🤝｜tech-support

[rigid warren]

Using directml webui and generating blurry images in img2img, turned off controlnet and the problem continues even when changing models.

[upbeat crane]

maybe try taking any loras out of the pos prompt, sure you've played with your denoising strength

[scenic saddle]

Thank you so much G

[rugged latch]

get stable :3

[fossil dew]

ty

i still gotta get the others to work first

i havent tried your advice quite yet

[upbeat crane]

you wont need that other extension if you use this one

[fossil dew]

is it possible theres a file in those folders i need to pull out?

oh cuz im tryina do 3

animateddifff, controlnet and wildcards

[upbeat crane]

you could just remove the 3rd one and install dynamic prompts, then still end up with working wildcards

[fossil dew]

🙂

[upbeat crane]

heres some good info on it, if you are interested

https://youtu.be/bQK5diN59NA?si=6i3endM2Lj1a4qA-

[fossil dew]

ok so I guess all I leanred is that wildcards arent what Im looking for 🤡 oopsie

I'm looking for

• ControlNet
• Whatever it is that people were using to make gifs (not deforum)

[upbeat crane]

ahh lol

gifs created from what exactly?

giffusion maybe?

you can also use SVD (stable video diffusion) to output as gifs #▶｜stable-video-diffusion

good luck

i also see that you got so far as adding ANIMATEDIFF you can use that as well

type "how to make gifs with animateDiff" into YT

[fossil dew]

ok well- I got ControlNet to work 🤷‍♂️

doing the instal by URL thingy so 🤷‍♂️ thats cool

[fossil dew]

ill check this out

ty

[upbeat crane]

you can always search youtube for just about anything you are trying to accomplish with SD, there is a tutorial for almost everything. np, yw

[ornate elk]

Animator, or Animatediff maybe

[ornate elk]

Is your webui updated?
What's your GPU?
And what's inside your webui-user.bat?
Do you tried to Inpaint maybe?

[zenith widget]

Good morning, is this warning something to worry about or can i ignore it? ive tried to install Onnxruntime and Onnxruntime-gpu and neither work. i ended up deleting the dwpose.py file and cant get it back 😅 if you have an idea on how to switch to opencv please let me know ❤️

edit: i found the dwpose.py file, added it back and still have the warning

D:\Apps\Stable Diffusion Web UI\ComfyUI_windows_portable\ComfyUI\custom_nodes\comfyui_controlnet_aux\node_wrappers\dwpose.py:26: UserWarning: DWPose: Onnxruntime not found or doesn't come with acceleration providers, switch to OpenCV with CPU device. DWPose might run very slowly
warnings.warn("DWPose: Onnxruntime not found or doesn't come with acceleration providers, switch to OpenCV with CPU device. DWPose might run very slowly")

[sly sparrow]

Do you guys feel any difference on using Comfyui or automatic 1111 on pinokio or external browser? for me automatic 1111 is way slower...

[valid gorge]

Placed an oder for PC parts, wish me luck lol. Haven't almost never used anything other than a Mac and this will be a major thing for me. Going to also install Linux (Ubuntu? maybe) alonside Windows11

Was able to get a "new" 3090 (old stock with full warranty)

[patent meadow]

any one knows what this means?

im trying to use reactor for face swap

[ornate elk]

Hey did it worked before?

[lost crescent]

Github is down??? At least here in London UK ...

cd..

OK - Github is back online!

[lost crescent]

How do I CLEAR CACHE in ComfyUI? - Okey Dokey

[dense sandal]

do you know what this is?

[sweet moat]

@ornate elk the image reward thing is not working for me, can you check if its working for you?
i can see the script but the score is not being shown
https://github.com/THUDM/ImageReward?tab=readme-ov-file#integration-into-stable-diffusion-web-ui

[ornate elk]

Nope, doesn't work. While loading it gets an error in cmd

[upbeat atlas]

Hi, I'm using stable diffusion directml on AMD windows, it was working fine but now I'm getting this error called

ModuleNotFoundError: no module 'optimum'

How can I fix this?

[sweet moat]

run pip install optimum

[upbeat atlas]

I did run it on cmd, it still gave me the same error

[sweet moat]

send the screenshot

[ornate elk]

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

[upbeat atlas]

Oki ima try it rn

[upbeat atlas]

Hey it worked, thanks a lot ❤️

[ornate elk]

no problem 🙂

[tepid wolf]

first try sd-webui-forge and give me this error, how to fix

[crimson rover]

Hey there, any idea why i get those weird generations and then it became completely gray?

Iam very new into this, is it because i dont have VAE file?

[nocturne topaz]

Nope - thats usually what you get when mixing architectures (it you have a sdxl model and a 1.5 lora)

[crimson rover]

how do i fix it? i just downloaded this and put 1 chckp. in it 😄

that's all

[nocturne topaz]

Are you trying to use it in cpu only mode?

[crimson rover]

oooh

i should probably use the command

--precision-- something

before it worked when i used it

[ornate elk]

What's your GPU?

[nocturne topaz]

Forge got rid of all the precision commands

Check the github page for full details on that

[crimson rover]

i5-10600

[ornate elk]

Thats a CPU

[crimson rover]

oh

1s

RTX 2070 Super

nvidia

[crimson rover]

iam not sure if i know how to work with it xd iam kinda newbie in this

also not native en speaker

[ornate elk]

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

[crimson rover]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=
git pull
python -m ensurepip

call webui.bat

before i had something after the "args"

and it worked

[ornate elk]

Add --xformers --medvram-sdxl --no-half-vae
After the args

[crimson rover]

ok

with space or w/o?

[nocturne topaz]

Its forge, not a1111

[crimson rover]

set COMMANDLINE_ARGS= --xformers --medvram-sdxl --no-half-vae

[nocturne topaz]

The no half args are ignored

[crimson rover]

how should i type it?

[nocturne topaz]

—xformers is a go, but others nope

[crimson rover]

not the medvram?

[tepid wolf]

I guess no, Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: f0.0.9-latest-51-g5bea443d
Commit hash: 5bea443d94f3a85f819cb8541c1bba0aac208d83
Launching Web UI with arguments:
Total VRAM 6144 MB, total RAM 16115 MB
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 3060 Laptop GPU : native
VAE dtype: torch.bfloat16
Using pytorch cross attention
ControlNet preprocessor location: D:\webui_forge_cu121_torch21\webui\models\ControlNetPreprocessor
Tag Autocomplete: Could not locate model-keyword extension, Lora trigger word completion will be limited to those added through the extra networks menu.
[-] ADetailer initialized. version: 24.1.2, num models: 9
Checkpoint realisticVisionV51_v51VAE.safetensors not found; loading fallback animagine-xl-3.0.safetensors
Calculating sha256 for D:\webui_forge_cu121_torch21\webui\models\Stable-diffusion\animagine-xl-3.0.safetensors: 2024-02-13 18:22:10,591 - ControlNet - INFO - ControlNet UI callback registered.
Running on local URL: http://127.0.0.1:7860

[ornate elk]

Oh he didn't mentioned it

[crimson rover]

i dont even know what forge or a1111 is xd

i just downloaded the diffusion and run it

[ornate elk]

Is your webui folder called
Stable-diffusion-webui ?

[crimson rover]

with some checkpoint i got from civic

stable-diffusion-webui

like this

yes

[ornate elk]

Okay yea then add the stuff I said

[crimson rover]

with space or nope

[ornate elk]

Also how large is the model file

With space

[crimson rover]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --xformers --medvram-sdxl --no-half-vae
git pull
python -m ensurepip

call webui.bat

kay lemme try

then it goes gray again

thats the whole page if that helps you anyhow

or installing the forge version instead might help?

idk whats the difference

i put this after args: --precision full --no-half --skip-torch-cuda-test

and it works now

[ornate elk]

no dont use them

that makes it run on your CPU

[crimson rover]

ye

but otherwise it doesnt work

[ornate elk]

the problem is that your mostly using a wrong file

[crimson rover]

xd

[ornate elk]

you didnt told me the model size

[crimson rover]

what model size?

u mean the one i downloaded?

[ornate elk]

yes

[crimson rover]

one was 5gb, other was 36mb, i was testing the 36mb one

[ornate elk]

yes thats the problem, the 36mb one isnt a model 🙂

thats a Lora

[crimson rover]

come to the voice xD

[ornate elk]

models are always larger than 2gb

[crimson rover]

u can tell me how to do it

[vocal burrow]

LoRa/Lycoris quick guide :
1/ Where do I put my LoRa : stable-diffusion-webui\models\Lora, Loras are NOT stable-diffusion model but rather patches for said models.
2/ How do I use my LoRa, you can either :

• Invoke it directly in your prompt by using the following syntax <lora:filename:weight> (or <lora:alias:weight> depending of the setting in setting->extra network). eg : <lora:animeTarotCardArtStyleLora_v20Offset:1> for using the Lora animeTarotCardArtStyleLora_v20Offset.safetensors
• Use the UI "Extra Networks" / card system by doing as described in my screenshot. Clicking an associated lora card should automatically insert the lora prompt liked mentioned above. https://cdn.discordapp.com/attachments/1002602742667280404/1191779742534348870/loraHowTo17.jpg
  3/ Be careful, LoRa are trained specifically for some SD version. Lora trained for SD1.5 (and derivatives) models will not load on SD2 models and vice versa (same goes for SDXL). Also even though Lora trained on SD1.5 models should work for every SD1.5 derivatives, they'll probably work best with the model that has been used for its training. Therefore, (by default) only Loras compatible with the model you currently have loaded will show up.
  4/ LoRa, LoHa, LoCon, what are these ? They're all commonly refered as "Lora" but differ slightly technically :
• LoHa : Lora with hadamard decomposition
• LoCon : Lora extended to redisual blocks.
• More explanations : https://stable-diffusion-art.com/lora/, https://www.sdcompendium.com/doku.php?id=lora_0014, https://www.sdcompendium.com/doku.php?id=locon_training_0071, etc

There you go

@crimson rover

[crimson rover]

yo

Aye thanks, ill check it, @ornate elk gave me a good explanation too

[true sphinx]

hey guys im getting an error upon launch and have 0 idea how to fix

[vocal burrow]

you're using python 3.10.3, it's not listed as compatible. Compatible python are [3.10.6 -> 3.10.13]

update python, delete the venv folder and then try again

[true sphinx]

ahh ok

[vocal burrow]

and please don't use windows store for that

[true sphinx]

no never use windows store

[vocal burrow]

that's the install folder of windows store tho

soooo

[crimson rover]

can you insert images in diffusion and work on them there?

[vocal burrow]

yes

[crimson rover]

how can you do that?

[vocal burrow]

Usually it's called img2img. But it can vary depending of what client/program you're using.

[crimson rover]

iam using the 1111 one diffusion

[vocal burrow]

img2img then

[crimson rover]

okay thanks

[true sphinx]

where can i find python 3.10.13 when i go to the official site it says no files for this version

[vocal burrow]

binaries are only available for windows up to 3.10.11

[crimson rover]

how to apply workflow?

usually i put the file into some folder but idk into which folder put a "workflows" file

[vocal burrow]

what is that "workflow" file you're talking about ?

[crimson rover]

it should change apperance

i have it as .json file

[vocal burrow]

where did you get it ?

[crimson rover]

civitai

[vocal burrow]

that's quite vague....

share the url

[crimson rover]

ill send u in pm

[vocal burrow]

bingo

it's all explained in its readme

that's a file for comfyui

[crimson rover]

i saw that

iam not sure if i know how to go step by step haha, so i have to install those 3 things

they are separated programs or i put those in diffusion folder?

[vocal burrow]

so far you're using auto1111-s stable-diffusion-webui

[crimson rover]

yes

[vocal burrow]

That's not comfyUI

[crimson rover]

oh

[vocal burrow]

it's two different programs

(both using stable diffusion)

[crimson rover]

is there a difference between those 2 programs?

like is it better to have that or that

[vocal burrow]

different workflow and UI

[crimson rover]

i mean, which is better xD

if i should just completely reinstall it

or just use the 1111

[vocal burrow]

There's no "better" tool just like there's no "better" car

it depends of your needs.

[crimson rover]

seems like my needs require those steps haha

[vocal burrow]

auto1111 is probably easier to understand.
comfyui is more for people who are ready to dig their hands in the dirt and do things themselves, experiment.

[crimson rover]

i see and is there something similar, not that hard to use as what i've sent you?

u can send me in pm if there is

[vocal burrow]

auto1111 has the biggesst community (thus more support, extensions, etc) but ComfyUI is close second.

[crimson rover]

then ill prob. stay at that xd, thanks for help

[feral mango]

hi how i restore comfyui to default again? loike when you delete the env folder in automatic1111?

[crimson rover]

is it normal that nothing is loading up? 😄 is the program laggy or whats happenin, this is loading for like 10 minutes+

i though its based on my net and pc speed

generating image doesnt even show any progress

odd, reloaded it with the blue button, changed checkpoint and it works now, probably bugged p_p

[true sphinx]

I updated my python to 3.10.11 and im still getting this issue

[sinful gate]

hey why is the training script for Cascade just idoling with no errors?

been like this for 15 min plus

[ornate elk]

Wrong Python

You need to install the normal python. Not the one from Windows Store

[true sphinx]

i didnt install from windows store went to the python site

this is where i went

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

[ornate elk]

Yea but it uses the one from Windows store

You should check under Apps if you can uninstall the other one

[true sphinx]

weird

[fair oxide]

this problem was never really fixed, pls help 💀

[loud gulch]

Probably been asked a thousand times now but is there a way to change the output location in Comfyui? I'd like to move all of my outputs to a Working Project Drive instead of them eating up my main drive.

[true sphinx]

ok so now i get this

even tho i have this

[ornate elk]

Oh you installed the webui the wrong way, by downloading it as zip

[true sphinx]

allright so do i need to change the set up

[ornate elk]

Yes you should follow my install guide. Its in the pinned messages of this channel

[ornate elk]

Can you show me a screenshot of your extensions?

[fair oxide]

ye

[true sphinx]

if i do the git clone will it wipe my loras models and so on

[ornate elk]

Make a new folder where you git clone in

Dont git clone in your existing folder

[ornate elk]

Looks normal rn.
What's your webui version?

[fair oxide]

1.7

[ornate elk]

Okay, make also sure the webui is whitelisted in any browser adblocker

[fair oxide]

hmmm alright one sec

ngl dawg, just realized i dont use adblocker LOL

[fair oxide]

should be good

[ornate elk]

Okay then next step is to disable all Extensions you installed. Then hit apply and restart

[fair oxide]

im just raw dawging the internet, no worries 💀

[fair oxide]

hmmm alright

[ornate elk]

Can recommend Ublock Origin

[true sphinx]

allright so i have git cloned to a new folder migrated my existing models and extensions ove

r

[fair oxide]

brb gonna download an adblocker and disable extensions LOL

[noble belfry]

adblocks are probably more active in stopping malware than antivirus software. i think this has been known for a while

stopping i say, not quaranteening

[ornate elk]

True. Even the FBI recommends using adblocker xD

[fair oxide]

didnt work

[ornate elk]

They don't show up in the folders ?

Make also sure you have an 1.5 model selectet

[noble belfry]

i assume its download an extension and stop extensions

[fair oxide]

ah

[old bane]

hay can someone explain to me how to fix this?

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.

[fair oxide]

that might be myissue

how to i do that

[ornate elk]

Just switch the checkpoint/model

Loras are made for 1.5 or sdxl. But they only show up when the model works with them (is 1.5 or sdxl)

[fair oxide]

explain like im dumb. Im dumb

[ornate elk]

1.5 loras won't show up when using sdxl models

Sdxl loras won't show up when using 1.5 models

[fair oxide]

so i should switch my a1111 to v1.5??

[ornate elk]

No

Auto1111 is a Programm

1.7.0 is the version of it

SD 1.5 is a model

That you use with auto1111

[fair oxide]

models.....

[ornate elk]

Checkpoints

[fair oxide]

oh

[ornate elk]

Same thing

[fair oxide]

i might be too high for this right now 🙏

[ornate elk]

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

[fair oxide]

im still very confused as to what i have to do LOL

[ornate elk]

You could show me the lora that won't show up and the checkpoint your using rn

[fair oxide]

thats the weird thing

its random

it changes every time i move them??

[ornate elk]

??

Move what xD

[fair oxide]

i have 3 folders in my lora folder, right??

each for like character loras

style loras

all that

when i move stuff there and refresh in a1111 it shows random ones

nothing in common

[old bane]

nvida and i dont know what you mean by inside the webui.bat

[fair oxide]

i move them back into the normal lora folder and everything is fine

but then i put the BACK into the, lets say character folder, now different ones are gone than last time

[ornate elk]

I mean the exact GPU name. And the content of the webui-user.bat file when you right click it

[fair oxide]

idk what to do dawg, tech hates me 😭

[ornate elk]

Probably some loras are for 1.5 and some are sdxl and it only shows the compatible ones

[noble belfry]

ppl dont know you can edit bats and bats are instruction notes and not programs

[fair oxide]

i make sure not to download the sdxl ones

i use civitai and it usually shows what loras are sdxl

all of them WORK

[ornate elk]

Yes exactly

[fair oxide]

they just vanish but only when put in subfolders 😭

[noble belfry]

put them in the stable-diffusion subfolder of models

[ornate elk]

Can you show me screenshots of this behaviour?

[fair oxide]

if i take them out of the subfolder they stop vanishing but like, thats the only way i know how to sort using this

[noble belfry]

if theyre models and not checkpoints

[fair oxide]

yeah one sec

[old bane]

where would i look at my gpu?

[ornate elk]

Task manager, performance tab, GPU

[noble belfry]

if youre windows, right click the start button/windows icon, system

[fair oxide]

alright so this is the very top of my loras in a1111 wthout me touching anything, im going to find the 2B loras and put them in my characters folder

[noble belfry]

ugh

[old bane]

also did you mean like go propertys of the file?

[noble belfry]

im not helping this

[old bane]

NVIDIA GeForce GTX 1650

[ornate elk]

Right click the webui-user.bat and click edit

[fair oxide]

alright, so ive moved all the 2b loras including the preview images to the character folder. Ive refreshed my a1111

everything actually seems to be in order this time except the preview images arent showing up on some

[ornate elk]

So it mostly works

Thats good

@old bane let me know when you have the file open, you need to do some changes in there

[fair oxide]

indeed, we're getting places

im gonna try to move some more stuff and see what that does

[ornate elk]

Okay, at the line COMMANDLINE_ARGS=
you add: --xformers --medvram --no-half

Then save. And relaunch the webui-user.bat
Then the error shouldn't appear again.

[true sphinx]

no difference

[ornate elk]

Open up a cmd and type
Where python

Oh wait it looks better than before

[ornate elk]

The issue is just that you some embeddings and hypernetworks that can't be loaded and giving errors

[true sphinx]

im lookjing and the json error

and that is odd why cant they be loaded

[ornate elk]

Idk

But I would try to remove all embeddings and hypernetworks mentioned in the errors

Maybe the files are corrupt, or really old, or falsely trained

@true sphinx you should check if they are Loras instead

If loras are placed in the wrong folder these errors are normal
Compare the file sizes in the folders

[fair oxide]

i decided to just delete them all and redownload them and hope that works at least for a little bit

they should make a way to tage your loras and sort by those tags

[true sphinx]

Could it be cuz i have some that are still lohas

[ornate elk]

They also go into models/lora

They are supported by default

The error is mostly because you put some loras into Embeddings and hypernetworks folder

So best is to compare the sizes

And everything thats not like the others isnt in the right place

For example, no embedding is 300mb

Or not even 100mb

[true sphinx]

ahhh

[ornate elk]

Im off now

[true sphinx]

cool ill chekck it

[old bane]

thanks it works now

[ornate elk]

Perfect no problem 🙂

[rigid warren]

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --precision full --no-half --no-half-vae --opt-sub-quad-attention --opt-split-attention-v1 --disable-nan-check --autolaunch --use-directml
git pull
call webui.bat

[dark viper]

---

TypeError Traceback (most recent call last)
Cell In[22], line 3
1 # SETUP MODELS & DATA
2 extras = core.setup_extras_pre()
----> 3 models = core.setup_models(extras)
4 models.generator.eval().requires_grad_(False)
5 print("STAGE C READY")

File c:\Users\PEY3C\StableDiffusion\StableCascade-master\train\train_c.py:128, in WurstCore.setup_models(self, extras)
126 effnet = EfficientNetEncoder()
127 effnet_checkpoint = load_or_fail(self.config.effnet_checkpoint_path)
--> 128 effnet.load_state_dict(effnet_checkpoint if 'state_dict' not in effnet_checkpoint else effnet_checkpoint['state_dict'])
129 effnet.eval().requires_grad_(False).to(self.device)
130 del effnet_checkpoint

TypeError: argument of type 'NoneType' is not iterable

Stable Cascade error when running Load Extras & Model tab

[nocturne topaz]

The models are all downloaded to the correct location??

[rigid warren]

nvm i got it, models were not the correct file size.

[dark viper]

all models are downloaded in \models\

https://github.com/Stability-AI/StableCascade/issues/11

nvm

the 11st and 6th Issue imply that you need to run the download_models.sh script in the models folder.

[nocturne topaz]

Poor colab setup then

[dark viper]

it is a local setup

this wasting me 1 hour in research

Oh if anyone encountered the same issues as mine

go into the .yaml, change the path to your proper file name. Turn the dtype to either float32 or bfloat16.

change the model version from 3.6B and 3B into 1B

my setup is big B small C

[nocturne topaz]

Thanks for the tips

[fair oxide]

@ornate elk have you read about this https://github.com/lllyasviel/stable-diffusion-webui-forge

[nocturne topaz]

I'm pretty sure he has, I have it running as my main now and the improvement is remarkable

[fair oxide]

amazing, i'm looking at a video on it

[nocturne topaz]

there are odd support issues (like you need a different animatediff extension, and supermerger and deforum don't work at all)

but I got like 70%+ more it/s out of it (like went from 7.7 to 12.8 on a basic 512x512 euler a render test)

[fair oxide]

yah thats what caught my interest about the article, btw does this work with controlnet?

[nocturne topaz]

absolutely.. controlnet is now built in to the system, as is kohyas hires fix and several other cool tools

[fair oxide]

sweet

[nocturne topaz]

And there are tabs across the top for Zero123 and SVD

[fair oxide]

so im curious, did you download the package or did you manually git pull?

[nocturne topaz]

I wiped and started clean

[fair oxide]

see i alraedy have a1111 installed with all other stuff

[nocturne topaz]

so git clone

[fair oxide]

oooh, why wipe ?

[nocturne topaz]

I backed up my folders first then copied them back into place

limited storage on the paperspace account 😛

[fair oxide]

oh

i think ill take the manual approachj with git

[nocturne topaz]

that, and at least at that point you know there isn't a hangover issue

ie something from the old one interfering

[fair oxide]

ok, this looks about right https://github.com/continue-revolution/sd-webui-animatediff/blob/forge/master/docs/how-to-use.md#you-have-a1111-and-you-know-git

[nocturne topaz]

yep 🙂

[fair oxide]

or maybe i could just run git clone https://github.com/lllyasviel/stable-diffusion-webui-forge.git and take it from there into editing webui-user.bat

that way i don't have to change anything to do with python and git

[nocturne topaz]

you could git clone it into a new folder, then use commandline args like --ckpt-dir to redirect to the folders used for your a1111 install

[lost crescent]

I am attempting to install torch 2.2.0+cu118 - this is the Pytorch.org text

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

Then when 2.2.0 install is complete - for some unknown reason - a 2.1.0+cu118 install automatically commences!!!!!

Xformers needs the 2.2.0 version, but it defeats me where this unwanted install originates?

[nocturne topaz]

no idea.. what extensions could be trying to load it?

[ornate elk]

Yes I did, but didn't installed it on my nvidia system so far.
On my amd system I got errors

Git clone into a new folder should be fine.

[fair oxide]

i got it setup and running now into a new folder using git clone but there is one warning and error i get ...
| WARNING:xformers:A matching Triton is not available, some optimizations will not be enabled.
| Error caught was: No module named 'triton'

[nocturne topaz]

did you manage to get zluda to go?

[nocturne topaz]

go into the venv and pip install triton - odd that it didn't come through with torch tho

[fair oxide]

do i need to run activate before intalling ?

[lost crescent]

Help, torch problem! 🙂

[nocturne topaz]

quick, use pitchfork instead...

oh wait, this aint a riot 😛

whats the error?

[lost crescent]

#🤝｜tech-support message

I install 2.2.0+cu118 - when that has installed - for some unknown reason 2.1.0+cu118 overwites it automatically!!!

[nocturne topaz]

I asked right under it what extensions did you just load?

or custom nodes if you are a comfy user

there will be something that is requiring the lower version causes the reinstall to happen

[lost crescent]

I have not just d/loaded any extensions

OK, but can I not suppress that? I need 2.2.0 for Xformers

[nocturne topaz]

in that case itll be torchaudio or torchvision requiring the lower version

[lost crescent]

So upgrade them?

[nocturne topaz]

xformers should be able to work on everything from torch 1.7 up

it should figure out which torch you have a build a whl for it

[lost crescent]

I get a message at startup

[nocturne topaz]

well if it was a fresh install those will be the latest versions of them and there isn't compatibility, but that seems odd

just go into the venv (using activate) and then pip3 install --upgrade torch --index-url https://download.pytorch.org/whl/cu118

[lost crescent]

Yes, normally there would be a message to confirm what extenson required 2.1.2 - but strangely it installs 2.2.0 then immediately overwites with 2.1.2 and no explanation as to what extenson requires it?!?!?

[nocturne topaz]

manual override then

[lost crescent]

I am using the pytorch.org page and this pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

So I hope not to get the message that Torch has Not Been Compiled With Cuda?

OK, I will try this pip3 install --upgrade torch --index-url https://download.pytorch.org/whl/cu118

But as you can see, I have exactly the same URL at pytorch.org

[nocturne topaz]

yep.. money on the other two wanting the lower torch version - easily fixed though

[lost crescent]

OK, it's d/loading/installing 2.2.0+cu118

[fair oxide]

getting some pesky errors, so i just grabbed the package and running it now

backed up the a1111 webui folder and instaling forge fresh

[nocturne topaz]

package dependency clashes for protobuf

?

ignore those, works fine without em

[fair oxide]

well i had it up running but the errors are nagging me

[nocturne topaz]

fair enough.. fresh venv never hurt anyone

[fair oxide]

doing the fresh install to get rid of those warnings

yeah

but im choosing to run forge within its own environment instead of having a1111 running on seperate folders

that will save some space on my 500gb SSD

holy smokes

that went pretty fast and up and running

and render is super fast

[nocturne topaz]

render is so smoking fast it aint funny

and if you haven't played with FreeU and LatentModifier before, its worth having a play with those

[fair oxide]

yes, gonna look into all those features 🙂 first thing first tho, gotta copy all the backup files into forge

controlnet models, checkpoints, lora etc

[nocturne topaz]

just use redirects in the commandline arguments

i.e. --ckpt-dir /path/to/old/StableDiffusion --lora-dir /path/to/old/Lora

[fair oxide]

they are all in the same SSD

[nocturne topaz]

exactly.. why put em there twice

[fair oxide]

not twice, im moving them

[nocturne topaz]

ah right

Do the redirect from the a1111 install

[fair oxide]

cause im not gonna use a1111

[nocturne topaz]

just in case something doesn't work yet (i.e. deforum/supermerger etc)

[fair oxide]

i know but that's fine for now, i waited on other stuff when comfyui got them implemented but a1111 didnt, so that's not biggy for me

[nocturne topaz]

👍

[fair oxide]

besdies i mostly need the txt2img and controlnet

[nocturne topaz]

I do miss the merger tool, but happy to wait for a while while making super fast renders!

[fair oxide]

checkpoint merger?

btw where is the venv folder inside forge webui? or does it not need it?

cause what i did was unpacked the zip file, ran update.bat, then run.bat and its working as it is with images but i dont see venv folder inside webui folder

[wheat helm]

anyone familiar with the "Error: 'types.SimpleNamespace' object has no attribute 'cn_1_weight'. Please, check your schedules/ init values" error?

[nocturne topaz]

it's there?!

[nocturne topaz]

not me, but that kinda suggests you have the wrong type of controlnet model loading

[fair oxide]

ok then it probably doesnt show on my end cause i see updates still running in the console

[wheat helm]

by that do you mean the wrong type of checkpoint model? I was attempting a deforum generation

[nocturne topaz]

cn usually refers to controlnet

[fair oxide]

funny i even rendered an image w/o the venv showing or the updates completing

[nocturne topaz]

at least it's working, huzzah!

[wheat helm]

whats controlnet?

[lost crescent]

Anybody now what to do about these two dialogs at ComfyUI Startup?

[nocturne topaz]

im guessing that comfy didn't have a venv?

and you've done the torchvision install in the main system one, and now it's got the wrong version for comfy

[lost crescent]

My ComfyUI is not the Standalone version

[nocturne topaz]

a way of controlling how an image is generated by using an input image as reference

have you made deforum animations before?

[lost crescent]

It's odd, as up until two weeks ago, these dialogs did not pop-up

No Deforum

[nocturne topaz]

ie is this a new break, or have you not tried this before

[lost crescent]

No Video at all

[nocturne topaz]

two help threads here chief

[wheat helm]

nope this is the very first deforum attempt.

ive done image to image and such

[nocturne topaz]

right.. are you in a1111?

[wheat helm]

yes

[nocturne topaz]

install the sd-webui-controlnet extension

then try again

[lost crescent]

???

[wheat helm]

if its not a hassle could u link it? this is what came up when i searched https://github.com/Mikubill/sd-webui-controlnet

[nocturne topaz]

that is the correct one

[nocturne topaz]

the deforum question wasn't aimed at you - I am helping you with the comfy errors on startup, as well as prodgrady with their deforum issue

although I now gots to fly.. good luck y'all

[valid gorge]

Are there any prompt extensions for A1111 or Comfy, that would interpet and "sweeten up" the prompts we make. Like the one in natively in Fooocus, or MJ

[nocturne topaz]

Rpg diffusion

[fair oxide]

@nocturne topaz if you look at the file tree here, it doesn't have venv folder and im still missing that in my local forge setup https://github.com/lllyasviel/stable-diffusion-webui-forge?tab=readme-ov-file

there are also other folders like output, ESRGAN are missing

[patent meadow]

can someone translate for anoobie please?

"I finally found the problem, it was because (controlnet) and (segment-anything) conflicted, after closing (segment-anything), everything worked fine."

[fair oxide]

@nocturne topaz my bad, those branches wont show up unless i run webui-user.bat, but when i tried to instal the all in one package i was missing those folders

[wheat helm]

It worked! I think sorta? lol. Now im getting a different error message but this time after 4 minutes of generating as opposed to last time the error being imediate. I think thats progress

the new error is "Error: [Errno 13] Permission denied: 'C:\Users\bmmgr\Downloads\dfehedtrh'."

[ornate elk]

Nope

Also its pytorch support Is limited

Disable or uncheck the Segment Anything extension

Then controlnet should work

[wheat helm]

oh were u replying to me? I think controlnet is working for me now... maybe. The error im now getting is "Error: '[Errno 13] Permission denied: 'C:\Users\bmmgr\Downloads\dfehedtrh"

"dfehedtrh" is just the name of the image im using for deforum init

ok i think i fixed that by dragging the image into the init image box, but now i just recieved "Error: '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.'. Before reporting, please check your schedules/ init values"

[patent meadow]

this error wont leave me

I reinstalled control net 3 times

https://github.com/Mikubill/sd-webui-controlnet/issues/108

tried all the suggestions in here none worked

i disabled all extensions nothing is interfering with it

[ornate elk]

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

[wheat helm]

my GPU is a Nvidia Geforce GTX 1660 Ti

and i tried that command that you gave to the guy who had the same problem yesterday and it didnt work for me

one thing im curious about tho is my webui-user.bat folder doesnt actually say .bat on the end. It is a batch file but it doesnt have .bat

otherwise it seems identical

[ornate elk]

Make sure you have --medvram --xformers --no-half to the webui-user.bat

[ornate elk]

Thats because Windows 11 hides the file Extension name

[valid gorge]

Great, thanks. Will check that out!

[wheat helm]

this look alright?

[ornate elk]

Yes

[ornate elk]

Go into controlnet settings and check the box:
Load Clip into CPU.
Then hit apply

That should fix it

[valid gorge]

Guys, how do you organize your LoRAs in Comfy? I have my checkpoint models placed under different folders but haven't have a system for LoRAs so far as A1111 recognizes are they compatible with SD or SDXL models, so I haven't have to worry about it.

Not sure if there's a similar filtering system within Comfy

[autumn lichen]

Hey guys, I am trying to run DynamiCrafter and am having an issue with XFORMERS.

It says I am trying to run torch on my cpu, and my version of python is anterior, yet I have update Python. Torch 2.0.0+cu118 is installed so it should be set to run on Cuda, which is also installed. I<ve attached a screen cap of anaconda.

I have used both anaconda navigator and conda powershell to update python (in case there was a double instance or whatever) and yet DymaniCrafter still thinks Python is the old version. Could it be that my repository has an old pytorch and python?

I've spent two days on this and feel like I am running in circles going mad 😅 Your help would be greatly appreciated!

[ornate elk]

Make sub folders, that should work

[valid gorge]

Yeah 🙂 thanks!

Is there a way to refresh loaded LoRAs without needing to close server and reload ui.

never mind, the refresh button seems to do that alongside other stuff. Just had to press it twice

[autumn lichen]

I wish my issue was that simple 😅

[valid gorge]

hehe

Hope you get it solved

[autumn lichen]

I found a solve for python

!pip install lmdb
!pip install torch==2.0.0+cu118 torchvision==0.15.2+cu118 torchaudio==2.0.2 torchtext==0.15.2+cpu torchdata==0.6.1 --index-url https://download.pytorch.org/whl/cu118

however using !pip at the begining of my .py gives me a syntax error

[patent meadow]

thanks but now i get a diffirent error

[ornate elk]

What's in your webui-user.bat?

[patent meadow]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --use-directml --lowvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling
call webui.bat

[ornate elk]

Okay change --no-half-vae to --no-half
And try again

[patent meadow]

ok

[patent meadow]

Discord is acting funny

[patent meadow]

Thank you very much that worked 🫶

[ornate ruin]

For an image with a 1:2 aspect ratio,(being twice as tall as wide)

What's the best way to go for? Currently I'm just generating at 512x1024 which.. works better than I expected but not great obvious. Should I stick to 512x768, go for a higher resolution than I need and crop the images?

[unreal valley]

Can anyone help with this error

I am just emulating what was shown in a tutorial

@ornate elk hi man, I am back with another problem lol

[lusty bay]

@lunar linden , did you find a solution? also facing the same problem...

[sonic hill]

Hi,

I'm unable to switch VAE in my auto1111 application. I'm getting the following error:
changing setting sd_vae to NameOfTheModel.safetensors: RuntimeError

To be precise, my checkpoint is realisticVision 5.1 and I'm trying to select the accompanying VAE

[lunar linden]

no. seems to be a known issue that has never been fixed: https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/9779

[lusty bay]

it sucks, then need to find some workaround to make this work, or use some serviced SD wrapper instead of raw one

[slender nova]

Hello, I just installed the dreambooth extension onto stable diffusion but for some reason the input tab is missing from the gui.
has anyone encountered this or know the fix?

[hollow wraith]

after installing the civitai extension i now get this error when using hires fix (2x):
RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 128 but got size 64 for tensor number 1 in the list.

anyone got an idea as to why? i reset my setup to yesterday's state in the meantime but i'd like to use the extension for the preview images

[nocturne topaz]

which civitai extension did you use? I recommend civitai browser +

[hollow wraith]

this one https://github.com/civitai/sd_civitai_extension

[nocturne topaz]

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

[lost chasm]

OK! 4090 installed! do I have to do anything to make SD work faster on 4090 using some special 4090 install shit that takes advantage of its features?
or just press go as normal

[zinc mirage]

Regarding stability.ai's text-to-image model, how to control character's pose? Is there anything like Controlnet extension function?

[lament fog]

For the most part the base models are better at certain resolutions, like sd15 at 512x512 or sdxl at 1024x1024. Certain other resolutions may work better. A list was posted for sdxl at one point and I saved it. 1024 x 1024
1152 x 896
896 x 1152
1216 x 832
832 x 1216
1344 x 768
768 x 1344
1536 x 640
640 x 1536

[steady thistle]

Is there a Stable Cascade install guide? It seems like it might be better for AMD.

[proper swift]

what makes it seem better suited to amd?

[nocturne topaz]

amd cards at the same price point tend to have more vram

[steady thistle]

I saw that it used fewer resources

[nocturne topaz]

the problem is cuda though; although zluda may have resolved that

[steady thistle]

I'm using AMD and it's not a happy experience.

[nocturne topaz]

have you installed zluda in the last day or two?

apparently it's not quite there for stable diffusion, but may be working in some setups

[steady thistle]

I don't know about zluda

[nocturne topaz]

its a conversion layer that allows amd cards to natively process cuda commands

[fair oxide]

This looks very promising, just hope its not behind paywall

[nocturne topaz]

the weights are publicly released on a non-commercial use licence

results are mixed, but finetuning has commmenced so give it 3 months to see what people come up with

[steady thistle]

This is amazing. I'll be working on this very soon.

[nocturne topaz]

if initial models go well, then it may take off, if not it will go the 2.1 route

[nocturne topaz]

fingers crossed for you

[steady thistle]

Do you have any pointers on an install? Everything I have done has been built on working around cuda

[fair oxide]

Sounds great, didn't know it was already publicly released, thanks for the update 👍 have you tried it out yet?

[steady thistle]

I might just try to install zluda then do an Nvidia install

[nocturne topaz]

Not yet, I'm waiting for the official compatibility with sd-webui-forge before launching in

[nocturne topaz]

i don't have AMD to play with it sorry

[fair oxide]

By the look of how cascade works, its supposed to be lot more efficient and less resource demanding which will make it easier for consumer use as well as training models.

[nocturne topaz]

except there are multiple models similar to sdxl size

[steady thistle]

That's why I was interested in it for AMD, but this zluda might solve my issues if alpha isn't too alpha. I have 12gb vram and I have to run -lowvram many times.

[fair oxide]

Ahh ok. Yeah AMD card doesn't have good support in using stable diffusion, i have a 12 gb card too but thats Nvidia

[steady thistle]

I'm seriously considering throwing $1200 into a tower with nvidia. I'm getting really annoyed.

[fair oxide]

Yes three models, but they are meant to be small in size, its based on the wurtschen architecture if you remember from last year

[fair oxide]

You could but technology might change in a big way when cascade comes out , so might wanna wait for it

[steady thistle]

Yeah, I think maybe so. I feel like I went with Beta over VHS. Again.

[neon ruin]

lol, came here to ask about this

[fair oxide]

Besides your 12gb AMD card is good performing card just that it lacks cuda support specifically for stability diffusion as of now

[neon ruin]

having all kinds of issues with new pc build (7900XT)

[steady thistle]

It's a really fine laptop, zero complaints except for SD, but this is my new hobby. heh

[fair oxide]

I wouldn't wanna throw away that card, I'd wait for cascade

[steady thistle]

I wonder if we could get a channel for AMD issues. I don't know how to keep up with the new releases otherwise.

[fair oxide]

Lot of people have been raising this issue about AMD incompatibility, stability AI would have to look into it

Ill have to check on cascade and the resources available on the internet when im on pc, im really excited about it

https://stability.ai/news/introducing-stable-cascade

[steady thistle]

Any ideas how to install?

[fair oxide]

Not sure, but i remember when wurtschen was talked about last year, you could run that on SDNext, i havent tried it myself tho.

[steady thistle]

ah, I like my local install

Oh, it is local. heh, I read that as Civitai

[fair oxide]

Sdnext runs locally its got similar interface as a1111

[nocturne topaz]

Sd next removed compatibility

[steady thistle]

Does it run ReActor?

[nocturne topaz]

Has to be done through a a1111 extension or a comfy node

Or from a seperate command line install

[fair oxide]

You mean compatibility for wurtschen?

[steady thistle]

I see it has FaceSwapLap

[fair oxide]

You might wanna take a look at this
https://twitter.com/cocktailpeanut/status/1757457604781978091?t=PLu4-bJkdvdzJZ0V8EFhuA&s=19

cocktail peanut (@cocktailpeanut) on X

Local Stable Cascade 1 Click Launcher

Just wrote a 1-click launcher for this Stable Cascade Gradio app. Works very well!

Works on all platforms: Windows, Mac, Linux

Twitter · Feb 13, 2024, 5:31 PM

[fair oxide]

👀
https://twitter.com/EMostaque/status/1757472480430616806?t=-ciGsDRVuxX6zX6E2JBiUQ&s=19

Emad (@EMostaque) on X

Hands and text from a base model? #stablecascade

Twitter · Feb 13, 2024, 6:31 PM

[cursive anchor]

im having this messege apared in the cmd

i can instal and run the program just fine

but when i close is and i try to run it for a second time it pop this

"Access Denied - Couldn't launch python - exit code: 1"

[nocturne topaz]

windows or linux?

[cursive anchor]

windows 10

nvidia card

it was fine for the last 10hs

my only way of running it now si to:

.Move outpost,embedding,models and extencion to a separete folder
.Delet everything
.Re-Instale it
.Run it
.and move outpost,embedding,models and extencion again
.ReInstal

Lo corri como Administrador y salio esto

[boreal bison]

Hey guys! After formattin my c drive, my portable a1111 stopped working which was on another drive. I reinstalled python etc but still no luck, anyone knows how to fix it?

[stoic cave]

Why are you using a portable version?

[boreal bison]

sorry I got it confused with comfy ui. So yeah not the portable... how do I recover it? I just finished installing another one on my c: and it works, I wanna bring back the one which lives on another drive

@stoic cave

[stoic cave]

Just move over the files in the other folder to the new one? Specifically what are you trying to keep just models and extensions?

[boreal bison]

everything

[lost crescent]

Your Huggingface Cache was re-established?

[boreal bison]

I don't think I have huggingface cache?

[untold flame]

currently at week 27 of trying to solve the Bitsandbytes issues with Kohya

[nocturne topaz]

Which optimizer you using??

[cursive anchor]

so nobody knows how to fix it? Im out of lukc it seems

[untold flame]

All of them honestly. I have had this issue for almost 7 months, and nothing has ever been able to fix it. My job kinda relies on me being able to train in Kohya, so this has caused a lot of issues for me

Only thing I have done that has kinda made it work is to use bitsandbytes-windows, but that is a really bad solution, as it's known to cause severe training deformities, which has been a serious issue for me

[nocturne topaz]

Last time I had that I had to swap between adamw and adamw8bit but that was on a 1060

[untold flame]

I use Adam8bit at the moment, are you suggesting I switch between them? I could give it a try, I am so defeated at this point

I have had it say it doesn't exist, and then that cuda failed to start, then it doesn't have 'optim' as a module, its always something

[nocturne topaz]

Yeah - if not surely someone over in #🔧｜finetune will have seen it

[untold flame]

I've been chasing this for months now with some pretty well educated people, but man this issue seems to be never ending. I have tried out other trainers and while some work, they seem to have other issues (mostly the lack of features I use a lot)

I will give what you said a shot, cause I am really desperate lmao

@nocturne topaz...

You are my FUCKING HERO

like OH MY GOD

Adam W is training. It never worked before.... oh my god

https://tenor.com/view/wtf-cat-grrrr-grr-grr-meme-gif-12171908424805132741

I feel so cheated, but infinitely grateful, thank you so much @@nocturne topaz

[nocturne topaz]

Most welcome 🙂

[ornate elk]

What kind of issues?

[ornate elk]

What's the error you get, when trying to launch it?

[boreal bison]

I think I fixed it, just reinstalled a new version and copied everything from the old version to the new version lol, and when I launched it started downloading things that were missing?

Had no idea it would do it, but seems like it did

[whole ember]

Accidentally deleted my Linux partition. Does anyone have a bash script(s) for reinstalling AMD Drivers/ROCm/Stable diffusion/dependencies, etc

I used to use Linux Mint but I don't mind shifting to Arch for OpenSuse if it makes the setup process faster/easier

[cursive anchor]

@nocturne topaz Could i ask for a bit of magic for my side to?

[nocturne topaz]

I'm not sure I have that level of magic for you sorry - I had seen that error the guy was having myself about 8 months ago so I remembered what the fix was.

That said.. are you trying to install sd next or a1111?

[fair oxide]

@nocturne topaz looks like triton doesn't work on windows?

[nocturne topaz]

It should ‘pip install triton’

Inside the venv

[fair oxide]

it did but what i mean i still keep getting warning that says triton missing something and some of the features will not work

i even grabbed the files from https://huggingface.co/r4ziel/xformers_pre_built/tree/22505e3edeead471f3801ff2c3d478ffa51be755

[nocturne topaz]

Ive not seen the error actually affect anything so far

[fair oxide]

as weird as it may sound.. something happened in between a few attempt to install forgeui and it started to run slower than regular a1111

[nocturne topaz]

Weird, but not totally out of the realms of possibility

In that case a1111 may just be right for you 🙂

Which card you got??

[fair oxide]

i was trying to figure out what went wrong but the only warning i was getting is about triton .. other thing i noticed the live preview looked horribley messed up but the final image showed up ok

rtx 3060

couldnt be card issue im guessing

[nocturne topaz]

Nah, 3060 usually sweet as

Could always delete the venv folder and restart but yeah shrug

[fair oxide]

the devs need to organize the release a bit, a lot of the technologies are scattered across different webuis

im not sure how cascade will be compatible with a1111 w/o completely changing everything about it and how it runs sd and xl

[nocturne topaz]

To be fair they could setup stage C (prior) as the regular model and stage B as the refiner (with stage a as the vae)

[fair oxide]

nice, hope so

im very eager for cascade to be widely released

interesting that in the 2023 CEO summit, google is saying opensource will be key to making progress with generative AI

[nocturne topaz]

Yep, but will be restricted by access to vram tbh

[fair oxide]

yeah, thats the concerning part, and kinda reflects on the economic model in ways we access resources

you know there are movements to implement UBI, not sure when but its in the talks

[inland zephyr]

Did anyone ever figure out why Automatic1111 infinitely stalls at 98% when using HiresFix? I went back through the history of this channel, and read proposed solutions for people who also had the same issue, but none work for me

About a year~year and a half ago, I had no issues

[fair oxide]

never faced that issue tbh, but that could be cause of your vram or ram too

[inland zephyr]

My PC is exactly the same as when I was using A1111 with no problems, other than regular updates to both my GPU and A1111 itself

[fair oxide]

also take note of your hires fix setup, like the value you set to upscaling on a given image size

[inland zephyr]

I always upscale x2, mostly I just use HiRes Fix for the quality and coherence improvements

[fair oxide]

upscaling by 2x is reasonable, and if your using a 8gb+ card you could apply that value fairly ok with 768 res

[inland zephyr]

Which is exactly what I do

768x512, x2 from hires fix, and upscale it separately later if I want it bigger

[fair oxide]

i would guess, there could be some bottle necking going on with your system

whats your vram, ram and cpu speed?

[inland zephyr]

RAM usage is at 95-100% when the stalling happens, and I'll get visual lag on both monitors until I forcefully close the cmd window

[nocturne topaz]

Sounds like a vae issue, tried a different vae??

[inland zephyr]

3070 8gb, 5700X with variable OC, but usually 4.5-5ghz, 16gb Trident Z at 3200

[inland zephyr]

Yeah, a bunch of different ones, and none

If I use none, it will stall at 48-50% instead

[nocturne topaz]

Have you tried using the tiled vae from the multidiffusion extension??

[inland zephyr]

Since I don't know what that is, I'm gonna say no

I'll install it now and try it out

[nocturne topaz]

It splits the latents up before firing it at the last step (vae) to not use as much vram

[fair oxide]

curious, could upscaler be related to this?

[nocturne topaz]

Upscale happens at 50% so if its past that the upscale happened successfully

[inland zephyr]

Hot damn, Tiled VAE worked

[fair oxide]

ah ok

[ornate elk]

What's inside your webui-user.bat?

[inland zephyr]

--medvram --xformers --medvram-sdxl

[ornate elk]

Remove --medvram
And add --no-half-vae

[inland zephyr]

Will do

Without Tiled VAE, that change still stalls at 98%

Although the process as a whole was much faster

[ornate elk]

Does it happen with every VAE?

[inland zephyr]

Yes

And with no VAE at all

Although without a VAE it still stop at 50% instead

[ornate elk]

What was the resolution you upscaled by 2 ?

[inland zephyr]

768x513

2

[ornate elk]

Hmm thats okay

[cursive anchor]

a1111

teh thigs is that i can instal it. i run in for the first time, i downlead cheackpoint, loras, styles. and it works worderfully. Now, it stop working the second time i try to run it. I close the tab and the cmd, run the run.bat and error

[ornate elk]

Did you installed the webui, with my install guide? Its in the pinned messages of this channel

Make sure to do this if you haven't before

[inland zephyr]

By the way, should I leave my command line with just xformers and no half vae, or is that something that was just for testing?

[ornate elk]

Your card should use --xformers --medvram-sdxl --no-half-vae

Then everything should work and sdxl will also be usable at a good speed

[inland zephyr]

Thanks, I'll ride out Tiled VAE until it stops working too

[nocturne topaz]

Only thing I can think of it there is some weird permissions applied to the venv folder?

[mighty topaz]

what thats mean?

no half vae? this help better to do what?

[cursive tendon]

anyone can tell me how to update this?

[vocal burrow]

to update, run: xxxxxxxxxx

did you try running xxxxxxxx ?

take the green stuff, copy paste it in a cmd, press enter

[cursive tendon]

what is xxxxxx?

[ornate elk]

To load VAE files for sdxl or if you have a GTX card

[mighty topaz]

i have geforge rtx 4070 so i need turn on this?

[cursive tendon]

thanks

[ornate elk]

Not needed if you dont get any errors, but won't hurt if its in there too

[mighty topaz]

ok thanks

[neon ruin]

Initially? It straight up wouldn't run. Had to add what was apparently an AMD specific statement to the webui-user commands: " --use-directml "

Within SD itself I've crashed at least three or four times now. Specifically it seems to happen if I try adjusting the batch-size, sampling steps, or even based on the model I use. It's hard to keep testing it when I'm not sure if my pc is just going to turn off at any given moment.

There have also been a few times where I'll go to generate something and it'll hit me with "insufficient memory", despite the fact that my gpu should have 20GB

[molten coral]

any idea how to get xformers installed with torch 2.0.1 for auto1111/forge?

[ornate elk]

Ah okay, yea, you should have followed my AMD install guide.
Its in the Pinned Messages of this channel. (There is everything needed to get it work)
There you also find the needed commands for AMD cards to get the best performance/usage

If you have any questions about what you can do or not with AMD and the webui, feel free to ask me. I'm running a 7900XTX

[night orchid]

@ornate elk the tiling option is no where to find, do you know where or how I can enable it?

I have red that it is a bug

I got it fixed

[ornate elk]

Delete the venv folder, and relaunch

[ornate elk]

Thats a known issue in the latest version. The dev will fix it today or tomorrow (hopefully)

[neon ruin]

Will definitely be checking this out, thanks dude. Yeah my heart sank when I started testing SD on this new build and I was hoping I didn't screw up by going with AMD

[neon ruin]

I edited the COMMANDLINE_ARGS with what you have pinned, tried a random prompt and increased the batch size/count just to see how it behaved. PC turned itself off in about 5 seconds.

[ornate elk]

So you didn't got the keras error again ?

[neon ruin]

I didn't have the keras error before, initially I was not able to launch webui-user until I added "use-directml"

Then what was happening was that I was getting an insufficient memory error in the webgui at times, and other times my pc would just turn off instantly.

I replaced the commandline_args statement with the one you have pinned, am able to still launch to the webgui but it still crashes if I try to generate anything with an increased batch size

[ornate elk]

Use batch count not size

Size won't work with the bad Directml vram optimisation

[neon ruin]

I see I see, well I haven't crashed yet 💀

[cosmic furnace]

any idea why 7900 xt not working with -medvram arg?

also --no-half --precision full ...

While other ppl seem to have them working 😒

[ornate elk]

Checkout my install guide for AMD in the Pinned Messages of this channel

There you find all COMMANDLINE_ARGS you need

Remove --precision full

[fair oxide]

@ornate elk have you tried faceID ip-adapter with multiple image inputs ?

[ornate elk]

Yes

[fair oxide]

i get bunch of errors :/

i think im missing some components for insightface

[ornate elk]

whats the error?

[fair oxide]

let me run the webui again i had to close it

[flat olive]

RuntimeError: Could not allocate tensor with 167772160 bytes. There is not enough GPU video memory available!

help

[feral mango]

and how i install formats on video helper?
videoh.264 and others

[flat olive]

i tried lowvram or medvram on command args but it doesn't make anything

[feral mango]

i shopuld see all this

[fair oxide]

@ornate elk it rendered the image but i get these messages in the console, not sure if that's something to worry about or harmless

[feral mango]

i shopuld see all this

[ornate elk]

whats your gpu again?

[flat olive]

rx 580 8gb

amd lmao

[ornate elk]

your on nvidia right?
if the image looks alright, (and like the input images) then you can ignore it

[ornate elk]

and whats the resolution you tried?

[flat olive]

512x512

[ornate elk]

and that gave you the error?

[flat olive]

yep

[ornate elk]

any loras or hires fix used?

[fair oxide]

yes rtx 3060, i think the images are coming out ok, but those messages in the console felt like a nag or something to iron out

[feral mango]

please help how to isntall video encodners ?

Dependencies
ffmpeg

skvideo

[flat olive]

nope, but i tried some parameters on arg lines from webui-user.bat

this ones --use-directml --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling --no-half --disable-model-loading-ram-optimization --opt-sdp-attention

i removed lowvram or medvram because it didn't work on any of the 2 ways

[ornate elk]

remove --opt-sdp-attention --disable-model-loading-ram-opt
and add --lowvram

[flat olive]

ok

i was reading about a --half-precision parameter what could help with this, is this like that?

[ornate elk]

getting the same error, but still works

[ornate elk]

you already have --no-half
in there so your fine

[fair oxide]

hmm i see, thanks for taking the time to check 🙂

[ornate elk]

no problem 😄

im still not shure if it looks like them

should have tried an other model haha

[fair oxide]

usually the input images should have pics of the same person that looks similar

[ornate elk]

ah right xD

[fair oxide]

but ofc you could try interesting variations but random images wont be lookalike

[flat olive]

still doing that RuntimeError: Could not allocate tensor with 4184866816 bytes. There is not enough GPU video memory available!

[ornate elk]

works, its a mixup xD

[fair oxide]

yes

[ornate elk]

can you show all the settings?

from txt2img

like steps etc

[flat olive]

also, im trying to use animatediff

[ornate elk]

wow xD

that thing wont work on 8gb vram amd cards

[flat olive]

lmao

ty

[ornate elk]

thats so heavy it struggles on 8gb nvidia aswell

[flat olive]

and, you know of any update for amd? atleast on wip?

[ornate elk]

np, if you want to make videos or animations you should use the Loopback Wavescript (but its more like deforum)

[ornate elk]

for better SD usage?

[flat olive]

yep, i was reading about sd use nvidia controllers

[ornate elk]

forget Zluda for now, (its early alpha, not made for SD)
Rocm on linux is your best bet.
Rocm on windows will come but wont support your card (to old)
Shark Webui from AMD, will work the fastet for your card on Windows.
And there is a new programm called stable diffusion cpp, but thats CMD only

[flat olive]

wow

cmd only

i will try with shark webui

to see if i can get somenthing good

ty cs1o

[ornate elk]

no problem, Shark is fast but the extension and lora support is limited (but it gets updated very often)

so the best thing you can use is Auto1111 on either Linux or Windows

[dusty blade]

Hello everyone, I installed automatic1111 via run.bat and finished it with update.bat, but I don't even know how to open it. Can anyone help?

[ornate elk]

hey, did you downloaded it as zip?

[dusty blade]

np man i installed phython on microsoft store and my problem fix it

error code 9009

now im still waiting

[ripe vessel]

Failed to import transformers.modeling_tf_utils because of the following error (look up to see its traceback):
No module named 'keras.internal'

whats this

i've been trying to install this crap for 2 hours straight now

yikes

[ornate elk]

you installed the wrong python version

also microsoft store python is not recommended

[ornate elk]

you better start a fresh and clean installation with my install guide.
its in the pinned message of this channel

[ripe vessel]

@ornate elk is there a way to fix that keras error

or do i have to wait

[dusty blade]

where i cant see

error again error code 1

[ornate elk]

in this channel, click on the Pinned Messages Icon

its a Pin Needle

at the top right mostly

yea pls use my install guide

then you wont get errors

[ornate elk]

i didnt updated yet, but you can try to add this:
--use-cpu-torch

[dusty blade]

So how can I delete the currently installed one?

im alredy install a automatic1111

[ripe vessel]

isnt that super slow

still the same anyway

[ornate elk]

delete the venv folder again

also i had a typo its --use-cpu torch

i would just wait until the Dev fixes the bug as he does that mostly in a day

[dusty blade]

@ornate elk Is deleting this folder enough to completely delete the program? clean installation will be done

[ripe vessel]

didnt help

yeah ill just wait

[ornate elk]

yes

wait no

[dusty blade]

ok...

[ornate elk]

ah that are two screenshots

so yes only the automatic1111 folder

[dusty blade]

ok delete automatic1111 folder

[ornate elk]

then its clean

[ornate elk]

let me know if you have any install questions

[dusty blade]

little help

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.

where to open cmd in this article

windwos q cmd?

@ornate elk its true ?

[ornate elk]

yes

but not with powershell

use cmd

[dusty blade]

how 😄

powershell starts automatic

[ornate elk]

like in the guide, click in the adress bar in explorer and type cmd

[dusty blade]

thx

[ornate elk]

no problem 🙂

[dusty blade]

@ornate elk any problem? im use rtx4060 8gb

[ornate elk]

nope looks good

[dusty blade]

finally

thx

[torn lotus]

hi, was wondering If I should keep comfyui and ditch a1111 or keep them both

[ornate elk]

that depends on you 😄 why would you ditch someone just because someone said to do

for what i do with SD, auto1111 works much better than comfyui, for others its the opposite

keeping both is a good option too, as you can link the models from auto1111 to comfyui. so no need for doubled files

[torn lotus]

I'm striving a little with the configuration yaml. I know checkpoints are only under comfyui

[ornate elk]

in which webui?

[torn lotus]

comfyui

the only folder with downloaded material under A1111 is /models/stablediffusion

[ornate elk]

what do you want to do?

link the models from auto to comfy?

[torn lotus]

I want to install SDXL

i only have 1.5

I have crosslinked already. is this corrrect?

[ornate elk]

yes looks good

now you only need to place the sdxl models in the models/stable-diffusion folder

[torn lotus]

should I add a checkpoints folder in stable diffusion /models=

I mean, In the end I'm using only one /models folder for both right

[neon ruin]

Well that was good while it lasted

Pc crashed again

[hybrid thorn]

Hey guys, just checkout StabilityMatrix: https://github.com/LykosAI/StabilityMatrix You can thank me later. 😉

The best thing: You have one shared Model folder and it loads the meta data from civitai.

[ornate elk]

how much ram?

also make sure to not run Wallpaper Engine or Epic Games Launcher in the Background

[neon ruin]

In the graphics card?

20GB if I’m not mistaken, it’s a 7900XT

I have 32GB on my pc, though that will be 128 soon

[ornate elk]

okay thats enough for sure

[ornate elk]

no dont add them into /models/

they need to be in /models/stable-diffusion
and loras in /models/loras

vae in models/vae

controlnet in /models/controlnet

[ornate elk]

how much wattage has your PSU?

Make sure you dont run any other GPU intensive tasks/programms at the same time as you generate images

[elder summit]

any update on the keras error? i just started getting it

[ornate elk]

Nope the dev hasn't been active yet. But the error is reported on github

[elder summit]

i dont know if this is the appropriate place to ask but is there i way i can reveret to a previous built

[ornate elk]

Yes with this command:
git checkout d500e58a65d99bfaa9c7bb0da6c3eb5704fadf25

[elder summit]

lol nice that worked thanks man

[ornate elk]

To update again you would need to use
git checkout master
When the patch with the fix is released

[flat olive]

i tried the shark version, but it dosen't work and the discord server is dead so im gonna try install ubuntu linux on my pc, any tips?

[slow garden]

I could really use some help, my install keeps breaking

This is what I have going on

[slow garden]

??

I havent got to that point yet

[dusty blade]

missing interrogate clip

image 2 image

[slow garden]

oh, you need help too?

[dusty blade]

yes

The trainer I watched has it, but I don't have it.

[dusty blade]

any idea?

[slow garden]

nah bro, I cant even get this installed lol

[dusty blade]

bruh

[slow garden]

fuck, maybe you could help me lol

look at that error I sent

[dusty blade]

I reinstalled following the pinned helper and it worked without any problems

[slow garden]

wheres that?

[dusty blade]

nvıdia or amd?

[slow garden]

amd

[dusty blade]

.

need alot vram

alot!

[slow garden]

sweet, thanks buddy

I have 8gb

[dusty blade]

ye me to and 7.9 usage

%100

[slow garden]

dayum

[dusty blade]

damn

ahahah

[slow garden]

well, Lets see what happens lol

I got a bad feeling it aint tho lol

swapped back from linux to windows just to use this in a easier way

and here we are

[dusty blade]

ahaha

[slow garden]

and it fucked up

wonderful

and I followed the guide

idk what any of this really means

[dusty blade]

can you look at image to image for me

@slow garden

[slow garden]

??

[dusty blade]

this page

[slow garden]

I cant access that page bc I dont have it installed

[dusty blade]

:/

my interrogate clip button are missing anyone help?

[unique juniper]

I have the error code 1 can someone tell me what to do ?

[tender gazelle]

I've never tried directml (or heard of it), so I don't know, sorry. Is there another solution you can try?

[slow garden]

what do you reccomend?

I used the guide for amd they had pinned

[tender gazelle]

Oh, for AMD GPU?

[slow garden]

yep

[tender gazelle]

I haven't heard anything, and never owned one. Hope you get it working though.

[slow garden]

yea, I dont want to ping people, so im just gonna have to wait

[small quest]

Hello, I'm trying to install/run automatic1111 on my Nvidia gpu, but I keep getting an error when I run run.bat
RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check
Would skipping the test still allow me to use my GPU or is it neccesary? If so how can I fix this issue?
I tried following this guide https://stackoverflow.com/questions/75910666/how-to-solve-torch-is-not-able-to-use-gpuerror
but I still get the same error.

[nocturne topaz]

usually thats the case if you've got Torch version issues.. .delete the stable-diffusion-webui/venv folder, then rerun run.bat

[small quest]

I don't see venv folder among the files

[nocturne topaz]

yeah, something weird with that installer then

[slow garden]

any chanve I could get some info on my error @nocturne topaz

[nocturne topaz]

delete the venv folder and run it again

[slow garden]

venv in the diffusion folder?

okie, its deleting, just a moment

[small quest]

I tried redownloading the installer but it's still not there, could you link the one you're using?

[slow garden]

@nocturne topaz it made no difference

[small quest]

double check which version of python you're using, turns out I was running 3.12 and downgrading to 3.10 seems to have fixed my problem