🤝｜tech-support

[dreamy lodge]

Are SDXL models more performance heavy?

[ornate elk]

Yes pretty heavy, not that good for 4gb vram gpu

But with fp8 mode you can lower the vram usage of these models

[rich garden]

hello, i watched this video https://youtu.be/onmqbI5XPH8?si=wW20BWH7zj-x3fB- on how to use stable diffusion to try and make some AI generated images like the ones posted on this twitter account for example https://twitter.com/RiasfanpageAi?s=21&t=pAoApEMSC1lC35feeFSzfw but the results i got weren't even related to the anime character i typed in the prompt. idk what i might be doing wrong as i'm still new to AI so hope someone can help me plz

RiasFanPage (@RiasfanpageAi) on X

We're the Busty AI Beauties, programmed to charm and delight, with looks that seem out of this world, we'll keep you up all night.

Twitter

[dreamy lodge]

Its because most models arent based on a single character or have knowledge of who that character is. Do some research on using Loras for the characters you want.

[rich garden]

whats loras?

[ornate elk]

Hey, you need to use other models.
You find them on Civitai.com

These have better quality and there are a lot for anime

[rich garden]

this? https://civitai.com/models

[dreamy lodge]

In the simplest way i can think to explain, loras are sub-models that can be trained off a certain character or other stuff. And yeah use other models than the base SD model

[ornate elk]

Yep

Good anime models can be
Sudachi, or divineanimemix

[rich garden]

i think the model i used was "wd-1-4-anime_e1.ckpt"

[ornate elk]

That's old and not good anymore

[dreamy lodge]

Paint and Nothing by dddokter

[rich garden]

ok ty ty

[rich garden]

do i just type those in the search bar?

[ornate elk]

@rich garden
You also used an old install guide. Better compare it with my guide from the Pinned messages.
You may need some adjustments for the best performance

[rich garden]

ah cool

did i type this right?

@ornate elk

[rich garden]

if you see this please ping me when u reply back

[ornate elk]

You need a space before --medvram

[rich garden]

Ok cool thx

And everything else is good?

Should I also remove “git pull”?

how's this?

[ornate elk]

Yes looks good.
You can leave git pull in there for auto updates of the webui.
But this can also break the webui if a bug is in the latest update

[rich garden]

how can i manually update if it breaks?

[ornate elk]

If it breaks trough an update you need to wait for a fix and then just run the webui-user.bat again

[rich garden]

oh ok cool

[ornate elk]

But you can also manually update by adding git pull one time and then removing it after that.
Adding
--update-check
Will leave a message at the start if there is a new version available

[rich garden]

under what line do i put --update-check ?

[ornate elk]

Same as --medvram

[rich garden]

ok

like this?

ok imma try sudachi real quick

or do i need to do something else first?

[ornate elk]

Nope for anime images that will work

[rich garden]

ok

i also downloaded divineanimemix too

[ornate elk]

If you want a specific character and the model don't know it, you would need to download a lora of the character to help the model knowing it

[rich garden]

uhh.. how can i do that? 😅

[ornate elk]

The more famous the character the more likely it can be created by the model

[ornate elk]

Lora's can be found on civitai.com
They are small files that go into the models/lora folder.

[rich garden]

oh ok

so do i just search a character name then on civitai.com?

[ornate elk]

Here is also a quick guide for Loras:
#🤝｜tech-support message

[rich garden]

ty

it's better but not 100%

just the face i mean

doesn't 100% look like rias gremory

[ornate elk]

Yea then you need a lora of Rias

[rich garden]

ok

same results for devineanimemix

also do u maybe know what model this twitter account uses for their AI art? https://twitter.com/riasfanpageai?s=21&t=pAoApEMSC1lC35feeFSzfw

RiasFanPage (@RiasfanpageAi) on X

We're the Busty AI Beauties, programmed to charm and delight, with looks that seem out of this world, we'll keep you up all night.

Twitter

[elder turtle]

Tone it down with the bikini pics, this is a SFW server

[rich garden]

oh oof 😅

[elder turtle]

If people would read the rules theyd know

[ornate elk]

Any semi realistic model

[rich garden]

guessing that they also used a lora for it?

[ornate elk]

For sure

[rich garden]

alright and thx for ur help btw

[ornate elk]

No problem

[rich garden]

[near acorn]

Howdy howdy, i was wondering if i wanted to generate the same face, in different emotions how would i go about that

A small sample of what i mean

[elder turtle]

Have a lora trained on that face or try IPAdapter or FaceID

[solemn atlas]

Hii, I have again two errors

[ornate elk]

Ip-adapter or inpainting

[near acorn]

ty ty

[ornate elk]

You need a .pth file as upscaler

[solemn atlas]

thats my upscaler

and somehow generating is very slow

[ornate elk]

Delete the 4x ultra sharp .pt file

And then set the hires steps to 10

[solemn atlas]

okay

so i dont need to download that upscaler?

and is there a reason why its very slow generating?

[ornate elk]

What settings did you used?

[ornate elk]

You have 2 version of ultrasharp in that folder. And one of them is broken (the .pt one)

[ornate elk]

Looks okay.
What's your gpu?

[solemn atlas]

Nvidia Geforce rtx4070

[ornate elk]

Okay, make sure you have --xformers in the webui-user.bat for the best performance.
Also if you use sdxl models these are slower

[ornate elk]

Make also sure to only launch the webui-user.bat and not the webui.bat

[ruby needle]

Hi. I am new here, and running stable diffusion on amd on windows 10 was reason why I came. I have already tried 2 different directml implementations but failed for different torch-related errors. I read somewhere that requirement is to have wddm driver model at 2.9 but mine is at 2.7 with latest radeon drivers and am wondering if anyone got this to run on windows 10 (it seems I have to update to 11 to get the wddm to 2.9 but I don't want to because of wmr portal not working properly there). If no one knows if it works on w10, I will try ZLUDA today and report back

[distant warren]

I figured it out, after i turn off the proxy, it work normally

[ornate elk]

I'm using Stable Diffusion with AMD on Windows 10 since last year.
Wddm driver has nothing to do with torch errors.
I'm using zluda rn but used directml the months before zluda was released.
For any questions and errors feel free to ask or send some screenshots here.

[ruby needle]

thank you, I tried directly, in conda and in wsl... is there "recommended" way?

[ornate elk]

Wsl is not supported. What's your gpu ?

[ruby needle]

6950xt

[ornate elk]

Then following my zluda install guide is the recommended way

[ruby needle]

ok my resource that said that wddm 2.9 is needed was for wsl

thanks, I'll try it now

[ornate elk]

No problem, feel free to ask if any step isnt clear

[vast apex]

how can you inpaint in comfyui/stabilitymatrix?

[ornate elk]

I think you get the fastest answer for that in #🧣｜comfy-ui

[vast apex]

asked like 2-3 times in the past few hrs :,)

[ornate elk]

Ohh okay

[ornate elk]

Maybe with these nodes:
https://github.com/Acly/comfyui-inpaint-nodes

@vast apex and here is the guide from the comfyui github:
https://comfyanonymous.github.io/ComfyUI_examples/inpaint/

[vast apex]

i dont have the node tree due to having installed the "simple" version, you got any idea how to get to the node based one :,)

[ornate elk]

Installing comfyui trough the github. There is a download.
https://github.com/comfyanonymous/ComfyUI?tab=readme-ov-file#installing

[vast apex]

i wanted to avoid redownloading but alr ima try

[ruby needle]

Ok I tried now... having same issue I had with other directml implementations
2 questions

1. I am using --medvram-sdxl on 16gb card?
2. Dont check the Pro Driver. - there is no option to check anything on hip download page... I see it is in setup. Did not select but I already have radeon pro drivers installed (will change to normal adrenalin if needed).
   the error is

...
SD-Zluda\stable-diffusion-webui-directml\venv\Scripts\python.exe" -m torch==2.2.1 torchvision --index-url https://download.pytorch.org/whl/cu118
Error code: 1```

I think the video card detection makes a wrong turn somewhere

ok maybe not detection, I have python 3.10.6 installed for this, probably need different version... I read that 3.10.6 is last one in which torch works

[vast crest]

hello

RuntimeError: CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

I really can't use animatediff which always gives me strange errors 😦

[ornate elk]

Hey, yes --mesvram-sdxl is needed, (even for my 24gb 7900xtx)
Idk if it will work with the Pro driver but you can try after the fix. If not switch to normal Adrenalin.

The error you have is a known issue since yesterday.

To fix it you need to navigate into the stable-diffusion-webui-directml\venv\scripts folder.
There launch a CMD via the explorer.
Then run the activate.bat inside the CMD.
Then copy and paste this command:
pip install torch==2.2.1 torchvision --index-url https://download.pytorch.org/whl/cu118

[tame carbon]

Hi, i downloaded comfyui as extension for my sd. but my models from sd are not linked to comfyui extension. Can i somehow specify the path to all of them? I dont really want to make my sd folder have twice bigger size by copying everything xd

[ruby needle]

since yesterday... when I decided to set this up 😂
ok many thanks

I manually installed pip torch wheel in other folder (different repo) but it worked on my cpu... that was only one that worked 😄
looks like this will work now

[ornate elk]

Hey, yes that's possible and pretty easy.
In the comfyui folder there is a
File called "extra_model_paths.yaml.example"
Rename it to "extra_model_paths.yaml"
Then edit it and only change the entry where you set the path to your auto1111 install folder.

[ornate elk]

The worst timing haha xD
I will make a github issue on the directml repo so that the dev can fix it fast

[tame carbon]

big thanks bro

[ruby needle]

  Downloading https://download.pytorch.org/whl/cu118/torchvision-0.17.2%2Bcu118-cp310-cp310-win_amd64.whl (4.9 MB)
     ---------------------------------------- 4.9/4.9 MB 11.2 MB/s eta 0:00:00
  Downloading https://download.pytorch.org/whl/cu118/torchvision-0.17.1%2Bcu118-cp310-cp310-win_amd64.whl (4.9 MB)
     ---------------------------------------- 4.9/4.9 MB 10.8 MB/s eta 0:00:00
Requirement already satisfied: pillow!=8.3.*,>=5.3.0 in c:\users\goran\appdata\local\programs\python\python310\lib\site-packages (from torchvision) (9.5.0)
Requirement already satisfied: MarkupSafe>=2.0 in c:\users\goran\appdata\local\programs\python\python310\lib\site-packages (from jinja2->torch==2.2.1) (2.1.5)
Requirement already satisfied: mpmath>=0.19 in c:\users\goran\appdata\local\programs\python\python310\lib\site-packages (from sympy->torch==2.2.1) (1.3.0)
Installing collected packages: torch, torchvision
  Attempting uninstall: torch
    Found existing installation: torch 2.1.2+cu121
    Uninstalling torch-2.1.2+cu121:
      Successfully uninstalled torch-2.1.2+cu121
  Attempting uninstall: torchvision
    Found existing installation: torchvision 0.16.2+cu121
    Uninstalling torchvision-0.16.2+cu121:
      Successfully uninstalled torchvision-0.16.2+cu121
Successfully installed torch-2.2.1+cu118 torchvision-0.17.1+cu118```

this was done but still can't start, same error

[tame carbon]

can i add previews to my loras somehow? some of them have strange names and sometimes its hard to get which one i need

[vast crest]

Hi

first click on the small icon shaped like tools

[tame carbon]

yeah when i press replace preview and press replace preview it says there are no images in gallery for some reason

[vast crest]

then a window will appear, go down to the bottom and you will find this

Go to botton: replace preview

And Save

[tame carbon]

i get this error

[vast crest]

close everything by pressing on the X at the top right

then you have to press the arrow icon to reload the Lora

and magically the preview will appear

of course before doing this procedure, you must first create a photo 🙂

[tame carbon]

ah, i thought i can put any photo there

[vast crest]

each Lora would be a different character.try it and let me know

[ornate elk]

Have you run it inside the venv?

[ruby needle]

inside the folder yes, I did not first activate environment

[ornate elk]

In the CMD should be
(venv) after running the activate.bat

If there is no (venv) in the front it's wrong

[ruby needle]

well activate doesn't wok

I'll see why

[ornate elk]

You can by renaming any image to the Lora's name and put it in the same folder

[ruby needle]

well... I switched to cmd 😄

I was in powershell

[ornate elk]

Oh lol

Yea that won't work

[tame carbon]

thanks, thats easier than creating a pic for each lora

[azure nimbus]

Is stability imploding right now? I'm getting lots of errors through the API that has been working until now.

[tame carbon]

i cant get any extensions rn for some reason too

[azure nimbus]

The platform website doesn't load either

[ornate elk]

Can you translate the error?

[tame carbon]

URLError: <urlopen error [WinError 10060] An attempt to establish a connection was unsuccessful because the required response was not received from another computer in the required time, or an already established connection was broken due to an incorrect response from an already connected computer>.

idk why it says another pc

[ornate elk]

Okay, something is blocking your connection, like an browser adblocker

Make sure to whitelist the webui in it

Also you may need to use a VPN OE change the DNS server

[azure nimbus]

Stability seems to be working again...

[violet inlet]

guys, i'm trying to weight merge 2 models 1.5 with comfy, but i need a clip model to do it.
someone know what model i need to use and where to find it?

[ruby needle]

ok I am back to old error... shall I add --skip-torch-cuda-test ?

[ornate elk]

No that would use cpu.
Can you screenshot the full CMD output?

[vast crest]

try to see if this article can help you - https://github.com/tgalal/yowsup/issues/319

[ornate elk]

Can you show me your environment paths?

With the two entrys

[ruby needle]

the one for zluda points to the folder where zluda is

moment

is there cmdline for zluda that I can check if it works with my driver?

[vast apex]

whats meant with "Dont use the same Folder where your DirectML webui is" ? what is directml and why would it be there

[ornate elk]

Its an the directml version of automatic1111.
So in your case don't install it in the same folder as stability matrix

[vast apex]

uhh ok

[ornate elk]

nope, which zluda version did you downloaded?

[ruby needle]

latest

[ornate elk]

stable ?

[ruby needle]

I don't know I will check

[ornate elk]

3.7 or 3.8pre ?

[ruby needle]

3.7

3.8 did not have download links

[ornate elk]

okay then do this:
go into the C:\ZLUDA\ Folder.
There copy the cublas.dll and the cusparse.dll and paste them inside it again.
Rename the copys to cublas64_11.dll and cusparse64_11.dll
Copy these two files to stable-diffusion-webui\venv\Lib\site-packages\torch\lib, and overwrite if asked.
then relaunch the webui-user.bat

[vast apex]

umm in step 3, "getting ZLUDA to run" segment.
i just copied the ZLUDA folder directory in my explorer
but im not sure what exactly im supposed to put for the amd driver thingy, copy what you wrote or find the folder and copy it myself in explorer?

[ornate elk]

you have to edit the system variables of windows and there make 2 new entry for zluda and HIP

at step 3 there is a link to a guide on how to edit the Path

[vast apex]

i got ZLUDA as you see

but as said im not sure about how im supposed to add HIP

just copy the %HIP_PATH%bin in there?

or do i gotta find the bin folder myself

and add directory

[valid pumice]

Dysfunktionel konstrktion

[ruby needle]

lol that is a lot of dead code... but it did not help still can't use gpu

I copied into torch/lib not the torch with version

ah the version is not lib just info

[vast apex]

ayo can you help me rq :,) do you just add the "%HIP_PATH%bin" to the variable entry? like just copy paste "%HIP_PATH%bin"

[ornate elk]

dont place it in downloads please
and yes for the other entry just add
%HIP_PATH%bin

[vast apex]

ah mb im stoopid

[vast apex]

gud?

[ornate elk]

what error do you get?

i think you might need to switch to the normal Adrenalin version

[ruby needle]

I'll do that first

Thanks

[rough nebula]

Hi I am trying to access the profile to manage my subscription (I want to cancel) but the page does not load

[vast apex]

uhh i thought i did everything but this comes up :,)

[ornate elk]

yea thats the bug i was talking about before. to fix it do the following:
Navigate inside the stable-diffusion-webui-directml/venv/scripts folder.
Open up a cmd by clicking into the Explorer bar (not searchbar) and type cmd and hit enter.
Type: activate.bat and hit enter.
Now copy and paste this line and hit enter:
pip install torch==2.2.1 torchvision --index-url https://download.pytorch.org/whl/cu118

[vast apex]

wait, is that in ZLUDA or my old comfyui folders

[ornate elk]

its in the SD-Zluda folder

[ruby needle]

mine started now

probably

[ornate elk]

perfect, remember the first gen will take 15-40 mins

[ruby needle]

webui doesn't work but something is happening

[ornate elk]

probably installing stuff

[vast apex]

the activate bat didnt have a reaction, is that normal?

[ornate elk]

no, dont click on it

[vast apex]

im talkin bout in cmd

[ornate elk]

looks good

[vast apex]

ok ty

[ruby needle]

ERR_CONNECTION_REFUSED in browser and in cmdline last message is Launching Web UI with arguments: --use-zluda --medvram-sdxl --update-check --skip-ort

[ornate elk]

show me the cmd output

[ruby needle]

ok it loaded now... there was more lines

[ornate elk]

looks normal

[ruby needle]

ok will leave it to generate something

and check if server is available from lan

[vast apex]

can i close or is it done?

[ornate elk]

from lan only if you add --listen
later to the webui-user.bat

[vast apex]

ayo wth

[ornate elk]

thats okay, as you didnt installed python 3.10.11 even if my guide said so xD

[vast apex]

um...

dam

[ornate elk]

dont interrupt it, if the generate button is grayed out, and no error in cmd, then let it run, it will take its time, even if nothing is showing up in cmd

[vast apex]

wai, so do i need that python version or should i leave it be

[ornate elk]

you can leave it be

[vast apex]

okeh :o

[ornate elk]

make sure you do step 4 correctly

[vast apex]

i thought once again i did it correctly

but the generation simply stops as soon as it starts

[ornate elk]

Any error in the CMD?

[vast apex]

odd

[ornate elk]

Latest amd driver?

[vast apex]

no but 24.1.1

do i gotta update?

ig ima try

[ruby needle]

thank you... sadge that without you internet wouldn't be a place where I could get this to work...

[vast apex]

same error after new drivers and restarting pc

[ornate elk]

Which zluda files did you downloaded?

[ornate elk]

No problem 🙂

[vast apex]

ehh wdym?

[ornate elk]

On the zluda link

Did you downloaded version 3.7 or 3.8

[vast apex]

you mean the amd thingy or like really zluda? i just used the command "git clone https://github.com/lshqqytiger/stable-diffusion-webui-directml && cd stable-diffusion-webui-directml && git submodule init && git submodule update"

[ruby needle]

is there issue with resolutions above certain? (default model that got downloaded automatically)

I tried to upscale a pic and it got corrupted, and another i tried to generate with higher res was similar

[ornate elk]

No I mean the zluda files

[ornate elk]

1.5 models are trained on 512x512 resolution.
Sdxl models are trained on 1024x1024
Also you get the models on civitai

[vast apex]

uh i downloaded that, i thought its the most recent

[ruby needle]

ah so you have to use resolution that is same like training set...

[ornate elk]

3.7 yes

[ornate elk]

Not exactly but near to it

Or using upscaling

[vast apex]

ye i downloaded that

[vast apex]

should be 3.7 so idk

[ornate elk]

Okay

[ornate elk]

Have you downloaded the library folder and replaced it ?

[vast apex]

ive renamed the original library folder and imported the new one

[ornate elk]

Okay good

In afk, can help later

[vast apex]

okay :o

[ornate elk]

But I have one fix you can try

[ornate elk]

go into the C:\ZLUDA\ Folder.
There copy the cublas.dll and the cusparse.dll and paste them inside it again.
Rename the copys to cublas64_11.dll and cusparse64_11.dll
Copy these two files to stable-diffusion-webui\venv\Lib\site-packages\torch\lib, and overwrite if asked.
then relaunch the webui-user.bat

[vast apex]

btw its fine if the zludas thingies are on different drives right?

[ornate elk]

Yea should work

[ornate elk]

Have you restarted the PC?

[keen summit]

my images keep being generated with this weird glitch/ blur effect.

uhm, any help would be appreciated

[ornate elk]

whats the model and the vae?

any lora used?

[keen summit]

https://huggingface.co/KBlueLeaf/Kohaku-XL-Delta
https://civitai.com/models/296576/sdxl-vae
And no lora

image generates fine, and gets distorted in the finished product, if that information helps.

(probably not)

[ornate elk]

what sampler, any upscaler used?

[keen summit]

DPM++ 2M SDE Heun, upscalers didn't change the outcome

[ornate elk]

can you try DPM++ 2M Karras?

[ornate elk]

also whats your gpu?

[keen summit]

GTX 1070ti

[ornate elk]

okay make sure you --xformers --medvram-sdxl --no-half-vae
in your webui-user.bat

[keen summit]

this seemed to be the answer I suppose

I appreciate the support man

[ornate elk]

no problem, still add the commands to the bat file for faster generation time ^^

[keen summit]

Already had xformers, medvram but I'll add the others in right now

[ornate elk]

change --medvram to --medvram-sdxl

[keen summit]

got it

[ornate elk]

so every model will work as fast as it can (1.5 and sdxl)

sdxl should take much less time then before

[keen summit]

huh, was wondering why it took like 30 minutes when 1.5 took 3 or so.

[ornate elk]

xD yea thats when not using --medvram-sdxl on 8gb vram cards

but also with dpm++ 2m karras it should took less time

and then use 25-30 steps and your good to go

also hires fix will add much more time to a generation

[vast apex]

yup

[ornate elk]

Hmm it should work

Have you moved the two zluda files manually ?

[vast apex]

i copied them and pasted/replaced them in the goal folder

[obsidian carbon]

Hiya! I may need quite a lot of help

Is SDXL possible to use with 8GB Vram on A1111?

[ornate elk]

Hey, with what task?

[obsidian carbon]

Well, its kinda similar to the previous person - except worse -- My sdxl is kinda slow, by kinda, its a lot

[ornate elk]

@vast apex Okay good, can you show me your zluda folder path?

[obsidian carbon]

Like, I may need to screen share and explain ://

[obsidian carbon]

Im a noob - whats that?

[ornate elk]

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

[obsidian carbon]

GTX 1070ti and cmd args are: --xformers --theme dark --autolaunch --medvram-sdxl --no-half-vae

[ornate elk]

That's good, and what settings did you used?

Like resolution, steps etc

[obsidian carbon]

lemme send image

That’s the settings

Like, when I use A1111 for SDXL, I turn everything else off to allow as much usage as possible

[ornate elk]

Don't use a refiner

That's not needed

[obsidian carbon]

Oh, okay, may I ask, when would one need one?

[ornate elk]

Also use batch count 4 instead of batch size

[obsidian carbon]

Alright, I will try those, and report back

[obsidian carbon]

What’s the difference between size and count ? I’m still learning this.

[ornate elk]

Mostly never xD

[obsidian carbon]

Fair enough 💀 I’m used to using it back in sdxl which may be also a huge reason why my pc was slow cause of it 😭😭

SD 1.5 not xl

[ornate elk]

Batch count: "generate each image after image"
Batch size: "generate x amount of images at the same time"
That needs a lot more vram an is not recommended for sdxl models.

[obsidian carbon]

Oh my okay, i understand . You explained that better than the yt videos I’ve been seeing then 😭😭💀

[ornate elk]

No problem 🙂

[obsidian carbon]

Just for reference, the way i do things is i do it in a small amount of steps in Euler A and then go to img 2 Img with a different model and add like 4 to 6 steps to get it too look better . That’s what i did in sd 1.5, is that necessary in xl?

[ornate elk]

Euler a and DPM++ 2m Karras are the recommended samplers.
For 1.5 you should use 20-30 steps for a good image.
With your card you can also use hires fix for upscaling the image to get better quality.

For sdxl you also should use 30 steps. If the sdxl model is a turbo or LCM model you need less steps.

You can also upscale in img2img with the SD upscale script to get a better quality and higher resolution.

[obsidian carbon]

Okay, I’ll keep that in mind.

Wow

[obsidian carbon]

You’re actually a god send

[ornate elk]

can you show me your zluda folder path and folder?

[half onyx]

Hey I am setting up stable diffusion using webui and every guide I look up says I should have a section under the train tab callled preprocess images. It is not there for me. I am runing v1.8.0. Is this a known issue and is there a solution?

[subtle quarry]

I'm getting an error when trying to run controlnet, and I don't know why it's doing so

*** Error running process: C:\Users\Quinn\Desktop\sd.webui\webui\extensions\sd-webui-controlnet\scripts\controlnet.py

oh i figured it out. i had the preprocessor turned on

[harsh saffron]

Hey, i got a question, what is a better model to make stencils for tattoos? Or just some random black and white drawing?

[neon swallow]

Website of Dream Studio is facing some issues. Help please!

CC @north cedar

[north cedar]

🙏🏼

[harsh saffron]

Anybody?🤣

[jolly gorge]

any images i make using faceswap lab in img-to-timg in automatic1111 dont save, anyone know how to fix this?

[ornate elk]

If your using Img2img Batch, don't set an output directory and it could work.
But faceswaplab is abandoned and you should use Reactor instead

[marble snow]

New to ComfyUI. I can't get SD Turbo to work on it. This it the workflow I copied and it seems to be working for others, but not for me.
WHat could be going wrong her?

Faced no problems with Automatic1111 with SD Turbo

The reason for using ComfyUI was to test out real time generation as that is very finiicky in Automatic1111 and to set up workflows which involved multiple post processing steps

[gloomy pivot]

Hey All! I've got a quick question - does anyone know how to setup any image generation model in a way where we can generate very similar characters in different scenarios?
let's say a knight and a princess, which could be used in 10 different scenes doing random stuff (running, fighting, dancing etc.) - the details don't have to be 100%, but at least somewhat similar

( I'm willing to pay for some consultations as well)

[marble snow]

2 common ways are Lora and Textual Inversion (not sure what it's called now)
LORA is very computationally expensive to train but gives great quality.
If you want something quick and reatively easier, then look into Textual Inversion.

If you're trying a specific face, then you can always simply use a FaceSwap. https://github.com/Gourieff/sd-webui-reactor
You can create face swap models for any face with as low as 1 image really, but it works great if you have say 10 ish images and can save a mean model of those 10 images for best results.
Super quick to create swaps and each face model file is like 4kb lol

[buoyant lotus]

the connect errors out

[ornate elk]

whats the model? any furry, yiffy mix? if so, these can cause issues

[buoyant lotus]

none of that

[ornate elk]

do other models work?

[buoyant lotus]

no

[ornate elk]

okay, then i would suggest you to install the webui the correct way, by following my install guide from the Pinned Messages of this channel
Direct linked here: #🤝｜tech-support message

if you then still get errors, feel free to share some information, like model size, gpu device etc

[warm gull]

different scenes doing random stuff (running, fighting

[normal silo]

but if it's like $520 then $520 would be fine

but it works great if you have say

[marble flax]

turbo or LCM model you need less steps

[elder turtle]

If youre running into issues, please conntact their support

[near acorn]

anyone knwo why my iamges come otu a bit bleak?

while its loading it looks fine but when its finished it seems bleakish

just seems to do it wit hanythign v3 for most part

[ornate elk]

You need to use a VAE file for color correction

[near acorn]

it prolyl didnt like the vae i was usign then

[ornate elk]

Kl-f8-anime2 is the best for anime

[near acorn]

is that a vae?

i taught that was a upscaler

[ornate elk]

Nope its a vae

[near acorn]

nvm was confusign ti with that

[near acorn]

a bit confused when i go to the site to download it onyl seee uhm, cpkt files

no idea what the yare do they work ?

[haughty zodiac]

Do you know if it's possible to train a lora directly on automatic1111 ui ?

I see that I can use khoya but would prefer to stick to one tool for now

[ornate elk]

They work the same as safetensor

Drop the VAE inside the models/VAE folder

[ornate elk]

Nope, you better use kohya

[haughty zodiac]

it's strange because the setup.bat doesn't work

[near acorn]

ty

[haughty zodiac]

it says that I don't have "No module named 'pkg_resources'"

even though setuptools is correctly installed

i'm trying with dreambooth extension on automatic1111 instead but it's not very clear

[ornate elk]

Don't use dreambooth

Its not for lora training

[haughty zodiac]

this?

[ornate elk]

Yes that's dreambooth

[haughty zodiac]

okay saw tutorials using dreambooth

but my issue with khoya is that the bat files doesn't run

[ornate elk]

Installing dreambooth only causes trouble for the webui

[haughty zodiac]

okay i'll uninstall

[ornate elk]

I never recommend it

[haughty zodiac]

but what can I do with khoya?

[ornate elk]

You also need to delete the venv folder after deleting the dreambooth extension

[ornate elk]

Make sure you have python 3.10.11 64bit installed and checked "add to path"

[haughty zodiac]

it is

[ornate elk]

Also you need an nvidia GPU for using kohya

[haughty zodiac]

3.12.1

[haughty zodiac]

alreday installed both cuda and cudnn

[ornate elk]

That's not supported by the webui and Kohya and will cause issues

[haughty zodiac]

oh crpa

I need the latest version

for other projects

[ornate elk]

You can still install 3.10.11 64bit
But then you also need to edit the webui-user.bat to set the path to the right python version

[haughty zodiac]

okay

[ornate elk]

For Kohya I can check if there is also a setting later

[vocal burrow]

worse case they can use py --3.xx to swap python version
( cf https://docs.python.org/3/using/windows.html#customization )

[haughty zodiac]

i have installed 3.10 and created a venv directing to 3.10 and not 3.12

i'm still getting this issue

i don't understand at all

[chilly knot]

oh so many sd 1.5 base models dont work for me

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

i try add no halfe

but get memory error

😭

[haughty zodiac]

do you know why it keeps blocking on the setup.bat?

i have latest version on github

[ornate elk]

What's your GPU again?

[chilly knot]

1050 ti

[ornate elk]

Idk, if you made a venv with 3.10 it normaly should work

[chilly knot]

is strange i try all solution

[ornate elk]

You only need --xformers --medvram --no-half-vae
And then check the model file sizes

[chilly knot]

now i have it

[haughty zodiac]

Oh just saw that it's already creating a venv in the setup bat

Maybe it's that

It's creating a venv but for the wrong version

[sweet hearth]

hi I have a question. I have stable diffusion and it works on my laptop, but i heard Forge is a better option if you got an older laptop. Somehow I was able to skip this error when using the normal a1111 stable diffusion, but I forgot how to fix it. and now Forge doesn't work. what should I do?

the normal stable diffusion still works btw

[ornate elk]

What's your gpu?

[sweet hearth]

Nvidea Quadro K3100M. I've tried updating it, but I also get an error message there

(nice luffy pf btw mate)

[ornate elk]

You get an error by updating the driver?

[sweet hearth]

this one

I downloaded the right driver update

If I open the normale stable diffusion. this is what I see in the cmd

somehow i disabled the check I think

[ornate elk]

Okay I would to the following:
Make all windows updates and optional updates (don't upgrade to win11)
Enable the checkbox for updates for other drivers on windows.
(Its under windows updates, Advanced Options, first checkbox)
Then make them too. Reboot. Search for updates again until no updates are found.
Then open up a CMD as admin and run
Sfc /scannow
That will check for system file corruption's.
If it found something and repaired it. Reboot.
Then try update the driver again.
If that fails you should get the tool DDU (Display Driver Uninstaller)
To remove the old driver and then install the new one.

[sweet hearth]

thank you I will give it a shot. I will give you a heads up if it did or didn't work

[ornate elk]

Sure 🙂 hope it helps

[sweet hearth]

which of the old drivers should I uninstall? the GPU 1 Nivdea quadro k3100m?

[haughty zodiac]

finally it works! Should I train a custom dreambooth model or a LORA?

[haughty zodiac]

please can someone help me step by step? I can share my screen, the training crashes at beginning, i'm following tutorials but it's not working

[ornate elk]

Do you have multiple?

[sweet hearth]

apparently, but those are gone now. only got the new one left

[dusty parcel]

Hey all, I was using the webUI, but let's say I want a more featured UI/programmable interface Is there a better one?

[haughty zodiac]

how is it possible that every notebook or tools i'm finding have dependency versions issue

https://imgur.com/a/mrTteIt

i'm using this khoya colab notebook to train a LORA and still getting these issues

i don't understand i'm following every step religiously

omg i had to check every dependencies to put them at the right version

[hollow ledge]

https://www.instagram.com/p/C5oO81AutCB/?img_index=1

uisato_

Had the pleassure of working alongside @maxcoopermax on his latest project, called “Seme”.

He asked me to experiment around the concept of renaissance era, giving life to a set of AI intervened custom-recordings [ft. @chinalabaig], experimental oscilloscopes [tbr, hopefully soon], and @touchdesigner systems, for his shows last week in collabor...

Likes

186

Instagram

[sinful halo]

no module 'xformers' message on cmd is normal ? (im using zluda)

[oak berry]

Has anyone ever been able to get the batch folder under control net to work? I have tried several times but it only creates images with the first pose in the folder and ignores all other poses.

[solemn atlas]

hey guys how can i activate the "SD XL" checkpoint and lora? it does not show in my sd

[fair oxide]

i am having a problem with my Lenovo Legion 5 3070 RTX Laptop, which i suspect may be related to running Stable Diffusion. About a month ago I started using SD heavily, and after about a week of this started to notice a major performance problem, namely frequent but brief (usually less than a second) freezing. Most noticeable when playing games but seemed to even be happening on the desktop.

I tried many things to fix it and even thought it might be a hardware problem. At one point I freed up a bunch of hard drive space because my SSD was almost full (95% full on a 4TB drive). I freed up about 500 gigs and that seemed to help some, but i moved on to trying other things because I assumed 500mb free HAD to be enough for virtual memory or whatever else.

After trying everything else I went back to freeing up more hard drive space. After freeing up a total of over a terabyte, the problem disappeared. It's been fine for about a week, I have continued using SD heavily and now i can see the same problem starting to come back again. And I still have over a TB free.

Has anyone else encountered this issue?

[ornate elk]

Yes, xformers is nvidia only

[ornate elk]

Hey, what's in your webui-user.bat?

[fair oxide]

ill check

@echo off

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

@REM Uncomment following code to reference an existing A1111 checkout.
@REM set A1111_HOME=Your A1111 checkout dir
@REM
@REM set VENV_DIR=%A1111_HOME%/venv
@REM set COMMANDLINE_ARGS=%COMMANDLINE_ARGS% ^
@REM --ckpt-dir %A1111_HOME%/models/Stable-diffusion ^
@REM --hypernetwork-dir %A1111_HOME%/models/hypernetworks ^
@REM --embeddings-dir %A1111_HOME%/embeddings ^
@REM --lora-dir %A1111_HOME%/models/Lora

call webui.bat

[fair oxide]

do you have some idea what might be causing this and/or have you seen anything like it before?

[ornate elk]

Your using forge?

[fair oxide]

yes

also played with comfy lately but mainly been using forge

[ornate elk]

Okay, how much ram do you have?

[fair oxide]

64 gb

[ornate elk]

Oh okay that should be enough xD

[fair oxide]

yeah

and 1tb hard drive free should be enough

500gb should be engouh

based on what i read 150 gb should be more than enough

for page file

i think they said ramx2

at the most

[ornate elk]

Also make sure to have enough free space on C drive (15-20gb)

[fair oxide]

thats what im talking about

theres only one drive on the thing that i see

shit is wack yo thought it was fixed then i was playing elden ring and all of a sudden it started to stutter again. noticed it started to happen right away this time

i was like

😐

[ornate elk]

And you run the latest nvidia driver?

[fair oxide]

yep

like i said i tried everything short of a full reinstall or having lenovo send a guy to my house

[ornate elk]

What you can try is to open up a CMD as admin
And run
Sfc /scannow

[fair oxide]

freeing up HD space fixed it but i had to free up a totally unreasonable amount and theres STILL over a TB free now

ok

Windows Resource Protection found corrupt files but was unable to fix some of them.
For online repairs, details are included in the CBS log file located at
windir\Logs\CBS\CBS.log. For example C:\Windows\Logs\CBS\CBS.log. For offline
repairs, details are included in the log file provided by the /OFFLOGFILE flag.

[ornate elk]

Oh rip

Broken system files

[dreamy lodge]

RuntimeError: CUDA error: out of memory
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

sorry if im impeding on the conversation

But how do i fix this

[fair oxide]

does this explain what's been happening then

[fair oxide]

youre not i think its just how it works here. come in and yell out your issue to whoever wants to help

[ornate elk]

Yes, a system error that can't be fixed needs a OS reinstall

[ornate elk]

What's in your webui-user.bat?

[fair oxide]

this machine didnt come with any reinstall thumbdrive or anything is it built in somehow? is that what we do now

[dreamy lodge]

--xformers --medvram --no-half-vae

[ornate elk]

You need a thumbstick with 8gb and then use the windows media creation tool

[dreamy lodge]

just what you told me the other day

[fair oxide]

ill figure it out thanks for your help

[ornate elk]

No problem, I guess the reinstall could fix your issue

[graceful girder]

hello I have a problem with stable diffusion I would like to do outpainting as well as inpainting I don't understand why but it seems that it doesn't work at all as much if I use OpenOutpaint or controlenet nothing works for in the case of openoutpaint I have black or glitch and for clear control an error

(screen of my openoutpainting and it same with control net but him he crash all m stable diffusion x))

[ornate elk]

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

[graceful girder]

my GPU is an AMD precisely an AMD 6750 XT and in my there is webui-user.bat set COMMANDLINE_ARGS= --use-directml --lowvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling

[ornate elk]

For openoutpaint its required to have --api
In the webui-user.bat

For inpainting on AMD directml you need --no-half aswell

But you should not use directml with that GPU.
You should use the zluda Backend for better performance and best compatibility

[graceful girder]

what does --no-half do ?

oh ? i need reintall an other stable diffusion for inpainting and better perf ?

[ornate elk]

It loads models as fp32, so inpainting will work with directml but it requires a lot more vram.
So its not good to use it for generating images.

[ornate elk]

Yes you need to follow my zluda AMD guide from the Pinned Messages of this channel

Its much faster and works better

[graceful girder]

ok I will do that if I have any problems I will turn to you

[ornate elk]

Sure! Let me know if something isn't clear

[graceful girder]

while I believe that it installs correctly (I have not yet received an error message) will the ZLUDA version be compatible with inpainting and outpainting despite the graphics card ?

okay i thinks i have an error look the error occurs when I change models

and on launch I have Warning: Failed to install handrefinerportable. Some processors will not work.
I don't know if this is really serious but I'll point it out.

[graceful girder]

(it's not for all model)

but idk why when i try to generate an img the generation don't work lmao

[ornate elk]

Let it load

First time will take 15-40 mins

[graceful girder]

realy ?

lmao

okay i try

[ornate elk]

If the generate button is Gray and and no error in CMD let It load

[ornate elk]

This is just the wrong lora for the model

[ornate elk]

Yes outpatient and inpainting will work

[graceful girder]

nice

[graceful girder]

ohhh okayyy

[graceful girder]

okay

[ornate elk]

Did it worked?

[graceful girder]

i thinks for gen yes i try other now

[subtle quarry]

Is there a reason why a Lora isn't working when I use controlnet?

Trying to use the deer antlers Lora and it's worked before, it's just struggling now

I think I figured it out, I had to I crease the weight of the trigger word

Increase*

[graceful girder]

for outpainting when i use openoutpaint idk why but my image as something wrong look

[ornate elk]

Hmm maybe change the model at the top also to the inpainting model

Also you can try using a VAE file

[graceful girder]

it is a model which is made for normally

and idk why too my control net make me an error befor make me an image lmao look

[graceful girder]

and what is a VAE file?

[ornate elk]

What control net model and preprocessor is that?

[ornate elk]

A file for color correction

[graceful girder]

how i can try to fix ?

[graceful girder]

this means what net and preprocessor

[graceful girder]

i thinks screen is better no ? x)

[ornate elk]

Oh yea you definitly need a VAE file

[graceful girder]

like u say x)

[ornate elk]

Switch the preprocessor to inpainting only or inpainting global harmonious, the lama one won't work

[graceful girder]

oh lama don't work ?

[ornate elk]

Here is the anime VAE.
Download it and put it into the models/VAE folder
https://civitai.com/models/23906/kl-f8-anime2-vae

Then go into Settings, User Interface, Quicksettings and there add sd_vae
Then hit apply and reload ui.
Then select the VAE in the new dropdown

[graceful girder]

ok i try

like that ?

[ornate elk]

Yes, you can also add there
Clips_stop_at_last_layer
To get Clip Skip

Some anime models work better with clip skip 2

Normal models need clip skip 1

[graceful girder]

okay i add clips

like thats

[ornate elk]

Then scroll up and hit apply and reload ui

[graceful girder]

okay i do that

okay now i try

okay for openoutpaint i thinks is now better no ?

[ornate elk]

Much better

[graceful girder]

i thinks too okay now i try control net

okay when i use controle net i don't now what he doing x)

my control net is good no ?

[ornate elk]

They are okay

But You need to be in txt2img or img2img, not inpaint tab

[graceful girder]

i m on img2img

[ornate elk]

Then you need to use a non inpainting model.
Only use the inpainting model when the inpainted mode is selected

[graceful girder]

oh ok i try

while I try to generate the extension of my image what is the use of the skip clip?

[graceful girder]

ok so the result, I'm not convinced, could you tell me how to adjust the poorly melted rating?

maybe i expand to much x)

[ornate elk]

For that you need control net inpaint

Then check resize and fill for the control net mode

[graceful girder]

okay i do that thanks

[dense flint]

I recently updated SDWebUI Forge and the "Apply and restartUI" on the extension tab has changed to "Apply and quit". Is there any way to change it back?

[buoyant lotus]

werfault.exe had a pop-up and then it died

[azure ivy]

hey guys does anyone knows how to fix
Error: The 'insightface==0.7.3' distribution was not found and is required by the application

[subtle quarry]

I don't know what this means, I was trying to use img2img ultimate sd upscale
RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

So when I transfered the image over via the button under the generated image, it spat out the error. However, manually putting it in doesn't spit out the error

Issue seems to have fixed itself somehow

[heady pine]

Could someone please help me with this, When I try and generate an image this message pops up, what do I do to fix this?

[grim iron]

I get this error can someone help? TypeError: 'in <string>' requires string as left operand, not int

[vocal burrow]

need full log

[heady pine]

I have fully reinstalled everything since nothing seemed to work, but I am still getting the same messages. Everything was working fine today until I closed and reopened Automatic1111 so I could refresh it. I'm guessing since I have git pull in my wordpad, it must have updated upon reopening the client. Could someone also please tell me how to get rid of this schedule type and what I should do to always have this downcast aphas enabled so it doesn't keep popping up in the override settings when looking back at my images since it never did before?

[marble flax]

this means what net and preprocessor

[normal silo]

What control net model and preprocessor is that?

[heady pine]

@marble flax & @normal silo

I also got this failed to install notice when installing ControlNet, so I'm unsure what to do to fix this. This is the folder it's talking about; am I supposed to have something that I don't?
I don't think I received this message when I installed Automatic1111 in the past.

[normal silo]

What control net model and preprocessor is that?

[heady pine]

I'm sorry @normal silo, is this the information you are after?

[nova ermine]

I downloaded thid model: https://civitai.com/models/23842/djz-robot-wars
and got this error. Any idea on what's happening?

[ornate elk]

In the settings is an option for the alpha_cumprod

[drifting kiln]

Hello, ive been using automatic1111 and when I try to generate anything about 540x x 540x, I get low vram error?, my GPU got 16gbvram, this only a amd issue? I've got the low-vram setting on my commands

[ornate elk]

Hey, what's your gpu?

[drifting kiln]

6950xt

[ornate elk]

Then you installed the directml webui without optimisations settings.
What you need is to follow my AMD Zluda guide.
Its in the pinned messages of this channel.
With that you won't have any vram problems and much better performance

[drifting kiln]

so should i delete?

my current one?

i use stabilty matrix if your familiar with it

[ornate elk]

Yes stability matrix doesn't support zluda

So you can delete it

[drifting kiln]

Make a new Folder on your Drive (not on Desktop, Downloads, Documents, Programms, Onedrive) and name it SD-Zluda.

then where should i?

google drive?

[heady pine]

I appreciate your help with this issue, if possible, when you are next free, could you please help me resolve my other issues.

[drifting kiln]

@ornate elk

yo

u here

[azure ivy]

do you where i can ask about Reactor extenstion not working , i get insightface errors and i did everything i could find to fix the error but it didnt got fixed

[ornate elk]

Yes

[drifting kiln]

hmm i made it work somehow

ty

[ornate elk]

Pls don't install it in any google drive or cloud

[drifting kiln]

HAHA

its currently running the cmd without any errors so

might be working

[ornate elk]

Okay, where did you installed it into?

[drifting kiln]

C:

i suffer with low iq

takes time to procesess stuff

[ornate elk]

If you managed to start the webui and pressed generate and the button is gray now and no errors in CMD. Let it load. First gen will take up to 15-40 mins

[ornate elk]

Did you followed the steps on reactor github ?

[drifting kiln]

Launching Web UI with arguments: --use-zluda --medvram-sdxl --update-check --skip-ort

[azure ivy]

yes precisely

[drifting kiln]

its stuck on that

oh nvm its loading

[azure ivy]

i installed , visual studio and everything it asked

[ornate elk]

Can you screenshot the error you get In the CMD?

[azure ivy]

when i open up the software

also my recator tab does not show up

this is a new pc i am working with , i installed stable diffusion recently

[ornate elk]

For the other issues please restart the webui and show me the CMD output

[ornate elk]

okay, can you try delete the venv folder and relaunch the webui-user.bat?

[ornate elk]

update python to 3.10.11 and delete the venv folder after that, then relaunch

[vestal plank]

How to solve it?

'NoneType' object has no attribute 'lower'

  File "f:\AI\ComfyUI_windows_portable\ComfyUI\execution.py", line 153, in recursive_execute
    output_data, output_ui = get_output_data(obj, input_data_all)
  File "f:\AI\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 "f:\AI\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 "F:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\efficiency-nodes-comfyui\efficiency_nodes.py", line 236, in efficientloaderSDXL
    return super().efficientloader(base_ckpt_name, vae_name, clip_skip, lora_name, lora_model_strength, lora_clip_strength,
  File "F:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\efficiency-nodes-comfyui\efficiency_nodes.py", line 151, in efficientloader
    model, clip = load_lora(lora_params, ckpt_name, my_unique_id, cache=lora_cache, ckpt_cache=ckpt_cache, cache_overwrite=True)
  File "F:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\efficiency-nodes-comfyui\tsc_utils.py", line 372, in load_lora
    lora_model, lora_clip = recursive_load_lora(lora_params, ckpt, clip, id, ckpt_cache, cache_overwrite, folder_paths)
  File "F:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\efficiency-nodes-comfyui\tsc_utils.py", line 363, in recursive_load_lora
    lora_model, lora_clip = comfy.sd.load_lora_for_models(ckpt, clip, comfy.utils.load_torch_file(lora_path), strength_model, strength_clip)
  File "f:\AI\ComfyUI_windows_portable\ComfyUI\comfy\utils.py", line 12, in load_torch_file
    if ckpt.lower().endswith(".safetensors"):```

[heady pine]

Sorry about all the images @ornate elk, I have forgotten how to screen record.

[ornate elk]

okay, what extensions do you have installed?

are they updated?

[heady pine]

I only have Adetailer and ControlNet, this extension is new as of today, (I'm pretty sure), so I don't know if this is what is causing the messages.

[azure ivy]

i did that , no difference

[ornate elk]

i dont think so, i think the extensions need an update for the new version maybe, as 1.9.0 is out for only 8 hours

[heady pine]

My bad then I thought it was new, I tried unticking that extension and generating another image, but I'm still getting the same messages, I'm a bit lost on what to do now

[azure ivy]

Download and put prebuilt Insightface package(https://github.com/Gourieff/sd-webui-reactor/raw/main/example/insightface-0.7.3-cp310-cp310-win_amd64.whl) into the stable-diffusion-webui (or SD.Next) root folder where you have "webui-user.bat" file or (A1111 Portable) "run.bat"

From stable-diffusion-webui (or SD.Next) root folder run CMD and .\venv\Scripts\activate

OR

(A1111 Portable) Run CMD

Then update your PIP: python -m pip install -U pip

OR

(A1111 Portable)system\python\python.exe -m pip install -U pip

Then install Insightface: pip install insightface-0.7.3-cp310-cp310-win_amd64.whl

OR

(A1111 Portable)system\python\python.exe -m pip install insightface-0.7.3-cp310-cp310-win_amd64.whl

Enjoy!

i found this in reddit and i did this]

its seems to work now but i am testing

[ornate elk]

you have a wierd error in your cmd "No image data blocks found"

[ornate elk]

thats also on the github of reactor, and is the simplest way to install it manually

[azure ivy]

i swear i did this before but i did not work

[ornate elk]

maybe you werent in the venv, because it wont work just using cmd alone

[azure ivy]

maybe because i needed python or venv yeah

[vestal plank]

How to solve it?

'NoneType' object has no attribute 'lower'

  File "f:\AI\ComfyUI_windows_portable\ComfyUI\execution.py", line 153, in recursive_execute
    output_data, output_ui = get_output_data(obj, input_data_all)
  File "f:\AI\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 "f:\AI\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 "F:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\efficiency-nodes-comfyui\efficiency_nodes.py", line 236, in efficientloaderSDXL
    return super().efficientloader(base_ckpt_name, vae_name, clip_skip, lora_name, lora_model_strength, lora_clip_strength,
  File "F:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\efficiency-nodes-comfyui\efficiency_nodes.py", line 151, in efficientloader
    model, clip = load_lora(lora_params, ckpt_name, my_unique_id, cache=lora_cache, ckpt_cache=ckpt_cache, cache_overwrite=True)
  File "F:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\efficiency-nodes-comfyui\tsc_utils.py", line 372, in load_lora
    lora_model, lora_clip = recursive_load_lora(lora_params, ckpt, clip, id, ckpt_cache, cache_overwrite, folder_paths)
  File "F:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\efficiency-nodes-comfyui\tsc_utils.py", line 363, in recursive_load_lora
    lora_model, lora_clip = comfy.sd.load_lora_for_models(ckpt, clip, comfy.utils.load_torch_file(lora_path), strength_model, strength_clip)
  File "f:\AI\ComfyUI_windows_portable\ComfyUI\comfy\utils.py", line 12, in load_torch_file
    if ckpt.lower().endswith(".safetensors"):

[drifting kiln]

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.

@ornate elk

should i do as it says?

[ornate elk]

what was the model you tried to use?

how bis is the model file?

[drifting kiln]

epicphotogasm

[ornate elk]

never add --skip-torch-cuda-test

[drifting kiln]

2gigs

[ornate elk]

okay that should work

whats in your webui-user.bat?

[drifting kiln]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --use-zluda --medvram-sdxl --update-check --skip-ort

call webui.bat

[ornate elk]

strange, you shouldnt get that error then, what settings did oyu used?

[drifting kiln]

thats it

i waited 20 mins for the first pic, then added loras and models to my files then resttared

and now it wouldnt generate

i added --no-half

lets see

looks like it worked

YOOOO ITS SOOO FAST

[ornate elk]

--no-half is normaly not needed

[drifting kiln]

idek what i does

it

but it works

but hey man thanks

i appericate it

[ornate elk]

no problem, should be still faster as directml ^^

[drifting kiln]

ITS WAYYY FASTER

u got no idea

mby u do

[ornate elk]

xD i do haha

thats why i made the guide

[drifting kiln]

xD

i can generate 1000x by 1000x

less go

[heady pine]

Is there no way you can think of to remove these messages? I am able to generate images with the messages still popping up, but I think with so many messages popping up, the client might crash without me knowing that it has. @ornate elk

[ornate elk]

what you can try is a complete reinstall by following my guide in the pinned messages of this channel.

[heady pine]

As of your first reply to me, I had done a full reinstallation of everything. So I'm unsure what I could do.

Git/Python everything apart from loras and the such

[ornate elk]

you can try update python to 3.10.11

then delete the venv folder again

[heady pine]

Same problem @ornate elk

[onyx solar]

Inbound: All the people with issues after upgrading A1111 to 1.9

[solemn moon]

Oh no

[drifting kiln]

@ornate elk sorry for bothering, but does the thing i just did alter the output?

[ornate elk]

Nope, it will just use more vram and slow down the generation time

[drifting kiln]

alr my badf

[tame carbon]

Hi, i just downloaded a lora and it doesn't appear in lora list. It weighs 1.7gb but this model's page on civitai tells that it's a lora. I think a cant drop a link here cuz its a nsfw lora

[zealous merlin]

what's this error I'm getting on automatic1111 colab at the last step: ModuleNotFoundError: No module named 'diskcache'

Also, the colab notebook I was using before for invokeai isn't working either... the models don't seem to be downloading or something

[chilly knot]

i have this error on many model but epicphotogasm work for me very good

[sullen hound]

I am getting these weird distortions on an img2img i am trying. I was using, juggernaught lightning 4 step to make janitor batman look realistic and would get the distortions most of the time if the denoise was low, i set the denoise higher and they seem to go away but i want the image to be the cartoon but realistic. What might cause this? The vae?

[willow river]

please help, after update i got TypeError: 'bool' object is not iterable in img2img tab after pressing generate

[chilly knot]

animdiff 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 --

[chilly knot]

oh it cant be fix

only change videocard

[gloomy bolt]

i've been runing into this error ever since the 1.9 update

although i don't see any visible errors with the text2img whenever i use it

ok it looks like it's for ad-detailer, whenever it's being used this message pops up, but it looks like ad-detailer is still doing it's job?

oh and it looks like img2img is also suffering, the images turn out to blurry no matter what i do

[ornate elk]

Try an other sampler

For example DPM++ 2M and as scheduler Karras

[chilly knot]

animatediff error

EinopsError: Error while processing rearrange-reduction pattern "(b f) d c -> (b d) f c". Input tensor shape: torch.Size([2, 4096, 320]). Additional info: {'f': 16}. Shape mismatch, can't divide axis of length 2 in chunks of 16

[chilly knot]

if i change batch size new error

NansException: A tensor with all NaNs was produced in Une

[loud marlin]

Hello. I discovered the world of AI generators a few days ago learning about it. I chose to got with the stable diffusion route even though i don't have a GPU. I know it will be painstakingly slow without GPU and maybe I won't get far but wanted to delve into this world that I am VERY interested in. If I am able to compute with this stuff I would invest the $ to get the proper hardware. With that all being said I am in need of help for a new beginner. I can't seem to get the stable-diffusion-webui.git to install the necessary files as I get a return error of https: is not recognized as an internal or external command.. I am the proper folder \Automatic1111> and copied that line in and this is where I am stuck. Thank you for your kindness and understanding of my struggle with it. I was following a YT video verbatim and there doesn't seem to be any rhyme or reason for my not being able to move forward. Thank You.

[ornate elk]

Hey, what's your CPU?

The webui can also work on AMD CPUs Integrated GPU

@loud marlin If you have an Intel CPU, you should follow my install guide from here:
#🤝｜tech-support message
When your at Step 5. you only need these args: --skip-torch-cuda-test --use-cpu all --opt-sub-quad-attention --no-half

[loud marlin]

Hey. Yeah I have an AMD CPU. After further looking into my laptop I guess I do have a GPU but it doesn't show VRAM or I don't know where to look. Its a NVIDIA GeForce RTX 3070. I think this is a dedicated Graphics card but unsure due to not seeing the VRAM. Does your guide not work for AMD? I followed a specific set of instructions from a YT video and it doesn't do anything but give me that error code. Thank you for your help!!

[ornate elk]

Oh so you have a gpu

A 3070 will work good for stable diffusion

My guides cover nvidia and amd

Follow exactly these steps and it should work: #🤝｜tech-support message

For any questions feel free to ask 🙂

[ornate elk]

And yes a 3070 is a dedicated GPU with 8gb of vram.

[normal silo]

[proud cargo]

anyone having this error on animatediff? video_length = mm_animatediff.ad_params.batch_size
AttributeError: 'NoneType' object has no attribute 'batch_size'

[proud zealot]

"size mismatch for model" can not change the checkpoiont, anyone know how to fix it

[loud marlin]

Thank you! Look forward to trying that out tomorrow.

[merry locust]

Hi Everyone. Not sure if this is the right channel for this. I'm looking for a stable diffusion / MidJourney professional who can assist with a project on digitally altering images of socks. I have PNG images and 3D files of the socks. The goal is to take these images, keep the socks unchanged, and completely transform the model and background to a design of my choice. I would also love to learn this process. If anyone is skilled in these techniques and is open to collaboration and teaching, please DM me. I've attached an example of the final result we want. Thanks!

[buoyant lotus]

just crashed after loading model for a few seconds

[native iris]

I didn't check this server for quite a few months and I can't find the channels to generate images anymore (or it's just the wrong server)

Where to use the art ai

[limber pier]

Request images in #🏞｜general-with-images

[native iris]

I just saw the bot status channel and it said all bot channels are downed

I guess that's the reason I can't find them

[limber pier]

Go ahead and just post a prompt in #🏞｜general-with-images

[grim iron]

I fixed it with a reload thanks though

[sinful halo]

@ornate elk man, i split my pc into 2 for my gf but now i have only 16gb of ram, i was trying to render a image in a resolution that i already had done before but i got this error;
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 7.43 GiB. GPU 0 has a total capacity of 7.98 GiB of which 0 bytes is free. Of the allocated memory 2.26 GiB is allocated by PyTorch, and 6.79 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

is it bcs of my memory ram ? orcould be something else ?

[limber pier]

that's vram... your gpu

not ram

if you were able to do it before and can't now it usually means it's right at the limit of what you have on your system

try closing everything except SD and trying again

[sinful halo]

everything is the same, i only split my ram in half an i noticed my ram gets to 100% when generating the image
and i noticed that it is rendering slowly than before
i remember one more thing, everytime i got max vram error before, this >>> GPU 0 has a total capacity of 7.98 GiB of which >>>0<<< bytes is free was never zero. was aways 2gbs free or something

i formatted my pc and now i have windows 11,
before it was half these numbers
100%|██████████████████████████████████████████████████████████████████████████████████| 35/35 [06:15<00:00, 10.74s/it]
tiled upscale: 100%|███████████████████████████████████████████████████████████████████| 40/40 [00:14<00:00, 2.80it/s]
100%|██████████████████████████████████████████████████████████████████████████████████| 13/13 [04:23<00:00, 20.25s/it]
*** Error completing request | 48/5000 [13:39<25:12:53, 18.33s/it]

[limber pier]

how are you splitting your ram in half btw? virtual machines?

[sinful halo]

sorry, my english is bad, i literally took half my ram out of my pc to make another one

im using zluda btw, rx 5700 xt

[limber pier]

oh damn

[sinful halo]

idk if its really my ram or something to do with the new windows, driver etc, i have no idea

i have the latest amd adrenalin drivers

im using a sdxl model

[limber pier]

it's not your ram

[sinful halo]

okook

[limber pier]

you're prolly using stuff that has pushed vram use up just a little bit

close literally everything on your comp including file explorer windows and try again

restart the program too

even just having a few images open will cost you vram

anything that does graphics costs vram

[sinful halo]

hmm, any chance windoes 11 use smore vram than windoes 10 ? is the major change ive made

[ornate elk]

You need to increase your windows pagefile.
But the error comes from the vram.
Make sure wallpaper engine or any other Programm is closed.
You also should install the tiled diffusion extension and then enabled only Tiled VAE.
That can help.
Also enable FP8 mode for sdxl in the settings could help

[tiny garden]

uhm... Can I sort my LoRA's in different folders in order to have a better overview of what I have?

[ornate elk]

Yes subfolders work

[tiny garden]

cool

I'm kind of losing track of what I have when scrolling through my LoRAs and although I have it all organized in spreadsheets, I still lack some organization in that

[ornate elk]

I made lora folders based on styles, characters, clothing, and then also an sdxl folder

Helps alot, also in the latest webui update we can have the good lora overview option back

[tiny garden]

you're giving me an idea there
I just sorted after characters, outfits, and stuff... might establish the same sorting for sdxl

[heady pine]

Is there a way to download an older version of adetailer?

[sinful halo]

i think pagefiles and fp8 mode fixed it, i think ill buy an nvme (using sata ssd now) and set a lot of pagefiles for it, more ram will help too i think

oh no, the speed is normal again but i gor out of vram error. the tiled vae makes my cmd (webui) stop working

[ornate elk]

For sdxl more ram will help yes, set the page file for C to 24gb Max and disable it for other drives

Tiled VAE shouldn't be a problem. Also make sure to use the sdxl fp16 VAE file

[sinful halo]

ok, thanks so much

[ornate elk]

Make sure you restarted the webui after installing the tiled VAE and set the fp8mode

Also you get the sdxl fp 16 VAE from here:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main

[sleek zephyr]

Hi everyone, no support received in the past few days. Stability AI has generated some inappropriate images that I would like to delete from my profile, or close the profile if this is not possible. Would it be possible to receive support? Do you know how to do it? I sent several emails but never received any replies thank you all

[wet ferry]

why doesn't torch download nvidia ?

amd is doing a torch download according to the integrated graphics profit of the processor, and I can't stop it... or maybe I'm getting it wrong and he's uploading the right one... My mind went blank. can you help me?

[grim iron]

is it a laptop?

[wet ferry]

...

[wet ferry]

desktop

stable diffusion was using the 4060ti graphics driver yesterday, but today it is seeing the integrated graphics driver of the amd processor.

It was fixed by uninstalling the AMD integrated graphics driver before turning on stable diffusion.
But I have to do the same process every time I restart the computer. I hope I don't encounter the same problem in the future... If a problem occurs, I will write here again.

[wet ferry]

again...

[ornate elk]

Disable the integrated AMD GPU in the Device Manager

[civic mirage]

i got a question, if i want to use pony xl, do i download the sdxl vae and put it into vae and the pony file into models?

[wet ferry]

I disabled it but the problem continues.

[pale echo]

I updated my Automatic1111 to the newest version (1.9.0), but now it just randomly stops generating. No Error message or anything, it just does nothing. Any idea how to fix it?

[chilly knot]

animatediff error
[3:49]
EinopsError: Error while processing rearrange-reduction pattern "(b f) d c -> (b d) f c". Input tensor shape: torch.Size([2, 4096, 320]). Additional info: {'f': 16}. Shape mismatch, can't divide axis of length 2 in chunks of 16

[ornate elk]

What's in your webui-user.bat?
And also are you running the latest nvidia driver?

[ornate elk]

Can you make a screenshot of how it looks when is stops?

[pale echo]

automatic tab or console?

[ornate elk]

Both

[pale echo]

@ornate elk seems like i found the culprit, tho i dont get why. if i dont use Hires.fix it just works fine, but with it active it freezes. never had that happen before with other versions

[ornate elk]

Make sure to set hires steps to 10.
Also what's your GPU and what's inside your webui-user.bat?
You might need some performance adjustments

[pale echo]

@ornate elk had steps normaly around 15. did they changed it?
my gpu is (sadly) a rx 6700 xt
--use-directml --no-half --no-half-vae --opt-sub-quad-attention --precision full

[ornate elk]

ohhh, then your using the old webui version (the slow directml one)

you should checkout my AMD zluda Guide in the Pinned Messages of this channel.
if you install that, it will perform much much better (in speed and vram usage)

[drifting kiln]

my auto1111 just stopped working

i restarted it

it wont generate nothing

or load loras, or anything

[ornate elk]

What does the CMD shows?

[drifting kiln]

Applying attention optimization: Doggettx... done.
Model loaded in 5.0s (load weights from disk: 0.5s, create model: 0.5s, apply weights to model: 2.0s, apply float(): 0.3s, load textual inversion embeddings: 1.0s, calculate empty prompt: 0.5s).

it wont do anything else

i have a prompt

[ornate elk]

Is the generate button grayed out?

[drifting kiln]

nope

[ornate elk]

And what happens when you press it?

[drifting kiln]

interrupt/skip and then it just resets after a while

[ornate elk]

And the CMD shows nothing?

[drifting kiln]

no it doesnt register anything

[ornate elk]

Okay then make sure to whitelist the webui in any browser adblocker.
Also if using opera or brave, try an other browser

[drifting kiln]

wow

it worked

tf

the adblocker?

how

u a mfn genius @ornate elk

[crimson summit]

😭

[drifting kiln]

was it that obvious?

[ornate elk]

Yea ^^ because when the CMD doesn't get any input something between browser and CMD is blocking it.

Could be the browser or an adblocker

We had such cases before

[tame carbon]

pls help. i used sd 1.5 checkpoints before and tried to download a sdxl one https://civitai.com/models/133005?modelVersionId=348913
but when i tried to load it i got this error

my user-webui file settings
gpu is gtx1660 6gb

[unique bluff]

I’ve been having an issue with stable diffusion (V1.5 on Mac intel). It goes as following when trying to generate an image: AttributeError: ‘NoneType’ object has no attribute ‘lowvram’
In the command I use to open stable diffusion it reads: File "/Users/nameofuser/stable-diffusion-webui/modules/sd_models.py", line 662, in send_model_to_cpu
if m.lowvram:
AttributeError: 'NoneType' object has no attribute 'lowvram'

I also cannot use a different checkpoint than the one that already comes in stable diffusion, despite it being on the list (I’m not sure I used the proper term, I’m referring to the files you can add on the models folder)

(Also, I’m at all not good with python and coding and etc. I also couldn’t find an answer elsewhere online)

[ornate elk]

whats your torch and xformers version?

you can check it at the bottom of the webui in browser

[unique bluff]

Would there be any way to reinstall stable diffusion?

[vocal burrow]

simply delete it and reinstall it

check the pinned messages for a mac sd install tutorial

[unique bluff]

Hummm

Okay

Hopefully that solves the issue

[tardy crow]

Hello! Quick Question! Is there a limit to concurrent calls to Stability API Service? Did a 'ctrl-f' on the documentation page for "concurrent" and "same" (for same time) and nothing came up.

[warped hemlock]

Can I run 2 instances of StableDiffusion on the same computer? I mean, I've done it. And it seems to work. I'm just curious if there's any drawbacks or important things I should know. Can I load 2 different models? One in each instance of SD? Can I use Deforum in one and render stuff while using txt2img in the other? Stuff like that is what I'd like to know. Running ASUS ROG STRIX B450-F MB, AMD Ryzen 7 2700X CPU, 32 GB Ram, GeForce RTX 4090

[civic mirage]

quick question, comfyui wont allow me to pick neither a checkpoint nr a vae, what am i doing wrong?

[unique bluff]

Okay, I installed stable diffusion and the problem persists. Is there any way to fix that?

Because I’m not able to generate any images at all

[signal sigil]

I have install Ella in ComfyUI. I have the Ella Model, in the Ella folder,
and i have downloaded all the files for (flan-t5-xl-encoder-only-bf16)
When i go to run the workflow, it just gives me this huge error.
Has anyone ran into a simular issue?

[ornate elk]

Yes you can do that without problems. You just should also have enough RAM for both instances

[warped hemlock]

Thanks!

[warped hemlock]

What checkpoint loader are you using? If you try the SimpleCheckpointLoader and click on the down arrow, do you see any models?

[civic mirage]

well yeah i figured that i had to put the checkpoint and vae files into the comfy folder and now i can chose them but the images coming out are terrible

[warped hemlock]

I've found that bad images are usually a problem with the VAE. If the VAE is baked into the model, loading an external VAE sometimes helps. If you're already using an external VAE, try a different one. I've had that same problem and I know I fixed it by experimenting with the VAE. Good luck. Hope this helps.

[vernal robin]

https://www.youtube.com/watch?v=rHOOJIV9JiY&t=3135s hey guys, im trying to get this up and running, and it seems like the models are being loaded by custom nodes, I'm still new to comfy for animation. I've downloaded the models and loras, but they dont seem to load in when I refresh, is there a special way to manually re-hook them in?

here's the link to the civit page to download the workflow: https://civitai.com/models/350524?modelVersionId=411119

[pale echo]

@ornate elk can i just copy the text which i need to add to the PATH?
%HIP_PATH%bin i mean

[ornate elk]

Yes

[pale echo]

@ornate elk and which sd i start then? from the SD-Zluda or from the SD.Next Folder? or did i did something wrong there?

[tawny flint]

hi. I'm trying to do multi face swap on an image with Reactor and I getting an error. If you have time please point me in the right direction

No source face(s) in the provided Index

[ReActor] 14:35:40 - STATUS - Working: source face index [0], target face index [0]
[ReActor] 14:35:40 - STATUS - Analyzing Source Image...
[ReActor] 14:35:41 - STATUS - Analyzing Target Image...
[ReActor] 14:35:42 - STATUS - Trying to halve 'det_size' parameter
[ReActor] 14:35:43 - STATUS - Swapping...
[ReActor] 14:35:43 - STATUS - --Done!--
[ReActor] 14:35:43 - STATUS - Restoring with GFPGANv1.4.pth
[ReActor] 14:35:44 - STATUS - Working: source face index [1], target face index [1]
[ReActor] 14:35:44 - STATUS - Analyzing Source Image...
[ReActor] 14:35:45 - STATUS - Analyzing Target Image...
[ReActor] 14:35:46 - STATUS - No source face(s) in the provided Index
the image has two faces

nm. i figured it out

[ornate elk]

Don't install SD next

Its written in the guide.
You only need to do the rocm library folder replace

And after that reboot the PC.
Then launch the webui-user.bat
If everything is done correctly, you don't have any errors in the CMD and when generating the first image it will take a while.

[severe sand]

@ornate elk any possible fix since i last checked here?

[pale echo]

i really didnt saw that, now i feel dumb ^^
testing it again, wish me luck. oh btw,do i keep the " " for the models commands?

[nova wren]

having an issue trying to generate using tensorRT and control net, i get an error about expending all tensors to be on the same device

[ornate elk]

If you added the custom paths for the model folder then yes

[balmy rivet]

Guys! I have RYZEN 5 5600H, 16GB RAM, Nvidia GTX 1650 4GB VRAM LAPTOP, can I run Stable Diffusion on it?

[drowsy creek]

how do I do this on arch?

pacman doesn't have that specific package, and it's not in the AUR

still somewhat new to linux, only switched like a month ago

[olive birch]

Is anyone using the Stability AI Developer Platform? How is your experience?

[drowsy creek]

found python3.10 in the AUR (it was called python310), but it isn't python3.10-venv

what do I do?

[ornate elk]

Yes you can. Follow the install guide for Nvidia in the pinned messages of this channel

[sinful halo]

Man, tiled var is that option alone ? Or is the control net tiled vae ? I activate it with hi res at the same time ? I'm not finding the right answer on Google

Bcs when I checked only the tiled vae option alone (the one slight above control net) it didn't nothing

[vocal burrow]

what are your settings for this one ? (screenshot the whole webpage)

[dapper basin]

trying to install stable diffusion using this method and running into error below

any idea what's wrong and how i can fixx it

???

[vocal burrow]

looks like a network error.

use cable instead of wifi if you can

[ornate elk]

Only tiled VAE. Then check the CMD

[ornate elk]

Network error. Retry when you have a stable connection

[drowsy creek]

is there an install guide for AMD arch?

the package the standard linux guide requires doesn't exist on arch

though, I guess I am just out of luck and going to have to use the huggingface space

since if I keep asking it's "begging" and I'm ignored, but if I don't my question gets buried

I just can't find any tutorials at all specifically for how to install SD on arch

[vocal burrow]

Because arch is for edgy users / those who like to live on the bleeding edge of technology. Which does not play nice with SD which use """"old"""" versions of python and stuff.

arch is designed for advanced linux users.

[drowsy creek]

"edgy users"?

you mean edge case

[vocal burrow]

it s a joke.

[drowsy creek]

or are you calling me an edgelord

[drowsy creek]

ah, ok np

[vocal burrow]

edgy user / bleeding edge

but yeah I would definitely not recommend arch for sd

As it will require you to compile a bunch of "outdated" (by arch standard) packages and make them live along with your other system packages

it s possible but it s definitely for advanced linux users.

[drowsy creek]

so there is no feasible way to run SD locally then?

since I don't wanna dual boot just to use one program

[vocal burrow]

btw since you have an AMD GPU, https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs#install-on-amd-and-arch-linux

@drowsy creek

[drowsy creek]

thanks

👍

[sinful halo]

i think it did it but its stuck when finishing (saving ?) the image, i dont see anything being used on task manager

oh shit, i did it with latent upscaler

[ornate elk]

Make sure you whitelist the webui in any browser adblocker (would recommend to delete that adblocker and install ublock origin, as its safer and better)

Your resolution is also a bit high for upscaling

[sinful halo]

thanks man, youre even helping me outside the "stable diffusion box" can i buy you a coffee ? i appreciate it.

[sinful halo]

my native resolution ? 1344 ?

what should i do ? render in a lower one ? even if thats an sdxl model ?

[ornate elk]

No just lower the upscale to 1.5 or 1.3

[sinful halo]

ok

[ornate elk]

Test if that works, then you can go higher

[dapper basin]

ok

[sinful halo]

any hint about the encoder and decoder tile size ? i put 1344 and 128 respectively

i seached about this but didnt find any information from someone with 8gb vram as well

[ornate elk]

Let them default. If you get an out of vram error then lower the encoder size

[sinful halo]

oks. what about your coffee bro

do you have it ? buymeacoffee ?

[ornate elk]

Nope didn't set it up, but I wanted to

[sinful halo]

when you do it ill be the first xD

i think a lot of people will buy it, im sure everyone appreciate you helping here

[ornate elk]

Thanks! 🙂

[sinful halo]

np man

[karmic nova]

my im2img images get red spots in certain spots, even though my source material has no highlights or overexposed parts (especially not in those areas). Is this a common issue?

It's almost always in that upper left corner of the forehead, if that helps... sometimes at the point where the neck meets the collar, as well, and often ears.

[dapper basin]

omg
now what's wrong?

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram-sdxl

call webui.bat

did I edit this wrong or something?

[pale echo]

@ornate elk works great so far i would say. no freezes and i even can use hires.fix freely (tested it with upscale by 2, normaly i had to use like 1.3). but my Additional Networks doesnt shows any usable loras. or is it not compatible?

[uncut jewel]

Hey guys! I dont really know where to ask that but let's do it there 🙂
I'm using ComfyUI to generate imagery and benchmarking different upscaler methods. Since a long time ago when i'm using the SD Ultimate Upscaler node i get this tiles issues appearing on my output ( kind of a grid on the final image where there's subbtle colors difference delimited by squares in the images which looks quite bad ). I tried to change the settings on the node to get rid of that issue but still have this tiling appearing at some point?
Anyone know how i could upscaler with that node without having the grid/tiles appearing on the output ?

[ornate elk]

Great to hear, The additional networks are compatible, they should show up.
Did you used the --lora-path option in the webui-user.bat or did you moved the Lora's manually?

[pale echo]

--lora-path option. to be honest, i dont even have a lora folder in my zluda model folder.

[ornate elk]

Hey, can you post the full CMD error log?

[ornate elk]

Then recheck if the path is correct. In my guide is just a placeholder.
Also Lora's only show up when a compatible model is selected.
Meaning 1.5 Lora's work only when using an 1.5 model and sdxl Lora's only when using sdxl

[dapper basin]

sorry i closed
wait i'll run it again if it shows the same i'll post
meanwhile can you check my .bat file if it's alright or not

[ornate elk]

The bat is correct 🙂

[dapper basin]

I have rtx 4060 8gigs

[dapper basin]

thanks

[pale echo]

the strange thing is, i can use loras in my prompts just normaly. and they showing up under the Lora Tab aswell. Only my Additional Networks cant see them

[ornate elk]

Do you mean the additional network extension?

[pale echo]

yes.

[ornate elk]

That thing is outdated since a year :/ so isnt supported anymore

Or maybe it only detects if the Lora's are in the lora folder of the zluda webui version

@pale echo something you could try, make a Lora folder in the models folder and place a lora in there

[pale echo]

tryed that, but then sd itself cant find the lora anymore. guess its really just the extension which isnt compatible anymore^^

[dapper basin]

so this time it happened right off the bat

[ornate elk]

@pale echo Okay, then no chance. Also there is no active fork of that extension

[ornate elk]

Seems like the default model wasn't correctly downloaded (maybe cause of a network error).

What to do know:
Go into stable-diffusion-webui/models/stable-diffusion folder
There delete the 1.5 ema pruned safetensor file.

Then download for example the Dreamshaper v8 model from here:
(Its a good starter model) and place it in the same folder where you deletet the other file. Then relaunch the webui-user.bat
https://civitai.com/models/4384/dreamshaper

[dapper basin]

roger that!

[gilded canyon]

Hey, I'm trying to use SDXL for the first time on Web Ui. I've tried just regular a1111 and web ui forge, but loading in my SDXL checkpoint seems to never work due to timing out or some ram error. I have a 3060 12GB, so I don't think vram is an issue.

any configurations or launch args I should be using for this specific GPU?

[ornate elk]

In auto1111 you need to add --xformers --no-half-vae to your webui-user.bat

Also what's the CMD error you get?

[gilded canyon]

not on my computer right now but ill try those arguments tomorrow and ask again with the error if I run into it, thanks for the speedy reply tho 💙

also, should i upgrade my RAM (not VRAM). I have 16gb, but idk if that affects a1111 or using SDXL

[ornate elk]

It affects sdxl. But depends on the error you get.
You may need to adjust your windows page file if the error is caused by RAM and not Vram

[gilded canyon]

ohh okay, thank u

[dapper basin]

is it done?

[ornate elk]

Yes, it should have opened a new browser tab

[dapper basin]

shit this was fast

thanks again CS1o

[ornate elk]

No problem 🙂

[ornate elk]

If the image looks unsaturated, you need to use a VAE.
Download the one from here and put it into the models/VAE folder.
Then go into Settings >User Interface >Quicksettings and there add sd_vae
Then hit apply and reload ui. Then select the VAE in the new dropdown.
https://civitai.com/models/276082/vae-ft-mse-840000-ema-pruned-or-840000-or-840k-sd15-vae

[severe sand]

Not home rn will get to it later

[silver wren]

I just sent in a support ticket, but I figured I'd drop a message in here too. It seems that on the v1 API, when a prompt fails the moderation check, the system returns a "{"artifacts":[]}" instead of the proper error message.

[tardy crow]

does anyone know if stability's web service has a concurrent calls per user limit?? couldn't find any doc to that on their page.

[haughty zodiac]

Hello i'm training a LORA but my s/it is way too high
I run on a 3080 10g VRAM
Here are my settings

[haughty zodiac]

please help me, I'm desperate right now