🤝｜tech-support

[karmic crown]

and rerun

[steady mountain]

deleted

[karmic crown]

if you run it as i told you it will update this one /Users/hong/stable-diffusion-webui since your home folder is /Users/hong

if you want to use it on some other place you can use -d option like -d a1111 and then it will use this folder /Users/hong/a1111

[steady mountain]

connection still errors out

same core dump

[fiery arch]

thank you. understood~

[karmic crown]

still reporting old ROCm?

[karmic crown]

what you get with this
pip freeze

[fiery arch]

@karmic crown If I want to uninstall SD, can I just delete the SD webui folder?

[karmic crown]

if you do not need anything from ther, yes, just remove that folder

you can use my script with same paras for a new install

[steady mountain]

shall I type pip freeze in the terminal folder of stable diffusion?

[karmic crown]

anywhere i want to see what you have in your global python installation

[azure sable]

anybody here use Regional Prompter? I just today started having a strange issue where it creates pics with weird artifacts, blurs, burnt red looking people, etc. Always had perfect results until today. Anyone know why?

[steady mountain]

pip freeze returns:

[karmic crown]

are you using model with lcm baked in by any chance?

[steady mountain]

ok message to long, screenshotting it

[azure sable]

how do I tell? I've only been using SD for a few months and I'm not sure what lcm is

[karmic crown]

which model you are using?

and which vae

[azure sable]

realistic vision 5.1 with baked vae

i've used that model with Regional Prompter for the last months with no issue.

[karmic crown]

in that case it is not what i suspected, i though your CFG is too high

[karmic crown]

what python --version gives? and what you get with venv/bin/python --version

[azure sable]

ah well, thanks anyways. Guess I'll have to just try different things and see if something specific is affecting it

[steady mountain]

Python3 -- version returns Python 3.10.12

[karmic crown]

does normal generation works fine with same model, same sampler, same number of steps? can you try?

[azure sable]

yes, works completely fine.

[steady mountain]

venv/bin/python3 --version gives the same return 3.10.12

[karmic crown]

in that case you probably only have that one version on system

[steady mountain]

ok

can it be ROCm that is the problem?

I swear I am considering formatting the partition and reinstalling Ubuntu, doing a fresh reinstall of everything

[karmic crown]

give me a second to prepare something

[steady mountain]

When I find someone actually running SD on ROCm Ubuntu 22.04 with XTX

[steady mountain]

inga problem

[karmic crown]

go to this folder

and run this

pip uninstall -r requirements.txt

that will uninstall every single package used by webui from your system python

[steady mountain]

this is the folder

[karmic crown]

yes

[steady mountain]

now I will run pip uninstall -r requirements.txt

[karmic crown]

yes

[steady mountain]

Found existing installation: GitPython 3.1.32
Uninstalling GitPython-3.1.32:
Would remove:
/home/mr/.local/lib/python3.10/site-packages/GitPython-3.1.32.dist-info/*
/home/mr/.local/lib/python3.10/site-packages/git/*
Proceed (Y/n)?

shall I proceed?

[karmic crown]

yes

[steady mountain]

Successfully uninstalled GitPython-3.1.32
Found existing installation: Pillow 9.5.0
Uninstalling Pillow-9.5.0:
Would remove:
/home/mr/.local/lib/python3.10/site-packages/PIL/*
/home/mr/.local/lib/python3.10/site-packages/Pillow-9.5.0.dist-info/*
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/libXau-154567c4.so.6.0.0
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/libbrotlicommon-92722cb2.so.1
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/libbrotlidec-db4b3db6.so.1.0.9
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/libfreetype-cb9caf6f.so.6.19.0
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/libharfbuzz-3543f599.so.0.60710.0
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/libjpeg-f2134fdd.so.62.3.0
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/liblcms2-12745711.so.2.0.15
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/liblzma-95592ee6.so.5.4.2
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/libopenjp2-78c47f58.so.2.5.0
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/libpng16-cb42a3c6.so.16.39.0
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/libsharpyuv-9340467d.so.0.0.0
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/libtiff-b9364ff1.so.6.0.0
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/libwebp-3ee34a0b.so.7.1.6
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/libwebpdemux-785dd7eb.so.2.0.12
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/libwebpmux-ddfacf1d.so.3.0.11
/home/mr/.local/lib/python3.10/site-packages/Pillow.libs/libxcb-3e83370d.so.1.1.0
Proceed (Y/n)?

[karmic crown]

yes

remove all

[steady mountain]

Proceed (Y/n)? y
Successfully uninstalled Pillow-9.5.0
Found existing installation: accelerate 0.21.0
Uninstalling accelerate-0.21.0:
Would remove:
/home/mr/.local/bin/accelerate
/home/mr/.local/bin/accelerate-config
/home/mr/.local/bin/accelerate-launch
/home/mr/.local/lib/python3.10/site-packages/accelerate-0.21.0.dist-info/*
/home/mr/.local/lib/python3.10/site-packages/accelerate/*
Proceed (Y/n)?

ok I will continue to the end

[karmic crown]

yep

when you finish

type this

pip cache purge

[steady mountain]

so

[karmic crown]

we are cleaning your system python first

[steady mountain]

mr@Nodex:~/stable-diffusion-webui$ pip cache purge
Files removed: 283
mr@Nodex:~/stable-diffusion-webui$

[karmic crown]

ok

now remove venv again

and run webui again

lets hope for the best

[steady mountain]

I accedently deleded venv from home instead of the stable diffusion folder

ok or shall I undo it

?

[karmic crown]

you do not need one there

[steady mountain]

ok

deleting from sd folder

deleted

[karmic crown]

venv is a "virtual enviroment" that will a1111 use

it will add packages there instead to system python

[steady mountain]

I will try to run the webui from the terminal as usual

[karmic crown]

yes

[steady mountain]

looks good or?

[karmic crown]

yea

[steady mountain]

downloading 1.5gb

I have been through this before. pray that it will be different this time

🙂

you speak scandinavian viking?

[karmic crown]

i speak serbian, croatian, bosnian, english, russian (but i forgot many things since i learn it many years ago), and some slovenian (i understand much more than i can speak)
i understand some bulgarian
i know some words and phrases in italian, german and spanish
i am learning korean (chinese is next)

is it enough? 🤣

[steady mountain]

It is enough. I was just curious because of the name.

I am from the place once called Svíþjóð myself.

[karmic crown]

ah... the land of Sabaton 🇸🇪

[steady mountain]

lol

yes. true

[karmic crown]

ouch, this is slow

[steady mountain]

and those lads with the synchronized headbanging

[steady mountain]

yes it is

I am in Greece currently, on the countryside

internet is not great

ready for some sleep soon?

I think I am

maybe I check on it tomorrow

it is 1:35 am here.

[karmic crown]

they are 1 hour ahead of us (time here is the same as in swe)

[steady mountain]

yeah, it is weekend, but still,

I will check on it tomorrow

I am off to bed.

[torn coral]

can someone explain after reinstalling a1111, faceswaplab doesnt work anymore?
it worked before but now the extension doenst work

[lofty patrol]

I making a brand new Installation of a1111. I installed Controllnet and Copied my Models from the old a1111 ves. 1.6.0 now i want to generate the exact same but this didnt work. Photo 1 Input Photo 2 Output. Same proccess with 1,6,0 version. But see in Picture 3 input und picture 4 output. Why its in the New version like Control net didnt have inpakt on the generation ?

[sonic ferry]

Is there a way to remotely manage one instance of ComfyUI? I turned on --listen but when I connect to it it's a blank slate, not showing what's going on on the instance that the PC is running.

[lavish meadow]

hey need some help, had to reset my pc and lost all my ai stuff is all i need the a111 web ui?

and what model should i use for a high fantasy face dnd style?

[strange thunder]

Regarding regional prompter and latent couple: Is anyone aware if the extensions can map overlapping regions appropriately? Or does the next region just overwrite any region underneath it?

Example: If I have a person lora and a clothing style and I map the person region first and then overlap the same region (minus the head) with the clothing style, will it be able to apply both? Can it do layers is basically what I'm asking. I can run some experiments myself but a) not in front of the system right now and b) it could potentially take a while and someone may already know. lol

[kindred tinsel]

Quick question: Am I allowed to change the name of the .safetensors model file to something a bit more readable if the naming from the author sucks or does it mess something up?

[strange thunder]

Yes you can, but keep in mind that that new name is what will be on the metadata. So 1) if you share it, they likely won't know the model unless you tell them verbally or the model file name is obvious (like realisticvision51.safetensor) and 2) if you wanna change the name of a model, make sure that's what you it to be called and do it right away so that when you use your own metadata they will match up. It's just useful to not have a scenario where older generations have a different model name despite being the same model.

[kindred tinsel]

Oh okay I see, well I dont really share a lot of my art tbh, I tend to feel overwhelmed by the lack of quality and yeah, just practicing rn. As for the other part, I have already used most models and some I havent even started to test so yeah.

As long as it doesnt break my stuff I am down for it haha thanks for the help!!

[strange thunder]

It should be fine. One generation with the exact same parameters with the same model just under the rename will tell you intantly if it's fine.

[lofty oracle]

I attemped to make on online link with --listen but it's been stuck on the same scren for like 10 minutes can anyone help

[kindred tinsel]

But that is already open as far as the eye can see

For the listen, comfy ui listens to the port it opened in your computer. It opens the port to the network. There is a link to the addresss and port up above the last message

[lofty oracle]

No it's for a1111 and also i am trying to get it running on my phone since it is already running on my pc

[kindred tinsel]

for a111 --listen? I think you might want to try out the --share option in your webui.bat file @lofty oracle that will create a link for your a1111 instance that you can access remotely!!

[lofty oracle]

i just did that and i got another error

[kindred tinsel]

Does it say it couldn’t create the file?

[lofty oracle]

'Could not create share link. Please check your internet connection or our status page: https://status.gradio.app.'
gradio is up rn 💀

[kindred tinsel]

Oop

I’m not fully sure how to handle that one chief. I know how to work around the antivirus software stopping it but not this one. Let me take a look at the GitHub

[lofty oracle]

Oh yeah I did get a virus warning
Fixed it working now thanks.

[kindred tinsel]

oh lmao then it was the antivirus blocking the windows server file np!

[heady pine]

Could someone please tell me where the Styles folder belongs

[simple tartan]

hey all, I reinstalled A1111 with DirectML and none of my images are saving the parameters (prompts, settings, etc). Is there somewhere i can fix this? i dont see it in any of the obvious locations

[steady mountain]

sadly, core dumped again

I think I will just format and reinstall on the partition in question

unless this all can get resolved today

[woven smelt]

Hi i have been trying to get openpose and controlnet to work and I cant figure out why they arent. Using automatic111 and depth/canny work fine. Ive tried loading an image, setting preprocssir and Ive tried just loading in a already processed openpose image but in every case SD doesnt even attempt to usecontrolnet. Using model thibaud_xl_openpose. Working with XL models like juggernaut, dreamshaper. Cmd prompt does not show any errors so any suggestions?

[ornate elk]

Did you copied the controlnet models too?
In the first image is says controlnet undefined. Thats why it didn't work

[ornate elk]

Have you checked the metadata of the image. Load one into the PNG-Info Tab

[simple tartan]

Yes, and it says parameters = none

[ornate elk]

Thats strange, you can try delete the config.json and ui_config.json to reset all the settings

[ornate elk]

Make sure to select "None" as preprocessor

For openpose if the image is a pose

[woven smelt]

I tried that as its a common answer but I think the instalaation is broken so im redownloading some stuff now

[ornate elk]

Okay

[simple tartan]

I'm going to redo the whole install, as when i tried to convert a checkpoint, nothing is working right since...

[woven smelt]

While that happens, do you need to use the openpose lora in the prompt for it to work or does it call it itself?

[ornate elk]

No need for a lora for ControlNet

[ornate elk]

Do you need my install guide for AMD?

[woven smelt]

I mention it because there is one on this page https://huggingface.co/thibaud/controlnet-openpose-sdxl-1.0/tree/main

[simple tartan]

if its shorter than wathcing a YT video, yes please. I'm on a 7950X with a 7900XTX

[ornate elk]

Perfect, same gpu as me, give me a second

[ornate elk]

Control lora can only be used in comfyui i think

But not needed for auto1111

[ornate elk]

Here is a quick Guide to install Automatic1111 Directml AMD Webui (Stable Diffusion)

You need to install Git 64bit: https://git-scm.com/download/win
and Python 3.10.11, 64bit: (any python above or 3.10.6) No 3.11, 3.12 https://www.python.org/downloads/release/python-31011/
And check "add python to path" when installing Python.

1. Make a new Folder on your drive (not on Desktop, downloads, documents, Programms, Onedrive) and name it Ai for example:
   C:\Ai\
2. 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.
3. Then you copy and paste this command:
   git clone https://github.com/lshqqytiger/stable-diffusion-webui-directml && cd stable-diffusion-webui-directml && git submodule init && git submodule update
4. Press enter and after its done you can close the cmd
5. 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
   After that save and launch the webui-user.bat inside the stable-diffusion-webui folder.
   After the Installation that can take a while. Youll get an URL http://127.0.0.1:7860
   Thats the webui you open in Browser.

[simple tartan]

i'm on py3.10.6 for now, any reason to go to .11?

I used this previous, trying your method now, thank you

https://www.youtube.com/watch?v=mKxt0kxD5C0

[ornate elk]

You can keep your 3.10.6 if you want

If you run into problems you can still upgrade

[simple tartan]

waiting on first run now

i'm glad you'r ehere now, lol

let me get tings working, and I'll have another question to fix 😄

[ornate elk]

Sure 😄 I can help, I know the limits of the gpu with directml

[simple tartan]

sweet

[ornate elk]

Also In a few months we get Rocm for Windows and everything will work better 🙂

[ornate elk]

20 minutes wtf xD I'll take a look at that later

[simple tartan]

speed it up, dude talks SLOOW

TLD..W, added --use-directml and --onnx to bat file, added torch-directml to requirements_versions.txt, and commented out low_vram option for the 1.5 model

[ornate elk]

Oh rip xD he used olive onnx

[simple tartan]

is that bad?

it was working great for me this morning

[ornate elk]

While that is faster, its so incompatible to all the features and your stuck on locked resolutions and specific models.
Directml is slower but is compatible with mostly everything and gets better outputs and you can use every community made model

[simple tartan]

ah, how much slower?

let me take it back off and try

[ornate elk]

Not really slow. But you get 3-4it/s
While on olive you get like 16it/s

[simple tartan]

yea, thats what i thought, thats REALLY slow by comparison

[ornate elk]

Yea you have to decide for now.
Best speed?: Shark webui on windows.
Best usage on windows: directml.
Best usage and speed on Linux: Auto1111

Wouldn't recommend anybody to use olive. As the Shark webui is much more advanced and is also faster

[simple tartan]

i've used shark a little, and i'd love to roll linux as my main OS, but my CAD software can't run in nix and cant run virtualized 😦

[ornate elk]

Can it run via Wine or bottles ?

[simple tartan]

not

Autodesk fails

[ornate elk]

Ah damn

Than you could dualboot

[simple tartan]

i hate shutting down more than i hate windows

[ornate elk]

Oh okay xD

[simple tartan]

restarting the process without onnx for now

[ornate elk]

Have you followed my install guide with the cmd args?

[simple tartan]

doing it now, i had added --onnx previously to your steps 🙂

[ornate elk]

Okay, you may need to delete the venv folder then to clear out the onnx stuff

[simple tartan]

i blew it ALL away

going to merge my extensions, LORA's, and outputs over next, waiting on first run still

i'll build a duplicate environment later that is onnx, and use it for testing, and leave this one alone once it works properly

woudl love to use ROCm and call it a day

[ornate elk]

Good idea

[ornate elk]

Yea rocm is the saviour haha xD

[simple tartan]

whenever we finally get it....

[ornate elk]

My guess is end of January. And then the webuis and ml stuff need to support it

[simple tartan]

true, so early feb 😄

[ornate elk]

Yea probably

But for me the directml version know is enough to play around with

Its still faster than my GTX1080

[simple tartan]

true, but we should be spanking 4090's with the 7900XTX

[ornate elk]

True, 20it/s is possible on linux

[ornate elk]

With --ckpt-dir "Path/to/models"
You dont need to copy models over

Symlinks are also supported

[simple tartan]

where are styles stored, I'm missing those still

[ornate elk]

Styles.csv

In the stable-diffusion-webui folder

[simple tartan]

got it

almost back to where i was, about ready to test

[ornate elk]

If you plan using highres fix later, you need the Tiled Diffusion Extension with Tiled Vae

But I can give you my screenshot of the working settings

[simple tartan]

lol, using it now and waiting for this really slow 8it/s

[ornate elk]

If it errors out you need to relaunch the webui to clear the stuck vram

512x768, upscaled by 2, hires steps:10, denois:0.5 should work without problems

For anything higher the tiled vae is needed

[simple tartan]

standby, you're being too helpful, too fast 😛

[strange thunder]

Regarding regional prompter and latent couple: Is anyone aware if the extensions can map overlapping regions appropriately? Or does the next region just overwrite any region underneath it?

Example: If I have a person lora and a clothing style and I map the person region first and then overlap the same region (minus the head) with the clothing style, will it be able to apply both? Can it do layers is basically what I'm asking. I can run some experiments myself but a) not in front of the system right now and b) it could potentially take a while and someone may already know. lol

[simple tartan]

is there a good extension to get a gallery like shark has?

[ornate elk]

There is this, but idk how good it is:
https://github.com/dr413677671/PromptGallery-stable-diffusion-webui

[woven smelt]

I am reinstalling everything. This page says to add some quick settings like backend but in my version they are not available https://github.com/vladmandic/automatic/wiki/SD-XL I want some way of verifying that I am using SDXL

[ornate elk]

What's your GPU?

[woven smelt]

2080 ti

[ornate elk]

My guide should be enough, sdxl is just a model you download and put into the webui

[woven smelt]

Ive done that with this new install along with the refiner

[ornate elk]

The link you postet is for Vladimanics fork of auto1111

[woven smelt]

I wasnt sure what was broken with the old one so I made a fresh one, added in controlnet, downloaded the contronet models including openposev2, the sdxl base and refiner and dreamwaker

[ornate elk]

My guide is for the normal automatic1111:
#🤝｜tech-support message

[woven smelt]

I didnt realise that guide was different I had just come across it on reddit but ill look at yours

Yes I cloned a new git repo, python is ver 3.11.7 and in path.

[ornate elk]

Thats a not supported python version

[woven smelt]

well shit

[ornate elk]

3.10.6 up to 3.10.11 is supported

If you follow my guide then you also need to edit the webui-user.bat
At the line COMMANDLINE_ARGS=
You need to add: --xformers --medvram-sdxl --no-half-vae
Then your ready for SDXL

[woven smelt]

Ok cool, chatgpt is telling me how to change pythoin version so ill do that first

[ornate elk]

Just uninstall the wrong version

[woven smelt]

yeah i use winget so thats easy but I wasnt sure if winget installs a specific version, but it does

[ornate elk]

And install the right version with "add python to path" checked

[woven smelt]

winget install auto does that

[ornate elk]

Good

Winget is really nice

[woven smelt]

yes. never have to dl an installer ever again. Although installing git from it wont add git to path, its per app as to wehether its set or not

[ornate elk]

Ah okay

[simple tartan]

well...i'm not getting the 77 token error i was getting on my onnx intsall, but the images are slow as dogshit rolling uphill, and the quality is horrific now

[woven smelt]

Had to change python paths in venv that it cached, now running into this

[simple tartan]

using teh 1.5 pruned full model

ONNX

new build

[woven smelt]

I think that model is dogshit right?

[simple tartan]

teh 1.5 one?

i have 128GB of ddr5 and 24GB of GPU ram, gimme the biggest baddie 😄

[woven smelt]

yeah i immediately got new ones from civit, tons better

[simple tartan]

why was is so good on onnx then?

[woven smelt]

But other people will know better which ones are the best. Juggernautxl and dreamshaper seem very popilar

i just assumed that the pruned one is like a placeholder

[simple tartan]

onnx didnt have pruned, it was 1.5, but i cant find the direct download

it was internal to the onnx plugin

[woven smelt]

no idea on that one sorry

[ornate elk]

Lol don't use the 1.5 pruned full model and refiner for 1.5 with hires fix.
You should use community made models that are much better in quality. Then dont use and a refiner while using hires fix.
Also you can't upscale to 2k with hires fix on Directml

[ornate elk]

Delete the venv folder and relaunch

[woven smelt]

I recloned the whole repo instead since its fresh anyway

[ornate elk]

Okay

Does it work now?

[woven smelt]

getting the big checkpints etc back in place first

[simple tartan]

i did with onnx

trying different settings now

what kind of gundam abomination is this?

[ornate elk]

Wtf xD

Getting stuff like this

[simple tartan]

wtf is happening on my install then?

[ornate elk]

idk what are your settings?

[simple tartan]

i went from instagram quality to wtf is this

[ornate elk]

Does the cmd shows any errors when switching models?

[woven smelt]

workss great now

[simple tartan]

none i can see

[ornate elk]

Can you try to set everything to default and prompt for a cat, then change the settings again starting with hires fix

[simple tartan]

Before hires:

after:

running out of gpu memory now

[ornate elk]

Activate tiled vae, encoder to 1024 decoder 128.
That works for 540x960 to get FullHD
But you won't be able to upscale 1024 by 2

That has to be done in img2img with the SD upscale script

[simple tartan]

hmm, a lot of stuff to figure out now 😄

[ornate elk]

Feel free to ask if you have any questions 🙂

[simple tartan]

only question is why did onnx break and give me all these new headaches today 🤣

i'm thinking i need to go back, its taking this version 15+ minutes to generate the same size images

ok..wtf... i rebuilt the onnx directory and its working now....

well....aside from not saving parameters in the png

[ornate elk]

15 minutes wtf

What settings did you used?

[ornate elk]

Onnx isn't good supported in the directml version. The dev also dont will improve the integration as he waits for Rocm on windows. So olive is dead end

Shark is more future prove as AMD officialy acquired Shark a month ago

[simple tartan]

In order, ONNX(pink hair): 7.6 seconds. DirectML(wtf pic):4m36s. Shark:15.39s(blob)

[lavish meadow]

where do i put loras again, i forgot...

[ornate elk]

Okay xD idk your settings but your directml isn't working correctly it seems.
But your cat image was good

[ornate elk]

Models/lora folder

[simple tartan]

what do you mean by settings, my prompt?

nothing modded

[ornate elk]

The txt2img settings like cfg, resolution, upscalers etc

[simple tartan]

just a few extensions and my LORA's

same as last image, one sec, i'll take curren pic

[ornate elk]

You should try figure out what is causing that issue. The cat with hires fix was fine before

[ornate elk]

Try Latent cubic as upscaler

[simple tartan]

bicubic?

[ornate elk]

Also 10-15 hires steps is faster and still good

[simple tartan]

i'll try that next run, just clicked generate

[ornate elk]

ok, also nice to know, when tiled vae is enabled, clicking interrupt is pretty fast and works better than without tiled vae

[lavish meadow]

thank you

[simple tartan]

clicked it, still waiting

there we go

and generate again

wont generate now, no gpu mem available

ffs

[ornate elk]

ah, restart required

[simple tartan]

yep, generating now

[ornate elk]

this image took me 1:23 mins

[simple tartan]

i think my prompt is too long and getting truncated maybe

the image gets better the simpler it is

i have a load of weighted values

copied right out of the lora

[ornate elk]

ah okay

yea dont overdo weighting

keep it below :1.5

[ornate elk]

keep the tokens below 225

[simple tartan]

no lpw for directml is there?

[ornate elk]

lpw?

[simple tartan]

long prompt weights

[ornate elk]

there is no prompt limit in auto1111.
onnx and shark have a 75 token limit

its just not usefull to use mega prompts xD

also more tokens = more vram

more loras = more vram

so if you go on 225 or more its more likely to error out

[simple tartan]

i'll have to tweak a bit then

bicubic made the images super small too

making icons now or something

[ornate elk]

smaller?

[simple tartan]

yea, i might do a web session tomorrow, watching new years stuff now, waiting for the ding

already ate my soba 😉

[ornate elk]

okay, here are my upscale settings rn, you can also switch to Resrgan4x+

[ornate elk]

soba 😄 nice!

i think you have to play around with your prompt, testing whats possible an what not

[simple tartan]

i just want to get back to the quality i was getting in onnx

[ornate elk]

with the settings of the screenshot

[simple tartan]

and back to 20it/s

[ornate elk]

onnx doesnt change quality, its just performance, you can get the same quality on a CPU and on a 4090, on cpu it would just take ages xD

[lavish meadow]

How do i turn an image to anime style as i wanna turn my bg3 character to that style?

ive never done img to img

[ornate elk]

put the image in img2img, then prompt for the character and sytle.
then play around with the denois.
the lower the denois the more it will look like the input image. the higher the more will change

[lavish meadow]

thank you, ill try it out

so denoise to 50%?

[ornate elk]

would be a good start

then adjust to your liking

also choose a anime model

[lavish meadow]

what models would you reccomend?

im looking for a fantasy dnd style like this

its not ai i believe but i love the syle and wonder of it

[ornate elk]

hmm, maybe try Cetus-Mix, Toonyou, Animerge,

[lavish meadow]

ill check those out now

thank you for the help

[ornate elk]

no problem 🙂

[ornate elk]

that style is pretty nice, you will need a good prompt for it too

[lavish meadow]

yeah itll be alot of trial and error

the style is downright gorgeous

[simple tartan]

dude, this tiled vae is soooo slow

[lavish meadow]

what is vae again?

n should i mess with it

[lavish meadow]

looks good

[ornate elk]

vae is for colors

[lavish meadow]

i like colours

[ornate elk]

you only need a few ones:

these are the most used and best. for realistic stuff use 84000-mse

for vibrant colors use the kl-f8-anime

[simple tartan]

i'm not using any specific one....

[lavish meadow]

how do i use them?

[ornate elk]

you download them and put them into models/vae folder.
then go into Settings-> User Interface ->Quicksettings, there add sd_vae
Then hit Apply and reload UI

then you have a dropdown to select them

[ornate elk]

why your running api stuff there too

or is that from an extension?

[lavish meadow]

anything i should mess with to make it run quicker, my pc has a 3060 and i310105f

or anything i should mess with to make it overall better in general

[simple tartan]

probably an extension

[ornate elk]

3060 with 12gb ? then you only need --xformers in your webui-user.bat

[lavish meadow]

yeah 12gb

[lavish meadow]

what is xformers?

[ornate elk]

its a performance boost, to activate it you have to edit the webui-user.bat
at the line COMMANDLINE_ARGS=
you add: --xformers

with that your generate faster, and be able to upscale larger images, and using controlnet faster or other heavy extensions

[lavish meadow]

ooo i see

[karmic crown]

are you using this one, or that is some other vae? https://civitai.com/models/118561?modelVersionId=131654

[lavish meadow]

how do i edit it?

[ornate elk]

ive downloaded them from their original upload places, on huggingface

[ornate elk]

right click, edit 😄

on windows 11 its under "more options"

[karmic crown]

this one i use is some kind of mix

[lavish meadow]

thank you, i aint that bright

do i add it next to it or underneath..

any spaces......

yknow what we ball it breaks it breaks

[ornate elk]

Next to the =

[lavish meadow]

bless you

do i need to restart it?

[ornate elk]

If your webui isn't updated you'll get an error

[lavish meadow]

gotcha gotcha

[ornate elk]

But every error is fixable if you get one

If youre new in SD then your webui should be already updated

[lavish meadow]

I aint exactly new had to wipe my pc for a fresh start

just been a while since i did this

but yeah i got the new version

[ornate elk]

Thats hires fix in the image, not tiled vae

The upscaling takes the longest

[simple tartan]

ah, well that tooo

lol

[ornate elk]

Tiled VAE kicks in the end

[ornate elk]

For good quality you should use hires fix

Also the extension boorutag autcomplete can help with prompting or finding words

[lavish meadow]

already do use hgh res fix

tho idk if im using the right

upscaler

[simple tartan]

its just not there, quality in onnx is leaps ahead, and faster. this is the exact same prompt, sampler, checkpoint, and settings

Happy New Year!

@ornate elk i found a difference, onnx is using torch 1.13.1 and your version is using 2.0.0

i think i'll downgrade and try again

[ornate elk]

Torch shouldn't be an issue for such a change

You've seen my images with torch 2.0

No artefacts

[simple tartan]

wasnt there an issue with torch2 not supporting diectml yet?

[ornate elk]

Nope

Onnx dont supports 2.0

[simple tartan]

ah

well wtf again

lol

two loras and 30 tokens total

[karmic crown]

can you post the result text for both?

lets see what is different

[simple tartan]

the what?

still learning here, where is that?

[karmic crown]

this thing

it must be different if you are getting different results with same prompt

i want to see what is different

[simple tartan]

the onnx doesnt really have that

Time taken: 5.0 sec.

A: 0.05 GB, R: 0.05 GB, Sys: 0.1/0.0546875 GB (100.0%)

[karmic crown]

post the one from a1111

[simple tartan]

Steps: 20, Sampler: Euler a, CFG scale: 6, Seed: 2934842649, Size: 512x512, Model hash: 1bb2108643, Model: perfectdeliberate_v5, Lora hashes: "Japanese-doll-likeness: 0450e1183d1a, gundam: ef694f6150c8", Version: 1.7.0

[karmic crown]

can you trye with cfg of 4?

is it better or the same?

[simple tartan]

i was going as high as 12 on onnx

here's CFG4

onnx at cfg4:

[karmic crown]

can you try without loras, just for test

letst try to get something decent first, than we will find the problem

[simple tartan]

a1111:

onnx:

[karmic crown]

ok, that is much better

[simple tartan]

agreed

[karmic crown]

try with cfg 6 now

[karmic crown]

btw, did you download another version of model or lora, or you coppied the existing ones?

[simple tartan]

copied over from other directory

[sturdy ginkgo]

if using SDXL definitely make sure the Lora is for SDXL .. you get an insane quality decrease if using SD1.5 Lora's

[karmic crown]

that is 1.5 model, so he should use 1.5 loras, but i agree with your statement

[karmic crown]

when you open loras tab, do you see both loras you are trying to use there or not?

[simple tartan]

here's one lora added back in:

yes

[karmic crown]

are both visible here?

[simple tartan]

yep

[karmic crown]

ok

show me your propmt please

[simple tartan]

pink eyes,beautiful detailed eyes,1 girl,smile,cute,hair bangs,long hair,wearing robot armor,red armor with silver accents,lora:gundam:1,

https://civitai.com/api/download/models/31131

that's the lora

[karmic crown]

btw i think we should continue this conversation in https://discord.com/channels/1002292111942635562/1011743094309396631 since your installation works fine (unless you add lora)

[simple tartan]

ok, lets go, tahnk you

[sturdy ginkgo]

someone know how to solve this (comfyUI, SDXL ip-adapter plus model, SDXL ClipVision model)
Error occurred when executing IPAdapterApply:

Error(s) in loading state_dict for Resampler:
size mismatch for proj_in.weight: copying a param with shape torch.Size([1280, 1280]) from checkpoint, the shape in current model is torch.Size([1280, 1664]).

File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 153, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 83, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\execution.py", line 76, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI_IPAdapter_plus\IPAdapterPlus.py", line 636, in apply_ipadapter
self.ipadapter = IPAdapter(
^^^^^^^^^^
File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI_IPAdapter_plus\IPAdapterPlus.py", line 272, in init
self.image_proj_model.load_state_dict(ipadapter_model["image_proj"])
File "D:\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 2152, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(

[lavish meadow]

whats the best sampling method to use for high quality like masterpiece stuff

[karmic crown]

depends on the model
unfortunately some models does not work with all samplers
in many cases sampler does not affect quality if there is enough steps

[lavish meadow]

gotcha gotcha thank you

[stone hawk]

I'm trying to use a Lora that requires "peft", but I could not find a way to install it for my needs. Can anybody help me out? I use Automatic1111

[sturdy ginkgo]

Lora's usually don't require anything extra

[karmic crown]

please send link to lora

[lavish meadow]

Anyone know a model that does dnd races like kobolds or dragonborn?

[sturdy ginkgo]

https://civitai.com/models/90/dungeons-and-diffusion-v3

[lavish meadow]

thank you

[lost crescent]

ComfyUI v0246 frozen - cannot zoom in/out/pan etc nor add any w/flows

All up-to-date

Win 10 8Gb VRAM GTE RTX 2070, 64Gb RAM

[lavish meadow]

is it possible i can make my own model?

so i could maybe replicate this style

[sturdy ginkgo]

https://www.reddit.com/r/StableDiffusion/comments/10cgxrx/wellresearched_comparison_of_training_techniques/

good luck 🙂

[lavish meadow]

dear god

ill try

[sturdy ginkgo]

for a style Lora and Embeddings are probably the easiest and fastest

[steady mountain]

Anyone running Automatic1111 on a ubuntu install with z790, I9 13900K and an 7900xtx feel free to DM me. I am curious how you made it work on your rig

[manic monolith]

excuse me, did you ever get a fix for this 'NoneType' object has no attribute 'lower'?

[gilded river]

does anyone know why ControlNet is not being applied?

I generated an image with the same seed with and without controlnet enabled and it outputted the same image

I am probably missing something obvious but I'm not finding it

[ornate elk]

you need to select a 1024x1024 resolution maybe, as thats what sdxl is trained on

[gilded river]

I set it to 1020, you think that's what's fucking with it?

[ornate elk]

In the image it says 512x512

[gilded river]

good catch but I don't think that was it

enabled and disabled, same seed

[ornate elk]

can you send me the input image?

ill try

[ornate elk]

hmm works for me

[gilded river]

its something on my end

I cant figure it out though

[gilded river]

did you do anything besides enabling it?

[ornate elk]

i set preprocessor to None, as my input image is already a pose

i also used no lora or anything

only an sdxl model + sdxl vae

1024x1024

[gilded river]

send me that image and I'll import the settings

[ornate elk]

here

[gilded river]

my controlnet is fucked

[ornate elk]

update your controlnet extension

also is your webui updated?

[gilded river]

I checked for updates

[ornate elk]

okay, whats your webui version?

you can check at the bottom in browser

[gilded river]

1.7.0

[ornate elk]

does the cmd shows any error?

for me it looked like that

[barren rover]

keep getting this error

[ornate elk]

amd or nvidia?

whats your gpu?

[barren rover]

nvidia

[ornate elk]

model?

[barren rover]

this is it right?

[ornate elk]

okay then you need to edit the webui-user.bat (right click)
at the line COMMANDLINE_ARGS=
you add: --xformers --medvram --no-half

thne save and relaunch

[gilded river]

I am redownloading controlnet gimi 1 sec

[barren rover]

i think i already have the xformers thing

wait a sec

[ornate elk]

okay looks good

oh wait i see the error is from RAM not Vram

[barren rover]

i already had it

[ornate elk]

how much Ram do you have?

[barren rover]

but im getting this error now for some reason

[gilded river]

that gpu has 4 gig

[ornate elk]

i mean RAM nor vram

[barren rover]

8 gb

[ornate elk]

yea thats on the edge

you need to increase your windows pagefile

also make sure to have 15-20gb free space on C drive

[barren rover]

how?

[ornate elk]

https://mcci.com/support/guides/how-to-change-the-windows-pagefile-size/

[ornate elk]

First make sure its set to System managed for C: and disabled for all other drives.
Then you can increase it for C: to 16000 min and 20000 max
Then restart the PC

Make sure to have at least 15-20gb free Space on the C drive

[ornate elk]

and does it load the controlnet correctly?

[gilded river]

oh my lord

oh wait I forgot to re-load the models

yea ControlNet is fucked

[ornate elk]

yesss

[gilded river]

I don't know what's going wrong but it is

[ornate elk]

then you should reinstall controlnet

[barren rover]

restarted the pc now the webui keeps shutting down

[ornate elk]

you have the wrong python version installed

3.11 is not supported

[barren rover]

where and how do i install the right one

[ornate elk]

3.10.6 up to 3.10.11 works

[ornate elk]

https://www.python.org/downloads/release/python-31011/

[barren rover]

do I have to reinstall it or just dorp the new stuff into the folder

[ornate elk]

uninstall the 3.11,
install 3.10 with "add python to path" checked,
delete the venv folder and relaunch the webui-user.bat

your webui is also not updated

you can do that by adding
git pull
into the webui-user.bat at an empty line before webui.bat

[barren rover]

is the 3.10.9 supported?

apparently i have two pythons

ill unninstal the 3.11

[gilded river]

@ornate elk

this is embarassing, when I pulled the model I pulled the wrong thing

I am a god damn idiot

[ornate elk]

Ohh what did you download?

[gilded river]

the link to the download screen

https://tenor.com/view/patrick-sundae-gif-25035574

[ornate elk]

It looked correct in your screenshot still

Good that it work's now ^^

[gilded river]

beacuse I did this

so it kept the same file name

[ornate elk]

Ohhh, well

[gilded river]

clown

either way thanks for the help homie

[ornate elk]

No problem 🙂

[shy hemlock]

Can anyone help me redirect my .bat file to the python location on my PC for stable diffusion...I had to move everything for space reasons, and the bat is lost

[barren rover]

did all of that and it keeps shutting down

[stark parrot]

ANYONE KNOW HOW I CAN USE MY STABLE DIFFUSION CHECKPOINT DIRECTORY FOR ''FOOOCUS''?

[shy hemlock]

How do i downgrade my Python from 3.12 to 3.10?

[simple tartan]

what OS?

[simple tartan]

ok, i'm trying REALLY hard not to stick with onnx, but the images are still extrely different, using the same ckpt and lora's, what gives. left is onnx and right is A1111

both windows with settings

changed to use same sampler and still.....

2 negatives for ONNX: no PNG info, and no output beneath render

[sturdy ginkgo]

get the trial of repo uninstaller (or the portable pro version from the high seas) and uninstall using that .. delete everything it finds, reboot pc, install python 3.10, make sure to select add to PATH, reboot pc to be sure it added to PATH (check in your system properties, advances properties, environment variables).

[simple tartan]

it really depends on his OS first though, there is no PATH in OSX or nix

windows, uninstall 3.12, then install 3.10, easiest solution

linux, apt/pacman/yum and install fixed version

OSX....buy a real PC

[sturdy ginkgo]

mac .. pay $4999 for a new one

[simple tartan]

gald we agree 😄

[sturdy ginkgo]

stand , power cable, and accessories sold separate

[simple tartan]

reminds me of a joke "if Apple sold cars, you'd have to buy the doors and seats separately"

[sturdy ginkgo]

to tank gas you need to buy the $699 proprietary gas adapter

[simple tartan]

lol

still trying to get my A1111 to work properly, colors and quality is still absolute shit

i've tried just about every sampler now

[sturdy ginkgo]

what VAE are you using?

[simple tartan]

"auto" aka, none specified

its at the top of the screen

I'm tempted to try a WSL build of SD and see how that performs

[sturdy ginkgo]

goto civitAI and grab any VAE

these fix color grading issues

no vae means it uses baked VAE if it exists or else just a grey image overlay

[simple tartan]

ok, doing that now, and will apply to both installs

[sturdy ginkgo]

vae-ft-mse-840000-ema-pruned_fp16 for realistic

kl-f8-anime2.ckpt for anime

sdxl_vae for sdxl

[simple tartan]

i grabbed the chillout since i'm using that model

but it reverts to "automatic" as soon as I select it

ok, used the vae ft one, and onnx reverts to automatic, but the a1111 keeps it, trying now

nope, even with defied VAE, its crap

[simple tartan]

I think when the experts are online in going to try and build a nix instance, as much as i hate shutting down, I really want to explore the capabilities of SD, as well as SDXL, and i can't get both to work in the same instance in windows currently

[wicked flint]

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.

anyone know how to fix this?
im usin AMD GPU

[ornate elk]

Are you sure the lora works on the onnx version?
Can you try the same seed and stuff just remove the Loras after

[simple tartan]

yea, even without the loras the quality is preemo on onnx

[ornate elk]

For me it looks like the loras dont work on the onnx version and break the image on the Directml

[simple tartan]

teh onnx is teh beautiful one, teh directml is the one thats garbage

funny enouhg, i disabled them on both, and now the directml has better quality

[ornate elk]

Try to lower the Lora weights of boath to 0.3 ans try again

On Directml

[ornate elk]

Thats what I'm saying. The loras are the Problem as they are overtrained.
You have to lower the lora weight to 0.5 or 0.3 for them to not break the image

Not every lora works good with :1

[simple tartan]

holy shit, that actually worked, 0.3 on both loras, now I'm getting this on the NON onnx version

and here's onnx

so directml is better now....

[ornate elk]

Yes, it was just the lora strength set to high

[simple tartan]

crazy, well, TDIL

guess i can shutdown onnx verison now

is it worth it for WSL?

my "work time" project is to build a ubuntu drive, got a spare 2TB NVMe, so i'll use that and try later this week

it's only gen4, first world problems 😄

[ornate elk]

I heard nobody that got WSL with AMD and SD working

[simple tartan]

Figures, hard to get ring zero when containerized

[ornate elk]

Who knows what's missing but you can get the LLM webui working on WSL

So I wonder why not auto1111

[simple tartan]

I'll just stick to the "next stage" being arch or Ubuntu directly

[ornate elk]

Yea thats a good idea

[simple tartan]

Trying to read if there's any gotcha between the distros for SD

[ornate elk]

Most use ubuntu, I would prefer arch. But there you need some additional steps I think

[simple tartan]

Are you my twin? 😂

[copper spoke]

Arch Gang Yewww

[ornate elk]

With symlinks yes

[copper spoke]

I use Arch BTW

[ornate elk]

What model did you tried to load?

[ornate elk]

You can edit the webui-user.bat
In there you can specify the path to python.exe at the line PYTHON=

[ornate elk]

What model did you tried to use?

[simple tartan]

can I set absolute path in the webui for the output dirs?

ie. T:/outputs

[ornate elk]

In the settings, yes

[simple tartan]

ok good, want to use my synced directories for all my PC's so i can hit one directory to view the images

thank you for all the help, now its down to tweaking my prompts, adding extensions, and tuning settings. Here's my typical output now, much closer to what I was trying to create:

[ornate elk]

Looks good 🙂 no problem

[wicked flint]

dreamshaperXL_alpha2Xl10.safetensors [0f1b80cfe8]

[ornate elk]

With loras or not?

[wicked flint]

without

[ornate elk]

Can you show me the txt2img settings when you get the error?

[wicked flint]

oh it was img2img, sec i'll open the ui again

[cloud lake]

I need help

Could anyone help me

[sacred sierra]

OK.. Reinstalled and . .. errors again on first run. (RuinedFooocus 1.28.2)

Using xformers attention in VAE
missing {'cond_stage_model.clip_l.text_projection', 'cond_stage_model.clip_l.logit_scale'}
left over keys: dict_keys(['cond_stage_model.clip_l.transformer.text_model.embeddings.position_ids'])
Base model loaded: sd_xl_base_1.0_0.9vae.safetensors
App started successful. Use the app with http://127.0.0.1:7860/ or 127.0.0.1:7860

[solemn moon]

whatsup

[ornate elk]

The resolution is to small

Try 768x768 or 1024x1024 for sdxl models

[solemn moon]

https://gofind.ai/stable-diffusion/sdxl-resolutions

[vestal viper]

need help, i finally got SD to run and generate image but this is what i got

this is cmd screenshot from that all black result

[ornate elk]

hey, whats your gpu and whats inside your webui-user.bat?

[vestal viper]

--medvram --precision full --no-half --no-half-vae --opt-split-attention-v1 --opt-sub-quad-attention --disable-nan-check --skip-torch-cuda-test

rx 6600, got those command lines from reddit + added skip torch

[ornate elk]

okay, these are not the correct commands for your gpu, also you installed the wrong webui (the one for nvidia). You can delete the whole folder and start a fresh install with my guide for AMD:

Here is a quick Guide to install Automatic1111 Directml AMD Webui (Stable Diffusion) on Windows:

Preparation:
You need to install Git 64bit: https://git-scm.com/download/win
and Python 3.10.11, 64bit: (any python above or 3.10.6) Not 3.11, 3.12
https://www.python.org/downloads/release/python-31011/
And check "add python to path" when installing Python.

Important:
If you tried to install SD before and failed (or installed the nvidia one) do this:
open up a cmd and type
pip cache purge
to remove the nvidia stuff.

1. Make a new Folder on your drive (not on Desktop, downloads, documents, Programms, Onedrive) and name it Ai for example:
   *C:\Ai*
2. 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.
3. Then you copy and paste this command:
   git clone https://github.com/lshqqytiger/stable-diffusion-webui-directml && cd stable-diffusion-webui-directml && git submodule init && git submodule update
4. Press enter and after its done you can close the cmd.
5. 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 you GPU has only 8gb Vram and you run into out of memory, you may need to change --medvram to --lowvram.
6. After that save and launch the webui-user.bat inside the stable-diffusion-webui folder.
   After the Installation, that can take a while is done, the webui will autolaunch in your Default Browser.

[vestal viper]

oooh alright, glad this time someone actually fast responded with detailed instruction. thank you very very much!

[uneven delta]

how to use sdxl

¿

[ornate elk]

download an sdxl model and put it into the folder with the other models

[uneven delta]

can i use multiple models in one image

[ornate elk]

nope

[uneven delta]

:(

[ornate elk]

no problem, feel free to ask if something isnt clear enough 🙂

[random idol]

Hey, a question
I'm tryna train SDV2 and one of the steps saying each instance of picture should have a unique identifier
Like this: https://i.imgur.com/d2lD3rz.jpeg
Tho my question is, for example I want to train a load of icons, but they are different icons and don't represent the same thing or meaning, but they all have the similar style.
Does this mean I should use same identifier for all of them while naming it or each should have their own identifier as name based on their meaning?

[cloud lake]

Could you help me to make a picture more details by using stable diffusion

[fair oxide]

anyone know a fix for this?

[ornate elk]

What's your python version?

[fair oxide]

3.10.6

i was told that is the one that works

[ornate elk]

Can you open up a cmd and type
Pip cache purge
Then delete the venv folder and relaunch the webui-user.bat

Then show me a full screenshot

[fair oxide]

ive had sd working before and then uninstalled everything including python before resetting my pc and now it dont work

yeah 1s

same thing

should i try a different version of python

[solemn moon]

whats your current methodology?

what pic?

[ornate elk]

Yes install 3.10.11

Then delete the venv folder again

And run pip cache purge again

Then relaunch

[fair oxide]

the exact same thing

[ornate elk]

And you ticked "add python to path"?

[fair oxide]

yes

[ornate elk]

Make sure you dont have a third party antivirus active. Also if your from Asia you need to use a VPN for install

[fair oxide]

nope just windows defender and real time prot is turned off

not from asia but i can try a vpn

same thing

this is my webui-user.bat file

i pur the right path for python no?

[ornate elk]

Wait

I know what's the problem lol

You installed python 32bit

Only python 64bit works

[fair oxide]

ah

okay

one second

[cloud lake]

Check your dm

[solemn moon]

https://tenor.com/view/shannon-sharpe-undisputed-nope-funny-face-gif-536487951695337499

[fair oxide]

@ornate elk ty it worked ill lk if i run into any more issues

[ornate elk]

Np

[ornate elk]

What's your GPU?

[fair oxide]

rtx 4060

[ornate elk]

Then add --xformers to your webui-user.bat

[fair oxide]

already did

[ornate elk]

For best performance

[fair oxide]

i used to use xformers and no half because my old gpu wouldnt run sd without using no half

i just upgraded my gpu yesterday

[simple tartan]

eh, people still run 32 bit OSes?

[woven smelt]

Is there logic to how loras interact for the art style? I have a 3d character lora and I was trying tags like realistic but it always looked 3d, when I loaded another lora in that is based on real life it suddenly changed to IRL photos without even needing further prompting? Does the order in which these are loaded matter? Can I force one way or the other somehow?

[simple tartan]

ok, trying SDXL now, aside from loras not working, expected since most of mine are 1.5's, but it seems hires.fix doesnt work either, keeps returning "RuntimeError:The parameter is incorrect."

[ornate elk]

Dont use --no-half for the new one

It makes it slower

[fair oxide]

yeah i wont

[woven smelt]

Seems if I lower the weight of the model it takes away the animeness, or if there are other models in there it must be a lower weight in comparison

[ornate elk]

You need to play around with the lora weight

[woven smelt]

Another question, what prompt should I use to get someons whole body in a shot, I put full frame shot but often it cuts off feet or sometimes head or lower half

[karmic crown]

i stoped using prompts fot that, just use hypernetwork, this one is good for what you want https://civitai.com/models/140117/latent-layer-cameras

[barren rover]

hassakumodel v13

[ornate elk]

Hires fix with sdxl will work but you need to remove 1.5 loras and maybe change the upscaler

[simple tartan]

using lanczos

and only one XL lora

[ornate elk]

Also you can't do upscale by 2 without running out of vram, so choose upscale by 1.5 and then set the encoder to 1280

[simple tartan]

even trying tiled vae

2048/256

[ornate elk]

Try esrgan as upscaler

[simple tartan]

ok, seems the prompt is also ignoring the ALONE part lol keep getting two people

[ornate elk]

In the screenshot it tries to load another model at the end

[woven smelt]

@karmic crown that is for 1.5 and im using xl but thanks ill look into equivalents for xl

[simple tartan]

1024x1024 image is great, 1920x1080 i get two people.... so much for making wallpapers

[ornate elk]

You need to use lower res and then upscale

Everything higher than native trained resolution will end in doublicates

[simple tartan]

ah

and thats using img2img right?

[woven smelt]

For upscaling I was using extras tab and some downloaded upscalers that did an amazing job. Alternatively can do in img2img but I think its was more resource intense as its generating new images.

[ornate elk]

Upscaling works with hires fix in txt2img and SD upscale script in img2img

[simple tartan]

ok

[ornate elk]

Extras can be used as last step

Its not for quality upscaling

But for resolution

[woven smelt]

what workflow do you reccomend then? 2x using one of those methods then extras after that?

[simple tartan]

so where do i go from a 1:1 to a 16:9 resolution?

[ornate elk]

Generate with hires fix > send to img2img and use SD upscale Script until you have your favour resolution > then send to Extras for max res like 8k or so

[simple tartan]

got it, thank you

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.

[woven smelt]

im going to give that a go now to try it out

you can get NaN if you just started SD and go right to img2img, if you generate a blank img in text2img first it fixes that

[simple tartan]

i've been running images

[ornate elk]

What are your settings?

[simple tartan]

I tried resize and fill to 10801080 from 10241024

[woven smelt]

should be resize or resize and fill?

and which model do you reccomend for the hires fix?

[simple tartan]

I don't want the main character size to change, I want the background and objects to fill though

[woven smelt]

you want to stretch the image then? then there is a script in img2img called outpainting

[simple tartan]

Ah, let me try that

[woven smelt]

im guessing I should be able to make a script that takes a set of generated images, reruns them through t2xtimg with upscale then to img2img, that makes me wonder if it exists in the UI somehwere

[hallow latch]

Hey, I got a question. I cannot generate anything when I got 2 mc client running? But once I close one it start generationg

Can I do something about that?

[simple tartan]

still getting nan errors with the scripts, calling it a night

[ornate elk]

You can show me your settings

I can check

[lost crescent]

My ComfyUI is not zooming in or out!! Manager works OK but cannot reposition nodes to work on them

[torn coral]

what kind of extension are u guys using for face swapping?

stable diffusion

[cold prawn]

ModuleNotFoundError: No module named 'encodings'

any idea what's this error and how to fix it?

[sturdy ginkgo]

what WebUi and are you using a Venv?

[cold prawn]

I'm using stable diffusion webui automatic 1111 v1.7

Venv?

also my python is v3.10.6

[sturdy ginkgo]

goto the webUI folder and open a command prompt or terminal: type:

python .\venv\Scripts\activate.bat

wait for the terminal to say: (venv): and type python pip -m install encodings

[cold prawn]

sec let me try

[sturdy ginkgo]

O_O

[karmic crown]

no no and no 🙂

remove python

venv\Scripts\activate.bat

run this

[cold prawn]

ok now it has the (venv)

what now?

[karmic crown]

whatever you intented to run, i see he wrote
python pip -m install encodings

[karmic crown]

that will install package encodings

but i didnt follow what you were talking about

[cold prawn]

I am getting this error ModuleNotFoundError: No module named 'encodings'

[karmic crown]

anyway you activate venv correctly now, and python commands will work

what was the error?

[karmic crown]

wait

[cold prawn]

python pip -m install encodings this should fix it?

[cold prawn]

ok I'm waiting

[karmic crown]

ok

forget that

close cmd

[cold prawn]

ok

[karmic crown]

remove venv folder

and just run webui again

it will create a new one

did that error popup out after upgrade maybe?

[cold prawn]

yeah had python 3.6 when I ran the webui then upgraded to 3.10.6

[karmic crown]

ok

[cold prawn]

working now

thanks bro

[karmic crown]

then it is expeceted

[sturdy ginkgo]

👍🏽

[karmic crown]

you had a mess in venv

[cold prawn]

should I install this? v2-1_768-ema-pruned.ckpt

download*

[karmic crown]

if you need that model

if you want to use 2.1

[cold prawn]

is there a model that gets downloaded with install?

like a default one

[karmic crown]

i recommend to use 1.5 or sdxl and to forget about 2.x

1.5 will be downloaded by default

[cold prawn]

what's the difference between 1.5 and 2.1?

is 2.1 like an unstable upgrade to 1.5?

[karmic crown]

1.5 is "old" trained on 512x512 images and alow creating of nsfw images
2.0 and 2.1 are newer trained on 768x768, but not trained on nsfw stuff
sdxl is newest trained on 1024x1024, aloso not trained on nsfw

but

almost noone use default models

we use user trained models

you can find many on https://civitai.com/

1.5 modelas are smaller and works faster, sdxl are larger and works slower (takes more time to generate)
sdxl is usually better in quality

but there are some tricks to make 1.5 looks better

[cold prawn]

where to download the sdxl model?

I can't find a .ckpt file

and I see sdxl base and refiner, which one should I go for?

[karmic crown]

here

and download safetensors not ckpt

ckpt in theory can execute malicions code, safetensores version is safe

[cold geyser]

Hello, does anyone have a free tutorial to use SDXL ? I watch many and they all say differents stuff, my SD is like the screenshot :
I already put 1024x1024 to show, but if you have the 512x768 BUT for SDXL i'm here to take.
I know i have to use REFINER, but when i render it gives me the pic i send.

[ornate elk]

You installed the webui the wrong way and also two times inside itself

Ah okay, its still downloading the needed stuff

[buoyant pawn]

oh wait that's not it

[ornate elk]

Hey, you dont need to use a refiner. The problem is that you dont have the sdxl vae selected. And your webui is also not updated.

[buoyant pawn]

sorry my new january screenshot folder wasn't set to /date order

[cold geyser]

Oh hello ! maybe can you help me for that ? i already hit the update button of SD x) (webui only)

[ornate elk]

Something is wrong the model gets send to cpu, are you on a laptop?

[buoyant pawn]

nop desktop

[ornate elk]

Okay can you try a 512x512 image, 20 steps ?
Also how big is the model you use?

Inside the models/stable-diffusion

[buoyant pawn]

yes i had that issue at the very begining it said that i had both cpu and gpu selected and i bypassed it by watching a chineese tutorial that added this "self.logvar = self.logvar.to(self.device)"
in /repositories\stable-diffusion-stability-ai\ldm\models\diffusion

[ornate elk]

Sure, what's your GPU?

[buoyant pawn]

that's probably the reason

i edited ddpm.py

[cold geyser]

Everything here x)

[ornate elk]

okay good, first close the webui and cmd.
then edit the webui-user.bat with right click, edit.
at the line COMMANDLINE_ARGS=
you add: --xformers --medvram-sdxl --no-half-vae
Then in an empty line above webui.bat you type git pull
Then save the file and relaunch the webui-user.bat

Then the next step is to go into Settings - User Interface - Quicksettings, there add sd_vae
then hit apply and reload ui.
Then you need to download a sdxl vae like the official one from here: (its a direct link)
https://huggingface.co/stabilityai/sdxl-vae/blob/main/sdxl_vae.safetensors
put the file into the models/vae folder
then click on the reload button near the vae selection in txt2img at the top and select it whenever you use an sdxl model.

[ornate elk]

yea thats the problem here

you have to undo that

[buoyant pawn]

i made a back up of the file befor edit just in case

change undone

[ornate elk]

okay now relaunch the webui-user.bat

[torn coral]

Why does many extension not show up in SD?

like Adetailer, faceswaplab, reactor

animatediff

[karmic crown]

wdym?

[torn coral]

im running SD 1.5

Like the wont show up

[sturdy ginkgo]

Have you restarted the webUI after installing the extensions/ restart PC maybe

[torn coral]

i did

i reinstalled SD

restarted my pc

[torn coral]

i tried everything

[karmic crown]

this is old version

[sturdy ginkgo]

update the webUI

[torn coral]

on SD 1.7.0 the didnt work either

[buoyant pawn]

i haven't left this chair for 7 hours straight, i'll try again after dinner tks for trying budy, right now it's not acting as if anything changed wish is woring me more coz he didn't revert to the old error

[karmic crown]

you need 1.7.0

[torn coral]

oke lets try

git pull

right

i downgraded

[cold geyser]

Thanks !

[torn coral]

what extension do u use for face swapping?

[karmic crown]

reactor

actually i do not use it, but i will use it 🙂

that one works nicely with new 1.7

[torn coral]

oke lets see i installed reactor

with 1.7

nothing

am i doing something wrong that it wont show up?

[torn coral]

i did brother

[karmic crown]

have you clicked this?

any errors in promt?

[torn coral]

i updated the probram

[karmic crown]

downloading something?

after you install any extension, you need to click that button

[torn coral]

i did

[karmic crown]

do you have any error in console (cmd)?

do you see that it installing or downloading something

?

[ornate elk]

np, if you want i can give you my install guide and you can start a fresh clean install

[torn coral]

somethign about insight

[karmic crown]

Building wheels for collected packages: insightface
Building wheel for insightface (pyproject.toml): started
Building wheel for insightface (pyproject.toml): finished with status 'error'
Failed to build insightface

you have errros

thats why it would not show

[torn coral]

im downloading insight rn

[ornate elk]

you need to follow the install guide for reactor on github, its not plug and play. For example it requires to delete roop, and installation of visual studio tools

[buoyant pawn]

Sounds great feel free to pm me, am geting some pizza outside befor i go on a mental breakdown lol

[karmic crown]

ok, you fixed your mistake 🤣 you arelady triggered me

[gusty solstice]

I have an issue updating SDW, as you can see. It doesn't go further than what you see. If someone can help me :/

[ornate elk]

haha i was scared you saw it hahah ;D

[karmic crown]

find folder where your webui is installed, open that folder from cmd
type git status and show us the result

close stability matrix first

[ornate elk]

Pleas no Messages here rn i will copy my install guides to Pin them

Installing ComfyUI:
https://github.com/comfyanonymous/ComfyUI?tab=readme-ov-file#installing

[ornate elk]

[AMD] Here is a quick Guide to install Automatic1111 Directml Webui for AMD GPUs (Stable Diffusion) on Windows:

Preparation:
You need to install Git 64bit: https://git-scm.com/download/win
and Python 3.10.11, 64bit: (any python above or 3.10.6) Not 3.11, 3.12
https://www.python.org/downloads/release/python-31011/
And check "add python to path" when installing Python.

Important:
If you tried to install SD before and failed, do this:
open up a cmd and type
pip cache purge
to remove the nvidia stuff.

1. Make a new Folder on your drive (not on Desktop, downloads, documents, Programms, Onedrive) and name it Ai for example:
   C:\Ai\
2. 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.
3. Then you copy and paste this command:
   git clone https://github.com/lshqqytiger/stable-diffusion-webui-directml && cd stable-diffusion-webui-directml && git submodule init && git submodule update
4. Press enter and after its done you can close the cmd.
5. Edit the webui-user.bat (right click), At the line COMMANDLINE_ARGS= You add: --use-directml --skip-ort --medvram --opt-sub-quad-attention --opt-split-attention --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.
• GPUs with less than 8gb Vram need: --lowvram
• RX 5xxx cards additional need: --no-half

6. After that save and launch the webui-user.bat inside the stable-diffusion-webui folder.
   After the Installation, that can take a while is done, the webui will autolaunch in your Default Browser.

[ornate elk]

[NVIDIA] Here is a quick Guide to install Automatic1111 Webui for Nvidia GPUs (Stable Diffusion) on Windows

Preparation:
You need to install Git 64bit: https://git-scm.com/download/win
and Python 3.10.11, 64bit: (any python above or 3.10.6) Not 3.11, 3.12 https://www.python.org/downloads/release/python-31011/
And check "add python to path" when installing Python.

1. Make a new Folder on your Drive (not on Desktop, Downloads, Documents, Programms, Onedrive) and name it Ai for example:
   C:\Ai\
2. 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.
3. Then you copy and paste this command:
   git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
4. Press enter and after its done you can close the cmd.
5. Edit the webui-user.bat (right click, edit), At the line COMMANDLINE_ARGS= You add: --xformers --no-half-vae
   Add the following command there too depending on your GPU Vram. (Check Task Manager ->Performance ->GPU ->Dedicated GPU-Memory)

• +11GB: Nothing to add besides the already added.
• 8-11GB: --medvram-sdxl
• 3-6GB: --medvram
• 2GB: --lowvram
• GTX 16XX cards additional need: --no-half instead of --no-half-vae
  6.After editing, save and launch the webui-user.bat inside the stable-diffusion-webui folder.
  After the Installation, that can take a while is done, the webui will autolaunch in your Default Browser.

[ornate elk]

[MAC] Installing Automatic1111 or Forge on Apple MAC OS:
https://github.com/viking1304/a1111-setup

[sacred sierra]

Anyone else getting this?!

[torn coral]

what checkpoint do u guys use for realism?

[ornate elk]

do you use a third party antivirus?

that gradio package is for making the webui accessible in browser by "hosting" it. mostly done local or if available as setting worldwide. So i guess some antivirus doesnt like its function

[vestal viper]

ah you're here sir CS1o, i want to ask about that pip purge thingy

[sacred sierra]

malwarebytes is what caught it.. that and I have windows defender

[vestal viper]

oh wait, it's CACHE not CHARGE LMAOOOOOOO

still got the same message tho, about pip not recognized as blabla

[karmic crown]

this is reason why i am saying not to run pip outside the venv

[ornate elk]

your python is not set in the windows path, did you checked "add python to path" when installing it ?

[karmic crown]

go to your webui folder and type
venv\scripts\pip cache purge

[vestal viper]

oh i see, alright then i'll do it again after i finish reinstalling everything

[vestal viper]

no sir i haven't done reinstalling

[sacred sierra]

When I installed Fooocus it asked if I wanted to share and I said no.. now I can't see any of my other installed lora and checkpoints etc. How do I correct this?

[karmic crown]

@ornate elk when you are lazy enough not to activate venv first, just run like this venv\scripts\command

i am doint that all the time on linux and mac (but it is venv/bin/command)

i was shocked when i learned it is scripts on windows and not bin

❯ venv/bin/python --version
Python 3.10.13

see? 🙂

[sacred sierra]

How do I get some checkpoints and lora back in to be seen by Fooocus ? I had said not to share because a lot didn't work, so I reinstalled. But now I can't add any.

[karmic crown]

i think i saw like 3 people who use focus here for 2 months

[buoyant pawn]

wait so me having it on the desktop could be the reason it's saying NansException: A tensor with all NaNs was produced in Unet. Use --disable-nan-check commandline argument to disable this check.

[vocal burrow]

what's the full error log ? what model are you using ?

[vocal burrow]

check for a file in Fooocus\config.txt it should indicates path a path for each category (checkpoints, loras, embeddings, etc)

[sacred sierra]

Any clue where exactly, not finding it..

[ornate elk]

No its not the cause for that error, but will prevent other errors

[vocal burrow]

what's that model ?

[buoyant pawn]

doll-likeness-by-edg

[vocal burrow]

so that's a lora, not a model.

you're trying to use a lora as a model

[ornate elk]

Exactly

As I said before check that your model is larger than 1.98gb

[lean laurel]

what version of stable difussion must i install? sorry the dumb question

[vocal burrow]

the most recent one

[buoyant pawn]

just downloaded a new one 1.9gb will try it now