🤝｜tech-support

[fair oxide]

thanks @karmic crown !

[karmic crown]

too pale, your cfg is probably too low

[fair oxide]

yeah i wonder cause cfg is at 2

[karmic crown]

it is ok with lcm

[fair oxide]

does that require mp4?

[karmic crown]

what is your lcm strenght?

0.3?

[fair oxide]

0.8

[karmic crown]

hmm

[fair oxide]

using LCM sampler

[karmic crown]

:0.8 and cfg 2 should give better colors

is that 1.5?

it if is try with 8400 vae

sometimes it gives better colors with lcm than anime vaes

[fair oxide]

can't lol im using sdxl

[karmic crown]

than i have one for you 🙂

[fair oxide]

only sdxl vae i have here is the sdxl_vae safetensor

[karmic crown]

https://civitai.com/models/152040/xlvaec

try this one

[fair oxide]

nice, it looks very well toned image.. getting it now

im gonna swap the model too.. instead of anime model im gonna try a more versatile one

[karmic crown]

see this #🍥｜anime message

i call that one "for emergency use only sdxl vae" 🤣

[fair oxide]

the comparison is very distinct

[karmic crown]

i am really intrested in what you will get if you use it, with same settings

[fair oxide]

i am

its rendering now

its generating pic frames atm

bit slower it seems

altho i dont think vae affects speed

could be the model i swapped

[karmic crown]

it should not affect speed

[fair oxide]

yeah it shouldn't

i also enabled mp4

so that might be it

[karmic crown]

mp4 would use ffmpeg (i guess) so it will probably take more time

[fair oxide]

once i enabled mp4 i didnt get any warning

but its taking some time

this is mp4

[karmic crown]

there is no reds, so i have no idea is that vae helped or not

but blue in the top right looks nice

[fair oxide]

it isn't as washed out as the previous one

[karmic crown]

yes i know, but it would be more obvious if there was something red on picture 🙂

[fair oxide]

this video was pretty straightforward and with your help on the motion module download it went smoothly ... i tried to make animatediff work before with oither guides but it didn't go so well in my past attempts https://www.youtube.com/watch?v=dAOODfDyelA&ab_channel=NextDiffusion

and btw that vae? its pretty crisp and toned

that vae needs to go mainstream for sdxl

[karmic crown]

it is experimental

read the full description

[fair oxide]

its pretty good so far

i know

i mentioned that to you before lol, but results are pretty good so far

[karmic crown]

i like it, to be honest, but someone said it make some artifacts

[fair oxide]

i will keep that in mind about artifact, but im gonna run more images with it 🙂

[strange thunder]

Could anyone familiar with the infinite image browsing extension help me in understanding something? I adore the "fuzzy search" option, but I can't seem to figure out how to get it to search more than one thing. If I want to find something with "sunset," no problem. If I want to find something with "sunset" AND "beach," it doesn't give me any results when I clearly have some that fit the criteria.

Presently poking around online to find the faq or answer, but I'm coming up dry..

[fair oxide]

any idea why my checkpoint merger is failing with the following error :

TypeError: argument 'metadata': 'dict' object cannot be converted to 'PyString'

[karmic crown]

drawing.*superman|superman.*drawing this will give you superman and drawing in any order, but you need to click regex button

[strange thunder]

This worked, but could you unpack that a little more? What is the formula exactly?

And does that mean that I'm not able to do a multi search that just filters images that contain all search elements? Is there no way to get an exaustive search of "superman," drawing," "skyline," and "full moon" without every permutation/combination of those terms?

[karmic crown]

i just checked the code. it can search only for one word

cur.execute("SELECT * FROM image WHERE path LIKE ? OR exif LIKE ? ORDER BY date DESC LIMIT ?", (f"%{substring}%", f"%{substring}%", limit))

let me try something bad

[strange thunder]

lol wut... because people only ever need to search one thing ever... lol

And ok

[karmic crown]

ok

looks like regex is the only option at this moment

use it like this

firstsearchitem.*secondsearchitem|secondsearchitem.*firstsearchitem

[strange thunder]

Ok. And for 3 search terms?

[karmic crown]

you will need 6 combinations

1st 2nd 3rd, 1st 3rd 2nd...

same thing

.* (any number or characters between including 0)

| = or

[strange thunder]

ohhhhh ok my brain was reading it as "." with the first term and "*" with the second term. silly me. makes sense now I think

[karmic crown]

first.*second.*third|first.*third.*second and so on

with 3 items you have 6 combinations

1,2,3
1,3,2
2,1,3
2,3,1
3,1,2
3,2,1

like this

[strange thunder]

Gotcha. Is there any particular reason multi term search isn't supported yet? Seems fairly basic in terms of a needed utility.. I won't complain, especially since there is an alternative (way better than being sol), but my brain just implodes when it sees stuff like this missing, especially when it's on such a useful function. Dx

Like how controlnet stopped recognizing png info data for the longest time and that was just life for a while.. lol

[strange thunder]

Also, I don't really have any script writing knowledge, but I imagine someone could just write some simple code where you can input "a,b,c,d" and the output would be all of those permeations that you could just paste into the search bar. How hard would that be to do?

[willow moth]

Higuys,how can i fix thisd

[carmine forge]

Hi, first time installing stable diffusion, does anyone know why?

[fast mirage]

a RX 6600!

sorry for the time on responding

[grim moon]

Hey

Where do I put this?

https://civitai.com/models/100191/ti-easynegativev2-textual-inversion-embedding

[solar sphinx]

Having trouble with x/y/z plot with checkpoint name. It seems to just end immediately with : attributeerror : module 'modules.sd_models' has no attribute 'reload_weights'

X/y/z works with every other option I chose except "checkpoint name"

[azure magnet]

How to solve this?

Error with controlnet

[bright light]

Guys, please help me. I don't understand why I no longer have tiling on my A1111. I need it to create a texture and I realize that I don't have it anymore... Do you know why?

[hallow jasper]

im trying to run stable diffusion with foocus and amd gpu but im getting this error: RuntimeError: Device type privateuseone is not supported for torch.Generator() api.

what could it be?

[bright light]

Note: The Tiling checkbox is now on the Settings page.

Can anyone explain why you removed this checkbox???? Now I have to reboot my A1111 every time I want to change this setting???? Seriously guys?

[karmic crown]

in the embeddings folder, next to Place Textual Inversion embeddings here

[karmic crown]

go to their github page and rise an issue
https://github.com/zanllp/sd-webui-infinite-image-browsing/issues
ask them to add support for more than one term in fuzzy search

I just gave you a workaround since I have no other idea how to help you. regex is there for complex searches and it is not intended to be used like that

[strange thunder]

And I appreciate it so much! Was mostly venting/probing for any other things I might have been ignorant to. lol. So thanks very much!

I was going to do that but wanted to ask first. Because that was something else brainbreaking: That no one else had had the issue when I did a search there XD

[ornate elk]

Then you need to edit the webui-user.bat and add --medvram --opt-sub-quad-attention --opt-split-attention-v1 --upcast-sampling --no-half-vae
To the Commandline_ARGS=

[fast mirage]

Thx! I'll try that, but what if the time to generate images continues very long?

[vocal burrow]

We are not auto1111, so your rant won't do much here.
Also I'm pretty sure you don't have to reboot A1111 just to toggle tiling.

[random pagoda]

cant use my lcm lora, any advice , i tried several steps / cfg / samplers

[karmic crown]

are you sure that you are using the correct lcm lora?
XL models - XL version of LCM lora
1.5 models - 1.5 version of LCM lora

[random pagoda]

yes 🙂

[karmic crown]

what cfg and lora strength did you use?

[random pagoda]

lora is on 1 but tried different stregth too

cfg from 1-2.5

and steps i tried 3/4/5/6/... 15

but its more about that the pics are completely messed up

like not a bad result , more of a wrong result

[karmic crown]

is this xl or 1.5?

[wraith steppe]

Hi, i have a problem, my automatic1111 continue to crash randomly without no sense... doesnt matter what version i use or what extensions i have on it, randomly it crash... in the middle of an elaboration can appear the text "select a button to exit..." and i dont know why cause no error show !! i restart and it crash but randomly ! i work a lot with deforum and i use the 1.6.1 versione and 1.7.0 rc version, it crashed also when i tried to install 1.7.0 ... no sense! is a problem of my pc? i notice also google chrome crash always is a ram problem??? pleas help

[random pagoda]

xl

[karmic crown]

@random pagoda try go genrate a simple image with default size 512x512 (if 1.5) or 1024x1024 (if it is sdxl), do not use upscalers, loras... just use lcm lora that is correct for your model (doublecheck that) use lora streng of 0.5 and be sure that you use the same name as the filename for example your filename is LCM_15 than you need to use lora:LCM_15:0.5 and not lora:pytorch_lora_weights:0.5 for example. paste the generated info when you get the result (as text)

use cfg of 1.5

and 8 steps

use lcm, euer a or unipc sampler

[verbal burrow]

What does it mean when it says not enough values to unpack?

[ornate elk]

Then show me your txt2img settings

[random pagoda]

ty its because of the 0.5 @karmic crown

8 steps

weird didnt know i cant use it at 1 weight

at least it seems the 0.5 fixes it

[ornate elk]

How much ram do you have?

[random pagoda]

@karmic crown still weird, seems like a lot of people use 1.0 and it works for them

[karmic crown]

i am found that this "math" works the best lora strenght = 1.125/cfg

you do not need to be precise

[random pagoda]

im still confused there are so many pics on civitai with 1.0 strenght that are not messed up

[karmic crown]

what is the cfg in those cases?

[random pagoda]

1,3

[karmic crown]

that is fine

[random pagoda]

or 2

like this here

[karmic crown]

cfg 1.3 + lora :1 is fine

[random pagoda]

if i use cfg 1.3 and lora 1 it ends bad again

ok with the lcm sampler its way less sensitive for errors

tried lcm sampler

[karmic crown]

what you get if you do not use lcm lora at all

with cfg of 2

[wraith steppe]

32gb ddr5 corsair 7200mhz

[random pagoda]

do u have any advice for getting better result with even lower steps like 3? or 5?

[karmic crown]

i never use less than 8

i use 6 for experimenting only

[random pagoda]

allright

[karmic crown]

if that is a new model, chek if it has lcm embeded

if it does, you do not need lcm lora

that can also be a problem

models with embeded lcm, do not need lora, that will just make a problem

[random pagoda]

got this with 3 steps

[ornate elk]

That should be enough but make sure your Mainboard supports 7200

[random pagoda]

looks fine now

[karmic crown]

try this one https://civitai.com/models/200061/hephaistosnextgenxl-lcmbf16hybrid

i love this one

it has lcm embeded

good anime model with embedded lcm is this one https://civitai.com/models/202108/bluepencil-xl-lcm

there are some others, but i didnt try them

[wraith steppe]

Is asus rog z790 asus hero

[random pagoda]

i just did a grid

i feel with lcm u can go higher with the weight

[ornate elk]

okay, whats your python version?

[hallow jasper]

how do i know if something is a lora or a checkpoint

[ornate elk]

everything larger than 1.98gb is a checkpoint

[wraith steppe]

3.10.6

[ornate elk]

okay thats good.

did you tried the stable webui version?

1.6.2

[wraith steppe]

I changed the clock of rams to 7000mhz seems work now but need to wait 1-2days first

[ornate elk]

okay hopefully that works

[spark moon]

hello anyone able to help out?

i'm trying to get into Dream Studio , and it is saying my password is incorrect, I've tried sending an email to reset password, but the email does not show up in my inbox

[fast mirage]

well i don't know where to find those, but one thing that i found out is while i'm generating an image, it doesn't anything from my gpu at all

it uses almost all my memory

forget what i just said, i can generate images at a acceptable rate now

like 1min per image

but they just all bug out xd

[ornate elk]

Dont use sdxl models with an resolution of 512x512 or lower.
Thats why the images look so bad.

Sdxl models got trained on 1024x1024

But that will likely not work with your gpu

So the best is to stick to 1.5 models as they are trained on 512x512
You can later upscale them for better quality and Highres resolution.

Also use batch size 1

[strange thunder]

https://github.com/opparco/stable-diffusion-webui-two-shot/issues/43 Could I get a little help understanding this comment/solution? I'm not sure what exactly I need to delete from the .py file. I've tried several combinations and I'm clearly not doing it right because the png info still isn't saving. I know it's the correct fork

[fast mirage]

oh alright.

can you give me an example of a 1.5 model?

and by model you mean... what? I'm really new into sd

[warped wyvern]

first time install and I got this message. Anyone know what might be causing this?

[azure magnet]

LMC works even to SD not XL?

[vapid ice]

Stable Diffusion is giving me this error when I try to open it.

Please someone help.

[onyx solar]

Do you actually have git installed?
Follow the installation instructions:
https://github.com/AUTOMATIC1111/stable-diffusion-webui#installation-and-running

[vapid ice]

Installed it, same error when trying to open via webui-user.bat

but I'm able to open it when clicking on run.bat

is that normal?

[ebon horizon]

I'm trying to pass an image from a sampler generating with SD v1.5 to another sampler refining with SDXL, but I'm getting weird noise when I pass the latent from one sampler to the next. Is there any way to get a clean output this way without decoding and re-encoding with VAE in between?

My current workflow is
512x512 > KSampler (SD1.5 )> Latent upscale x2 > KSampler (SDXL) > KSampler (Refiner) > Latent Upscale x1.5 > KSampler (SDXL) > KSampler (Refiner) > KSampler (SD1.5) > RealESRGAN x2 > Output

[pure viper]

Every time I try to launch Juggernaut in my Comfy workflow I get a screen full of red all kicked off at the top by:
Error occurred when executing Efficient Loader:

'model.diffusion_model.input_blocks.0.0.weight'

[ebon horizon]

Did you download the .yaml file and place it in the folder alongside the model?

[pure viper]

AH! Let me do that. Thanks.

I guess I don't know where to find that. Don't see anything on CIVITAI

[ebon horizon]

https://civitai.com/models/46422?modelVersionId=127207

Make sure you select the right version at the top, then under download options there should be one called config

[pure viper]

So I was here because this is where I got the checkpoint. And I don;t find one there. I guess I'll download that one as I'm so new to this stuff that it doesn't matter which one I use. I thought I was getting the XL. Cheers.
https://civitai.com/models/133005?modelVersionId=240840

[ebon horizon]

Version 5 of XL has it. You may be looking under the later builds that were made for RunDiffusion.
https://civitai.com/models/133005?modelVersionId=166909

[pure viper]

So, that fixed it. One thing I learned is sometimes the config/yaml is a link in the window and sometime it's in the pull down window with the checkpoint and the vae. Cheers.

[verbal burrow]

How do I get SDXL?

Been using SD V1.5, but can't seem to get SDXL to work, at least for the checkpoint I'm using it never loads it

[desert grove]

I'm getting the first error followed by the second when trying to use a depth controlnet all of a sudden.

Welp, it randomly went away after disabling the controlnet, swapping to another checkpoints and swapping back(tried this on it's own at first and still had the error), generated without it, then enabled it again and now it's back to working.

And now it's doing it again.

What gives?

It seems to be caused by using a refiner, but I'm not doing anything differently from the last times I've used stable diffusion.

It won't let me load the model I'm trying to refine with. Was there an update or something to the webui that might've broken the model? I haven't changed anything manually.

[ornate elk]

I mean checkpoints = model.
Get Dreamshaper v8
Thats a good 1.5 model for example

[delicate urchin]

Why does it read that as a string and not float?

this workflow has always been working until I moved all comfy files to a new pc

If I put float to string and string to float between those two then it works again

but that should be not needed

FIXED: the math node in comfy is full of bugs. I just replaced the nodes with the same identical nodes and then it worked

[elder turtle]

What UI are you using?

[quiet lion]

Can anyone advise me on how to set up img2vid with comfyui on colab? I am getting this error.
I have comfyui manager - shall I install model direct using that?

[quiet lion]

I think I a managed to solve that but now running a prompt I get this error:

ERROR:root:!!! Exception during processing !!!
ERROR:root:Traceback (most recent call last):
File "/content/drive/MyDrive/ComfyUI/execution.py", line 153, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
File "/content/drive/MyDrive/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 "/content/drive/MyDrive/ComfyUI/execution.py", line 76, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
File "/content/drive/MyDrive/ComfyUI/comfy_extras/nodes_video_model.py", line 45, in encode
output = clip_vision.encode_image(init_image)
AttributeError: 'NoneType' object has no attribute 'encode_image'

[unique anchor]

Kohya SS can only run on Linux?

Cuz I am trying to run on Windows and I get bitsandbytes error and upon further inspection, I found out that

The bitsandbytes library supports CUDA versions from 10.2 to 12.0. However, it's important to note that this library is currently only supported on Linux distributions, and Windows is not officially supported at this time​

[elder turtle]

https://github.com/bmaltais/kohya_ss#installation
Should run on windows without issues

[unique anchor]

I get this error, let me get the link

https://hastebin.com/share/emoroparim.javascript

[elder turtle]

I havent used it myself so cant say too much, did you follow the instructions for installing it?

[unique anchor]

Yes, step-to-step.

Even followed the tutorial from Not4Talent and Aitrepreneur

🥹 I was getting a CUDA error before but when I reinstalled, it disappeared and gave me this new error.

[uncut jewel]

Hey guys, just here to ask a quick advice, what's the shortcut to make pop the searchbar in comfyui, to get nodes way faster ?

[elder turtle]

Double left click

[uncut jewel]

no way it's that simple -_-

I tried so many stuffs and it's just that

Thanks ahah

[verbal burrow]

Automatic11111

Also I just got am RTX 3060, do I need to get anything for it?

[karmic crown]

#🤝｜tech-support message

follow this guide

use the command line from there

ooops wait, there is no command line ther

[jagged plover]

did you try to use the bnb 8bit optimizer? if so, you'll have to not use that on windows i guess? just swap to any of the other optimizers.

(bnb should work on windows iirc but a lot of research software gets wonky when you're on windows so idk)

[karmic crown]

For 3060
Use:
--xformers --medvram --no-half-vae

[ornate elk]

Not medvram for 12gb

[karmic crown]

listen to CS1o

i can help with Mac 🙂

--medvram for 8GB?

[ornate elk]

--medvram for 6gb or less

[karmic crown]

ill try to remember (if you are not here next time) 🙂

[ornate elk]

--lowvram for 3gb or less or if they want to use heavy extensions with 4gb

[unique anchor]

May I ask what that optimizer thing is?

[verbal burrow]

I already have the webui though

Been using it when I had installed my 1060

[karmic crown]

my guide for Macs has a command line, so I though his PC guide also has it, but then I saw that there is no command line there

[lost crescent]

Torch not compiled with Cuda? Help, anybody 🙂

[elder turtle]

Did you update the UI anytime recently?

[burnt lily]

Hi guys, i am using civitai extension to get the preview images for LoRAs, i think the extension makes my SD to calculate sha256 constantly (constant looping from filename a-z, A-Z, then back to a-z, even the files with preview and json files). Found similar issue on github, but tagged as non reproducible (https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/12826) and the guy used civitai-helper instead of civitai extension. I've tried to only include 1 LoRA file in the directory, and it the scanning loop did not happen. Maybe the hashing process failed in one or more LoRA, i don't know. It's so frustating, and for temporary remedy, i just use the --no-hashing command. wonder if anyone here have solved it, thanks!

[jagged plover]

optimizer selection in kohya trainer is, uh, which provider of magic math you want to do the trainy with?
... sorry it's hard to explain without explaining a lot of math that i don't even understand well to begin with.
Main practical point is just use whatever the standard recommended one is, and don't do the bnb/8bit one

presumably either adafactor or one of the adamw variants is the recommended one

[unique anchor]

Okiee I will look into it tomorrow

Kohya is on my work laptop which is at work xD But thank you so much. I will surely look into it tomorrow <3

[karmic crown]

temporary disable the plugin and check what happens

you can also try this one instead https://github.com/BlafKing/sd-civitai-browser-plus

[ornate elk]

You need to edit the webui-user.bat and at the line COMMANDLINE_ARGS=
You add: --xformers --no-half-vae

[ornate elk]

What's your GPU ?

[crimson leaf]

@vocal burrow finally, i replace the rx580

No more out of support issues hopefully

[verbal burrow]

I'm not sure if I did

[elder turtle]

You should, one update around the XL release implemented the support

[verbal burrow]

How can I check the version to see if it did update?

[verbal burrow]

When I launch the webui-user.bat I see Version: v1.6.1

[elder turtle]

Do you have git pull in your webui-user.bat?

[verbal burrow]

Just "git pull" yeah

[elder turtle]

Then it should be up to date

[verbal burrow]

`@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=
git pull
call webui.bat`

That screenshot I took just now pops up everytime I launch the webui-user.bat now

[elder turtle]

Does it spit any error while loading?

I mean the XL model

[verbal burrow]

Let me try to load it

[elder turtle]

Did you reinstall torch?

[verbal burrow]

I have to reinstall it?

[elder turtle]

It says so in the UI

Also your python version doesnt seem to match xformers

[ornate elk]

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

Python is ok

[elder turtle]

Well nvm

[ornate elk]

Torch is just to old for xformers

[verbal burrow]

StableDiffusionXL_v30Safetensors loads now

[ornate elk]

Have you deleted the venv folder ?

[verbal burrow]

In the process of doing it rn

[elder turtle]

Try again after

And restart the UI

[ornate elk]

Also 1.6.1 is not the latest version

So git pull maybe dont work

[ornate elk]

okay thats old stuff to fix that:

Close Stable diffusion.
open up a cmd and run
pip cache purge
After that close the cmd.
Go into the stable-diffusion-webui folder.
Click in the File Explorer bar. (not in the search bar).
then type cmd and hit enter.
after that type and run:
git reset --hard HEAD
and then
git pull
Then delete the venv folder and run the webui-user.bat

Then your updated and SD will be much faster

[verbal burrow]

Also want to mentioned my models for Lora or embeddings didn't load either

[ornate elk]

try my fix and then they should work too

[verbal burrow]

Yep, doing that rn 👍

[verbal burrow]

File Explorer bar?

Nevermind

[elder turtle]

Where your path is

[verbal burrow]

The venv folder is not there (webui folder)

[elder turtle]

Let it run, it should rebuild

[verbal burrow]

Oh, just popped up

Also, is it worth using the civiai extension?

[elder turtle]

The one that imports covers? I like that yeah

[verbal burrow]

That's all it does?

[elder turtle]

Well I only use it for that

[verbal burrow]

https://tenor.com/view/heavy-tf2-the-rock-rock-the-dwayne-johnson-gif-22149536

[elder turtle]

You can also download models through it, but I dont see the use

[verbal burrow]

I don't have a token for civiai lol

Alright, the webui launched, gonna try to generate something

All my Lora models except 1 aren't showing up, same with embeddings

[vocal burrow]

Are those compatible with the model you currently have loaded ?

[ornate elk]

You need to edit the webui-user.bat

[verbal burrow]

Would they show up if they weren't?

[ornate elk]

To use xformers

[vocal burrow]

no

[verbal burrow]

Same as before?

[ornate elk]

--xformers --no-half-vae at the COMMANDLINE_ARGS

[vocal burrow]

loras/embedding are made for a specific version of SD (1.5, 2.0, 2.1, SDXL, etc)

[verbal burrow]

I had a model that said it was StableDiffusionXL but even that doesn't show up

[elder turtle]

Did you download it from a shady source?

[verbal burrow]

I switched my checkpoint and it the models aren't show up still

[verbal burrow]

Civiai

Nevermind, I'm just dumb

I didn't hit refresh

[elder turtle]

[verbal burrow]

Should I add git pull under it as well?

[elder turtle]

If you want to keep it updated, yes

[verbal burrow]

It did generate an image, but got this in the command prompt

[ornate elk]

You should update your extensions too

And Anylora is not a lora

[verbal burrow]

Huh.mp3

[ornate elk]

Remove it from lora folder

Its a model for loras

[verbal burrow]

So it doesn't go into the Lora Folder under Models?

Where do I put it or do I not have anywhere for it?

[verbal burrow]

I'm this far into the rabbit hole, might as well

[rancid spindle]

Hey guys, I been playing with stable diffusion on my m1 Mac it’s a beefy M1 but I was told I should be looking at a windows desktop to get the best performance when running locally is there some specs of hardware recommended

[ornate elk]

models/stable-diffusion

[elder turtle]

Id at least get a 30 series GPU and a fitting PC
For the best performance right now youre probably looking at a 4090

[rancid spindle]

Does it matter laptop vs desktop

[elder turtle]

Laptop GPUs tend to run slower and on less power, so id go desktop if you want the performance

[rancid spindle]

Thank you much …

[elder turtle]

No problem, but be sure to not just throw money at the first best system you see. I advise people to build their own PC all the time

[lost crescent]

GeForce RTX 2070 8Gb VRAM

[verbal burrow]

I might need some help with figuring out if I have my Loras and checkpoints correct. I assumed that anything with the name Lora goes into the Lora folder within Models

[sand trail]

is there anything else i can do to stop getting this message

[rancid spindle]

Yep that was exactly what I was going to do … ok so I need to do more researcher on hardware and building myself thank you

[elder turtle]

Its also cheaper to build yourself, just set yourself a limit cash wise

Else youll get lost

[verbal burrow]

Someone told me it's more expensive to build your own pc

I built my own anyway

[lost crescent]

Cannot get Torch to be Compiled with/by Cuda?! RTX 2070 8Gb VRAM

[rancid spindle]

I was looking at building a pc but the GPU seems more expensive than having it built already

[elder turtle]

You usually pay a fee for assembly and some extra for the company building it

[elder turtle]

Well it depends on the GPU, a 4090 obviously is more expensive than a 3050 for example

[rancid spindle]

So you recommend just pick the hardware and having someone else build it …

[lost crescent]

I bought a fantastic PC - but I'm convinced that the GPU was second-grade stock (I sent it back 3 times for video faults!!!)

[verbal burrow]

Oh, I assumed you meant build it on your own without any help

[elder turtle]

I meant that yeah, just saying if you buy one prebuild you pay extra

[ornate elk]

For a good SD ready PC you should get a nvidia card with at least 12gb vram.
16gb or more RAM.
And a CPU with 6-8cores, 4 would work too but for future i would go for more

[elder turtle]

Only if you dont believe you can do it alone

[lost crescent]

Personally, my 8Gb VRAM is working the Pixart-alpha (via lowvram bat file) really good

[elder turtle]

Also pcpartpicker is a great source for compability of parts

[lost crescent]

70 seconds per 60 step output

[sand trail]

can I do that to my 6gb

[rancid spindle]

Yeah time is my issue building myself so I will look at parts and have someone build it

[ornate elk]

It can

[sand trail]

i was adviced to use medvram

[lost crescent]

Not sure - the Pixart-alpha I'm using came from patreon (SECourses) and includes a lowvram option, 8Gb seems to be the minimum)

[elder turtle]

I mean then you can also buy it prebuild, saves you all the looking parts up and stuff

[lost crescent]

I'm not sure how to do it?

[verbal burrow]

If you're going to build it yourself, please keep in mind dimensions of each component compared to the casing

[ornate elk]

Auto1111 does it at install

[lost crescent]

How do I compile my Torch using CUDA?

[rancid spindle]

Do you have a recommended site for prebuilt and yes I am looking at the 4090

[lost crescent]

Yes, but I'm trying to setup comfyUI

[elder turtle]

Depends on where youre located tbh

I only know a few german sites

[ornate elk]

Comfyui should work by default with nvidia cards

Are you on a laptop?

[lost crescent]

This is what's happening with my nVidia Card

Torch not compiled with CUDA enabled?!

My Cuda Toolkit is v12.3

My Torch is 2.1.0

(Or is it 1.6.0?)

[verbal burrow]

Should I be worried about this error?

[sand trail]

any one tried token merging

[lost crescent]

At Github - most people assume that you are not using an nVidia GPU!!!

[lost crescent]

A company in London UK (MESH, of Wembley) built me a 64Gb DDR5 RAM, RTX 2070 8Gb VRAM, 2x1Tb SSDs ... and for 5 years it has worked almost flawlessly

MicronPC of Boise, Idaho also do customer-spec builds

I wonder if I just swap out the 2070 - for a 4090?

What could go wrong?!

[lapis merlin]

how do I fix privateuseone error if using --directml with AMD GPU?

[lost crescent]

torch.cuda.is_available() = false

where torch

[heady pine]

I am trying to reproduce an image I made with the same seed, but every time I do, the image is different, does anyone know how I can fix this issue?

[ornate elk]

Can you show the complete cmd output?

[vocal burrow]

ddr5 ? five years ago ?

[elder turtle]

Probably underpowered PSU

[vocal burrow]

Also check if you've got enough physical space to swap for a 4090, those are chunky

[elder turtle]

Or get the MSI Liquid one with the 240 rad

[vocal burrow]

1/ --xformers (and some other flags but mainly this one) will make stable-diffusion non deterministic. Meaning for the same prompt+settings running on different hardwares you'll get different results. The output should still correspond to whatever the prompt is.... It might just be slightly to very different if you're unlucky. So yeah, some command line arguments can alter the outputs of your prompts and your hardware might also do the difference.
2/ Do you have all the loras, embeddings mentioned in the prompt ?
3/ Are you really using the same settings ? Copy pasting manually what's inside the "Generation Data" boxes is not enough. Civitai is hiding a bunch of data there. If you want to get the exact same settings you'll have to click the "Copy Generation Data" button at the bottom of the image page, paste its content into Auto1111's prompt field and then click the blue arrow under generate to automatically apply each value to the correct field
4/ Did you have any overrides set beforehand ? They should show up at the very bottom of the page (some will probably show up too if you use data from civitai like I mentioned in 3/)
5/ Maybe it's using some extensions that does not record its settings in the metadata
6/ Maybe you're using different versions of some extensions/auto1111/lora/models/etc that yield different results

[lost crescent]

Prolly DDR4

[lost crescent]

Upgrade my PSU

[gray oyster]

get windows x nvidia for ease of use or linux x amd if you like to suffer and also to safe a few bucks

[lost crescent]

So still stuck with "Torch not compiled with CUDA enabled!" 😄

Github also has little info on this matter ...

[gray oyster]

what gpu you got?

and what cuda version

[lost crescent]

RTX 2070

Cuda 12.3

At least CUDA Toolkit 12.3

[gray oyster]

try 11.8

that should definitely work

i heard that for 20 series cards 11.8 is recommended, might be outdated news though

[lost crescent]

Yes, but how to enable CUDA to compile Torch?

[gray oyster]

in auto1111?

[lost crescent]

ComfyUI

[gray oyster]

ah

then delete the venv folder and relaunch comfyui after installing cuda 11.8

[lost crescent]

My error is AssertionError("Torch not compiled with CUDA enabled")

[gray oyster]

also run nvcc --version in cmd

[lost crescent]

OK, will try

[gray oyster]

it could be that it tries to install PyTorch for the wrong cuda version

you would need to find the line in one of the many files for that

[rancid spindle]

Ok I an in the US let me do some research to see if there’s a company that specializes in building machines for this purpose .. thank you

[gray oyster]

also if you wanna use cuda 12.x use 12.1

that might be the issue

[lost crescent]

OK, 11.8, 12.1 - any more? 😄

[lapis merlin]

I have --directml as the only command

[gray oyster]

try 11.8 that should definitely work

[lost crescent]

Going to try 11.8 ...

[gray oyster]

the cuda toolkit i mean of course, not pytorch

pytorch is being installed through comfyui itself

[lost crescent]

Yes, CUDA Toolkit 118

[gray oyster]

yea

[ornate elk]

Then your using not auto1111 but SD.Next I guess

[verbal burrow]

Is it normal for the AI to start randomly making weird images even though I didn't change the prompt?

[gray oyster]

depends

[lost crescent]

if you have a low cfg then maybe ..

[verbal burrow]

I've kept it at 16, but randomly I get these images that don't make sense

[elder turtle]

I have that sometimes with running batches, my first image is fine and the other 3 are gibberish

[elder turtle]

16 is mighty high

[gray oyster]

16 seems way too high

higher cfg at lower resolutions usually causes oversaturation and just not good looking images

14cfg at 1024x1024 is already pretty high and should only be used with highres.fix imo

[verbal burrow]

I seem to get better results when I set it to 16 compared to say 7

[gray oyster]

do you use negative prompts?

[verbal burrow]

Yep

bad resolution, text, (worst quality:2), (low quality:1.8), jpeg artifacts, ugly, duplicate, easynegative, morbid, mutilated, out of frame, (mutated hands), background, bad hands, (poorly drawn face:2), mutation, deformed, blurry, (bad anatomy:2), bad proportions, extra limbs, cloned face, disfigured, malformed limbs, missing arms, missing legs, extra arms, (extra legs:1.0), (fused fingers), extra fingers, long neck, (distorted face), overexposure, underexposure

[gray oyster]

worst quality:2 is too high

try not to use anything higher than 1.5

[verbal burrow]

Oh, okay, I usually see people do 1.8 for prompts

[gray oyster]

also try to always use n.x format

so you always have a float

[verbal burrow]

What's that?

[gray oyster]

instead of 2

you do 2.0 for example

[verbal burrow]

Oh

[gray oyster]

at the very least it looks a bit better imo

[lost crescent]

lcm can be boosted for detail by using FreeU

[gray oyster]

@lost crescent does 11.8 cuda work?

[verbal burrow]

So having anything:2.0 is simply too high?

[gray oyster]

yea

1.8 is usually the max before it looks bad

of what i found

[lost crescent]

I don't have a VENV folder!

[gray oyster]

what folders do you have?

worst case you need to reinstall comfyui entirely

[lost crescent]

VENV is for A1111

[gray oyster]

usually python based programs use a venv folder as its the virtual environment so your global libraries that are installed stay unaffected

say you have a project which needs pytorch pre 2.0.0 but another uses post 2.0.0

[verbal burrow]

Somehow the preview images during generation have a good start but by the end of the process come out completely different

[lost crescent]

ComfyUI_portable has its own python_embedded folder

My ComfyUI (not portable) uses main 310 Python/global

[gray oyster]

thats bad but that makes it slightly easier to fix

alright so you have 11.8 installed?

cuda?

[lost crescent]

Yes it has installed - I will reatart CUI and see what happens ...

[gray oyster]

i dont think it will fix anything since you probably need to reinstall pytorch

[heady pine]

Sorry for the late reply, I made the image myself, I removed xformers from the webui-user.bat to see if that would work, but sadly no. I made this image about 10-15 minutes before I started having the problem of not being able to remake it, I'm trying to see how use it as a base image with other models to see which model I should use going forward. Usually I just put the image in and click -Send to txt2img- and the image at the time works, but now, I'm unable to remake any of my past images doing that.

[gray oyster]

https://pytorch.org/get-started/locally/

go on this website and then select your required options

[lost crescent]

Its working so far ... 🙂

[gray oyster]

alright

if it works then thats awesome

if not you will have to use the website to generate the installation command

then run that in your cmd

[lost crescent]

So 12.3 to 11.8!!! Way to go

[gray oyster]

tipically yea

awesome

[lost crescent]

Got to the UI - now says this

[verbal burrow]

@gray oyster Do you use refiners?

[gray oyster]

not me no

[gray oyster]

for that i honestly have no clue

might find something online though

[lost crescent]

I re-booted - something seems to be working... 🙂

Yesssss. Back in business, but not ComfyUI_portable!! BIG ComfyUI

This image via JPS.json - a vey busy and intricate w/flow

Now to try the Pixart-alpha ...

Thank you for all the help!

[gray oyster]

no problem

[crimson leaf]

At what amount of ram is --medvram suggested?

12GB vram enought not to do it?

[lapis merlin]

anyone here use comyfi ui with amd what are your command line args besides --directml?

[gray oyster]

--medvram was probably thought for 8gb

[crimson leaf]

RuntimeError: Detected that PyTorch and torchvision were compiled with different CUDA versions. PyTorch has CUDA Version=11.7 and torchvision has CUDA Version=11.8. Please reinstall the torchvision that matches your PyTorch install.
Uh, how do i fix this? Is torchvision in the venv or what?

[thorny horizon]

hi, im trying to train a SDXL model with 180 images with 1500 images in regularization folder, mt vram usage reaches 12 gb and crashes (using 3080ti), and even when it doesnt crash, it working very slowly 27.5s/it, im using kohya to train, is there any parameter im putting wrong ?

this is my first time training a model and i followed this video https://youtu.be/N_zhQSx2Q3c?si=Ia8oOr9DHJH0MAji

[lost crescent]

Uninstall 11.7?

[solid fossil]

yo i installed the stable diffusion from automatic1111 since im using a amd card, and after launching the webui-user i get this error: ImportError: DLL load failed: The specified module could not be found.

any ideas? tried everything foundable in the internet

[ornate elk]

Medvram is for 6gb or less or AMD

[ornate elk]

Hey, what's your python version ?

[solid fossil]

3.10.6

[ornate elk]

Okay,
then close the webui.
Open up a cmd and type
Pip cache purge
Then hit enter.
Then delete the venv folder and run the webui-user.bat again

[solid fossil]

aight gimme a second

tried this before but without the pip cache purge

btw idk if this matters but its quite weird, downloading it normally works "fine" but when i run it as a administrator i dont even get till the dll error it crashes before that

how is it possible that administrator works worse than running it normally

[ornate elk]

Dont run it as admin

It should work normaly as user

[solid fossil]

yeah im not, was just a try

[solid fossil]

same error

[crimson leaf]

how do i do that? There is no pytorch under pip list

must be in venv, but i already deleted it as i got my nvidia gpu(i used rocm)

[ornate elk]

Okay looks like a bad install.
Follow my directml install guide

hey, you installed the webui the wrong way:
use my guide instead:

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

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

1. Make a new Folder on your drive (not on Desktop, downloads, documents, Programms, Onedrive) and name it Ai for example:
   C:\Ai\
2. You go into the folder you created in this case Ai, then click in the File Explorer bar (not searchbar) and type cmd then press enter.
3. Then you copy and paste this command:
   git clone https://github.com/lshqqytiger/stable-diffusion-webui-directml && cd stable-diffusion-webui-directml && git submodule init && git submodule update
4. Press enter and after its done you can close the cmd
5. Edit the webui-user.bat (right click), At the line COMMANDLINE_ARGS= You add: --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling
   After that save and launch the webui-user.bat inside the stable-diffusion-webui folder.
   After the Installation that can take a while. Youll get an URL http://127.0.0.1:7860
   Thats the webui you open in Browser.

[solid fossil]

brb i think i found the solution

[ornate elk]

Okay let me know

[solid fossil]

worked

media feature pack was missing

windows 10 N has some dll issues ig

[lapis merlin]

what settings should I use for ComfyUI (amd gpu) ? I get 96% ram usage and then it crashes 32gb ram, 6700xt.

[frosty pawn]

Who knows the solution to the problem: "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]

Oh yes its the buggiest and worst windows to use

[ornate elk]

Did you followed the instructions for the directml installation?

[ornate elk]

What's your GPU ?

[frosty pawn]

Geforce GTX 1660

[ornate elk]

Then you need to edit the webui-user.bat
At the line COMMANDLINE_ARGS=
you add: --xformers --medvram --no-half
Then save and launch webui-user.bat

[crimson leaf]

@ornate elk are those args good for nvida gpu? I used it with rocm/amd unsure if it is still good (replaced -medvram with -xformers)
export COMMANDLINE_ARGS="--xformers --no-half-vae --opt-sub-quad-attention --opt-split-attention-v1"

[ornate elk]

What's your GPU?

[crimson leaf]

rtx 4070

[ornate elk]

8 or 12gb ?

[crimson leaf]

12gb

[ornate elk]

Then you only need --xformers --no-half-vae

[crimson leaf]

alr thank you

[frosty pawn]

Thank you

[crimson leaf]

i still get this error, after reinstalling the venv:
RuntimeError: Detected that PyTorch and torchvision were compiled with different CUDA versions. PyTorch has CUDA Version=11.7 and torchvision has CUDA Version=11.8. Please reinstall the torchvision that matches your PyTorch install.
But it says in the download
Successfully installed torch-2.0.1+cu118 torchvision-0.15.2+cu118

in the pip list i have those, but nothing called "pytorch"
torch 2.1.1 torchaudio 2.1.1 torchvision 0.16.1

[carmine forge]

does anyone know why this keep happening at the start?

[strange thunder]

Would upgrading from windows 10 home to windows 10 pro logistically do anything to mess with/break my a1111 setup?

[ornate elk]

Open up a cmd and run
Pip cache purge
Then delete the venv folder again

[solid fossil]

how can i avoid this?

[ornate elk]

Edit the webui-user.bat (right click), At the line COMMANDLINE_ARGS= You add: --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling

[solid fossil]

will the quality suffer from this?

[ornate elk]

Nope

Its needed to get the best performance on amd directml

[solid fossil]

aight thanks

[lapis merlin]

yeah I installed following their instructions on the github page

it gets to 98%+ ram usage then says "Could not allocate tensor with 6553600 bytes. There is not enough GPU video memory available!"

[jade herald]

i try to connect this node to other and i dont know what node/nodes to use, is there any one here who now this? look at image

[ornate elk]

What's your GPU?

[lapis merlin]

6700xt and directml is only detecting 1gb when launching

i did the windows install instructions btw

iirc i have SDXL installed and it works fine but I want the ComfyUI node stuff

[thorny horizon]

im unable to start kohya, can anybody identify the error

[ornate elk]

Did you also this?:
https://github.com/comfyanonymous/ComfyUI#directml-amd-cards-on-windows

[lapis merlin]

yeah, I installed this way:

git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI
python -m venv venv
venv\Scripts\pip install torch-directml
venv\Scripts\python -m pip install --upgrade pip
venv\Scripts\pip install -r requirements.txt

then I launch ComfyUI with

cd ComfyUI
venv\Scripts\python main.py --directml --lowvram --use-split-cross-attention --auto-launch

[ebon horizon]

Make sure to activate the python venv, then run 'pip install bitsandbytes-windows'

[lapis merlin]

after it launches I click load defaults > select model > and hit queue then it will just keep using ram til it crashes

the crash is always either

"Torch not compiled with CUDA enabled"

or

""Could not allocate tensor with _____ bytes. There is not enough GPU video memory available!"

[hexed crest]

hey guys may i ask a question here

[ebon horizon]

Yes, we're here to help.

[hexed crest]

thanks, here goes: i want to ask if anyone knows what kind of problem i seem to be running into
i'm trying to do that anime photo to realistic photo conversion thing
and the results keep coming back very strangely shaded
like the controlnet lineart is keeping the shapes correct, but the colors are just not following the shapes at all and are just random

like this

my txt-image is generally fine but also has this problem if i ever dare to make the picture any size other than 512x512

so i'm wondering if this means some part of my installation is broken or if i forgot to install some vital component

[thorny horizon]

im still getting similar error, i activated vnev from scripts folder and i installed 'pip install bitsandbytes-windows' in cmd

is there any particular folder where i need to install 'pip install bitsandbytes-windows'

in git bash it says requirement already satisfied

[hexed crest]

any chance you might know what's causing this coloring issue...? thanks so much

[tame yoke]

what are some ways to speed up cpu generating

[ebon horizon]

ControlNet lineart focuses more on composition. Try using controlnet canny in combination with img2img and denoise 0.5 or lower.

[heady pine]

I’m having a problem when I try to remake images that I have made previously. If I make a new image, I’m able to remake an exact copy of it again, but upon closing Automatic1111, and re-opening it, the images are always different.

I reinstalled Automatic1111, but the same problem keeps happening.

Does anyone know what I could do to fix this problem?

[hexed crest]

thanks, i already solved the problem though: seems like the model i used (realistic vision) just can't color this for some reason, switched to other models and it works perfectly...no idea why that happened though

i'll try using canny to see if realistic vision can still be salvaged on my end... maybe the file is corrupted or something?

[solid fossil]

can i somehow make my stable diffusion darkmode

[ornate elk]

If your windows is set to dark mode the browser and webui will be too.
Or you can add --theme dark
To the webui-user.bat

[solid fossil]

aight thanks

[crimson leaf]

i did but still issuing, here is the log:
https://pastebin.com/YJ9gL1NX

Detected that PyTorch and torchvision were compiled with different CUDA versions. PyTorch has CUDA Version=11.7 and torchvision has CUDA Version=11.8. Please reinstall the torchvision that matches your PyTorch install.

[sand trail]

can i add aurgments to comfy ui ?

https://www.reddit.com/r/comfyui/comments/15jxydu/comfyui_command_line_arguments_informational/

[crimson leaf]

Also my pip cache currently:
`$ pip cache list
Cache contents:

• antlr4_python3_runtime-4.9.3-py3-none-any.whl (144 kB)
• ffmpy-0.3.1-py3-none-any.whl (5.6 kB)
• font_roboto-0.0.1-py3-none-any.whl (2.4 MB)
• jsonmerge-1.8.0-py3-none-any.whl (18 kB)
• lit-17.0.6-py3-none-any.whl (93 kB)
  `

[sand trail]

found this online but somehow i cant make it work

[crimson leaf]

meh forget it, now works by using v1.6.1 sd

dam, that's hella fast

5s for img generation

rx580 used like around a min or sth

[carmine forge]

does anyone know why queue is stuck on load checkpoint?

i selected the model

[ornate elk]

Can you show your whole txt2img settings?

[strange gust]

hello guys! i have a new laptop and im trying to run stable diffusion locally. i have a 8gb VRAM but im having the cuda out of memory error with 512x512 photos. it says that pytorch its taking 6gb. can somebody help me please!

[ornate elk]

Hey what's your gpu?

[strange gust]

NVIDIA DetForce RTX 4060 Laptop GPU

[ornate elk]

Okay, then you need to edit the webui-user.bat
At the line COMMANDLINE_ARGS=
you need to add:
--xformers --medvram-sdxl --no-half-vae

Then save and relaunch

[strange gust]

IT WOTKS, thank you so much!!!

[crimson leaf]

uh, anyone got a fix for PLMS distortion? in v1.6.1 it's totally messed up, but in v1.2.1 it worked fine (earlier used amd, since switched to new gpu, i can't go back to that version(torchvision issue), also it distorted on amd aswell after going to 1.6.0, so i could go back to 1.2.1 and had it working again).
Here is the distortion: (1.2.1 works fine, in 1.6.1 it's all messed up AGAIN)

I already asked once this issue, and no it's not the VAE, we tried out already so often

using Ubuntu 22.04.03, ryzen 5 2600/24GB ram/rtx 4070 12gb edition, issue only appears to be only on the newer version
Args: --xformers --no-half-vae

[ebon horizon]

If model/sampler/etc are all the same, this is likely your seed. Use the same seed with a non-converging sampler for reproducibility.

[crimson leaf]

My guess is the Fix faces option, as there is no option to on/off it in the newer UI

is it?

[ornate elk]

What's the size of the checkpoint you try to use?

[solid fossil]

the one i always used

around 3 gigs?

[ornate elk]

Okay thats fine, any extensions used?

Or what are your txt2img settings when you get that error

[solid fossil]

everything worked fine, but i just added some lora models

now it doesnt work anymore

should i just add the commandlines?

--no-half

[ornate elk]

What was your GPU again?

--no-half is needed for any rx5xxx series gpu

[solid fossil]

6700xt

its fine its working after restarting stable diffusion

[ornate elk]

Okay your card dont need --no-half
Only if you want to use inpaint or adetailer

But using --no-half will require so much vram that will run out of it fast

So I wouldn't recommend using it

[solid fossil]

okay

delete the whole thing?

[ornate elk]

--no-half-vae is needed

[solid fossil]

oh okay

so is there no way how to increase the resolution?

am i capped by my vram

[tame yoke]

my asuke is different and i wanna ask if theres like a setting that can make cpu make images like gpu

first image is how its supposed to be

3rd is mine

[ebon horizon]

What model series you using? Normal or XL? What resolution are you trying to generate?

[solid fossil]

1920x1080

normal

well lets say i tried to use 1920x1080

512x512 is the best working for me and my casual workspace atm

[ebon horizon]

1.5 and v2 models are trained on 512x512, larger sizes will be less performant

[solid fossil]

but to publish all those pictures i need better resolution than 512x512

like the quality isnt bad its just rly compressed

[ornate elk]

With a 6700xt you can do 512x768 upscaled by 2 with hires fix.
And you can do 540x960 upscaled by 2 but for that you need to install the tiled diffusion extension and then only enable tiled VAE.

[ebon horizon]

You can try generating a smaller image with the desired aspect ratio and then upscaling. For 16:9 a common starting size is 1024x576

[solid fossil]

maybe i should add that i installed stable diffusion 5 min ago lol

appreciate all the tips il look into it

[ebon horizon]

It's also worth trying out ComfyUI. I find it tends to be more optimized.

[solid fossil]

couldnt i just make like a img2img2 with 0% denoising strength and just change the resolution that way?

should take less ressources

[ornate elk]

xD sry but you asked haha
Stay at first to 512x768 for portraits. If you want to know how to upscale I can guide you

[solid fossil]

my focus are instagram posts

if my informations are correc it should use 1080x1080?

[ornate elk]

Idk instagrams resolution

[solid fossil]

yeah so could u help me upscaling without losing quality?

working with the done pic i just sent before

[ebon horizon]

Setting img2img with a larger output size than the input, in most UIs, will just do an upscale before encoding. It's doing the same steps on the backend.

[solid fossil]

ok found the solution in extras i think

it has its own upscaler

looking good

upscaled without any loss ig?

still kinda shocking if i get the reality checks that this is not a real person

and i just invented her basically

[solid fossil]

yeah didnt like the result

[ebon horizon]

0 denoise means the sampler isn't doing anything. I don't even think it would run with 0 denoise.

[solid fossil]

it gave me a output, just not sure if it changed the resolution

didnt check

[ornate elk]

Extras upscaler is a lazy upscaler that dont fix the image nor enhance it (it can enhance but not as much as hires fix or img2img upscale script)

[solid fossil]

would u guide me through it?

[ornate elk]

here is an example of upscaling with hires fix:

[solid fossil]

what is different with ur sampling method?

is it better than the default one

[ornate elk]

oh there are a few samplers, some are better than others. Euler a, dpm++ sde and 2m karras are the most used

[solid fossil]

and what exactly is happening here?

what are hires steps good for?

[ornate elk]

thats how often the upscaler goes to improve the image
10-15 is enough

[ebon horizon]

The thing with samplers, you want to understand which ones converge and which don't. Converging (like 2m) will generate mostly identical images with more steps, just with more detail. Non-converging (like SDE and Euler) tend to yield radically different images depending on step count

[solid fossil]

i see isee

but isnt this just enhancing the picture?

talking about the hires steps now

[ornate elk]

yes it enhances the image and makes it larger in resolution

[crimson leaf]

uh, is it common to get spammed by those errors while dpm adaptive generating?

[solid fossil]

can i do this with an already existing picture?

[ornate elk]

yes in img2img

[crimson leaf]

oh, gnome crashing, all fine

[ornate elk]

but if your image was created in txt2img you can also put it into png info and click to send to txt2img, then rerun it with hires fix

[solid fossil]

i dont see any hires steps option in img2img tho

[solid fossil]

ahh

[ornate elk]

its under scripts (sd upscale script) and its a bit more to explain

so first use hires fix

[solid fossil]

well the issue with that is im not getting the same picture

its just copying my prompts

[ebon horizon]

What's your denoise set to?

[solid fossil]

0.5

[ornate elk]

normaly it will set all the settings as used for that image

[solid fossil]

yeah but isnt it still a random ai picture?

more than close similar isnt possible

i mean its still generating but i think u can see that it wont be the same output

[ornate elk]

yep

so some setting is different

[azure magnet]

LMC works even on SD normal or only on XL?

[ornate elk]

you also can try use Esrgan4x as upscaler and set denois to 0.35

that shouldnt change the image to much

[solid fossil]

wow but i can see the quality difference

looks awesome

except the teeth..

like always

why is this a thing that all persons i create in stable diffusion have weird looking eyes, hands or teeth at the beginning?

barely generates any first try that look actually human

[ornate elk]

you need to use quality tags and good negative tags

[solid fossil]

aight

why is my masking black afterwards?

i thought im only redoing the teeth with this tool

[ornate elk]

like i said you would need to use --no-half for inpainting

[solid fossil]

ohhh

so i cant actually use it aslong as i dont add the command line

[ornate elk]

yes

[solid fossil]

but u also said u shouldnt add it since it takes too much ressources

[ornate elk]

true. some people only add it for inpainting and then remove it

[solid fossil]

mehh

[ornate elk]

i mean you can try

as for 512x512 images it will work

but it could go out of vram when using hires fix

[solid fossil]

yeah took me quite a while now to render this one with 10 steps

[ornate elk]

also a tip. whenever you get the out of vram error you need to restart SD

[solid fossil]

yeah found out

[ornate elk]

with the settings from above

[solid fossil]

dangg

which model u using?

[ornate elk]

that was with realisticvision v6

[solid fossil]

why does it has two options?

is one of them xl or smth

[ornate elk]

go for the smaller one

[solid fossil]

this looks so much better than epicphotogasm

[ornate elk]

most model creator prune their models to make them smaller. that wont affect the quality in a noticeable way.

[ebon horizon]

I ran your original through my HiRes workflow and was able to get a similar output. You may need to tweak the settings a bit for each image until you find what works

[solid fossil]

aight thanks

[solid fossil]

is the cfg in comfyui the same like in the normal stable diffusion?

im watching a video rn and the cfg beyond 1.2 isnt really realiable(in comfyui)

[solid fossil]

is there like a faceswap option in stable diffusion?

[desert garnet]

is there a version of illusion diffusion that can combine 2 images? I want to create a QR code with another image in it, without using a text prompt (preferably a huggingface space)

please ping me if you have what im looking for

[supple meteor]

Hallo, how to invite BOT into own server? Thanks

[ebon horizon]

cfg scale in Comfy should be identical to core SD

[thorny horizon]

i think i found my issue, its due to wrong version of bitsandbytes

[solid fossil]

can i use checkpoint merges as normal checkpoints?

[coarse parrot]

any english or turkish guide like this?

https://www.youtube.com/watch?v=eiMP7513UUE&t=113s

[ebon horizon]

For this you'll want sketch2img
https://www.youtube.com/watch?v=hEBVps8rnRk

[coarse parrot]

thx ❤️

[ebon horizon]

Yep, they're functionally the same as any other model. You can even merge them with other merged checkpoints to create a franken-model.

[misty sparrow]

I have automatic1111 stable diffusion webui setup and running in WSL. I am looking for a way to make it run using DirectML instead of using the CPU as I have an old AMD DX12 card. Can anyone point to a way I could do that? I was following this tutorial https://sakari.niittymaa.com/blog_install-stable-diffusion-web-ui-using-wsl-and-anaconda but didn't do the Cuda step.

[ornate elk]

Hey, what's your GPU?

[misty sparrow]

An old one, it's a r9 290X

But is DX12

[ornate elk]

Ohh I dont think that will work for SD

[misty sparrow]

Not at all? I was reading somewhere that DirectML does work for any DX12 card, but I'm no expert

[ornate elk]

You can try to install the directml fork of auto1111 but thats not done via anaconda or wsl

[misty sparrow]

I saw that one after I set this up. Maybe I can give it a try. I like the Anaconda and WSL setup though

[ornate elk]

@misty sparrow
You can also use my guide for amd GPUs here:

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

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

1. Make a new Folder on your drive (not on Desktop, downloads, documents, Programms, Onedrive) and name it Ai for example:
   C:\Ai\
2. You go into the folder you created in this case Ai, then click in the File Explorer bar (not searchbar) and type cmd then press enter.
3. Then you copy and paste this command:
   git clone https://github.com/lshqqytiger/stable-diffusion-webui-directml && cd stable-diffusion-webui-directml && git submodule init && git submodule update
4. Press enter and after its done you can close the cmd
5. Edit the webui-user.bat (right click), At the line COMMANDLINE_ARGS= You add: --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling
   After that save and launch the webui-user.bat inside the stable-diffusion-webui folder.
   After the Installation that can take a while. Youll get an URL http://127.0.0.1:7860
   Thats the webui you open in Browser.

[misty sparrow]

OK, perfect, thanks for your help, I will have a play around

[ornate elk]

No problem and good luck 🙂

[ornate elk]

That guide isn't for directml as the auto1111 webui dont support it. Only the fork of my guide and SD.next and comfyui support it.
If you want to use it on Linux your option is to use the Rocm driver but thats not supported for your GPU I think.

[misty sparrow]

I don't mind using Windows if there is a way, just this was the way I found that worked for CPU at least as I couldn't get the Windows version working. What's the difference if not webui? Sorry I'm now. I've only been using Roop via commandline

[karmic crown]

@ornate elk may you please add command line args to your nvidia tutorial with remarks about VRAM size 🙂

[ornate elk]

Yes, I will rewrite it soon 🙂

[karmic crown]

yesterday i said to someone to look at your guide, since i thought you have them there

and he told me "i have webui installed" 🙂

[ornate elk]

Auto1111 is the name of the webui. Its an easy to use interface to use Stable Diffusion.

[unique anchor]

I am using Adamw8bit, so that's the issue?

[ornate elk]

If you want to run it on CPU I can guide you too

[misty sparrow]

And the other option that might work is just a commandline version or something?

[ornate elk]

My guide is for the webui. There isn't a commandline version for the directml

But It has an API support

[misty sparrow]

Everything is working in webui on CPU so far already, thanks, but I was wondering iof I could get it to go faster. It takes like 5 mins per image on my i7 8700K

[ornate elk]

You can't do much about CPU speed.
Maybe use --medvram --opt-sub-quad-attention --upcast-sampling
In your webui-user.bat

[misty sparrow]

Yea, I meant, that's why I was looking into GPU acceleration to make it go faster. Thanks I will take a look at that option also

[ornate elk]

I have an old r9 280x around maybe I'll test it later if it works

[misty sparrow]

Thanks, don't put yourself out too much, but if you find anything let me know. I will continue playing around with your suggestions in the meantime. Thanks for you help

[ornate elk]

Yes i doubt i have time to test it today but ill let you know.
No problem 🙂

[misty sparrow]

No rush

[thorny horizon]

I trained my first lora using a sample of ~200 images of a face and 20 repeats with only 1 epoch (to test it out first), but its generating images like this as soon as i apply it, it doesn't matter what prompt i put

[unique anchor]

Anyone has a recommended config file for Kohya SS for Low End GPU like GTX 1650

I am getting this error now :/

RuntimeError: NaN detected in latents: C:\Users\-\Desktop\Output\img\40_comic page style\comicsteps (1).png

[lament tusk]

Hello, I am developing dreambooth model using 1 training image.
And I am not sure what is the proper hyperparameters to train lora model
I am currently using realistic_vision_5.1_no_vae as based model.
And lora rank is 4.
Please help me.
🙏

[small urchin]

hello im back, i was just wondering if a fix for the terminal and UI desync. like webui showing 99% and termal shows it as completet but nothing happens

[kindred cave]

Has anyone been having issues with the Custom Node "Reactor" in ComfyUI on RunPod?

It keeps crshing each time when I use it, iv updated all the nodes, updated comfy tired unistalling and installing it again via ComfyUi manager but still having the same issue. Log wise all I can see that thee crash occurs when reactor is analysing the source image.

2023-12-06T16:50:39.783388007Z [ReActor] 16:50:39 - STATUS - Working: source face index [0], target face index [0]
2023-12-06T16:50:39.787583130Z [ReActor] 16:50:39 - STATUS - Analyzing Source Image...

[limpid igloo]

Hello guys, anyone can help me to obtain prompt of people without face distorted? Is there a method to not obtain this distortion

[alpine lichen]

import torch
from diffusers import StableDiffusionXLPipeline

pipe = StableDiffusionXLPipeline.from_pretrained(
"./stable-diffusion-xl-refiner-1.0",
torch_dtype=torch.float16,
)
pipe = pipe.to("mps")

prompt = "a photo of an astronaut riding a horse on mars"
image = pipe(prompt).images[0]
print(image)
image.save("/tmp/output.png")

---

Loading pipeline components...: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 5/5 [00:16<00:00, 3.21s/it]
Traceback (most recent call last):
File "/Users/xxxxxx/xxxxxx/test.py", line 12, in <module>
image = pipe(prompt).images[0]
^^^^^^^^^^^^
File "/opt/homebrew/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/opt/homebrew/lib/python3.11/site-packages/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", line 1062, in call
add_time_ids = self._get_add_time_ids(
^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/homebrew/lib/python3.11/site-packages/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py", line 666, in _get_add_time_ids
raise ValueError(
ValueError: Model expects an added time embedding vector of length 2560, but a vector of 2816 was created. The model has an incorrect config. Please check unet.config.time_embedding_type and text_encoder_2.config.projection_dim.

---

I am new to sd and encountered this problem. Can anyone help solve it?

[karmic crown]

Python on Silicon Mac

[alpine lichen]

yes, sdxl img2img succeeded but sdxl text2img throws error

[karmic crown]

i have made a tutorial for automatic1111 (web ui for SD) for all macs, but i didn't play much with SD from python itself

try to use python 3.10 and nightly build of CPU torch

see what you can use from here https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/5461#discussioncomment-7738578

for start brew install python@3.10

make a new venv with python 3.10 instead of 3.11

[alpine lichen]

ok, i will try

[karmic crown]

pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cpu

this will install dev version of torch

btw

are you using Python directly because you want to use it like that, or did you didn't know that it is easier to use automatic1111?

if you just want to generate images, better install a1111, but if you actually want to use python directly, that is another topic 🙂

[alpine lichen]

my company asked me to study how to apply it to business, so I had to learn how to use python to call directly

[karmic crown]

may you please send pip3 freeze > req.txt

and which git repository you are using? huggingface?

[alpine lichen]

accelerate==0.25.0
certifi==2022.12.7
charset-normalizer==2.1.1
diffusers==0.24.0
filelock==3.9.0
fsspec==2023.12.1
huggingface-hub==0.19.4
idna==3.4
importlib-metadata==7.0.0
invisible-watermark==0.2.0
Jinja2==3.1.2
MarkupSafe==2.1.3
mpmath==1.2.1
networkx==3.0rc1
numpy==1.24.1
opencv-python==4.8.1.78
packaging==23.2
Pillow==9.3.0
psutil==5.9.6
PyWavelets==1.5.0
PyYAML==6.0.1
regex==2023.10.3
requests==2.28.1
safetensors==0.4.1
sympy==1.11.1
tokenizers==0.15.0
torch==2.2.0.dev20231207
torchaudio==2.2.0.dev20231207
torchvision==0.17.0.dev20231207
tqdm==4.66.1
transformers==4.35.2
typing_extensions==4.8.0
urllib3==1.26.13
zipp==3.17.0

https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-1.0

txt2img succeeded with webui, but still fails with direct python call

[alpine lichen]

oh, maybe i should use https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0 for txt2image instead of refiner

[karmic crown]

you can try this but do not try to use xformers - that is for nvidia

this might be interesting too (generation description from image)

i would also check this https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/API

you could have one pc with strong GPU, like nvidia 4xxx and then you can use that same instance over API from other machines

and i would probably use linux instead of windows in that case

response = requests.post(url=f'http://127.0.0.1:7860/sdapi/v1/txt2img', json=payload)

easy as that

[fading fable]

how do i fix error code 2

[karmic crown]

send full error, with more details

[fading fable]

ok wait

so do you know the fix ?

[narrow reef]

i'm not an expert but it looks like it failed to download the file, maybe spotty internet?

[karmic crown]

download failed

[fading fable]

what to do ?

[narrow reef]

close it and try again

[fading fable]

did that like 10 times already

[karmic crown]

do you have nvida or amd?

[fading fable]

nope some potato laptop

i mean its not that bad

[karmic crown]

wich gpu it has?

intel?

[fading fable]

let me chek

how do i check

?

[karmic crown]

https://www.howtogeek.com/414201/how-to-check-what-graphics-card-gpu-is-in-your-pc/

[fading fable]

amd ryzen 5 3500u

[karmic crown]

that should be Radeon RX than

[fading fable]

yeah

amd radeon

[karmic crown]

check this #🤝｜tech-support message

but i am not sure does it work for your radeon

at least python git and the basic install is fine

[fading fable]

ok

[karmic crown]

you can try

i use pc like 3 times a year 🙂

i am using macs and linux servers 🙂

[fading fable]

i mean you guys get it easy in the us like you all have mega pcs

[karmic crown]

let me know if that doesn't work, since i am not sure if ml version cam work on that hardware

you will just have to remove 1 folder and to redo some steps if it doesnt work

[fading fable]

ok

[narrow reef]

not sure where/when the problem cropped up, but at some point during my training and merging I lost the ability to save as safetensors, I get an error about a non contiguous tensor.. I haven't been able to find any way to fix it, hoping someone here has an idea

[fading fable]

one more thing by the way, can i convert a safetensor file to a ckpt?

[karmic crown]

why would you do that?

[fading fable]

nevermind

[karmic crown]

ckpt can contain malicious code, sefetensor can't

[fading fable]

how do i upgrade pip ?

[ebon horizon]

pip install --upgrade pip

[karmic crown]

but... wait

you need to activate venv first than

which one do you want to upgrade? global one or the one in venv?

[fading fable]

ok i did myself i think

[karmic crown]

anything about v20 is still good

do not worry too much about it

[fading fable]

can i use safetensor as a model ?

[ebon horizon]

Yes, Safetensor is the recommended format for models

[fading fable]

then how do i use it i dont know how to use it

[ebon horizon]

Are you using a web gui? There should be a folder where you place it to load it into the ui

[fading fable]

yeah im using a web gui

what folder is that ?

[ebon horizon]

What UI are you using?

[fading fable]

web ui

[ebon horizon]

For A1111 they go in .\webui\models\Stable-diffusion
For ComfyUI they go in .\models\checkpoints

[fading fable]

ok tnx

ok what is this now

[ebon horizon]

Git failed to checkout, unexpected disconnect. Check the stability of your internet connection.

Likely a blip with your ISP connection

[fading fable]

ok

[fading fable]

so i have to put the safetensor in the models\ stable diffusion file ?

[misty sparrow]

When using Automatic1111 webui in CPU mode, how do I make it use the full CPU? When running, my CPU only goes up to like 70% Is there a way to increase the worker threads?

[karmic crown]

inside of the webui folder (where you installed it), you have models folder, and there is a folder called Stable-diffusion inside, put your models inside

[karmic crown]

what is your command line?

[fading fable]

ok got it

[karmic crown]

"Put Stable Diffusion checkpoints here"

[misty sparrow]

Just the default

[karmic crown]

do you have gpu, is that a laptop?

[misty sparrow]

I have a GPU, but it's an old AMD R9 290X and was told it wont work, so just stuck with CPU mode

[karmic crown]

try this

--skip-torch-cuda-test --upcast-sampling --no-half --use-cpu all --opt-sub-quad-attention

[fading fable]

is this a network issue ?

[karmic crown]

i am using that on my linux VPS

[karmic crown]

yes

[misty sparrow]

Oh yea, shit I did have --skip-torch-cuda-test --no-half already just not in the config. Thanks, I will give that a try

[karmic crown]

how long it takes for you?

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Size: 512x512
Time taken: 2 min. 53.9 sec.

on my linux VSP

[misty sparrow]

6mins59secs before adding you commands. I'm just doing another one now with your commands now, but still only using 63% CPU

I have i7 8700K

[karmic crown]

#🤝｜tech-support message

as you can see it is in range from 60 to 90

[misty sparrow]

OK, so that is normal I guess. I was hoping I could push it harder to make it go faster

Just seems like it isn't using 12 threads

When I run Roop. It uses all 12 threads and CPU is 100%

[karmic crown]

as a1111 plugin? or some other way?

[misty sparrow]

Standalone

Both installed in WSL

6mins15seconmds this time with your commands, so a little faster

[karmic crown]

i found this

:: Boost thread priority
SET command=<program.exe> <options>
start "" /REALTIME /B /W %command%

you can try like this

SET command=webui-user.bat
start "" /REALTIME /B /W %command%

in a new bat file in same folder

hmm, will it work? or cant use bat like that, and need to call it... i forgot 😦

[misty sparrow]

Not too sure, I'm very new lol

[karmic crown]

i am using mac 90% of the time, the other 10% are linux servers

[misty sparrow]

I'm OK at coding, but not done much ML

My screenshot of the config earlier was the BAT file, but as I'm using WSL I think it's using the webui-user.sh not the bat I made a mistake, I got confused becasue I was trying to set it up in Windows without WSL but couldn;t get it working

[fading fable]

what is this error?

[karmic crown]

wait, wait... do not use wsl

you are adding another translation layer without reason

[fading fable]

how do i fix this

[misty sparrow]

I couldn't get it working in Windows natively, I was getting lots of errors, so I just did it in WSL as that was the only way I could get it working

[karmic crown]

#🤝｜tech-support message

follow this guide

when you make it run, let me know, so i can tell you what to do next

[misty sparrow]

OK, thanks, I will give that a try

[fading fable]

can anyone help me with this

[sturdy cedar]

Say my original video file I'm feeding into AnimateDiff has greenscreen with one subject. Is there a node or workflow that can mask the subject in the scene from the green screen throughout the sequence? Any tutorials on this anyone recommends?

[karmic crown]

it looks like you have network issues, so the installation fails in different stages of installation

[fading fable]

oh ok

[karmic crown]

try with cable instead of wifi

[fading fable]

ok

[karmic crown]

remove venv folder

it will be created again when you run webui

[fading fable]

me?

[karmic crown]

yes

[sharp lichen]

hello! What is the best way to outpaint in auto?

[fading fable]

can i try using a different network ?

[karmic crown]

as long as it works better than your current one

[fading fable]

ok

how long does the {installing requirements} take?

is this a network issue ?

[ebon horizon]

Yes, this indicates you're experiencing HTTPS timeouts. You may want to try installing from another network; once all the components are installed you can run it completely offline if needed

[fading fable]

ok

so how long does installing the requirements take ?

[tranquil nacelle]

if u alrealy have the stable model it will take like 10 min but if u dont have u have to download 4 gb file that take long depends on your internet i suggest if u have downloaded the app ones go to models and copy it to a folder in that case when u want to download it again just copy it to the models folder to not download that 4gb again and again or download it from hugging face

[ornate elk]

Make sure to whitelist the webui in any browser adblocker

[crimson leaf]

tf is this?

[ornate elk]

Thats from the dreambooth extension

[wet token]

It's for training models

[crimson leaf]

how did i got it? Was it bc i loaded the sdxl base model?

[rich lynx]

hello ppls. i hope i am in the right place. I just joined the discord channel. I've been using stable diffusion for a couple months and as of 2 weeks I can't seem to get it to work anymore.

I keep getting this error, RuntimeError: Failed to import transformers.modeling_utils because of the following error (look up to see its traceback):

    CUDA Setup failed despite GPU being available. Please run the following command to get more information:

    python -m bitsandbytes

    Inspect the output of the command and see if you can locate CUDA libraries. You might need to add them
    to your LD_LIBRARY_PATH.

i tried to get bard the ai to help me and i just dont seem to understand. I can't find any file named libcudart.so . i was using stable diffusion before i even had the CUDA toolkit which i have downloaded recently. anyyways, if that makes sense to anyone and you are free, I could use some help. Thank you.

[ornate elk]

no its because you installed the dreambooth extension

[ornate elk]

hey, whats your gpu?
i suggest you update your graphic card drivers and then delete the venv folder and relaunch the webui-user.bat

[rich lynx]

i updated the graphics card through the nvidia experience recently and i think thats when things messed up. otherwise i was just learning to use animatediff in stable diffusion. im running a rtx 4070

ive deleted the venv folder a couple of times and ran webui-user.bat and the same problem comes up. ive even reinstalled my Unreal Engine in the epic launcher. I'm honestly completely lost in this prob.

[ornate elk]

whats your webui version?

[rich lynx]

i dont know how to check that (>_<)

im new to alot of this, sorry