

i have this

[vocal burrow]

click on GPU 0 and screenshot the whole window

[vestal viper]

currently installing, will inform you again after it's done

[lean laurel]

give me a sec

[ornate elk]

It has 2gb vram

[lean laurel]

thats bad?

[vocal burrow]

So that's the amount of ""vram"" you have. Putting into quote here because to my knowledge "Vega 3" gpus are (mostly) chipset / iGPU. Meaning they're not "proper gpu" but gpu integrated within your cpu. Making them way less powerful and they don't actually have proper vram but use part of your system ram.

[ornate elk]

@lean laurel Its not much but you can try following my install guide, as directml can use shared memory

[lean laurel]

what is shared memory??

[vocal burrow]

so tldr, it will work but it will be :
1/ slow because it's an igpu
2/ kinda limited because you're using AMD hardware on windows which is not fully supported

[vocal burrow]

"shared memory" is the part of "system ram" dedicated to your chipset/iGPU

regardless one solution for you is to do as cs1o said and install the "directML fork" of auto1111

[lean laurel]

thats why it say 0, so i can share memory so i have more vram

[vocal burrow]

(there are other solutions that can be discussed later on if needed)

[lean laurel]

oh the one that sent me in the guide?

[vocal burrow]

Sharing "more memory" won't make said memory faster.

[lean laurel]

can i install more vram??

[vocal burrow]

and it cannot necessarily be done, depends of the hardware, the bios. I wouldn't recommend tweaking that if you didn't know what an igpu was 5 minutes ago.

[lean laurel]

sorry i m pretty bad with computers

[vocal burrow]

no, that's hard wired in your laptop.

[lean laurel]

ohh ok ok, thanks @vocal burrow

ohhh ok ok

so anyway, should i follow the tutorial that @ornate elk sent?

[ornate elk]

Yes 👍

[lean laurel]

ohh thanks a lot ❤️ !! so the only diferent of my low vram is that will take more time making the pics?

[vocal burrow]

You will have some incompatibility and little quirks because you're using AMD hardware. But overall yes, hardware does not affect the quality of the output.

[lean laurel]

thanks a lo @vocal burrow and @ornate elk for your help, u two are really kind, i m glad i talked here

[ornate elk]

No problem, feel free to ask here again if you have any upcoming questions

[buoyant pawn]

could that be the issue ?

[ornate elk]

Do you get any error now?

[buoyant pawn]

it's been like this for 45 min

[ornate elk]

And you reinstalled the whole thing with my guide ?

[buoyant pawn]

not yet i had to take a break

[ornate elk]

Ah okay

[buoyant pawn]

was hoping for a breakthrough lol

[ornate elk]

A reinstall is faster than 45 minutes waiting

😄

[buoyant pawn]

i'll try again with an other version of python

[vocal burrow]

saving you some time, it won't fix it.

[buoyant pawn]

i might have something to do with how cpu and gpu are assigned beside that i have no clue followed everything thing as it should

[vocal burrow]

I don't know what you currently have installed, but if it's not the directml version, it's no use trying to make it work

[ornate elk]

He has nvidia 1660 ti

[gusty solstice]

E:\StabilityMatrix-win-x64\Data\Packages\Stable Diffusion WebUI>git status
HEAD detached at v1.7.0
Untracked files:
(use "git add <file>..." to include in what will be committed)
modules/textual_inversion/preprocess.py

[vocal burrow]

oh right, I'm mixing up cases, ignore what I said @buoyant pawn

[buoyant pawn]

u scared me lol

can i copy the vnv library ? to avoid redownloading 6gb

[vocal burrow]

vnv ? do you mean venv ?

if so, better not.

and you won't redownload everything as some files should still be cached by pip.

[undone kindle]

how i can download it ? or use it

[buoyant pawn]

i got the Python 3.10.11 installer 64bit

i'll reboot first see you later

[vestal viper]

@ornate elk it's done installing and showing the usual UI on browser. how do i purge pip cache? cmd from webui folder? do i close webui cmd first

[buoyant pawn]

install was succesfull but i forgot to launch it as admin, i hope it won't mess up anything...

[ornate elk]

Open up a cmd and type
Pip cache purge

[vestal viper]

done. what's the next thing to do?

[ornate elk]

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

If there is no venv folder, then also launch it

[vestal viper]

it says creating venv in directory when i launch it

[vestal viper]

IT FINALLY WORKS FLAWLESSLY! THANK YOU SO MUCH!

man, had been trying to set SD up for a week now, pretty happy that it finally runs

[ornate elk]

Nice no problem 🙂

[buoyant pawn]

Congratz i hope i will be able to say the same thing by the end of the night

[vital token]

how do you guys stop nudity and pedo images from being generated... i have tried blocking all bad words i can find... and also have NSFW FILTER ON...

[buoyant pawn]

New issue my oc freese while the cmd is downloading torch, am not sure its compatible with 3.10.11

[ornate elk]

How much RAM do you have?

[buoyant pawn]

16

Froze twice while downloading torch

[ornate elk]

Let it load it can take a while

Make sure to have 15gb free space on C

[ornate elk]

Its doing stuff in the background

Not frozen

[buoyant pawn]

The mouse not moving

[ornate elk]

Oh

Let it run

Should be not take to long to install

Give it 15 minutes and check if its still stuck

If yes, then you need to increase your windows pagefile

[buoyant pawn]

The SD journey ain't an easy one...

[vocal burrow]

how good/bad is your internet ?

[buoyant pawn]

1000ko/sec

add python to environment variable should be check right

[vocal burrow]

yes

[vocal burrow]

how much ram do you have ?

[buoyant pawn]

16

[vocal burrow]

ok should be enough as long as you're not doing crazy things on the side.

and how much free space on c: ?

[buoyant pawn]

87 go

ssd

[vocal burrow]

plenty enough too

[buoyant pawn]

am starting from scratch with 3.10.6

5th time the charm XD

Hmmm he didn't redownload torch thats cool seems like he is using a cache somewere

1hour saved 😂

There is a new version of pip i should update that right ?

Ouch

[barren rover]

is it maybe bc Im using a refiner?

[ornate elk]

Yes that could be the cause

Restart and try without it

[ornate elk]

Looks like a connection issue

[buoyant pawn]

deleted the repositiries since it failed few time and restarting

seems like it's progressing as intended now

really sorry that it took so long guys...

[ornate elk]

No problem, good that it works now ^^

[buoyant pawn]

the ui is up is that the right config to use ?

also should i download any extension befor starting,

[kindred tinsel]

Hello, does anyone know where to change the default colors for nodes in comfy, like where can I add my own colors to it?

[ornate elk]

Thats the webui, Dreamshaper is the model. Prompt for something to test

[buoyant pawn]

i need to use one of dreamshaper prompts right

[ornate elk]

Nope, you can also just type Cat

Or anything you want

[ornate elk]

Boorutag autocomplete extension can help with prompting or word finding

[buoyant pawn]

same as befor stuck in loading T_T

[ornate elk]

What does the cmd shows?

[ornate elk]

Hmm it should start

[buoyant pawn]

it said something about updating pimp at some point

while isntaling repositories

[ornate elk]

Updating pip is not needed

[buoyant pawn]

i have no clue at this point

am using this : --xformers --autolaunch --medvram --no-half-vae --no-half

[ornate elk]

You can remove --no-half-vae

But that won't fix it

What does your task manager shows?

Anything at Max usage ?

[ornate elk]

Okay and the cmd, still the same?

[buoyant pawn]

yeah 0%

do you think it might have something to do with cpu/gpu code for usage ?

[ornate elk]

What code?

[ornate elk]

Can you restart the webui-user.bat and test again?

Also make sure the webui is whitelisted in any browser adblocker

[buoyant pawn]

ok

[kindred tinsel]

Hello, does anyone know where to change the default colors for nodes in comfy, like where can I add my own colors to it?

[buoyant pawn]

i had this red error apear 5 times and at the botom it's loading at 0%

while the gpu vram is in full use

[ornate elk]

Is your graphics driver updated ?

Also what browser do you use?

[buoyant pawn]

i haven't done it lately but i mean i can run most games, am using opera gx

[ornate elk]

Can you try on edge ?

[buoyant pawn]

it went from 0% to 10%

it's loading but very slowly

this time

[ornate elk]

Make sure to not have any games or other graphical heavy programms open

[ornate elk]

Do you use opera VPN?

If so disable that

[buoyant pawn]

nop

awefully long for just cat lol

[ornate elk]

Would still suggest to also try it in edge

Yea thats to slow

[buoyant pawn]

and will probably fail at the end with some black screen

shouldn't it be using my gpu processor ? or just the vram

[ornate elk]

It should use your vram of the gpu

You can close the webui, cmd and opera and then try again in edge

Also let me know if you get the Asgi error again

[buoyant pawn]

ok

it's working though even though slowly lol still on opera

first picture generated locally and it only took 20 min and a full day of hassle XD

[dusty meteor]

so on Mac Intel, when I run stable-diffusion-webui (A1111 normal install), I always see the "Warning: caught exception 'Torch not compiled with CUDA enabled', memory monitor disabled" notification. Is there any way to force PyTorch to monitor memory, even with CUDA disabled? I believe the memory monitor being disabled is consistently leading to MPS Backend Out of Memory issues after 5-10 generations - see https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/5461#discussioncomment-7989294 for context

[kindred tinsel]

So like anyone know where to set custom colors in comfy UI?

:(

[wicked flint]

how to uninstall certain extensions?
tried to install roop but its not working somehow, might because of im using amd gpu :/

[kindred tinsel]

No irs not about amd. Just go online and look for Microsoft visual download. It’s a Microsoft ide. Or look online for the redistributable they are asking for

In the error it says it needs a visual c++ compiler since ROOP uses some c++ thinfies to work.

If you install Microsoft Visual Studio you will get those if you select the component that says development in c++

[barren rover]

Im getting this error now: RuntimeError: [enforce fail at ..\c10\core\impl\alloc_cpu.cpp:72] data. DefaultCPUAllocator: not enough memory: you tried to allocate 9437184 bytes.

[barren rover]

and this one too: NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

[kindred tinsel]

Hey ppl one question, I noticed in my comfy ui that it says

VAE TYPE: torch.float32

If I’m on amd shouldn’t I use float16 ? Or is that alright

[proper oracle]

When attempting to load embeddings such as this https://civitai.com/images/2574679 and loading onto a pipeline, how do we configure it so that we can use all the triggers (tokens?) that are in the embedding? I am really struggling to reproduce the results on SDXL.

If I understand the process for loading the embedding you must define a token, but when running the prompt it does not appear to take affect. Any pointers here would be greatly appreciated.

Oh my. I just realized that it's not multiple tokens in one embedding but each embedding is a separate download. 🤦‍♂️

[simple tartan]

ok, i'm back, good morning, time to get the fill and outpaint working

[kindred tinsel]

Hey ppl one question, I noticed in my comfy ui that it says

VAE TYPE: torch.float32

If I’m on amd shouldn’t I use float16 ? Or is that alright

repost for information

[azure sable]

anyone know what would cause photos to come out like this when using Regional Prompter?? Never had this issue until the last two days.

pics generate normally when regional prompter is turned off.

[open raven]

hey guys what do i do with a .safetensors file

[simple tartan]

depends on what it is

thats basically the file extension for all SD stuff, loras, vae's, models, checkpoints, etc

[ornate elk]

What are your txt2img settings?

[simple tartan]

welcome back brother 😄

is there a way to "hide" lora's in the UI? or better yet, a way to tag them like civitai does? i know that SD and SDXL auto-update based on the model selected, but if I'm doing robots I dont need to see all the anime ones, etc

[ornate elk]

You can make folders inside the lora folder to sort them. Then you can click only on the category you want to see them

I made folders for sdxl, anime, clothing, styles and characters for example

[buoyant pawn]

hi there ¨¨

[simple tartan]

all of my lora's are already in folders, and it seems you cant nest them

[ornate elk]

Have you clicked the show directorys button in the webui?

[simple tartan]

yes, but i've been storing the models into their own folders, so for example, i have 20 lora's for robots, and thats in 20 folders, so its either Show All or Show One

[ornate elk]

Yea there is now way to hide single loras, it only shows whole folders

[simple tartan]

rats

I gotta scroll through a LOT of idols and maids to get to the robots, lol

should be easy to on/off the folders i would think....maybe i should have GPT write me an extension, lol

[ornate elk]

Wait, why are they in the same folder then?

[buoyant pawn]

i don't wanna interupt CS1O, but I looked online a bit and i found out that i don't have a src folder, also pip should have alot of requierments in the environment file and i have none

[simple tartan]

each one is in its own folder, Show ALL is the only option I can really use unless i want to flatten my folder structure

[ornate elk]

This is for wsl, normaly you see all in the requirements.txt

[ornate elk]

Why single loras in their own folder?

[simple tartan]

it was cleaner i thought, set the subfolder to model name in settings

[ornate elk]

Hmm in don't see an advantage there over just sorting multiple in one, like styles, characters, anime etc

[buoyant pawn]

seen some ppl using extensions to manage their models folder it can get crowded in there lol

[ornate elk]

How does that look like now?

[ornate elk]

That looks normal, have you tried the Edge tests? Also did you get the Asgi error again?

[buoyant pawn]

yeah same results, asgi didn't come up again though

[ornate elk]

Hmm

[buoyant pawn]

i think it's using my cpu instead of my gpu or i am missing some critical components

[ornate elk]

Try updating your GPU driver and then delete the venv folder

[buoyant pawn]

i did update the gpu driver

[simple tartan]

a ton of folders each with 3-4 files.

[buoyant pawn]

i saw someone installing all of those

[ornate elk]

Afk, driving

[buoyant pawn]

gotta go to work anyway, i'll try this evening again may be

have a safe drive

[ornate elk]

Okay, yea in the evening we can do further troubleshooting

[upbeat atlas]

Hi, I have a probelm. I have stable diffusion directml on windows. It was working all fine, but now it keeps giving me the error "torch is not able to use gpu" when I launch. It told me to add a command and I did but it was SO much slower to run. I reinstalled and it gave the same exact error. How can I fix it?

[woven smelt]

Im suing XYZ plot to try different prompts for adetailer. My adetailer prompt has commas in it so I cant use commas to seperate the prompts out. What syntax should I use to seperate the prompts?

[ornate elk]

You need to add --use-directml to your webui-user.bat

Remove the --skip-torch-cuda-test

For a reinstall with the best performance commands checkout my install guide: #🤝｜tech-support message

[upbeat atlas]

2 hours, 4 youtube videos, 7 reddit threads and a bunch of open tabs on chrome to find a solution and you just got it fixed with a single line
Y'all tech support people are built different istg

[ornate elk]

You should have checked the issue tab on the github, there you mostly find any info to new errors 🙂

[upbeat atlas]

Good to know, thanks again ❤️

[ornate elk]

No problem 🙂 dont forget the other commandline args you need

[granite crown]

I'm using Stable diffusion on windows with a 6900XT. After updating for the first time in a long time it's now taking 10-15X longer with the same prompt and model

From what i've read this seems to just be an issue with the newer versions? I'm wondering if there's something I can try to help it or if I should just go back

It also does not seem to be a big fan of using your args

Fixed it not liking your args, it's back to its previous speeds

Thanks for the help

They're certainly speedy

Though it does seem to run out of memory, slightly odd since it doesn't seem like I should need --lowvram with my 16gbs of vram

[ornate elk]

So the args work now?

You dont need lowvram but you should know that your limited on the resolution with directml.

[granite crown]

The args do work, I just had to delete venv and let it fix itself

I'm using 448x448 and a 2x upscale hi-res fix

Running out of memory

[ornate elk]

You should be able to do a:
512x768,
30 steps,
hires fix: upscale by 2
Denois: 0.5
Hires steps: 10
Upscaler: Resrgan4x+

Without getting out of memory.
Best is to disable all heavy background Programms and closing all other browser tabs. Also make sure you dont run Wallpaper Engine in the background

You also should install the Tiled Diffusion Extension.
The Tiled VAE function of it helps against the out of memory error.

[granite crown]

I cannot currently use those settings without getting the out of memory error, I will try with the extension

[ornate elk]

Okay, make sure you dont use sdxl but use 1.5 model to test

If you installed the extension. Restart the webui. Then only enable tiled VAE and set encoder to 1024 and Decoder to 128

Then try my settings again

[granite crown]

Still getting the error

What sampling method are you using with your settings?

[ornate elk]

Mostly using DPM++ 2m SDE karras or Euler a

But its strange that you get that error still.

Can you download the Dreamshaper v8 model and try again?

[uncut jewel]

I seem to have the same issue when i updated my ComfyUI and now, i can't use it 😢

[ornate elk]

@granite crown Also you need to restart SD every time you get an out of vram error so the vram gets cleared. Before trying stuff again.

[granite crown]

I'm not using ComfiUI and this is a fresh install of stable diffusion

[granite crown]

Ah.

I was under the impression it was automatically purged, that is probably it

[uncut jewel]

Got this screen now since i updated ComfyUI, any idea on how to solve that ?

[ornate elk]

What's your GPU?

[uncut jewel]

3090

[granite crown]

I restarted stable diffusion and I am getting the same error, would you happen to have a link to the dreamshaper V8?

[ornate elk]

Update your GPU driver.
If its the latest version. Then run the comfyui update+torch

[ornate elk]

Sure:
https://civitai.com/models/4384/dreamshaper

[granite crown]

DreamShaper was also unable to provide me my turtle. I got another out of memory error

[ornate elk]

Restart the webui-user.bat and make a screenshot of it for me pls

[granite crown]

Of that bat?

Or the error?

[ornate elk]

Of the whole cmd before you press generate

[ornate elk]

Looks okay

[granite crown]

Well, I have no idea what changed, but I have been able to run it without getting out of memory

Performance is somewhat worse than when I started, but I'll take it being a bit slower if it actually works

[uncut jewel]

Worked with the update+torch, thanks! 🙏

[granite crown]

It's roughly the performance I had before I updated stable diffusion, so it's good enough

Though that does make me wonder what performance I could get with your args on the older version

[ornate elk]

You should get a much better performance with that GPU and my args

Is your driver updated?

[granite crown]

It is up to date

My performance dropped by about 10-15x after I updated, and you got that back to normal

Though my previous args probably would have made you cry

[ornate elk]

Its very strange, because on a 6700xt it works without getting out of memory

[granite crown]

Quite odd indeed. But I have also never been able to get the performance of other cards, it's a bit odd

[ornate elk]

Yea someone with a 6900xt had the same problem a while ago

[granite crown]

That may have been me, I left and rejoined

[ornate elk]

Nope, it was on the Shark discord

[granite crown]

I was also on the shark discord lol

[ornate elk]

Ah okay xD

[granite crown]

Seems to be my curse, hopefully one day I can figure it out!

[simple tartan]

ok @ornate elk i moved everything into a root directory for now, trying to pen out the best path forward now, the obvious folder is the "explicit" one, rofl

[granite crown]

Through all this I went from 9-15s/IT to around 6s/IT, so I'm happy

[ornate elk]

Wait can you try a 512x512, 30 steps?

Without hires fix

[granite crown]

No hi-res?

Gotcha

[ornate elk]

You should get it/s now

[granite crown]

Dreamshaper?

With tiled VAE?

[ornate elk]

It doesn't hurt to have it active

As it only gets used when the resolution is high enough

[granite crown]

ah, alright

Looks like i'm getting just below 3it/s

[ornate elk]

Okay thats normal for directml

[granite crown]

🙏

[ornate elk]

Do you have any extension installed ?

[granite crown]

Just the one you told me to

[ornate elk]

Or do you use any long prompt or lora for testing?

[granite crown]

No lora

What I'm comparing are two decently long prompts though

Presumably that's the reason for my poor performance?

[ornate elk]

Can you try again with just
Cat as positive and
Blurry as negative
Then activate hires fix and tiled vae.
Hires steps 10

512x512 upscaled by 2

Then Open your task manager and check the vram usage of the gpu

[granite crown]

It's using all 16 gigs

[ornate elk]

You can also send a screenshot of the webui with the error, if it won't work

[granite crown]

15.8/16

Prompt generated fine

[ornate elk]

With hires fix?

[granite crown]

Mhm

[ornate elk]

Okay good

Then you can try 512x768

[granite crown]

Not too shabby

[ornate elk]

Yea hires fix takes always long

But its okay

10 hires steps is enough, dont go over 15

30 normal steps is mostly enough too

[granite crown]

Alright, I generally run lower so that works

[ornate elk]

Also dont go over 225 tokens

More tokens = more vram needed

[granite crown]

Tokens?

[ornate elk]

The counter on the top right of the prompt box

[granite crown]

Ohh

No

Good heavens

That would be a long prompt

It generated fine at 512x768

[ornate elk]

With hires fix enabled ?

[granite crown]

Yep

[ornate elk]

So it should have a resolution of 1024x1536

[granite crown]

That is correct

[ornate elk]

Perfect 🙂

[granite crown]

Wooo

[ornate elk]

If you want a FullHD image then use
960x540 with upscale by 2

[granite crown]

I can run that?

[ornate elk]

But thats the limit for AMD cards with hires fix

[ornate elk]

Yes with tiled VAE enabled

[granite crown]

Neat

[ornate elk]

You can later upscale images further in img2img with the SD upscale script to get 4k resolution

[granite crown]

Solid enough for AMD, bit sad I don't have NVIDIA, but they also wouldn't fit in my little toaster so it's fine

Gotcha

Thank you for all your help

I truly appreciate it

[ornate elk]

No problem 🙂
If amd releases Rocm support for Windows this or next month hopefully, then it will be a lot easier and faster

If you need specific settings for img2img upscaling let me know

[granite crown]

Every night I sleep dreaming about that, but it never seems to happen

I definitely will

Oh, is it worth messing around with the optimizations under stable diffusion in the settings?

[ornate elk]

Nope, these are already set trough my args in the webui-user.bat

[granite crown]

Ohhh, alright

[ornate elk]

Also xformers won't work for AMD

[simple tartan]

is it too much to ask for stable extensions? why does the civitai logo randomly disappear from my loras and i have to do a full reload of the LLM?

[granite crown]

Much to my monumental disappointment

[ornate elk]

We will get Flash Attention v2 when rocm and torch are ready

[ornate elk]

LLM? Did you installed the civitai helper extension?

[granite crown]

I cannot wait for that day. I'm sure i'll also be back in here asking for help lol. But i've taken enough of your time, thank you for your help. It's past 2AM so I shall be taking my leave, have a wonderful rest of your day!

[simple tartan]

LLM, SD, whatever you want to call the whole engine

i only have teh civitai browser+ extension

[ornate elk]

No problem, feel free to ask for any AMD stuff with SD. I've tested a lot on different cards. But for now you have the right settings to generate good images without problems 🙂

[granite crown]

👍

[ornate elk]

Hmm okay, I dont use it but I can imagine that it maybe timeout sometimes as Civitai in general is not a stable site

[simple tartan]

maybe that's it, because the icons just randomly appeared again mid generation

[simple tartan]

should i add xformers on amd build? or just keep ignoring that error on startup?

[ornate elk]

Xformers won't work for amd

Only for nvidia at the moment

[simple tartan]

ok, so ignore it

[fair oxide]

Hey please someone look at my problem in #🏞｜general-with-images

[idle stump]

Is it normal to get an uninterruptible power supply for my RTX3090 rig? My computer restarts once in awhile when I use SD or play games with it

[rotund lava]

if it restarts when the gpu is under heavy load like gaming or sd then its a PSU issue prob not enough watts,the UPS only helps when theres a lot of power outages in your area

[simple tartan]

yep

shoudl have a UPS if you run windows as main too, its REALLY hates hard power cycling

[idle stump]

Sorry what really hates hard power cycling?

[idle stump]

Another problem is that my monitor sometimes black out when someone switches appliances on, like the aircon. But my kid's weaker PC doesn't suffer from this problem

[rotund lava]

try to connect the computer and the monitor to another outlet,one thats free and doesnt have anything else connected

[idle stump]

They're from different outlets but maybe they are connected behind the wall?

[rotund lava]

also does your lights flicker/go really dim when the computer with the 3090 is under heavy load?

[simple tartan]

windows, as in most likely your OS

[idle stump]

I mean what or who doesn't enjoy power being cut off?

[rotund lava]

id say your house electrical system cant handle all the devices you have connected,maybe its a problem with the thickness of the cables they used or if you live in an apartment your landlord did it on purpose for w/e reason

[simple tartan]

linux can handle it surprisingly well, thats why most SBC's use some form or distro of nix

[idle stump]

So... would a UPS help in my situation?

3 situations that happens: when someone switches an electrical appliance, when I use stable diffusion after having not used it for a while, when I play games after not having played for a few hours

[simple tartan]

from what you've said, only partially, it sounds like you have two issues, 1)your PSU isnt powerful enough for your components, and 2) your home has bad wiring or low amp circuits

[idle stump]

My PSU is 700w i think...

[simple tartan]

a UPS will assist with issue 2, but it wont fix it, only an electrician can do that

[idle stump]

I see

[simple tartan]

and what CPU, mem, storage, gpu, etc?

[idle stump]

i7-10700, 32GB, 300GB, RTX 3090

[simple tartan]

300gb being a what? spiiny death wheel, sdd, nvme, usb drives?

[idle stump]

SSD

We're two captains speaking to each other

[simple tartan]

i mean personally, on a "recent" pc, i wouldnt even look at anything under 750W, I like expansion room. some team green will need to chime in, but i think that the 3090 max TDP is only like 400W, so that would still leave 300 for a .....130W cpu, i think, trying to thing off of memory

llol

what PSU is it, quality matters there too

[idle stump]

Huh. Never crossed my mind the CPU competes with the GPU for power as well...

I don't remember. Not convenient to check since I'm handicapped

[simple tartan]

everything does, number of ram sticks, number of pcie cards, number of drives

i'm running a 1KW, for a 7950X, 7900XTX, 128GB DDR5 (4x32), and 3 NVMe's

got a 1.6KW (dual 800W) on my TR Pro server in my other place

[karmic crown]

please type

git fetch --all

and then

git reset --hard origin/master

that should fix and update your webui files

than stability matrix check for update should work
let me know if it worked

[idle stump]

You building an AGI or something?

[simple tartan]

i like to dabble in ALL things 😉

[karmic crown]

trust me, i am frustrated by your installation almost as much as you

please show me your current version info (bottom of webui)

and send me result of
git status
and
your current webui-user.sh

[gusty solstice]

Doesn't seem to work :/

[karmic crown]

what git status gives you now?

[gusty solstice]

HEAD detached at v1.7.0
Untracked files:
(use "git add <file>..." to include in what will be committed)
modules/textual_inversion/preprocess.py

nothing added to commit but untracked files present (use "git add" to track)

[tame yoke]

is there a way i can see the generating progress
cuz all i can see now is 5%, 50%, 100%

[karmic crown]

please close stability matrix and do not run it until we check a few things

run git fetch --all and show me what you got

[gusty solstice]

Nothing

[karmic crown]

run
git checkout master
and show me what you got

[gusty solstice]

Switched to a new branch 'master'
branch 'master' set up to track 'origin/master'.

[karmic crown]

nice

git status

[gusty solstice]

On branch master
Your branch is up to date with 'origin/master'.

Untracked files:
(use "git add <file>..." to include in what will be committed)
modules/textual_inversion/preprocess.py

nothing added to commit but untracked files present (use "git add" to track)

[karmic crown]

git reset --hard origin/master

[gusty solstice]

HEAD is now at cf2772fa Merge branch 'release_candidate'

[karmic crown]

good

git status
again

[gusty solstice]

On branch master
Your branch is up to date with 'origin/master'.

Untracked files:
(use "git add <file>..." to include in what will be committed)
modules/textual_inversion/preprocess.py

nothing added to commit but untracked files present (use "git add" to track)

[hallow latch]

I always go with 20 hires steps 😄

[karmic crown]

please remove this file modules/textual_inversion/preprocess.py
and do those two things again

git reset --hard origin/master
git status

[karmic crown]

you "head" is not "deatached" anymore, which is good thing (it even sounds good 🙃). now we need to remove untracked file and to get all proper files

[ornate elk]

You can of course ^^ it just increases the vram usage

[gusty solstice]

I don't remember how to remove a file in CMD XD

[karmic crown]

do it from file manager

do not bother

[ornate elk]

You mean a preview image?

[gusty solstice]

On branch master
Your branch is up to date with 'origin/master'.

nothing to commit, working tree clean

[karmic crown]

perfect

now it should work properly, we cleand up the mess

[gusty solstice]

I try the update again.
Seemed that because local and remotes branches weren't the same, happened this conflict. Right?

[karmic crown]

it is updated now, but you can try to see if it works now

it should work without error now

[gusty solstice]

I'm watching the real time log. Wait and see.

[karmic crown]

somehow your local version was messed up, that was the cause of issue

[gusty solstice]

But I'm affraid that it'll still blocked at this line: HEAD is now at cf2772fa Merge branch 'release_candidate'

[karmic crown]

that is fine

Merge: 4afaaf8a 0dfffe53
Author: AUTOMATIC1111 <[e-mail]>
Date:   Sat Dec 16 09:58:07 2023 +0300

    Merge branch 'release_candidate'```

that is the last commit

[gusty solstice]

Maybe it needs time to continue the update, but now it doesn't run more lines

[karmic crown]

that is the last line

there is nothing to do anymore

you are on the latest update

[gusty solstice]

There's still the loading circle :/

[karmic crown]

that is strange

when you run webui it should work

[gusty solstice]

I know... And I don't why I still have this bug

[kindred tinsel]

@karmic crown do you knew if there is a way to store custom colors in comfy?

[karmic crown]

those are two last commits

[unreal marsh]

im having an issue trying to reinstall stable diffusion, it keeps giving me error code 128 and idk how to fix it can anyone help? :((

[lean laurel]

hi, sorry this is gonna be a dumb question, i should click where says windows installer 64 here in python page??

[vocal burrow]

unless you're still stuck in the 2000s then yes you want the 64 bits windows version.

windows installer (64 bits) is the cleanest way to go cause it will register itself in windows registery and a bunch of other things.

Thus why it's the "recommanded" option

[lean laurel]

yeah sorry, i just wanna be sure i m downloading the right version for the IA

[vocal burrow]

np

[lean laurel]

thanks a lot @vocal burrow ❤️ i already installed git

[unreal marsh]

🎉

[vocal burrow]

what's the full log ?

[lean laurel]

the folder must be in the same place where is the window folder??

[vocal burrow]

your internet connection went for a walk during the initial download

[unreal marsh]

yeah, its always bad

[vocal burrow]

delete this folder and try again

use cable instead of wifi if possible

[simple tartan]

if I want to keep a character and use over and over in different poses, environments, model to 3d, etc, should I generate a load of photos and train my own lora, or is that not necessary?

[unreal marsh]

unfortunately thats not something i can do

[vocal burrow]

theorically you can generate a lora even with only one image, but it's way more tricky.

[simple tartan]

is that what i should be doing?

[vocal burrow]

cf https://civitai.com/articles/903/training-an-oc-lora-with-a-single-base-image-part-1

[vocal burrow]

Can't answer for sure because as with any creative process it depends of your needs, what you're trying to achieve exactly and all the jazz....

controlnet might be enough if you just want to change the pose / background, etc

I would redirect you to #📝｜prompting-help tho as it's not really a tech support question.

[lean laurel]

Edit the webui-user.bat (right click), At the line COMMANDLINE_ARGS= You add: --use-directml --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling
If your GPU has only 8gb Vram and you run into out of memory, you may need to change --medvram to --lowvram.

hi, i dont have the 8gb Vram, should i use lowvram?? whats the different??

[unreal marsh]

@vocal burrow so if it dosent work and i get the error code again what do you suggest?

[simple tartan]

I want to create "character" and then be able to write a prompt of "(Character:1), at the beach, in a blue swim suit" and then do an other "(Character:1), sitting on a sofa, glare from the sun, inside an apartment"

[simple tartan]

I tried asking there yesterday and no one is chatting there

[vocal burrow]

then yes lora is the way to go (or embedding, it's easier to create but less effective)

[simple tartan]

guess i'll study on "WTF is controlnet"

[simple tartan]

whats "more effective" then? controlnet?

[vocal burrow]

if you want to go "I put (character) in my prompt and it gives me a picture of my character" then controlnet is not gonna help you.

[simple tartan]

yea, thats what i want, lol

[vocal burrow]

controlnet gives you more control (ah ah) on the output, mainly for stuff like posing, image composition, etc... It will not help you create new worlds/concept/character that the model can understand.

[vocal burrow]

Then I'd say try to install from scratch in another folder :/
You might even want to run pip cache purge to be sure you're not using any cached / already downloaded files.... But then you'll be waiting for everything to download again

[proper oracle]

newb question, i have a pytorch_lora_weights.safetensors file that is the output of dreambooth training. When loading this onto a pipeline via diffusers it works. I cannot figure out how to get it to show up in A1111 UI however. Anyone have pointers on where I'm going wrong here?

[vocal burrow]

if SD asks for more vram than you have, it will crash. As simple as that.
That's what --medvram and --lowvram are for. To indicate SD "yo actually my rig is not that great so maybe let's go with the slower but safer approach"

[karmic crown]

that is a proper place to ask questions like yours (or mine, which is also not answered yet)

[lean laurel]

ohhh thanks a lot, so better i type lowvram so i can use SD safe??

[vocal burrow]

what's your gpu ?

[karmic crown]

https://www.youtube.com/watch?v=mmZSOBSg2E4

but PLEASE, keep questions like those in https://discord.com/channels/1002292111942635562/1011743094309396631

[vocal burrow]

or https://stable-diffusion-art.com/controlnet/

[karmic crown]

@vocal burrow can you move those to https://discord.com/channels/1002292111942635562/1011743094309396631 ?

[vocal burrow]

there's no "move message" functionality on discord, best we can do is nudge them in the correct channel.

[lean laurel]

the one we talked yesterday, i have 2 gb vram

[vocal burrow]

sry kinda forgot, doing lots of stuff. Then yeah definitely go with the --lowvram

[karmic crown]

i was sure that is an option like that, ok, ill try to keep conversations like those to minimum here and to direct them there

[lean laurel]

dont worry, u are so kind, everyone i open this channel u are helping people ❤️

[vocal burrow]

and don't ask for high resolution.... and use light / 2gb model as much as possible. @lean laurel

[lean laurel]

ask for high resolution? thats after i installed?? will i have low quality outpots?

[vocal burrow]

Lora should go in stable-diffusion-webui\models\Lora

[vocal burrow]

Yes that's after installation don't worry.
Also yes with lower resolution picture will inevitably be of lower """"quality""".

for starts I'd stay it's best for you to stick to SD1.5 models and to 512x512 resolution

[lean laurel]

oh i see, but is a significant diferent?

[lean laurel]

do that models look fine??

[vocal burrow]

once you've got the hang of it you might start to look at way to upscale safely with only 2gb of vram

[lean laurel]

oh ok ok, thanks!! ❤️

[vocal burrow]

you tell me :

• https://civitai.com/models/4384/dreamshaper
• https://civitai.com/models/88895/alpha-lyrae

In those examples people are obviously generating at higher resolution and/or upscaling but the model they use is still SD1.5 (therefore based upon 512x512 dataset)

[lean laurel]

in git, should i create a start menu?

[lean laurel]

omg it look awesome ❤️

its ok?

[vocal burrow]

as you wish, it's just like any other programs. Do you want git to create shortcuts in your start menu ? yes or no ?

[proper oracle]

I placed the file there but it won't show up in the UI. (even after restarting)

[lean laurel]

i m sorry, idk what is start menu, sorry

i saw what it is, i wont create it

[vocal burrow]

it's the menu that pops up when you click the "start" / windows logo button in your bottom bar.

[lean laurel]

thanks @vocal burrow, sorry for do too many questions, click next?

[karmic crown]

it should be like this

[vocal burrow]

I'd say it's easier to pick "nano" but it doesn't really matter for you. you can safely spam "next"

[lean laurel]

thanks, and when say:

[vocal burrow]

next next next

[lean laurel]

oh lovely pics ❤️

[karmic crown]

can you show me what you have?

[lean laurel]

just next till end lol??

[proper oracle]

in this case the JuggerCineXL2 shows in Loras but mine does not.

[simple tartan]

basically, and then go install GitKraken

[karmic crown]

i know what it is

the first one is SDXL, and the other is 1.5 lora

and you have SDXL model active

[proper oracle]

They are both trained on SDXL

[unreal marsh]

i think fate dosent want me have this downloading :((

[karmic crown]

you probably downloaded 1.5 version, insted of sdxl version

https://civitai.com/models/195519?modelVersionId=219782 SDXL

actually, is that lcm or something else?

[proper oracle]

Interesting, I suppose it's possible. Wouldn't it not work with an SDXL model though?

[lean laurel]

and here?

[karmic crown]

only compatible loras are visible

[proper oracle]

I can use it with a custom pipeline, just not A1111 UI.

[karmic crown]

when i load sdxl model i only see sdxl loras, or compatible loras

[lean laurel]

do i click something??

[proper oracle]

I'll switch to 1.5 and see if it shows up

well, i'll be plum tickled. Apparently I have been training 1.5 and not SDXL as when i switch to 1.5 I see it. Thanks for the help!

[vocal burrow]

LoRa/Lycoris quick guide :
1/ Where do I put my LoRa : stable-diffusion-webui\models\Lora
2/ How do I use my LoRa, you can either :

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

[lean laurel]

do i check something??

[karmic crown]

i guess you should pin that

[vocal burrow]

it's in my copypasta.txt file ^^ I try to not clutter pins with informations that are only relevant under certain specific situations.

[vocal burrow]

you guessed it, next

till the end

[lean laurel]

oh but i dont enable nothing??

[vocal burrow]

don't need to

[karmic crown]

can you pin that in promting help, not here 🙂

[vocal burrow]

hmmm maybe, I'll think about it. It might be a good idea indeed.

[lean laurel]

ok thanks i m installing it!

[proper oracle]

Apparently using train_dreambooth_lora_sdxl.py and specifying an SDXL model doesn't guarantee an SDXL Lora? Well, back to the drawing board. Really appreciate the help!

[vocal burrow]

maybe you can borrow someone else's connection ?

[lean laurel]

Git installed!!! ❤️ ❤️ now python!!

should i also click next next till end, justo check "add python path"?

[vocal burrow]

definitely make sure to check the "add python to path" checkbox https://cdn.discordapp.com/attachments/1002602742667280404/1075947280307331132/image-49.webp

other than that, meh, you can install it somewhere else if you prefer.

[lean laurel]

oh thanks
install somewhere else?

thats two things must be checked??

[vocal burrow]

everything as default except for the "Add python.exe to PATH"

[lean laurel]

ok, thanks!! i going to install it ❤️

[vocal burrow]

Like, install it on a different folder / drive. Useful if like many you're trying to keep things organized properly and don't install everything to their (and sometime very obscure) default path.

[lean laurel]

thanks, thats why later the tutorial say that must make a new folder called AI?

setup was succeful!! it was quick

[vocal burrow]

it's to avoid people simply copy pasting blindly the git clone azeazeazeaze command in cmd and then complaining about "why doesn't it work ?" when by doing so they just installed SD in system32 / oneDrive / any other specific folder with restrictions.

I mean .... they're still gonna copy paste stuff blindly because much to my demise we can't force people to actually learn what they're doing. But by telling them to create a new AI folder we're minimizing the mishaps.

[lean laurel]

ohh so if i do in a new folder there no restrioction, ok i installed the 2 programs, now i should go to the AI folder, and click file explorer, right?

[lean laurel]

thanks that kind, i actually dont know what i m doing thats why i m so grateful for helping me

[vocal burrow]

follow the rest of the tutorial.

[lean laurel]

yeah i m working on it ❤️

the thinks that i must copy, just copy past and type enter key??

[vocal burrow]

yes like indicated

[lean laurel]

ok sorry this is another dumb question, how i open the file explorer??

yeah i m pretty bad with computers

[vocal burrow]

file explorer is a fancy name for the program that is used when you double click a folder

I'm guessing you're at the You go into the folder you created in this case Ai, then click in the File Explorer bar (not searchbar) and type cmd then press enter. part ?

[karmic crown]

come to the dark side (mac) we have a script (that i made) so not need to follow any guide and to install anything manually

[lean laurel]

yeah i m in that part!

ok so click the file explorer, thats the one in the left of the searchbar??

[vocal burrow]

go to your stable-diffusion-webui folder then do as follow

[lean laurel]

oh thanks a lot, i ll do it ❤️

now copy and paste?

[vocal burrow]

yes

[lean laurel]

enter??

[karmic crown]

if that is an AMD, than yes, those are the right commands

[lean laurel]

thanks ❤️ now is done

thanks @karmic crown

so now this

Edit the webui-user.bat (right click), At the line COMMANDLINE_ARGS= You add: --use-directml --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling
If your GPU has only 8gb Vram and you run into out of memory, you may need to change --medvram to --lowvram.
RX 5xxx cards additional need: --no-half

i should use the "lowvram"

[vocal burrow]

yes

[lean laurel]

and i ignore the RX 5xxx thing??

[vocal burrow]

for now at least yes

[karmic crown]

if your file extensions are hidden, webui-user.bat is the one without extension that looks like webui-user

[lean laurel]

ohh ok ok, let me see how should i do it

to be honest, idk what is the webui-user.bat, is just the right click? where that line where should i click?? sorry i got lost

[vocal burrow]

webui-user.bat is a file in your stable-diffusion-webui folder

like viking said, by default windows hide file extensions (for some reasons) so you might only see two "webui-user" files

[lean laurel]

thats exactly what i see, two webui-user files

[vocal burrow]

The correct one's content should look like that

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=

call webui.bat

[karmic crown]

the other should be webui-user.sh

[lean laurel]

should i click them to see if look like that??

[vocal burrow]

open them with notepad

[lean laurel]

ok i ll do it

[karmic crown]

if you see webui-user.sh its the other one

i would make a screenshot, but i am on mac

[lean laurel]

ok let me open it, idk sometimes "open with" dont show on the options

[karmic crown]

you might need to go to show more first

windows 11 if i remember correctly has that

[vocal burrow]

yes windows 11 hide """non trivial""" stuff in the "show more" (or something like that) submenu

[vocal burrow]

I can't screenshot it either because I moded mine and removed that stuff :p

[lean laurel]

ohh thanks thanks!!

this one isnt it?

[vocal burrow]

nope

that's webui-user.sh

[karmic crown]

the other one

that one is for us from the dark side (linux and mac)

[vocal burrow]

like I said its content should look like that

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=

call webui.bat

[lean laurel]

oh ok ok, the other one is a "batch file" i dont know how is called in english sorry

[unreal marsh]

so its impossible then

[karmic crown]

thats the one

[lean laurel]

i cant open it in with notebook but is that

ok ok

[karmic crown]

you need to open it with Notepad

or to choose Edit if that exists

[lean laurel]

ohh i need to open with notepad to edit it??

[karmic crown]

yes

[lean laurel]

is just i dont see the option "open with"

[karmic crown]

edit

[lean laurel]

dumb me lol

haha thanks ❤️

oh yeah is like the one u sent

[karmic crown]

now add that as instructed

--use-directml --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling

[lean laurel]

Edit the webui-user.bat (right click), At the line COMMANDLINE_ARGS= You add: --use-directml --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling
If your GPU has only 8gb Vram and you run into out of memory, you may need to change --medvram to --lowvram.

i paste

[karmic crown]

show me what you have, before save

or edit again

[lean laurel]

You add: --use-directml --lowvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling

[karmic crown]

add
--use-directml --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling

to line that looks like this
COMMANDLINE_ARGS=

[karmic crown]

almost

[lean laurel]

is it right?

[karmic crown]

remove :

=--

COMMANDLINE_ARGS=--use-directml --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling

like this

show me what you have

[vocal burrow]

like this

[lean laurel]

there one space between the = and the -

[lean laurel]

oh so i delete the space and is fine?

[karmic crown]

yes

it would work with sapace, but remove it to be like Aryetis sent

[lean laurel]

so i close it and is done??

[karmic crown]

save

and close

[lean laurel]

ohh should i click something for save or close will show the save option?

[karmic crown]

it should ask for save on close (i hope)

ctrl+s will save it

[lean laurel]

is saved!!

let me open it again to see is everything is right

[karmic crown]

it will install bunch of stuff now

[lean laurel]

is saved!!

[karmic crown]

when you run it

[lean laurel]

ok install bunch of stuff now?? so should i run it in order to install that

where i run it??

[karmic crown]

double click that file you edited

it will run and install what is needed

you will just need a little patience (depending of your internet and computer speed)

[ornate elk]

hey 😄

[lean laurel]

oh hi @ornate elk ❤️

[lean laurel]

oh wish me luck

this is... normal??

[ornate elk]

i see you already got help with the install Steps 🙂

[lean laurel]

yeah!! @karmic crown and @vocal burrow helped me a lot!

[karmic crown]

nope

[ornate elk]

oh that damn error, "no module named pip" any AV active?

[karmic crown]

do you have mcafee?

[vocal burrow]

unlucky, looks like you might have to reinstall python with your anti virus turned off

[karmic crown]

that thing itself beacame worse than any virus

[vocal burrow]

just to confirm, run those commands in cmd where python, python --version and py --list-paths

and screenshot the result

[lean laurel]

i have an anti virus dont remember the name

[lean laurel]

in any folder??

[lean laurel]

so i close this??

[karmic crown]

yes close that

and check what antivirus you have

if it is mcafee, please get rid of that c....

[lean laurel]

ok ok haha let me check it

[lean laurel]

so first i type this in any cmd??

i think is the window protector

[karmic crown]

you type those commands in cmd window

some antiviruses do not allow proper python installation, and mcafee is the worse of all

[lean laurel]

sorry the late

[lean laurel]

what should it say when i enter that??

[vocal burrow]

a bunch of stuff, screenshot it and we'll check

[lean laurel]

ok one sec, sorry for too mutch throubles

[halcyon swan]

wait what

[kindred tinsel]

I’ve got mcaffee and nothing wrong has happened with it .-.

[halcyon swan]

ok so sticking with microsloth defender was the right choice after all? 🙂 👍

[lean laurel]

C:\Users>py --list-paths
Installed Pythons found by py Launcher for Windows
-3.10-64 C:\Users\r\AppData\Local\Programs\Python\Python310\python.exe *

C:\Users>where python
C:\Users\r\AppData\Local\Programs\Python\Python310\python.exe
C:\Users\r\AppData\Local\Microsoft\WindowsApps\python.exe

C:\Users>python --version
Python 3.10.11

is something wrong @karmic crown ??

[vocal burrow]

sometime it messes things up, sometimes it doesn't....No idea why tbh. What I know is that sometimes pip is not installed correctly due to antivirus

[vocal burrow]

ok screenshot your C:\Users\r\AppData\Local\Programs\Python\Python310\Scripts folder then

[ornate elk]

McAfee is bloatware and spyware like most free antivirus

[lean laurel]

let me check it

[kindred tinsel]

Mcfee isn’t free though

[lean laurel]

this??

[ornate elk]

it comes free for 3-6 months if you buy a new windows pc so that the normal users think they need it xD

[kindred tinsel]

That seems about right for pip

[lean laurel]

so, whats the problem?? should i try open again what gave me wrror??

[halcyon swan]

does the same happen with avast?

[ornate elk]

avast is even worse since it got bought by a adware company

[halcyon swan]

i forgot how to check path on windows, but isn't this a path thing?

[lean laurel]

so the problem is not the pythom, and if i disable the real time protection from my window defender, and them i click the webui-user?

[halcyon swan]

wait they got bought out? damn

[vocal burrow]

hmmm looks correct, that's weird

[lean laurel]

should i get afraid??

[ornate elk]

yep, the main company has bought also CCleaner, Avira, and norton360, and infectet their own software with cryptominers

[vocal burrow]

afraid might be too strong of a word :p

[karmic crown]

windows defender should not disable anything related to python, in 99.99% its the other antivirus

[lean laurel]

may this work??

[lean laurel]

ohh forget my last message

so... if afraid is too strong word, should i be a bit scare??

[vocal burrow]

try deleting your stable-diffusion-webui/venv directory, turn off McAffee and then launch webui-user.bat again

[lean laurel]

i dont have McAffe i think, i have norton security ultra

[vocal burrow]

same thing.

[lean laurel]

where is the stable-diffusion-webui/venv??

[vocal burrow]

try deleting your stable-diffusion-webui/venv directory, turn off [insert whatever antivirus you have here] and then launch webui-user.bat again

[lean laurel]

ok i ll turn off the anti virus thanks lol haha but where i delete that u said?

[karmic crown]

here #🤝｜tech-support message

[vocal burrow]

wherever you installed sdwebui (so in c:\Ai if you followed the pinned tutorial)

[lean laurel]

yeah i did!!

[karmic crown]

inside you should have venv too

[lean laurel]

sorry i need to go a moment, i ll back asap, really sorry and thanks for all your help, wont take long!!

[vocal burrow]

(forgot the -directml, my bad)

[karmic crown]

i am off too, bye guys

[lean laurel]

bye @karmic crown thanks for all!!

i m back, sorry i had to go for a while

[lean laurel]

so if i do that maybe the program is finally ready??

[ornate elk]

yes, your antivirus is mostly the problem

[vocal burrow]

most likely yes

[lean laurel]

i should delate the whole folder with venv name??

[vocal burrow]

yes

[lean laurel]

ok my anti virus is turn off for a while

just to learn, what are this files inside venv and why should i delate it?

[vocal burrow]

it's bsically a copy of everything python related that stable diffusion needs to run.

It makes a copy of those files that way it can edit and modify them without modifying your actual system python

[lean laurel]

ohh i see, thanks!!

look everything fine??

[vocal burrow]

looks good so far, let it do its stuff

[lean laurel]

ok thanks!!

❤️

done??

[vocal burrow]

no it's still installing

[lean laurel]

oh ok ok i will not hurry it

will take a while, right??

[vocal burrow]

depends of your system, between 2 and 20 minutes I guess

[lean laurel]

is ok if i still use the computer while installing it or better i close all the other windows?? like chrome and discord

[vocal burrow]

should be fine if you have a solid connection

[lean laurel]

ok just i was about to start an anime movie so it will have to wait lol ahaha

i dont wanna force my connection

and after that, is done? i can start making pics? off course i would need to download a model

i just want a model to transform my photos into art style

[vocal burrow]

it's gonna download the 1.5 vanilla model automatically

[lean laurel]

oh that one is nice??

[vocal burrow]

not really ^^

it's the basis of all other models. It's relatively basic compared to what we have now

[lean laurel]

lol hahaha, which one do u recommend me if i wanna turn my photos with people in anime or videogame style??

and, i should always check that the model can be run by 2gb vram?

[vocal burrow]

the larger the model, the more likely it is to not run on your "weak" 2gb vram gpu

with 2gb models you should be safe but you can try bigger ones

[lean laurel]

ohh ok ok, there any info i should specially focus when i wanna download a model??

[vocal burrow]

stick to 1.5 based model as those are the models with the lower vram requirements

[lean laurel]

thanks ❤️
do u recommend me one for anime or 3d videogame style?? again, i wanna transform photos into that style

[vocal burrow]

nyanmix, anything, AbyssOrangeMix, MeinaMix, etc are good lightweight models for anime, there are plenty others. You can find most of them on civitai

for "rendering like" output I used to play with redshift diffusion, but most "all in one" models should be fine such as dreamlike, dreamshaper, protogen, etc

[tame yoke]

yea like when its generating
5%
10% and so o n
it only shows the generating image on 5%, 50%, 100% percent mark

[lean laurel]

so is done?

it opened a page on my chrome

[vocal burrow]

yup

[lean laurel]

yupiiiiii!!!

i dont have to do nothing more?? can i close the cmd window and start using it??

[vocal burrow]

No, the cmd holds stable-diffusion local server.

start it everytime you want to use sd and keep it open (it won't reinstall everytime :p)

[lean laurel]

ohh so how should i open it everyone i wanna launch SD?? how i open SD?? now i have the chrome window open but in future??

[vocal burrow]

you simply run webui-user.bat

[lean laurel]

ohh the one i edited and them open it to install all??

[vocal burrow]

yes

[lean laurel]

ohh thanks thanks, and for install models??

[vocal burrow]

you download and put them in models\stable-diffusion folder

[lean laurel]

oh and nothing more to do??

where do i switch model??

[vocal burrow]

top left of the webpage "Stable Diffusion Checkpoint"

I'll go for now, bye

[lean laurel]

bye @vocal burrow and thanks for everything!!!

[ornate elk]

you can define in the settings under preview count, how often you want to see an preview

but that will use mroe vram

[lean laurel]

lol i still need to learn how use the AI, i tried ti make a cat dancing with a hat and well...

lol

[devout pecan]

im using colab for this, was working fine just a while ago but is stuck like this now for some reason

[dusty meteor]

so when I type git status I get fatal: not a git repository (or any of the parent directories): .git

[devout pecan]

anyone know why?

i tried creating a new notebook but it still wont go past fetching, did the huggingface token thing too

[dusty meteor]

This is my webui-user.sh

`#!/bin/bash
#########################################################

Uncomment and change the variables below to your need:#

#########################################################

Install directory without trailing slash

#install_dir="/home/$(whoami)"

Name of the subdirectory

#clone_dir="stable-diffusion-webui"

Commandline arguments for webui.py, for example: export COMMANDLINE_ARGS="--medvram --opt-split-attention"

export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-split-attention --upcast-sampling --no-half --medvram --use-cpu interrogate"

python3 executable

#python_cmd="python3"

git executable

#export GIT="git"

python3 venv without trailing slash (defaults to ${install_dir}/${clone_dir}/venv)

#venv_dir="venv"

script to launch to start the app

#export LAUNCH_SCRIPT="launch.py"

install command for torch

#export TORCH_COMMAND="pip install torch==1.12.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113"
export TORCH_COMMAND="pip install torch torchvision torchaudio"

Requirements file to use for stable-diffusion-webui

#export REQS_FILE="requirements_versions.txt"

Fixed git repos

#export K_DIFFUSION_PACKAGE=""
#export GFPGAN_PACKAGE=""

Fixed git commits

#export STABLE_DIFFUSION_COMMIT_HASH=""
#export CODEFORMER_COMMIT_HASH=""
#export BLIP_COMMIT_HASH=""

Uncomment to enable accelerated launch

#export ACCELERATE="True"

Uncomment to disable TCMalloc

#export NO_TCMALLOC="True"

###########################################`

[devout pecan]

please🙏

[ornate elk]

also try the dreamshaper v8 model, thats a good basic one

[lean laurel]

oh thanks, is ok if i use it with low vram??

[halcyon swan]

you might need to use the low vram flags, i never got to use local SD (iGPU with 512mb vram sucks, just like 10 mins for cpu render) but i think these are --low-vram and --cuda-attention-slicing

[ornate elk]

Sure its 2gb

[lean laurel]

local SD?? what u mean??

[ornate elk]

He means the thing you installed ^^

[lean laurel]

thanks a lot ❤️ i wrote the low vram thing

[halcyon swan]

yes

[lean laurel]

oh thanks thanks haha lol

[halcyon swan]

i can only use things like nightcafe and tensorart ;-;

(PSA: nightcafe doesn´t prepare you for the real world, i still don´t know what steps does "short" runtime use)

[inner zinc]

Hello friends! I'm very new to SD and I've run into a serious issue where every time I try to generate an image, all I get as a result is a bizarre mixture of black and grey textures. I was able to create images with no problem until yesterday but I have no idea why it's not working anymore. I've tried uninstalling and reinstalling Automatic1111 but nothing works... Here's an example of what I'm getting:

[ornate elk]

what yre your settings for this?

[inner zinc]

Now if anyone here can help me with that and you need anything else, please let me know. ☺️

[inner zinc]

Let me get a screenshot for you and I'll be right back 🙂

[ornate elk]

also whats your gpu and whats inside your webui-user.bat?

[inner zinc]

My GPU is GTX 3060 with 12GB of vram so I don't think that's the issue. I'll come back with a screenshot of my webui-user.bat in a few. And thank you so much!

So this is what's inside my webui-user.bat : @echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --xformers --autolaunch

call webui.bat

I have to wait for SD to load before I can provide a screenshot of my settings. 🙂

[ornate elk]

Yup, webui looks good

Maybe its a specific setting

[inner zinc]

Here we go:

[sturdy ginkgo]

try using only 3 negative embeddings at most

[inner zinc]

You mean, the negative prompts? I'm sorry but I very, very new at this and I'm still trying to learn the vocabulary. 😊

[sturdy ginkgo]

with 10+ you basically saying .. don't gen 99% of what i am asking

[inner zinc]

Ah... makes sense... Thank you!

[ornate elk]

your word strenghts are to high, dont go over :1.5, doing like 2.0 is forcing it so hard that random crap will come out

[sturdy ginkgo]

https://cdn.discordapp.com/emojis/1102316663598223390.webp?size=48&name=this&quality=lossless

[inner zinc]

Now that makes sense because I started having problems after I increased the word weight. Thank you so much for your help, guys!

[ornate elk]

no problem 🙂

[sturdy ginkgo]

ugh..this workflow experiment is giving me a headache

[buoyant pawn]

CS1o i got some news

when i reduced the size of the picture generation to 212*212 the generation worked in 15 sec

more or less

[simple tartan]

i'm attempting to outpaint my first image, and gettings Nans errors, what should I be doing different?

[ornate elk]

hm thats interesting, and you updated your gpu driver and deletet the venv folder again?

[ornate elk]

you may need --no-half
for inpainting thats needed

i think for outpainting may too

[simple tartan]

ok, thanks. Is outpainting only available as a script?

i saw a video where it was more like the inpaint tab, but I dont have any "outpaint" options other than Outpainting MK2 and Lazy Man Outpainting

[ornate elk]

yes or if you install the OpenOutpaint extension

[simple tartan]

ok

why not another extension at this point, lol

[buoyant pawn]

the venv folder is impacted by the gpu update ?

[ornate elk]

yes

[buoyant pawn]

i'll delete it right now

[sturdy ginkgo]

might as well use any of the realtime ai image generation plugins for krita and photoshop etc, that work with comfyUI 😄 , just draw and edit in a photoediting app and comfy will generate the changes in a second

[kindred tinsel]

what are you trying to do? impaint and region prompting?

[sturdy ginkgo]

just saying that that might be better than 500+ extensions 😄

[simple tartan]

i'll eventually get to trying comfy

[worthy portal]

hi
I'm getting this error, is it because of my lack of GPU? Or is there a way to solve it?
When I produce images less than 400px in 400px, it works, but for more than that, this error is displayed.
please help me
.
NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
Time taken: 3.3 sec.

A: 2.16 GB, R: 2.30 GB, Sys: 4.0/4 GB (99.2%)

my pc is
win 10
Gpu quadro T2000 4G
Cpu core i7 9gen
ram 32 G
SSD 512

[ornate elk]

Hey, you need to edit the webui-user.bat and at the line COMMANDLINE_ARGS=
you add: --xformers --medvram --no-half-vae
Then save and relaunch

That will help to improve performance and using less vram

[ornate elk]

Also make sure the model is larger than 1.98gb

[ornate elk]

Has anything changed ?

[lofty salmon]

Hey does anyone know much about the watermark? I am concerned after reading that the watermark contains your ip address in images 🤔

[ornate elk]

"The watermark" ?
Where? In what product?

[lofty salmon]

Using the SD art repo to generate images, naturally in A1111.
I randomly caught a thread about some sort of new watermark tech that SD may be using - ostensibly to simply track if images are AI generated, but then saw that the library being used also seems to imprint identifiable data like your IP address?

Using various sd 1.5 and sdlx models if that factors in

[ornate elk]

If your using auto1111 everything is fine. Its an open source webui to locally generate images. It doesn't apply invisible watermarks at all.

Other local webuis are safe to use too. But I can't speak for every webui service that is closed source. But its very unlikely they would implement it.

[lofty salmon]

Ah ok
I would have thought the base SD would be what implemented it, but that doesn't watermark it either?

[ornate elk]

As watermarking is done trough code/scripts any model is free of watermarks.

As it just stores basic training data

[lofty salmon]

Ah sorry I meant the generated output images, not the models

Or does a1111 do the image outputting? I am still new to this btw sorry

[ornate elk]

Like I said it depends on the software you use. Auto1111 does not use watermark code or scripts

Auto1111 is a community made Software to make Stable Diffusion easy accessible and usable

It uses the core stuff of stability and compvis stable Diffusion functions and enhances it with a ui and more features

[lofty salmon]

Ahhh so SD itself is literally just a model + bits, there's no code that comes with it to make it make the images, things like A1111 are for that

[ornate elk]

Stable Diffusion is a process, and its coded. You can check that code for example on github, its uploaded by CompVis. This barebone code can generate images when used together with a trained model. But its a commandline tool only.

Auto1111 builds up on that code, enhances and modifies it to make it user accessible.

The models we all use in the webui is the training data, its just basic information.
These models can be loaded and used only with Programms like auto1111 or comfyui etc to generate images. They use Diffusers for that.

To add a watermark you would need to code a script that applies it after an image is generated. And that doesn't exist in any of the opensource community made webuis.

@lofty salmon here is a good explanation on how Stable Diffusion works in general:
https://stable-diffusion-art.com/how-stable-diffusion-work/

[lofty salmon]

Thanks v much for the info above And ya I will read more on this actually.
So if A1111 and the SD code isn't watermarking, makes me feel better - IDK where this worry of IP watermarking comes from, maybe another UI or service

[ornate elk]

No problem 🙂 and yea its pretty much safe to use. Also I never heard about IP watermarking in regards of SD.

What you need to know is that the output images you generate contain so called Meta Data, like photos from your Phone.
In the Meta Data of SD images is listed what settings you used to generate the image. Nothing personal. You also can remove Meta Data with tools.

You can load an generated image into the PNG-INFO tab of auto1111 to view the Meta Data

[lofty salmon]

That is super useful because I've suddenly found out I didn't need to save my prompts in text files lmao

[vocal burrow]

note that not everything is saved in the metadata, for instance extensions specific settings might not be saved in there.

[lofty salmon]

I can't see anything like this in the a1111 info tab, but is there any metadata to be wary of when posting stuff?
I initially came asking about this sort of thing e.g. the watermark because I was in a little ai art group and some members had some hate comments from anti AI people. I came across something about IP watermarking and was v concerned about harassment

[gray oyster]

Is there a way to set CFG to 0 in AUTO1111?
I wanna try out the "new" SDXL Turbo model but they claim that it doesn't use CFG so it's recommended to disable it completely

[vocal burrow]

just set cfg scale to 1 and use 1 or 2 sampling steps.

[gray oyster]

yea i just did that and it works amazingly good

[gray oyster]

but i would still need a good webui which regenerates the image as soon as the prompt changes

[mental locust]

I am trying to do tiktok dance - anime... using bb95furry... and animatediff is ruining the image quality, but without animatediff it's a flickering inconsistent mess. Attached are examples, with ComfyUI workflow in the PNGs. I've got as far, with community help, as knowing that vpred is a thing that necessitates modelsamplingdiscrete, rescalecfg, and freeu_V2 nodes. Any help or suggestions welcome, I've exhausted the possibilities of altering the available settings, as far as I can figure out.

[zenith stump]

Face Swap

[mental locust]

That would be half the battle, and great suggestion I didn't even consider using it with a non-photo model lol. Animatediff is still massively reducing the quality of the body too though :/

[gray oyster]

So uh, I tried playing around with AUTO1111 again and for some reason xformers is much slower than sdp now despite them being the same in the past (1024x1024)

just thought i would mention it

[proper oracle]

Any ideas why A1111 recognizes my XL Lora as 1.5 when it was trained on XL? If I load the Lora via the diffusers library it works, however when I attempt to load with a model in A1111 it will only show for 1.5 models and does not work.

Using the train_dreambooth_lora_sdxl.py script, here is the config.

[remote ivy]

not sure if this is the right channel, but does anyone know if the multidiffusion extension still works on the latest version of automatic1111's webgui?

[worthy portal]

Thank you very much, this seems to be helpful, but I keep getting this error
.
NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
Time taken: 4.4 sec.

A: 1.75 GB, R: 1.92 GB, Sys: 3.6/4 GB (89.3%)

[ornate elk]

What is the model size you try to use?

[ornate elk]

Yes, it still works

[remote ivy]

Both tiled diffusion and tiled vae?

[ornate elk]

Yes both

[worthy portal]

i have these
animatrix_v20 1.99 gig
dreamshaperXL_turboDpmppSDE 6.5 gig
v1-5-pruned-emaonly 4.4 gig

[ornate elk]

Animatrix should work, can you show your settings + error when you try to generate?

[worthy portal]

this is
NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
Time taken: 4.3 sec.

A: 1.75 GB, R: 1.92 GB, Sys: 3.6/4 GB (89.3%)

[ornate elk]

Show me a screenshot of your txt2img settings

[worthy portal]

OK, wait a minute
I can't upload a photo
Now I will do OK and send

[ornate elk]

Did you added --xformers --medvram --no-half-vae to your webui-user.bat ?