🤝｜tech-support

[slow garden]

Sweet, let me clear space and ill report back

Thanks man

I now have this error sadly

Gpu is a 1650 super

I added in the arguments for the card, prob solved, thanks man

[ornate elk]

okay, make sure you have --xformers --medvram --no-half

[vagrant notch]

Do the models SDXL TURBO and SDXL lightning work on A1111?

[agile burrow]

I will try to run a distrobox with ubuntu

[ornate elk]

yes

[vagrant notch]

I put the models in the model file but they don't show up on A1111... Is there something I should do differently than with classical models?

[ornate elk]

nope they should be then selectable

make sure the collab version runs auto1111 1.7.0 or newer

[gilded nacelle]

So for whatever reason, animagine isn't working well. However, I'm currently trying different models and getting good gens. Thank you once again for helping me with the amd setup. It is very truly appreciated!

[ornate elk]

no problem 🙂 try out the AAM XL AnimeMix Model, its great!

[gilded nacelle]

Will do!

[ornate elk]

also thanks to you i now have added the 7700S into my guide so others know it can work

[gilded nacelle]

Awesome! Glad I was able to be a successful test subject

[ornate elk]

The latest Stable Diffusion install Guides for Nvidia and AMD:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[bleak mirage]

@ornate elk your a God send it works now 😁

[thorny valley]

hey guys I have a really dumb question. I just started using comfyui. how do I see my generation speed?

[ornate elk]

hey, its the it/s or s/it in the cmd

[ornate elk]

perfect! no problem 🙂

[bleak mirage]

👌

[thorny valley]

oh lmao I didn't even check cmd. does it tell you the time it took to generate too? I'm not in front of it right now

[ornate elk]

it actually does!

[thorny valley]

:O

[thorny valley]

thanks for checking! I wanted to compare some generation times vs my old a1111 "workflows"

[ornate elk]

no problem 🙂

comfyui is mostly a few seconds (5-10) faster

[wheat helm]

does someone know where to put commandline arguments when you get an error

for example

--disable-nan-check

ive done this before but just cant remember where lol ik its probably simple

[ornate elk]

whats your gpu? and when do you get that error

no one should ever use --disable-nan-check, it wont fix errors, it ony hides them

[jagged talon]

I paid for a month of stable diffusion and am not getting credited.

[brave frigate]

anyone know why when using Hires. fix it keeps giving me RuntimeError: Not enough memory, use lower resolution (max approx. 576x576). Need: 0.4GB free, Have:0.1GB free

nvm mi fixed it

needed an embedding

[shy charm]

i'm trying to figure out how to install forge on top of A1111, do i just run these commands in git bash opened in the stable-diffusion-webui folder?

i don't want to accidentally mess something up so i just want to make absolutely sure this is what i'm supposed to do

[wheat helm]

ahh ok well that was close lol glad i couldnt remember. Ive never actually used that exact line before just meant using the place the arguments go

and my GPU is a GTX 1660 Ti

i got the error when I attempted to generate deforum

and also when i attempted to generate img2img

[south vigil]

hey so everytime when i want to create multiple pictures with the batch count and also add an upscaler from hiresfix the pictures get more and more destroyed until there is only 1 overlayed color (the batch count would be normally on 10)

4th pic looks like this

[ornate elk]

If you do it like the guide says then you mess it up anyway. Better make a separate folder for torch install

[ornate elk]

Then make sure you have --xformers --medvram --no-half
In your webui-user.bat at the line commandline_args=

[shy charm]

i ended up having forge as a separate installation

[ornate elk]

What's your webui version?

[agile burrow]

@ornate elk
dosent work on distrobox

[terse lichen]

from where can i get inference core dependencies or how to go about it ?

[ornate elk]

no module named pip is the cause

something wnt wrong when installing python. on windows its caused when the AV blocks python

[ivory bramble]

Getting this error:
RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check
I read the sticky carefully and followed instructions for NVIDIA users.

My web-ui.bat currently looks like this:

@echo off

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

call webui.bat```

It's a fresh install. It was working about 10 minutes ago. I installed a few extensions:
<https://i.imgur.com/Pv404KF.png>

*Then* I got an error in my CMD saying that SD didn't have permissions to write to the folder, so I changed security settings. I have no anti-virus, and only use Windows Security.

I also downloaded a few safetensors models. Other than that, I started getting the RuntimeError: Torch error again. It was initially the reason why I did a fresh install in the first place.

I also looked around here on Discord, and I saw users saying NOT to add "-skip-torch-cuda-test" so I'm not sure what to do. I do not want to use my CPU.

Full error in my cmd

Also, I looked around here, and I saw people saying NOT to add "-skip-torch-cuda-test" so I'm not sure what to do.

Full error in my cmd ```C:\Stable Diffusion\stable-diffusion\stable-diffusion-webui>git pull
Already up to date.
venv "C:\Stable Diffusion\stable-diffusion\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.9.4
Commit hash: feee37d75f1b168768014e4634dcb156ee649c05
Traceback (most recent call last):
File "C:\Stable Diffusion\stable-diffusion\stable-diffusion-webui\launch.py", line 48, in <module>
main()
File "C:\Stable Diffusion\stable-diffusion\stable-diffusion-webui\launch.py", line 39, in main
prepare_environment()
File "C:\Stable Diffusion\stable-diffusion\stable-diffusion-webui\modules\launch_utils.py", line 386, in prepare_environment
raise RuntimeError(
RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check
Press any key to continue . . .

Line 48 of pytorch is main()
Line 39 is prepare_environment()

My PC specs are here:
https://i.imgur.com/JuffFfQ.png

I just deleted the venv folder as suggested in another post here. It's downloading again. Will see what happens.

Sorry for the flood of posts

[honest tendon]

R-ESRGAN AnimeVideo

how do i add this in

nvm

[opaque sphinx]

why i get this msg when i open the open audio ai thing ?

[ivory bramble]

I'm not support here, but what browser is that?

[vocal burrow]

what is "open audio ai thing" ?

[ivory bramble]

Again, I'm not support, I'm sure someone else will chime in later but;

I searched Audio AI thing on Google and I couldn't find anything having to do with Stable Diffusion.

What I would do is find their github page and see if a Discord is listed to ask the devs, or post on their GH page.

From what I've read about Revoked certificates so far, it could be from either end. You could might have some anti-virus running, maybe a VPN, or the browser (or browsers) are trying to protect you from something that is deemed a threat.

[ornate elk]

also make sure you have the latest nvidia driver installed

[ivory bramble]

Ah, good idea. I don't think I've opened GeForce experience in quite some time.

But adding onto Leven's thing; it could be something on their end. You can use this site to see if it's revoked for all: https://www.ssllabs.com/

[ivory bramble]

Do you use a NVIDIA gpu?

[ornate elk]

i have an old GTX1080 and a newer AMD GPU

[ivory bramble]

That's what I have

Mine in 1080ti

How do you use a NVIDIA and AMD together?

[ornate elk]

the error you got most likely will disapear after a driver update

[ivory bramble]

Downloading it now

[ornate elk]

2 different PCs

[ivory bramble]

Oh, gotcha

Is AMD the way to go these days? My 1080ti can still handle most gaming but it's still dated

[bitter bridge]

how to generate an image please help

[ornate elk]

it depends. For gaming they offer more vram for less money. but for AI tasks they are behind nvidia

[bitter bridge]

I don't
Understand

[ivory bramble]

I see. I'll let you take that.

😄

[bitter bridge]

......

[ornate elk]

i switched from my GTX1080 (8gb vram) to the AMD RX 7900XTX (24GB Vram) xD no problems with any game anymore haha

[ivory bramble]

[ornate elk]

where? localy or on this discord?

or online?

[bitter bridge]

on discord

[ornate elk]

on this discord you need an active subscription at https://stability.ai/stable-artisan
(there is also a trial option)
then login in on that site with your discord to connect both accounts.
then you can generate images in #1237459938901491852
More info in #artisan-faq

[bitter bridge]

Thank you

[ivory bramble]

Ahhh unfortunately didn't work. Same CMD error. I might just do a completely fresh install again.

I'm not sure why it happened, though. It was working fine. The only setting I changed was to add vae on the top of the UI and installed some extensions, added some models.

Even after deleting the extensions and models, it still gave me the error

Is there a way to delete all config that will regenerate fresh when I launch web-user.bat again?

[south vigil]

v1.9.4

[honest tendon]

im gonna be installing comfy ui, do I have to make a clean reinstall of everything?

gpu 6950xt, 12700f

[ornate elk]

Hey for that checkout the first link in the pinned messages of this channel.
There follow the Comfyui Zluda guide

[ornate elk]

Better reinstall it clean by following my guide from the pinned tab

Then you can move your models over

[honest tendon]

okay

https://youtu.be/8rB7RqKvU5U?si=udub2ff3vVxRQs9c

btw will this guide also work

[ornate elk]

not really, it will be very difficult and anoying (and it has wrong steps and outdated information in it)
conda isnt needed at all and not recommended. Also directml version will be much slower for your GPU than with ZLUDA

thats why i made a better guide

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides#amd-comfyui-with-zluda

[opaque sphinx]

it is the windows main antivirus

[opaque sphinx]

i meant this

thanks a lot i ll search

[ornate elk]

im getting the same error

so its on their end

[opaque sphinx]

oh i see, thanks a lot

[ornate elk]

if you still want the model then here is the link:
https://huggingface.co/audo/stable-audio-open-1.0

[opaque sphinx]

oh thanks a looot

[ornate elk]

np, it can be used with this tool:
https://github.com/diontimmer/audio-diffusion-gradio

or in comfyui with an extension:
https://github.com/lks-ai/ComfyUI-StableAudioSampler

[opaque sphinx]

that is really so helpful you the best T^T

[ornate elk]

no problem 🙂

[azure magnet]

Hi, Is there someone that could help with ComfyUI?

Why don't load the checkpoint that i have in automatic1111 folder?

[ornate elk]

hey, have you editet the models_path ymal file?

[azure magnet]

Yes, this is my line: C:\Stable diffusion

I think i make a mistake in a point but idk where

those are the other lines: checkpoints: models/Stable-diffusion
configs: models/Stable-diffusion
vae: models/VAE
loras: |
models/Lora
models/LyCORIS
upscale_models: |
models/ESRGAN
models/RealESRGAN
models/SwinIR
embeddings: embeddings
hypernetworks: models/hypernetworks
controlnet: models/ControlNet

[ornate elk]

you need to edit the base path to the folder where the models folder is in

like: C:\Stable-Diffusion\stable-diffusion-webui\

[azure magnet]

I'll try rn

[azure magnet]

Don't work

[ornate elk]

thats how it looks for me

[azure magnet]

This is mine. Look identically

[ornate elk]

and your file is called extra_model_paths.yaml ?

[azure magnet]

yes, but i have even extra_model_paths.yaml.example, maybe should have i to delete?🤔

[ornate elk]

yes

only have one file

[azure magnet]

Still don't work

[azure magnet]

maybe In found an error: Prompt outputs failed validation
CheckpointLoaderSimple:

• Required input is missing: ckpt_name

[honest tendon]

okay thank you

[ornate elk]

What model have you selected?

[ornate elk]

NP, feel free to ask if you have any question

[azure magnet]

No one, if i click on the arrows don't work

[azure magnet]

I even put the path of folder of the models but doesn't works

Ok now work the problem was the backslash in end of the string

Ty

[ornate elk]

Oh, np

[honest tendon]

wheres the add to path?

@ornate elk

wait nvm its Enviroment Variables right?

wait your guide is actually really easy to follow

[ornate elk]

yes

[ornate elk]

i hope so ^^ glad to hear

[honest tendon]

also what is Fooocus?

is that a new UI?

i had 50gb worth of recycle bin trash

[ornate elk]

Its an other webui for people who come from Midjourney

Has much less features but is easy to use

[desert grotto]

What I can generate with my own stable diffusion is much worse than what civitai can with the same prompt, settings and resources.
Does anywone know why that might be?
Maybe some auto include loras or upscalers that I am not privy to?

[honest tendon]

oooo, it shouldnt be an issue if i download that in tandem right

i wanna try both

comfy looks more hackery

which i like

[desert grotto]

Go with comfy, with mods, it is so convenient

[ornate elk]

no problem, you can have every webui installed at the same time

if you installed zluda once then its easier to setup for the others

[cosmic furnace]

Any known improvements with amd lately?)

[honest tendon]

quick question

is this right? there wasnt a check box for pro drivers

its sending.. omg wifi pls work

[cosmic furnace]

I wonder just when zluda will be at the same speed as linux rocm >_>

[honest tendon]

nvm!

it works

[honest tendon]

how do i open the file without having to open it as admin all the time? if i dont it doesnt open its weird

is there an automatic way to do it

oop now it doesnt work

ah shit

i think i fucked up somewhere mostly on this step i think

did i make a mistake?

WTF

i followed this and downloaded all the right files though, thats odd

@ornate elk

[ionic grotto]

guys i get this error when i downloaded an upscaler and when i use it i get this : TypeError: argument of type 'NoneType' is not iterable

generations only go to half and then i get the error

[ornate elk]

yep thats rght

[ornate elk]

have you restarted the PC after installing hip sdk?

[ornate elk]

yes here is the fault

you have zluda folder inside zluda folder

then your environment path may be incorrect

[honest tendon]

ah gotcha thanks!

have a good night or day

[ornate elk]

Thanks you too.
Hope it works now. But recheck your Environment paths

Also restart the PC one time after Installing the hip sdk

[honest tendon]

yay

JEEZ THIS THINGS QUICK

after the 40min Q time though lol

btw I installed comfyui manager if thats alright, idk what its for but im installing it lol

do these work for all cards?

[ornate elk]

yep comfyui manager works. Its for installing extensions

not all extensions will work, you have to try
for reactor i found a workaround to make it work

[ornate elk]

also if you used auto1111 before but with directml, you should replace it with the zluda version of it to get the speed and better vram usage like in comfyui

[ornate elk]

do you have the full cmd log?

[teal adder]

Hello

stupid person here

im trying to train a model using the webui

I added the Dreambooth extension, but I cannot find the input tab.

Am I missing something?

seems like its been broken for a while, I cannot find much about this.

[ornate elk]

Hey, I would suggest to not use Dreambooth but Khoya_ss for training

Its a seperate tool that won't break your webui like dreambooth does

[teal adder]

prefacing, I am very much a noob lol

I am trying to train a model after a specific artsyle

ive heard dreambooth is good for artsyles

i will look into Khoya_ss

ty

[ornate elk]

For styles you should train a lora

Much easier to do

[teal adder]

I thought people added Lora on top of Dreambooth

[ornate elk]

Khoya_ss is perfect for training loras

Lora goes on top of models

There are a lot of styles and character loras on civitai

Much easier to use, and to train

[teal adder]

I thought civitai had already made models

and excuse my ignorance here, would that muddy my data set that I am trying to train it on?

[ornate elk]

Civitai is the largest database of community made models, loras, embeddings and more

If you have a large dataset like hundreds of images then you can train a model if you really want to.
If you have like 15-75 images you can easily train a lora on your artstyle

[teal adder]

I have around 1300 images

in the specific style I am looking for

[ornate elk]

Ah okay

What's your GPU BTW?

[teal adder]

I have a 4090 from work

[ornate elk]

Thats perfect for training

[teal adder]

the problem is I am having a real hard time getting the tools lol

or rather using the tools

[ornate elk]

Yea training is not an easy task

There are a lot of outdated tutorials out there

[teal adder]

Thats exactly it

its like when I was first learning linux

there is always this one guy who knows exactly how to solve everything and his steps are not documented anywhere lol

Do you do captioning for images ?

[ornate elk]

Yes its needed

[teal adder]

what do you use?

[ornate elk]

But Khoya and dreambooth can do auto camptioneing

I only used Khoya

But will use OneTrainer soon

[teal adder]

can I upload my own if I generated my own before?

[ornate elk]

Yes

Caption files are only txt files

Can be loaded and used in any train tool that supports it

[grave hornet]

does stable diffusion really not work with amd graphics cards

[ornate elk]

Thats wrong. It works with AMD very well, depending on the card

I'm using it with my AMD GPU

And if you check the Pinned messages of this channel you'll find my AMD install Guides

[grave hornet]

Alr thank you

[fair oxide]

question

i got a really good char that i like. i tried using the same seed and input it into a new txt2img and wanted to change the position of the char, but nothing really changed after changing the prompt a good bit. how can i use the same char?

but in different scenarios/different features

do i play with the variation strength?

[grave hornet]

i have a few questions

i cannot find the %HIP_PATH%bin

and i opened the web-user.bat and it closed instantly without showing anything on this step

also is this right?

i tred opening it again after completing the steps but without that file i couldnt find still nothing

[subtle timber]

Been struggling so bad having issue after issue with kohya,

ValueError: torch.cuda.is_available() should be True but is False. xformers' memory efficient attention is only available for GPU

and

                             captions for these images. If class token exists, it will be used. ```

I've gone and checked and rechecked that the caption file extension checker is set to .txt and all my files are .txt Also the direction is set correctly, it should be looking in the right place as its finding the images just not the txt files

Any suggestions would be super appreciated, 😅 I'm going on day 3 of agony.

Going to try oneTrainer here and see if that works otherwise I'm out of options.

[grave hornet]

its saying it cant find python

i did add the hip_path but still not working

[subtle quarry]

Do what it says, if you haven't already. You need to disable the microsoft store python in settings.

[ivory crest]

I found everydream2trainer easier to learn then how to use discord 😛
Btw, just ask Gemini or Claude if you get stuck 😉

[teal adder]

Dream2trainer ?

[ivory crest]

Soooo what will the min GPU be for SD3 do you all think?

[ivory crest]

Every dream trainer apparently. I have low GPU, so went for that one instead

[teal adder]

I'm in the midst of a training a model rn. We'll see if I messed up or not

[ivory crest]

Which one are you using to train the model?

I trained a model, with only 8gb GPU. It worked! Took a few hours. It sucks hahahahaha

[teal adder]

SB2 with LORA

It's got like 1250 images to work with

Hopefully it looks like anything

[ivory crest]

Now for loras I just cheat and use civita haha

I should learn how to do loras on my own system

[teal adder]

It was a headache so many sliders

And text fields

[ivory crest]

Apparently the tagging is the important part, fortunately there are auto taggers online

[teal adder]

I used auto tagging and reviewed a few

Its 1250 images

It would take days

[fair oxide]

is there a way to use like generative fill using impaint

or do you have to match the prompt and make it specific to what you wanna change

[grave hornet]

its stuck on this now is this normal

nvm

😦

what do i do about that

[subtle quarry]

fix this

if you manually put in the directory, you put \ not /

[grave hornet]

i dont think i did i just open the web user and everything does automatically

[subtle quarry]

Well, my help has been drained 😆 I know jack nothing about AMD.

Or what that error is.

[grave hornet]

sad

[subtle quarry]

You'll have to wait on someone who knows.

[grave hornet]

i tried something if it doesnt work ill just wait until someone who does know shows up

[teal adder]

I tried training usijng kohya_ss and the model it spit it out does not work lol

[grave hornet]

something different is happening it seems good

[grave hornet]

how do you add new checpoints i got one from civit ai

so i saw somewhere that if you had less than 8 gigs of ram to change medram to lowram and i tried to change it to lowram but it didnt work

also trying to generate an image doesnt seem to be working

itll say waiting than itll dissapear no loading bar or anything

and if i press interupt nothing happens

[orchid maple]

Trying to run the program and download it through the Nvidia way but it stops at this

Don't know what to do.

[ornate elk]

You need to disable the igpu

If you have an amd CPU + gpu

[grave hornet]

i fixed that issue but now the images just wont generate

[ornate elk]

What did you fixed?

[grave hornet]

the issue where it wouldnt open

[ornate elk]

And how?

Also what's the error in CMD now?

[grave hornet]

i mean its getting past it and the actual page is opening

hold on let me open it again

i try to put in prompts but when i press generate the loading bar doesnt show up and it doesnt load

and pressing interupt doesnt do anything

[ornate elk]

I need a full cmd log

[grave hornet]

i thinkthats it

oh wait its working

it wasnt earlier my b

[ornate elk]

Looks good

First gen can take 15-40 minutes

After that, its really fast

Just let it load

[viral knot]

40 minutes 😮

[ornate elk]

Hey, please follow the install guide in the pinned messages of this channel

[subtle quarry]

Hey, why is my inpainting removing parts of an image instead of adding to it?

It's happening in both a1111 and comfyui and it's frustrating me.

Might've been using the wrong nodes in comfyui.

[solemn moon]

anyone else not SD'ing?

[pseudo prairie]

bit random whenever i use discord call thing to communicate and play game at the same time, The game noise stops coming from my headphones. only the call sounds comes from the headphones as for the game it comes from from monitor

how can i fix that

[candid shale]

how can I install Xformers on a windows? most guides seems to just tell me to select it on the list but I don't see it.

[solemn moon]

add to yr command args

[candid shale]

Add what?

[solemn moon]

--xformers

next to yr --med vram

[solemn moon]

you get it?

[candid shale]

Yep, it worked, thanks

[subtle quarry]

wrong kinda tech support my guy.

[austere quartz]

Hey

Is there any chennal where i can write prompt and get images ??

[dark viper]

Uhm... #1237459938901491852 but it is a subscription service

[austere quartz]

Any free one ??

[dark viper]

You can try other online image generator though

Like Civitai and PixAI

[austere quartz]

Please tell me wich one is better

[austere quartz]

Ok thanks

[obsidian carbon]

Anyone else having seed stuck on fixed in comfyui regarding the efficiency nodes? I’m trying to do a Lora test with xy plots but the seed is stuck on fixed and basically just made the same image over and over again ://

[ivory crest]

This sounds extremely obvious, but you have it set to random after each generation right? Also, for upscaling, if it also uses a seed, sometimes that can make things far too similar when not set to random.
Also, have you done a hard refresh using comfy Manager?

To answer, in the past yes, today no.

[obsidian carbon]

There is an issue with the seed is the problem - recently, whenever I use -1 or chose randomise in efficiency nodes, the seed keeps straying fixed to the point I had to change it to a seed input and keep it on randomised. In the upscaling, I may change that then to randomise also to see if that changes it

Also, mind my noobness, what do you mean a hard refresh?

[obsidian carbon]

like heres an example of how it should look vs whats happening -- mind you no settings changed and the workflow is the same

[ivory crest]

That restart button

[obsidian carbon]

that I did in the past but I will try it now and see what happens // should take around 10 mins for a xy plot to appear

[ivory crest]

Closing the window, restarting with Manager, and even restoring my computer if need be, has so far fixed all of these sorts of problems for me.

[obsidian carbon]

I will try those in a bit. I know I’ll be getting a new SSD soon as that will mean I have to get things hard reset but it’s just being poopy of late — I’ve noticed online some people have mentioned that K Samplers have been acting up lately with the seed stuff. I can send my workflow and show you it to see if I’m missing something that’s causing it?

thats my workflow - let me know if theres something here thats not working right :

[ivory crest]

A screenshot of your comfy workflow as an image would be better for me. I'm lost when it comes to coding, but also I don't dl random stuff online.

[azure magnet]

what is control after generated in comfyui?

[obsidian carbon]

its a .json file, download it and drag it in

[civic canyon]

Hi, Can someone explain to me why when trying to change color in img2img inpaint of something on a picture I'm getting a burned out result? Like this one on screenshot. Same thing happens with hair too. My settings on second screenshot. When trying to do "fill" it gives a different result with "washed out wavy" image, if like it tried to generate the whole image and then shove it under the mask. I've seen XriCT doing this with an extension, but it has to be possible without extension. Happens with these models btw: default one, the one on screenshot and realisticvisionv6. Didn't try the rest but pretty sure it's me being an idiot

[ornate elk]

Your using AMD.
If your on the directml webui you need to have --no-half in your webui-user.bat to make Inpainting work

Also set the inpaint area to "only mask"

[civic canyon]

Ooooooooooooh man. Okay I'll try it in a while, currently afk. Didn't know that.

[ornate elk]

Yea its only needed for the directml webui version. And it will sadly increase vram usage by alot

With your 4gb gpu I would create two webui-user.bat files.
One with --no-half for Inpainting.
And one without it (but with the args from my guide) for the better speed

[civic canyon]

Ummm, I've got a 4Gb amd gpu and when I'm generating it's already using 99% of it. Is it gonna crash? I know there sre flags to run in low-vram mode, so do I need that? Currently if I try to use hi-res to up it to 1024 it crashes with couldn't allocate memory to gpu

[civic canyon]

Thanks
Dunno what guide you're referencing, can you share the link? Dm is fine

[ornate elk]

Make sure you use the recommended cmd line args from the Guide in the pinned tabs

[civic canyon]

Oh, pinned tabs. Alright, thanks mate!

[ornate elk]

@civic canyon this one:
#🤝｜tech-support message

[civic canyon]

Will ZLUDA work for Radeon Pro 5300M btw?

[ornate elk]

I checked it and it could work

But can't guarantee it

[civic canyon]

Ok. Worth trying I guess

[ornate elk]

Sure, at step 4 from zluda make sure to download the RocmLibs.zip

[civic canyon]

worked like a charm, thank you!

[ornate elk]

No problem 🙂

[ornate elk]

Also install the tiled vae extension

Enabling that can prevent oom errors when upscaling to high

[civic canyon]

I'll try. I got only ~6 gb left on disk😁😁 running a bootcamp on work macbook hehe. Plus I have Forza installed soo..forza....or sd...Fooorza......or sd...

I've seen a portable version of SD, will try doing that soon I think. Will move it to usb stick

[ornate elk]

Ahh rip storage xD

[civic canyon]

what if I want to generate higher-res stuff but GPU gets out of memory? What can I do to achieve a result in a longer time? For example, I'm trying roop face swap, but the result is blurry and unnatural. When I try to upscale it the gpu fails

oh, and just for future: do GPTs have answers to such noob questions? Don't wanna bug you every time

upd: ok found through search I should us reactor

[ivory crest]

@paper crag

[acoustic saddle]

trying to use xyz prompt, search and replace for this,
so that it generates an image with (human 1girl brown_hair white_shirt from_front standing looking_at_viewer:1),(anthro crab four_arms crab_arms crab_legs crab_pincers decapod pincers not_furry:0), then (human 1girl brown_hair white_shirt from_front standing looking_at_viewer:0.99),(anthro crab four_arms crab_arms crab_legs crab_pincers decapod pincers not_furry:0.01), and so forth, but it's treating the groups of parenthesis separately and making 200 images instead of 100 images with a single replaced group

tl;dr goal is to have a series of images to make a carcinization gif with, but can't get the search and replace function to replace the whole string.

formatting like this still makes it create 202 images where the human and crab parts of the script are replaced separately instead of together

looking into the "Prompts from file or textbox" function, may just have to concatenate the whole prompt

hm, that works, but it just generates noise with the extremely low prompt weights

[signal sigil]

Hey guys. I was wondering if someone could help me out with this.
I'm running Pixart Sigma in ComfyUi, and The T5 xxl encoder appears to to using 25GB of Vram when in 4Bit quantization?
I have no idea why this is the case. Should it be using that much Vram when quantized?
The T5 loader node says it's using Model 00001 of 00002 also.

[sturdy kestrel]

hey i really want to get into making ai art but really stupid when it comes to computers and im having a really hard time just staring up SD. i tried a bunch of methods of fixing this on reddit but i just kept getting errors.

[acoustic saddle]

using webui, I just hit save and a setting I have not interacted with has changed, "multiple_tqdm", where is this in settings and what does it do?

[ornate elk]

Yea use Reactor instead of Roop.
And like I said you need the tiled vae extension to get less out of memory errors

Also stay with 1.5 models, these are mostly 2gb in size

[ornate elk]

Hey, please follow the install guide from the pinned messages of this channel.
Your error is caused by your Antivirus Software breaking the Python install.
You need to uninstall python, reboot, disable the AV and install it again.

[little bough]

Hi, does anyone know the cause of this in A1111? Errors all over .. I had the issue a while back when I was moving models around between A1111 and Forge but I resolved it iirc by simply moving models back into the correct folder but I can't remember exactly.. I'm not sure what is going on but I gather it might be a known issue with an easy fix, any ideas? On boot up I get the cmd window as can be seen, no major issues other than a1111 and then i get a massive stream of errors and gibberish as seen in the next cmd windows , any ideas? thanks

[vocal burrow]

update your a1111, update your extensions. And if you moved SD, delete venv

[ivory crest]

I don't get charged for these right?

[little bough]

had already updated it but couldnt directly update extensions since as you can see the extensions page was garbled but then i thought maybe the animatediff error and another deforum error might be causing it, removed them and it is working again. cheers

[teal shadow]

Hello. As I have some problems with comfy I’d like to reinstall everything and start “clear”. For that can I just delete the whole SD folder, deinstall python, git and start everything from new? Also would it be possible to keep SD on external hard drive? Thank you

[azure magnet]

Hi, i need an help. Need I a vaeXL for SDXL or no? Can I use even a normal vae?

[ornate elk]

For sdxl you need to use a sdxl vae

Or none. Most models have them integrated

[ornate elk]

You can delete the comfyui folder.
Install python 3.10.11 64bit
No need to uninstall git or python

[azure magnet]

Ok, last question for XL need i a refiner? What is?

[ornate elk]

Nope, it was made for the base sdxl model to enhance quality but its not needed for community made models

[azure magnet]

ok, thank you

[frail umbra]

Where would this comfyui node be looking for ipadapter? Apparently not models/controlnet

[vernal notch]

if i signed up for a paid account with gmail login am i able to tie my discord to it?

or do i have to make a new account

to use the discord bot

[frail umbra]

The answer was models/ipadapter

[rapid moon]

is it possible to get email when we use for example 90% of the credits?

[hot coral]

Hello all!
Is anyone experiencing image blurring degradation with the inpainting API?

I have a SPA for inpainting with a base canvas for the uploaded image and another layer where the user can paint a mask. The base image and the mask are sent in an api request. The return image then replaces the base image so the user can paint a mask again.

The return image is always degraded, I'd say about 10% blur on each iteration. If no one else is seeing this, then it must be something on my end.

Thanks in advance,

[primal glen]

I'm trying to get inpainting working. Can someone tell me what I'm doing wrong? Here's the console log: https://pastebin.com/Yg0bmQah

I

I've tried using different prompts and different models and nothing seems to work

[azure magnet]

Hi, i need an help, what is Clip set layer and how should i set?

And other questions on comfyui

[viral knot]

it depends on the models, most models use default clip of 1. if a model is better with 2 or higher, usually it will be mentioned on the model page.
in comfy, you can connect it right after the load checkpoint node, by connecting the clip part to the clip set last layer node and then from the clip
last layer node, you do the usual connections to the prompts, etc

and in comfy, it's negative numbers, but it's basically the same thing

-1 is 1, -2 is 2, etc

[azure magnet]

I was standing an workflow, and i find something that i don't understand. How should i use Vae? I don't understand Vae decoder (tiled), Vae encoder (tiled), Vae decoder

[viral knot]

tiled just means it will do it tile wise instead of one shot, so by doing tiles, it saves memory, depends on your GPU
i mean you can try both and see what works

[azure magnet]

I have 12GB Can i avoid to use the tiled version?

[viral knot]

yes

but then again it also depends on the dimensions in question, if you are dealing with huge dimensions, you might need tile, it really all depends on the workflow

[azure magnet]

Ok understand, last question, how to use Vae Encoder and Decoder?

[viral knot]

well encoder is usually for image to image stuff and decoder is basically always needed to decode from latent space to normal (image) space that you actually see

if you are using comfy, i highly suggest you check some of the basic examples

https://comfyanonymous.github.io/ComfyUI_examples/

you can learn a lot from this alone

[azure magnet]

So basically, after a KSampler, should i use an decoder, and after an upscaler an encoder?

[viral knot]

like i just said, check the examples i linked, it will answer all your questions i think 🙂

[azure magnet]

Ok i'll read it. Thanks for your help.

[viral knot]

np

[ornate elk]

Hello please ask in #🗣｜artisan-support-feedback

[ornate elk]

Hello please ask in #🗣｜artisan-support-feedback

But I guess you don't need to make a new account

[formal trout]

got this when trying to run SUPIR example workflow

I have to put the image size there, I don't understand

I have to calculate the image size?

[ivory crest]

Wth?!

[formal trout]

I'm trying other SUPIR workflow, having trouble with this node

I changed the method from "false" to "keep proportions" and it started

got this error then

[fathom spruce]

is the "Upscale Latent" node in comfyui the same as using HiresFix in auto1111?

[storm hearth]

In the A1111 webui, users can make styles here, can I find these user-made styles in my webui directory or where are these saved?

[ornate elk]

In the stable-diffusion-webui folder.
Its called styles.csv

[storm hearth]

Found it, thank you!

[balmy shore]

is there any special info or anything regarding these samplers

i havent tested extensively but they seem to work just like the regular ones but im wondering if theres somethign im missing

[vapid sluice]

Hey, am I doing something wrong with clip skip? I am using a Python script not the web ui

output = pipe(
    prompt=prompt,
    num_inference_steps=25,
    guidance_scale=7.5,
    scheduler=scheduler,
    seed=1,
    clip_skip=2
)

I tried both 2 and -2 but get blobs both times...

[dapper briar]

Hello guys, can you help me please, why is my renders are looking like this?

[onyx solar]

You're using Pony Diffusion, which is an SDXL model, but you're trying to render at 512x512, which is only 1/4 megapixel. You need to be at 1 megapixel (1024x1024 or similar).

[dapper briar]

it's the same, what else I could do wrong?

[onyx solar]

Where to begin...

1. You need to read up on how to prompt for Pony XL because you're missing tags that should exist.
2. You should also not use highres fix/upscaling until you have a working set of parameters.
3. Don't use refiner.
4. I can't see what clip skip setting you're using, but you should read into that and ensure it's set appropriately.

There's more, but that's a place to start.

[dapper briar]

clip skip I set on 2

[vapid sluice]

Some more details: I can correctly use clip skip with the base SDXL model as follows:

import torch
from diffusers import StableDiffusionXLPipeline

pipe = StableDiffusionXLPipeline.from_single_file("./sd_xl_base_1.0.safetensors",torch_dtype=torch.float16)
pipe.to("cuda")
pipe(prompt="cat", num_inference_steps=25, clip_skip=2).images[0].save("cat_sdxl.png")

That produces correct image of cats. However when I swap base model with Pony diffusion, the result is blob.

pipe = StableDiffusionXLPipeline.from_single_file("./ponyDiffusionV6XL_v6StartWithThisOne.safetensors",torch_dtype=torch.float16)
pipe.to("cuda")
pipe(prompt="cat", num_inference_steps=25, clip_skip=2).images[0].save("cat_ponydiffusionxl.png")

Note both models work fine when clip_skip=0. I also used the recommended VAE with pony diffusion but no luck,

[proper pebble]

anyone know how this was made?? thankss in advance
https://www.instagram.com/p/C781eUDoJ2h/

visualsk2

[Vision IV/Part.6] Thanks so much for 170.000 Followers ✨🙏🏻
Only a few days left until the tutorial is released.

#grainisgood #idea #reels #framebyframe #photography #blurry #explorepage

Likes

13873

Instagram

[ornate elk]

Your trying to use SDXL stuff with Automatic1111 version 1.5.1
But you need 1.7.0 or newer (latest version is 1.9.4

[onyx solar]

Good catch. I didn't even look at the versioning.

Not that the other stuff was not also helpful, but versions do matter.

[dapper briar]

Oh, I see, thanks, I'll try to update then

[ornate elk]

Np, if the update don't work, look in the pinned messages of this channel for the latest install guides.

[dapper briar]

Seems like it's work now, thank you guys!

[teal adder]

so I trained this model, supposedly.

Where do I place the output to use it in the sdwebui?

[teal adder]

it is not appearing in the sdwebui when I place it under thelora tab

I did i mess up something?

This lora took a long time to train lol

the Kohya lora verifier fails

[azure magnet]

Hi, where should i put an upscaler?

[unkempt junco]

Hi All, what is a better stable diffusion UI than Fooocus?

[subtle timber]

What the heck is thisss 😢 I've restarted and reloaded and updated everything and it doesnt go away. It doesnt give me any log files or anything 😭

TypeError: this is undefined

Solved by reinstalling sigh :/

[obtuse crag]

HELL NO

zzzzzzzzzzzzzzzz

make people pay for this.. It's Frodulus

[alpine pawn]

Hey there, I made an model in fooocus and now I want to generate different poses and different backgrounds forthat model, but whenever I am clicking face swap option and writing a prompt it is generating a differnt model, can anyone please help :)

[glacial citrus]

I cant figure out why no matter what model I use, a CLIP skip of 2 or more is always a mostly solid color, with little shade

[vocal burrow]

screenshot your whole webpage

[alpine pawn]

any solution?

[ornate elk]

Stable-diffusion-webui is more advanced and works better for every GPU

[drifting pendant]

Hello there, does anyone tried to run two tesla P100 or two tesla P40 ?
If yes, what are the motherboard / CPU / RAM, minimal requirements ?
I've read somewhere a good rule of thumb is to run double RAM, compared to the VRAM, it is diffucult to run both of this card together in a Linux environnement ?

[neat prairie]

Guys, who can tell me how to teach ai to draw a certain anime character?

[azure magnet]

Hi, where should i put an upscaler?

[solemn moon]

Train your own Lora, never done it but that’s what you’re after

[ornate elk]

Hey, in the models/ESRGAN folder

[ornate elk]

Download a lora of that character from Civitai

Or if its a not well known character, you can train your own lora on it

[azure magnet]

Another questions, how work the extension on comfyui? In automatic1111 in the folder extension i have the extension. Do they works even with comfyui?

[ornate elk]

No, for comfyui you need the comfyui manager. With that you can install and manage the extensions

Auto1111 extensions won't work in Comfyui

[azure magnet]

ok thanks

[unkempt junco]

any thoughts on this build to run stable diffusion: https://www.bestbuy.com/site/ibuypower-y40-gaming-desktop-intel-core-i7-13700f-32gb-ddr5-ram-nvidia-geforce-rtx-4060ti-16gb-2tb-nvme-ssd-black/6557595.p?skuId=6557595

[ornate elk]

Hey, the GPU in this build has only 8gb Vram.
Thats something you shouldn't go for in 2024.
12gb vram should be your minimum

Also the CPU in this build is to pricey and not needed for Stable diffusion or that gpu

[unkempt junco]

its says 16gb gpu mem: Q: QuestionIs the GPU 16GB or 8GB? The listing says the GPU is GTX 4060Ti 16GB but under the "features" tab it says its an 8GB card. Which is it?(1 answer)

A: AnswerIt is the 16GB card.

[ornate elk]

Okay, so 16gb would be good

But make sure they don't trick you there xD

The only thing that matters for Stable-diffusion and other Ai stuff is the GPU Vram.
Second is the GPU performance

32gb System RAM is perfect

[rancid nymph]

i keep getting this error even after adding the variables to the environment variables

[hallow saddle]

How do I get access to the Dreamer Community "art-movements"?

[rancid nymph]

when i try to go to %HIP_PATH%bin in run this happens

this is stupid why cant you just make it point directly at zluda and hip

[south vigil]

hey guys so I somehow cant come to a solution of prompting specific objects. In my case right now i would love to do pictures of a nintendo switch, since there are no loras or checkpoints i can find on civitai, i somehow hit a wall and dont even know if its possible to somehow get to prompt specific things, can you maybe help me with telling me what i need to learn for doing such tasks?

[azure magnet]

Sorry to ask again, but i need an help. Is it possible to install adetailer on comfyui?

[ornate elk]

Then you can also add the direct path to the rocm bin folder instead of %hip path

[ornate elk]

Adetailer isnt available for comfyui, but there are other extensions that can work. Idk which ones

[rancid nymph]

like this in path right?

[azure magnet]

Is controlnet available?

[ornate elk]

Yep

[ornate elk]

Yes

And reactor too

If you have the comfyui manager installed you can search in it

[azure magnet]

Last question there is some change that i could do to improve output quality, or it all just set?

[rancid nymph]

i restarted after adding that and im still getting the same error

[ornate elk]

It depends on the model and prompt. Resolution etc. So all stuff you have to set yourself

[ornate elk]

Can you show the full cmd log?

[ornate elk]

Very strange that ; thing in the path

[rancid nymph]

i noticed that earlier

[ornate elk]

Open up a cmd as admin and run
sfc /scannow
Then uninstall hip SDK. Reboot the PC.
Then install hip SDK again.
Then reboot the PC

[rancid nymph]

ok wait i found out where it is

[ornate elk]

Ahh remove it there

The ;

[rancid nymph]

weird

[viral knot]

but how was that even added there with the ; to begin with lol

[rancid nymph]

i never edited that

i saw it earlier and tried removing it and it inserted itself back

so i thought it was supposed to be there

[ornate elk]

Very wierd

But now its gone?

Recheck after saving

[rancid nymph]

yeah its gone

sd is starting now

this is a little concerning but i dont think it should affect anything since ive seen this a few times

[azure magnet]

Sorry another question, adetailer and controlnet work even on SDXL or have their own version on purpose?

[ornate elk]

Thats normal as xformers is for nvidia only

[ornate elk]

Controlnet needs the sdxl Controlnet models to work

Adetailer works on every model

[rancid nymph]

only took 5 hours in total to get it to work

[viral knot]

could be worse 🙂

[ornate elk]

Ez ^^ but you can add +15-40mins to it. (For the first image generation)

[rancid nymph]

i dont think its over

it wont proceed from this 😔

either that or its taking an extra long amount of time to load

[ornate elk]

Just let it load

Looks good

[rancid nymph]

i wonder if that semicolon couldve been caused by a memory error in some way

bc i had to put in my 32gb of ram thats going bad to be able to run this

i did memtest86 on it and one bit got 250 errors but im too lazy to figure out which stick its on

[ornate elk]

xD

Then you should do an SFC /scannow once in a while to fix system errors

[rancid nymph]

probably

considering ive never done that

my ram is 10+ years old

im working with hand me down stuff that i got for free

except for my gpu

[ornate elk]

What's your GPU?

[rancid nymph]

its a mircale stable diffusion generated using my cpu last night

[rancid nymph]

Radeon 6600 XT OC Edition

[ornate elk]

Ahh nice that will work with zluda

[rancid nymph]

thats why i was trying so hard to get it to work with my gpu

[ornate elk]

5hours is still good.
We had people here finding my guide after trying for days to get it work xD

[rancid nymph]

technically i tried a few months ago and gave up after 30 minutes

also it takes a VERY long time to load models and checkpoints

[ornate elk]

So the webui is launched now?

[rancid nymph]

yeah

[ornate elk]

Then do the first image generation.
It can take 15-40 minutes.

[rancid nymph]

i cant bc the models wont load

[ornate elk]

What does the cmd shows?

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker

Also show the txt2img page

[rancid nymph]

its weird because the checkpoints load but not the models

[ornate elk]

Checkpoint = models

Loras are not models

[rancid nymph]

oh

loras arent loading then

[ornate elk]

We can fix that later

Better generate the first image

Loras only show up when a compatible model is selected

But the first image shouldn't be modified

[ornate elk]

Oh you loaded an sdxl model

[rancid nymph]

i just realized the main model i downloaded is an nsfw model oops

now its doing what it was doing last night

[rancid nymph]

to be clear it didnt make it obvious on the download page there was no nsfw in the images

[ornate latch]

Hello! Is this the place I would post to get help? I've got an issue with SDXL freezing up while upscaling images. I'm fine waiting if you guys are in the middle of helping someone else

[ornate elk]

Okay perfect. If the cmd doesn't show any error its working

[rancid nymph]

oops

[ornate elk]

Generate button should be grayed out

[rancid nymph]

last night this happened and it sat on this for about 30 minutes and did nothing

[ornate elk]

Hey, yea thats the right place.
What's your Webui, GPU and RAM?

[rancid nymph]

there was no cpu activity at all

[ornate elk]

What does the cmd shows right now?

[rancid nymph]

still the same thing

[delicate citrus]

I'm thinking in rent a cloud pc, is this good configurations for sd?
@ivory crest

[ornate elk]

Okay good then yea let it load.
It can take 40 minutes because you used an sdxl model.

[rancid nymph]

oh

huh better find something to do then ig lol

[ornate elk]

Yea, watch an anime haha.
After that long first image generation it will be really fast
And even faster on 1.5 models

[rancid nymph]

last night itd show me the progress of the generation on the progress bar

[ornate elk]

Yea it will do that when it starts generating. Right now its doing zluda cuda stuff in the background

[rancid nymph]

ig that would explain why its using 8gb of ram in the background

[ornate elk]

Thats because you loaded an sdxl model in. These are heavier than 1.5 models

They stay in Ram and then get loaded into the GPU VRAM

[rancid nymph]

praying my ram doesnt suddenly crash my pc bc one of those sticks are hanging on for its life 🙏

[ornate elk]

Yea I would recommend then to just let your PC running and doing something else xD

[ornate latch]

Thank you! I'm using A1111, I have a NVIDIA GeForce RTX 3060 with 16gb installed ram. Basically whenever I use Highres fix it has a chance to freeze my pc at 70% rendered. After 1-3 minutes I get control back slowly and the image starts rendering again, but usually much slower. Sometimes this happens in image2image as well but not I don't use it as much. I've tried reinstalling A1111 and using different XL checkpoints but it didn't change

[ornate elk]

Okay, a 3060 with 12gb vram? Or a laptop version with 6gb

And What's in your webui-user.bat?

[ornate latch]

Right now it's just the default from download since I reinstalled yesterday
@echo off

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

call webui.bat

[rancid nymph]

the ram did the thing

and sliced my fingers

[ornate elk]

Crashed?

[rancid nymph]

yeah and now my pc won't post past the ram check

wait nvm

I reseated it

I think there's something wrong with the contacts

it seems to be very sensitive to any type of movement

[ornate elk]

At the line commandline_args=
You have to add: --xformers --medvram-sdxl --no-half-vae
Then save and try again

And always launch the webui-user.bat and not the webui.bat

[ornate latch]

Alright, my old .bat had the first and last line not the middle, I'll run with it and check back after I make some images

[ornate elk]

Sure if that won't help I have two other improvements you can try 🙂

[ornate latch]

set COMMANDLINE_ARGS= --xformers --medvram-sdxl --no-half-vae like this?

[ornate elk]

Okay so you rerun the webui-user.bat and try you first image generation again?

You also don't need 32gb RAM.
16gb will work too

[rancid nymph]

i cant switch sd checkpoints now

im trying to go to the default one just to test it out and it wont let me switch

[ornate elk]

Try an other 1.5 model

The base one is bugged

[rancid nymph]

could you suggest one

[ornate elk]

Dreamshaper v8

Its 2gb

[rancid nymph]

so happy civitiai has fast download speeds

otherwise this would be even more miserable than it is now

how long might this take

[ornate elk]

Imagine if they would add a speed limit like Nexusmods xD

[rancid nymph]

ugh god

i hate nexusmods

[ornate elk]

Its fine, but I don't like how the kicked out the NexusModManager in favor of Vortex.
And now they will replace vortex too

[ivory crest]

I don't see Nvidia gpu mentioned?

[rancid nymph]

love how a chromium based browser is still using more ram than sd

i think its generating now but its not using my gpu

well somethings happening

[ornate elk]

Perfect

Let it load

Its on the road to finish

[rancid nymph]

i wish itd give you another command prompt window with info on what its doing

bc all i have to know that it hasnt halted is task manager

[ornate elk]

Yea would be nice

[rancid nymph]

imma go do something that takes time and hope that my pc doesnt die while im gone

[ornate elk]

Good luck ^^ let me know when it worked

[delicate citrus]

not in this service, any other to recomend me?

[balmy shore]

i downloaded more ram but its not working

but foreal, when using __wildcards__ , is there a way to make the output of the wildcard get sent to the img2img instead of the wildcard in the original prompt when sending an image from txt2img to img2img?

[ivory crest]

I don't know how old this article is though: https://gist.github.com/devinschumacher/87dd5b87234f2d0e5dba56503bfba533

Have you tried google collab btw? It's free with gspace account is enough to run SD (also SD3)

[rancid nymph]

it worked

[ornate elk]

Perfect!

[rancid nymph]

it runs 900x faster using my gpu too

i gets results in seconds

[ornate elk]

Nice 🙂

Also the vram consumption should be less

[rancid nymph]

and the loras work now too

[ornate elk]

Yep that's what I thought

You have to check if a lora is made for 1.5 or sdxl

They only show up with the compatible model selected

You also can make sub folders in the lora folder

To manage them

[rancid nymph]

im gonna try and get that other model to work now

it works but i dont think its supposed to use img2img lmao

[ornate elk]

xD

[rancid nymph]

i made a mistake and now sd is using 11gb of ram

[viral knot]

im sure it was just a happy little accident

[delicate citrus]

yes I use colab pro, the free there's a long queue and it often crashes, but each time the colab is one way, sometimes it works, sometimes it doesn't. But I chose to pay the pro

@ivory crest thanks for the article I you read it

[ornate elk]

What was the mistake?

[rancid nymph]

i set the resize image to x2 on accident

[ornate elk]

For upscaling first generate in txt2img with hires fix. If you then want it larger use the SD upscale script in img2img

[hallow gust]

anyone have any idea why whenever i select a vae it just goes back to none?

[ornate elk]

Check the cmd for any errors

[hallow gust]

and then after that its just a giant wall of text

[ornate elk]

Thats not a vae

Chilloutmix is a model

Check its file size

VAE files are smaller than 1.98gb

Everything larger than 1.98gb are models

[hallow gust]

got it thanks

probably copied the wrong file over then never checked

[ancient plover]

Hi, generating images with Zluda webui is suddenly slowing down my whole pc, especially as it nears completion... I just updated it to version: "v1.9.3-amd-23-gd23a1724 •  python: 3.10.11  •  torch: 2.3.0+cu118" after not touching it in the last month or so. I never had this issue previously with 1.8, 1.9.0, or before, but now, my whole pc nearly crashes as I get close to finishing generation. When it happens my monitors sometimes freeze, my mouse might freeze or become unusably slow and laggy, etc., and Im basically forced to just stop everything and wat until it passes. My PC is fairly beefy: 32gb RAM, decent cpu, 6900XT GPU, and this is happening for images that i previously wouldn't have any issue with. I tried watching Task manager live and found that my cpu, gpu, and ram utilization all briefly reached 100% during the lag, so I dont know what exactly the bottleneck is. Does anyone have a fix for this?

[frosty agate]

Whatg does this mean

[toxic moth]

are these the same:

[solemn moon]

yess

[delicate urchin]

https://github.com/lks-ai/ComfyUI-StableAudioSampler

I donwloaded this custom addon with the manager and now ComfyUI wont load anymore. The cmd window shows for a second then dissapears. I have tried to delete the custom node folder and run comfy but same issue.
What can I do to fix this and run comfy again?

[viral knot]

i heard that comfy is actually working on native implementation of stable audio

[ornate elk]

Hey, thats mostly caused when using sdxl with hires fix.
Make sure you have set hires steps to 10.
Then make sure you have the tiled vae extension installed and tiled vae enabled.
Then you also can download the sdxl vae fp16 Version. Then it shouldn't lag or freeze.

[wind helm]

Hi! Not sure if this is the right place to ask for advice - please redirect me, if not! – I want to apply a new mono-colored background to a portrait foto using "inpaint anything" in automatic1111. 2 things bother me, though:

1. I have no clue how to apply simply one color (or maybe a gradient) to the background. What happens is always "full of fantasy".
2. The general colors of the original image are always "respected". See examples: there is a dark and a bright area in the image. And even though everything is masked, the bright area is always respected
   I've used various sampling methods (Euler, Euler a, DPM++ 2M Karras, ...) and various models (all base SD1.5). Any idea what I am doing wrong?

[ornate elk]

You can try run in the install bat again

[delicate citrus]

anyone knows how I turn on the T4 on Kaggle?

[junior eagle]

Can I make pictures right here on Discord? Can someone help.

[ivory bramble]

No, but you can look into this something like this; https://github.com/lllyasviel/Fooocus

Or try https://civitai.com/

You can look into something like Midjourney, but that's paid and certainly doesn't have the features and tweaking power that Stable Diffusion does.

[fair oxide]

I get this when I run web ui.bat

I have a radeon 6650xt

[grim scaffold]

Having the SD3 Medium models been released officially, when will it be made available in the DreamStudio online application? Currently, there are only two models available: SD1.6 and SDXL. Thank you.

[floral flame]

hey I can't figure out how to start stable audio open again

I opened it automatically when downloading it and the interface popped up

but then I closed it and idk how to open the interface and application again

Also how does one allocate more vram to the program so it doesn't crash?

[summer owl]

Anyone know where i get these clip models for the basic SD3 comfy workflow?

[fair oxide]

I am getting this error trying to clone the hugging face repo for SD3 git remote set-url origin https://USERNAME:[e-mail]/stabilityai/stable-diffusion-3-medium
fatal: not a git repository (or any of the parent directories): .git

[vocal burrow]

same place as model https://huggingface.co/stabilityai/stable-diffusion-3-medium/tree/main/text_encoders

(haven't tried yet, I'm kinda at work :p)

[vocal burrow]

not sure why you're trying to git clone the model repository but it's clearly stated in the url that you have to put your USERNAME and TOKEN in there

[fair oxide]

i removed it to obsfucate

[vocal burrow]

yup they work (from my 2070 Super; 8gb vram; 1.3it/s using the basic workflow)

[ornate elk]

Only if you subscriped at
https://stability.ai/stable-artisan

[ornate elk]

Then you need to follow my AMD Zluda guide from the pinned messages of this channel

[digital radish]

OMG Help!

just updated to latest comfy too

[fair oxide]

update comfy ui

[vocal burrow]

update your comfyui

[digital radish]

omg nvm

[vocal burrow]

did you restart then ?

[digital radish]

forgot to refresh 😐

[fair oxide]

😅

[digital radish]

🤦‍♂️

[digital radish]

thanks

I had a moment of panic, lol

[waxen badger]

I'm getting error loading sd3medium safetensors into A1111. Anyone else?

[digital radish]

Okay, now on this one... where to find?

[digital radish]

https://huggingface.co/stabilityai/stable-diffusion-3-medium/tree/main/text_encoders ... found it! Thanks reddit!

[ornate elk]

Auto1111 needs updates, it can take a few days until it works there

[timber turtle]

where can i download these files?

├── LICENSE
├── sd3_medium.safetensors
├── sd3_medium_incl_clips.safetensors
├── sd3_medium_incl_clips_t5xxlfp8.safetensors
└── ...

[vocal burrow]

I've sent you the link already https://huggingface.co/stabilityai/stable-diffusion-3-medium/tree/main/text_encoders

filenames are not exactly the same

[floral flame]

hey I can't figure out how to start stable audio open again
I opened it automatically when downloading it and the interface popped up
but then I closed it and idk how to open the interface and application again
Also how does one allocate more vram to the program so it doesn't crash?

pllzz I need help

[timber turtle]

these are other files, aren't they?

[vocal burrow]

the models ? https://huggingface.co/stabilityai/stable-diffusion-3-medium/tree/main

[timber turtle]

thank you

[digital radish]

https://huggingface.co/stabilityai/stable-diffusion-3-medium/tree/main/text_encoders you also need these.

not those exact ones from the workflow, but the corresponding ones on the link to download

lol, got like 20GB of stuff...

can I even run sd3 on a 6GB?

[vocal burrow]

won't know for sure until you try it :p

[old ether]

Can anyone tell me why my comfyui nodes are red for SD3? I've updated it.

dont worry apparently clicking refresh worked

[digital radish]

deleting shit from my drive, lol

[junior eagle]

Once I’ve subscribed (I’m on a 3 day free trial now), how do I go about making pictures on Discord? Do I go in the SD3 room and type a prompt? Or another room?

[vocal burrow]

cf #artisan-faq

[summer turtle]

I'm getting an error using the fp8 version of t5

imma try with fp16 version once I finish downloading

[fair oxide]

Its stuck on this now

Nvm

As soon as I say somthing

[dry ivy]

I get the same issue, fp8 on clip and in the model itself

Error occurred when executing CheckpointLoaderSimple:

module 'torch' has no attribute 'float8_e4m3fn'

File "E:\ComfyUI\ComfyUI\execution.py", line 151, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
File "E:\ComfyUI\ComfyUI\execution.py", line 81, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
File "E:\ComfyUI\ComfyUI\execution.py", line 74, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
File "E:\ComfyUI\ComfyUI\nodes.py", line 516, in load_checkpoint
out = comfy.sd.load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, embedding_directory=folder_paths.get_folder_paths("embeddings"))
File "E:\ComfyUI\ComfyUI\comfy\sd.py", line 464, in load_checkpoint_guess_config
sd = comfy.utils.load_torch_file(ckpt_path)
File "E:\ComfyUI\ComfyUI\comfy\utils.py", line 14, in load_torch_file
sd = safetensors.torch.load_file(ckpt, device=device.type)
File "C:\Users\Divirel\miniconda3\lib\site-packages\safetensors\torch.py", line 313, in load_file
result[k] = f.get_tensor(k)

[timber turtle]

which samplers work well?

[dusky nebula]

Hey, how can I activate my account, (vinculate it with my subscription)

[summer turtle]

fp16 fixed it for me

[dry ivy]

oh nice

[summer turtle]

who said sd3 will have less requirements than sdxl lol

everything is being loaded in lowvram mode

[dry ivy]

In medium or medium incl clips?

[rotund musk]

am i the only one getting 'OSError: stabilityai/stable-diffusion-3-medium does not appear to have a file named model_index.json.'

[waxen badger]

I keep getting this error when trying to do pip install -r requirements.txt

[summer turtle]

https://huggingface.co/stabilityai/stable-diffusion-3-medium/tree/main/text_encoders the t5xxl_fp16.safetensors

[dry ivy]

Isn't that for the tripleclip?

I mean the model

[merry hearth]

Probably the wrong place to ask this, but I've just installed the windows self-contained ComfyUI which apparently supports SD3 today. It didn't come with any checkpoints, can I use the same checkpoints I had on StableDiffusion or do I need specific ones that are developed for sd3?

[summer turtle]

oh sorry, its medium

[dry ivy]

Nice, works for me

Now to figure out which samplers work or not..

[ornate elk]

Hey, you can use the same as with the other Webui

[timber turtle]

90% of the times he writes the text wrong, is it my problem or does it happen to everybody?

[dry ivy]

Likely everyone

[fair oxide]

I ran the instrutions for the amd installation

Why do I need a nivida driver?

[astral turret]

so i saw someone make a nice looking visual novel with anime style sprites they made in ai and i set up stable diffusion on my pc and thought id give it a try, but i just cant get it to work. i've tried so many different methods to try and generate poses and character turning and character sheets and it seems to just ignore most of the text i put and there is no style consistancey it just gives me really shitty looking images, and i just dont know why its not working for me. ive tried a few loras and things but it just hasnt seemed to help.

i dont get how its possible to make character sprites when it seems to impossible to even get a single pose of any character to generate.

[glad fiber]

which one should i download?

[rotund musk]

did they forgot to upload the model_index.json?

[dry ivy]

First or second

fp8 likely doesn't work for you

[sudden helm]

I figured first or third (first if you bring your own text models?)

[dry ivy]

I had to use Clip g, clip l and the t5xxfp16, to make the incl clips model work

[sudden helm]

but I am not having luck with either, so what do I know 😄

[solemn moon]

excuse my lack of understanding.. is 3 avail for Auto?

[vocal burrow]

A111 is not compatible with sd3 yet

[solemn moon]

thanks for saving me space by not dwnlding all the models and trying them 🙏

damn it looks good though o0

[dry ivy]

Well.. depends :P
Negative prompt is a hit and miss, then add the fact that commonly used samplers and schedulers, can corrupt the image

[glad fiber]

is there a workflow to use sd3 on comfyui yet?

[solemn moon]

omg, so a new model to learn then..

[glad fiber]

trying to use it directly gives this

[formal trout]

hi, how can I edit the extra models path so I can put all I need for SD3 in another directory in another drive? like the clip

[solemn moon]

pls pls pls can the code bro's get Control Nets done, lineart to be specific 😅

[sudden helm]

so did you guys download the text encoders for clip_g and clip_l and the t5 that are referenced in the basic workflow and then use the sd3_medium (which doesn't include text encoders) to get it to work?

[onyx solar]

Just create subdirectories in your models folders

[dry ivy]

yes

[formal trout]

I tried but got errors

[formal trout]

for examaple I got

` other_ui:
base_path: /media/mario/Ultra-Zen/SD3/
checkpoints: models/checkpoints

gligen: models/gligen

custom_nodes: path/custom_nodes`

it loads fine, the checkpoint

[onyx solar]

No...don't change the configuration file from what you had originally at all.

Just make subfolders within the actual checkpoints folder for SDXL, SD3, etc.

Same with the other folders inside models.

[formal trout]

oh nevermind, I don't what ait was but I got ti to work

other_ui: base_path: /media/mario/Ultra-Zen/SD3/ checkpoints: models/checkpoints clip: models/clip/

I cannot do subdirectories, I neeed it in another drive

[sudden helm]

what directory is it searching these models for? Same directory as the Checkpoint ?

[fierce hollow]

Guys, I'm not really fumbling for SD, if I install a new version, SD 3, will the models from my previous version support the new version?

[formal trout]

no, it is clip: models/clip/

[merry hearth]

Nice, thank you!

[balmy frost]

Anyone get past the "Error while deserialiing header : InvaildHeaderDeserialization" error

[analog kernel]

where does one get these clips?

[sudden helm]

https://huggingface.co/stabilityai/stable-diffusion-3-medium/tree/main/text_encoders

[analog kernel]

oh thanks, didnt expect it to be labled text encoder

[wet magnet]

do I just download sd3_medium_incl_clips_t5xxlfp8.safetensors and place it in my comfyui models folder?

right now I'm trying
from diffusers import StableDiffusionPipeline

model_id = "stabilityai/stable-diffusion-3-medium"

pipeline = StableDiffusionPipeline.from_pretrained(model_id)

but I'm getting syntax errors (running from python command shell)

[analog kernel]

place it in your clips folder in the same directory your models folder is

[kind sphinx]

j'ai installé Comfuy.
I have install Comfuy.
From : https://github.com/comfyanonymous/ComfyUI?tab=readme-ov-file#installing
There is a portable standalone build for Windows that should work for running on Nvidia GPUs or for running on your CPU only on the releases page.

J ai mis sd 3 medium checking dans le bon répertoire mais j ai cette erreur :
I installed Comfuy. I put sd 3 medium checking in the right directory but I got this error:

Error occurred when executing CLIPTextEncode:

'NoneType' object has no attribute 'tokenize'

File "C:\SD3\ComfyUI_windows_portable\ComfyUI\execution.py", line 151, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\SD3\ComfyUI_windows_portable\ComfyUI\execution.py", line 81, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\SD3\ComfyUI_windows_portable\ComfyUI\execution.py", line 74, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\SD3\ComfyUI_windows_portable\ComfyUI\nodes.py", line 57, in encode
tokens = clip.tokenize(text)
^^^^^^^^^^^^^

[wet magnet]

thanks

[snow void]

Hey! Attempting to run SD3 in Comfy and continue to get this error each attempt at queuing a prompt, anyone able to help me out?:<

[wet magnet]

prompt?

[sudden helm]

"A cute hamster professor, holding a sign that says "it works!", glasses, tuxedo" I think

[sudden helm]

I think the sd3_medium model doesn't include the text encoders it needs to handle the prompt. You need to download the three encoders as shown in the basic workflow from hugging face and place them in your clip folder.

[honest tendon]

where do i put this file in?

is it literally on the control net here?

do i literally just copy paste the pip command in my terminal?

@ornate elk

[ornate elk]

Nope

Just download the .safetensor file from here and put it into the models/Controlnet folder

https://huggingface.co/lllyasviel/sd_control_collection/tree/main

[fierce hollow]

The sd3_medium_incl_clips_t5 california model is not loading. What could be the problem? It just throws it over to the previous model
Here is the terminal that shows

[snow void]

Will try this, thank you!!

[ornate elk]

Auto1111 needs an update before it works

[fierce hollow]

How do I upgrade?)

[ornate elk]

With git pull

[glossy lynx]

is there an update for forge also?

[ornate elk]

But Auto1111 doesn't released the update. So we have to wait maybe a few days

[fierce hollow]

Can I go into more detail? I'm stupid;(

[ornate elk]

Forge is deprecated

[glossy lynx]

why? its better?

[ornate elk]

Because the Dev of Forge decided to use it as experimental project for the future.
He recommends to switch back to Auto1111 or Comfyui

[glossy lynx]

have any of the improvments that forge has been merged into A1111?

[ornate elk]

A lot of them, yes. They are in the dev build and in the next release of auto

[glossy lynx]

can i pull the dev build?

[ornate elk]

Sure

[glossy lynx]

ok cool thanks

[ornate elk]

Here is the full information about Forge:
https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/801

[tame yoke]

how do i make it show me whats its rendering like every 10%?
for now it only shows me after its past 50% and when its finished

[ornate elk]

In the settings you can change the preview steps slider

[wet magnet]

Error occurred when executing CheckpointLoaderSimple:

module 'torch' has no attribute 'float8_e4m3fn'

File "c:\ComfyUI\execution.py", line 151, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
File "c:\ComfyUI\execution.py", line 81, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
File "c:\ComfyUI\execution.py", line 74, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
File "c:\ComfyUI\nodes.py", line 516, in load_checkpoint
out = comfy.sd.load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, embedding_directory=folder_paths.get_folder_paths("embeddings"))
File "c:\ComfyUI\comfy\sd.py", line 464, in load_checkpoint_guess_config
sd = comfy.utils.load_torch_file(ckpt_path)
File "c:\ComfyUI\comfy\utils.py", line 14, in load_torch_file
sd = safetensors.torch.load_file(ckpt, device=device.type)
File "c:\comfyui-env\lib\site-packages\safetensors\torch.py", line 313, in load_file
result[k] = f.get_tensor(k)

I just uninstalled/reinstalled pytorch

[errant wren]

may i ask what are the maximum value of these params in sd3?
height
width
inference steps

[tame yoke]

what do i set to what?

[ornate elk]

It depends on the steps you use.
Right now its set to 5 so if you use 30 steps you get 6 previews.
One at every 5th step

[tame yoke]

well that doesnt sound right

[ornate elk]

Lower it so you get more previews. But that also makes the image generation longer

[tame yoke]

big difference?

[ornate elk]

Depends. If you set it to show a preview every 2-3 steps it will really take longer

But you have to try, idk what your GPU is

[glossy lynx]

is it possible to put progress bar on the comfy-backend step? been minutes and thought it died

but as i was typing this it progressed further

[dry ivy]

Use FP16 T5

[wet magnet]

where do I insert this? I can screenshare if you have a moment.

[dry ivy]

Swap the t5xxl_fp8_e4m3fn with https://huggingface.co/stabilityai/stable-diffusion-3-medium/blob/main/text_encoders/t5xxl_fp16.safetensors

[wet magnet]

okay thanks

[brittle lantern]

so if i was gonna use sd3on my m2 pro macbook ill have to wait or is there anyway else

[wet magnet]

I place this in the checkpoints or clips folder?

[dry ivy]

These go in the ComfyUI\models\clip

Models -> ComfyUI\models\checkpoints

[brittle lantern]

so as of now comfy ui is only supported with sd3? cuz my auto1111 doesnt work

[solemn moon]

@brittle lantern

[wet magnet]

I didn't expect to dive into comfyui today either.... no time like the present

[brittle lantern]

ik hence why im asking what its comp with

but idk how to completely optimise the download of comfy ui on mac

[wet magnet]

I don't think it is

compatible with A111 yet

[solemn moon]

the readme listed 4, cant remember which

[wet magnet]

just use chatgpt4o and ask it to walk you through installing comfyui on mac

I'm on PC otherwise I'd be more help

[brittle lantern]

is 4o like £18 a month correct?

[wet magnet]

probably

but the free version is just as good

[dry ivy]

A111 might get support in like a month

[brittle lantern]

when was 4o free?

[wet magnet]

you just can't do multi-modal things like upload images or documents to it

[brittle lantern]

[wet magnet]

its free to chat with it

some features are blocked but the LLM itself is free

probably slower inference time, but the membership is worth it IMHO

I can already tell though comfyui is what you want to learn

[spare spruce]

Hello everyone I need help, I don't have this pencil in my SD, how can I add this?

[brittle lantern]

bro already read my aura but yh i would like to learn its just so complicated my autism gets triggerd seeing so many stuff

[analog kernel]

is there a way to train sd3 with any software like kohya or do we gotta wait?

[wet magnet]

its just different.... I don't view it as being any more complicated

[fierce hollow]

I decided to try SD 3, what's the problem? Why is there such a cartoon format on the promt about a realistic girl?

[wet magnet]

put cartoon in your negative prompt

[brittle lantern]

what sampler do i use for sd3?

[brittle lantern]

true guess time to full learn it also @wet magnet what cool stuff do i get with 4o from the free veriant

[fierce hollow]

And why can there be poor image quality? Settings, promts are all the same new Automatic1111 everything was much better

[dry ivy]

DPM++ 2m or Euler (not ancestral), with sgm uniform

[brittle lantern]

i see ty

would fp8 be better for m2 pro 16gb ram or the normal

[sudden helm]

maybe try the provided workflows and work from there?

[wet magnet]

@dry ivy getting closer I swapped the clip for fp16 and placed the 2 other files in my checkpoint folder and now it says....

Prompt outputs failed validation
CheckpointLoaderSimple:

• Value not in list: ckpt_name: 'sd3_medium_incl_clips_t5xxlfp8.safetensors' not in ['sd3_medium.safetensors', 'sd3_medium_incl_clips.safetensors']

[fierce hollow]

Unfortunately, sd3 does not work in automatic111

[dry ivy]

Inside clip folder
Clip G
Clip L
T5 FP16

Inside checkpoint folder
sd3_medium_incl_clips.safetensors OR sd3_medium.safetensors

[wet magnet]

I forgot clip g and clip L

[fierce hollow]

WTH??

[sudden helm]

in comfyUI you must have set the checkpoint to be "sd3_medium_incl_clips_t5xxlfp8.safetensors" which doesn't exist in your checkpoint folder. So just use this "sd3_medium.safetensors" if you have the other text encoder models placed in the clip folder

[fierce hollow]

Why is this happening? What am I doing wrong?

[dry ivy]

Use CFG 4 to 6

and dpm++ 2m or euler on sgm uniform

[fair oxide]

Hey

[fierce hollow]

And how many steps is it better to put?

[dry ivy]

No idea yet, 50?

[fair oxide]

Can someone help me qwickly?

[dry ivy]

Maybe 👀

[fierce hollow]

Would DPM++ SDE be suitable?

[dry ivy]

Nope

[fair oxide]

I followed the download for amd

[dry ivy]

[brittle lantern]

any steps or cfg scale i need to use

[dry ivy]

Might be a question for CS1o, whenever he gets back

[dry ivy]

20-30 are usually good

But people still think the model is bad, so you're still likely to get bad results

[brittle lantern]

30 steps 7 cgf 2m karres

not bad

[fierce hollow]

SD3?

[brittle lantern]

yh

[wet magnet]

what do you have in your mac m3?

[brittle lantern]

m2 pro

[wet magnet]

how long did generation take?

should say in the terminal

[brittle lantern]

prompt executed 146 seconds but i was tabbed out for some time so idk if that affects anything

[rose ferry]

Is SD3 compatible with the MacBook Pro M1? It seems not because I receive static images with provided workflow.

[brittle lantern]

guys i kinda messed up how tf do i open up the side menu on comfy ui 💀 i got rid of it accidentally

[brittle lantern]

downloaded comfy ui and it worked on m2 pro

[rose ferry]

interesting.

[wet magnet]

m1 might be stretching it

but I'd give it a try

probably can do 512x512

[brittle lantern]

but i cant gen anything cuz idk how to get sidebar back

[rose ferry]

It has nothing to do with static

[wet magnet]

reload the page

[rose ferry]

reboot, yo can get it back

relead page sometimes doesnt,.

[brittle lantern]

iirc my m2 pro is sig faster than m1 correct?

[wet magnet]

m2 yeah

m3 is a jump from m2

[brittle lantern]

it didnt so do i have to restart terminal?

[wet magnet]

probably

[rose ferry]

you need argument. Open webui-user.sh in text or notepad
Change #export COMMANDLINE_ARGS="" to export COMMANDLINE_ARGS="--skip-torch-cuda-test --no-half --use-cpu all".

[brittle lantern]

my m2 pro was alr expensive idk if i wanna go higher

[fair oxide]

I want to use my gpu

[rose ferry]

bah, if higher, i am going PC

[brittle lantern]

my pc has rx 6600 xt in it

[rose ferry]

Apple is intergrated GPU