🤝｜tech-support

[ornate elk]

loras are smaller files, that got trained on specific styles or characters to support a model to get that specific style or char

they have to be used together with a model

[blissful nimbus]

What's the best Sampling method?

[willow zodiac]

I could really use some help. When I run ComfyUI, my models in a1111 do not load in.

I have updated my extra_model_paths.yaml file to point to my base path. I assume it is typed correctly as I copy/pasted from the file explorer. I'm not sure what I am not doing right.

[glad fiber]

guys, when I train a lora, all models have the exact same size

and they dont work (the character is not applied to the image)

the paramters used in kohya_ss ^

[wraith bolt]

which settings did you use?

[glad fiber]

I used the default ones

[willow zodiac]

CS1o, Thank you. I selected the YAML extension when I saved as, and didn't notice the .yml in the editor. That was what I needed.

[wraith bolt]

without --xformers --medvram --no-half ?

[glad fiber]

you meant command lines

I used those

[wraith bolt]

yea just settings in general

cause i think my gpu is way too slow

[patent spear]

I am about to upgrade my graphics card from AMD RX 5700 XT to a nvidia RTX 4070. I'm assuming I have to uninstall the webui since I installed it using the AMD settings. Does anyone know the process to convert it to nvidia GPU settings?

I installed the Automatic11111 AMD DirectML Webui using the steps in the pins

[flint acorn]

do i need to re-install stable diffusion if i have updated my gpu driver?

i keep getting errno22 - invalid argument

in cmd when trying to launch

[compact falcon]

START OF TRACEBACK
Traceback (most recent call last):
File "C:\Stable diffusion\stable-diffusion-webui\extensions\deforum-for-automatic1111-webui\scripts\deforum_helpers\run_deforum.py", line 116, in run_deforum
render_animation(args, anim_args, video_args, parseq_args, loop_args, controlnet_args, root)
File "C:\Stable diffusion\stable-diffusion-webui\extensions\deforum-for-automatic1111-webui\scripts\deforum_helpers\render.py", line 206, in render_animation
mask_vals['video_mask'] = get_mask(args)
File "C:\Stable diffusion\stable-diffusion-webui\extensions\deforum-for-automatic1111-webui\scripts\deforum_helpers\load_images.py", line 106, in get_mask
return prepare_mask(args.mask_file, (args.W, args.H), args.mask_contrast_adjust, args.mask_brightness_adjust)
File "C:\Stable diffusion\stable-diffusion-webui\extensions\deforum-for-automatic1111-webui\scripts\deforum_helpers\load_images.py", line 80, in prepare_mask
mask = load_image(mask_input, mask_input)
File "C:\Stable diffusion\stable-diffusion-webui\extensions\deforum-for-automatic1111-webui\scripts\deforum_helpers\load_images.py", line 71, in load_image
image = Image.open(image_path).convert('RGB')
File "C:\Stable diffusion\stable-diffusion-webui\venv\lib\site-packages\PIL\Image.py", line 3236, in open
fp = builtins.open(filename, "rb")
PermissionError: [Errno 13] Permission denied: 'C:\Stable diffusion\stable-diffusion-webui'
END OF TRACEBACK\

what do i do?

[sour plank]

What is this error and what does this mean?

[wraith skiff]

any solutions?

[ornate elk]

Click on Available, click on Load from then you get a list. There search for sd-webui-controlnet and click install

[wraith skiff]

this right?

[ornate elk]

Hey, you can follow my install guide in the pinned messages of this Channel. Then you only have to copy the models, embeddings, ans outputs folder to the new webui

[ornate elk]

You can first try to delete the venv folder. If that dont help then please show a full cmd screenshot

[wraith skiff]

still same error

[ornate elk]

Can you show the cmd log?

[ornate elk]

You installed the webui the wrong way, by downloading it as zip and also placed it in the downloads folder. That can make problems.
Better to use my Install guide in the pinned message of this Channel

[wraith skiff]

okey how can i delete all and start from 0

[ornate elk]

You only need to delete the sd.webui folder

[wraith skiff]

thanks i will try again

[ornate elk]

If you have any question about the install steps, feel free to ask

[amber copper]

Hey, I got some problem with my Kohya_ss (trying to train a lora) I always get this error. I got python 3.10.9 version so everything should work, but it just doesn't... I also tried a fresh reinstall here's the error:

Does anyone know what the problem could be?

[wraith skiff]

i cant launch

[ornate elk]

Because you PC Username does contain unsupported characters

[wraith skiff]

😐

[amber copper]

yeah, the maximum in the paths should be an empty space. You literally have an nightmare in there xD

[ornate elk]

You would need to change your Username. (Recommended). Or you install python on a different drive where the path doesn't contain the Username

[wraith skiff]

i have only that drive im tryin change username

[split sorrel]

Hello, I wonder why I'm getting these results when using

[ornate elk]

When using? Inpaint?

[patent spear]

Thanks. Just to clarify, I don't need to delete or uninstall the old folder? I can just follow the steps for the nvidia guide?

[ornate elk]

Yes you dont need to delete it. But after the install and moving the models, you can

[patent spear]

Oh okay I think I understand, I am a complete beginner, never used Python before. I guess the new folder will be called stable-diffusion-webui instead of stable-diffusion-webui-directml

[ornate elk]

Correct 👍

[patent spear]

Thank you, I was worried about doing something wrong 🙂

[ornate elk]

No problem 🙂 it should all work without problems. But If you get one we can fix it here

[ornate elk]

because for inpaint on AMD you need to add --no-half to the webui-user.bat
But that will increase the vram usage for everything resulting sometimes in out of memory errors.

[split sorrel]

set COMMANDLINE_ARGS=--use-directml --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling Here, it's written that I'm actually using this

[ornate elk]

yes i know, for inpaint to work you need to add --no-half

then in the img2img tab
select inpaint area "only masked" and set the resoultion to 512x512

[split sorrel]

I didn't understand, it's already there in the last code. If I need to change it, please edit and send it to me. I've done the other settings, but the result is the same."

[ornate elk]

no its not, there is --no-half-vae
no --no-half

use these: --use-directml --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half --upcast-sampling

[split sorrel]

it work 😄 so thankyou

[wraith skiff]

i installed again but..

[ornate elk]

whats your GPU?

[wraith skiff]

rtx 4060

i update my driver

okey i think its worked

[ornate elk]

Good, don't use --skip-torch-cuda-test or it will use the CPU

So if you get that error again you need to delete the venv folder

[wraith skiff]

I formatted my computer and changed the name. I'm downloading again now

[ornate elk]

Ah okay, then everything should work 👍

[wraith skiff]

okay now im tryin extensions

again

[ornate elk]

Make also sure the webui is whitelisted in any browser adblocker if you use one

[wraith skiff]

all off

[ornate elk]

Let me know if it worked

[wraith skiff]

Installed into C:\all\stable-diffusion-webui\extensions\sd-webui-controlnet. Use Installed tab to restart.

hahah

thanks dude

big respect+

[ornate elk]

No problem, for ControlNet you also nees the controlnet models. You can get them from here:
https://civitai.com/models/38784/controlnet-11-models

Each model is 700mb + its config file (there are multiple files you can select at the top)

Put them into models/controlnet

[wraith skiff]

👍

[steady thistle]

@ornate elk Good morning! I'm still working with inpainting, I have a pic of a man holding a sword, and the angles are wrong. I want to mask the hand and sword, so I mask it, replace the prompt with a sword, keep the size and everything, but everywhere I paint it screws up the background too. I'm using the same model, same sampler... not sure what to do. Do I need an inpainting model?

[ornate elk]

Hey Ohhs, Inpainting models can help. But it also works when using the same model that you used for the initial image.
What you have to check is to set
Inpainting Area to "mask only"
And then set the resolution to 512x512

These settings are essential for inpainting

[hazy wigeon]

getting this when trying to generate txt2img > NotImplementedError: Cannot copy out of meta tensor; no data!

[steady thistle]

The resolution needs to go to 512x512 rather than match my original? I tend towards using taller images rather than square.

[ornate elk]

When you select inpaint area to "only masked" the resolution then is not the image output resolution. Instead its the resolution of your mask that get inpainted.
Inpainting is like generating an image into an image.
So thats why you set it to 512x512 as the 1.5 based models are trained on that resolution. 768x768 works too.

[ornate elk]

Can you send a screenshot of the whole cmd?

[steady thistle]

Thank you! I'm sure that's my issue. 🙂

[ornate elk]

Np, so yes you can use non square images as input but the inpaint area needs to be a square.
The output will have the same resolution as the input

[steady thistle]

This is really cool. lol

[hazy wigeon]

i literally turned it on & off again and its fine now... but here's another error when i try to use SAM: >> OSError: [WinError 1314] A required privilege is not held by the client: 'D:\stable-diffusion-webui\extensions\sd-webui-controlnet/annotator' -> 'D:\stable-diffusion-webui\extensions\sd-webui-segment-anything\annotator'

[ornate elk]

What's SAM?

[hazy wigeon]

Segment Anything

[ornate elk]

Ahh

[ornate elk]

You may need to change the read and write permission for the extensions folder.

[hazy wigeon]

i see

[valid gorge]

Performed a super simple (and non-scientific) test of the speeds I'm getting with my current system in order to have some comparison data for later when possible getting a dedicated GPU for SD.

Would be interesting to hear your results on something similar. The prompt I used and details can be found in the picture.

Especially interested in RTX3060 12GB, RTX4060 TI 16GB and similar cards.

[vocal burrow]

feels like it might interrest you https://vladmandic.github.io/sd-extension-system-info/pages/benchmark.html
It's vlad/sd.next data, so not auto1111 but close enough.

[valid gorge]

Looks very interesting, thanks!

This is very interesting! Looks like Linux in general increases the speeds significantly

[vocal burrow]

depends of the hardware, for AMD definitely yes.

[valid gorge]

I'm specsing new system, thinking of getting Ryzen 5700x so I guess I need to install Linux too

alongside windows that is hehe

[steady thistle]

I have 12gb vram on this: Processor AMD Ryzen 9 5980HX with Radeon Graphics, 3301 Mhz, 8 Core(s), 16 Logical Processor(s), Radeon RX 6800M GPU with 12GB GDDR6... I wish I had Nvidia for this stuff.

But I don't want to touch Linux because just no, I've been there and don't want to go back. If you're up for it, maybe you can do better than my results.

[valid gorge]

I've been more or less my whole life on mac and I feel like a total alien on Windows but I'm willing to test both haha

[steady thistle]

Oof! lol... good luck!

[valid gorge]

Thanks haha

How your GPU compares to my M1 Max @steady thistle ?

[steady thistle]

I don't know enough to really say.

I just know I have to fight with medvram or lowvram when I have 12gb of vram, because I'm using Radeon.

[valid gorge]

Sounds like a hassle 😦

[glad fiber]

still needing help here #🤝｜tech-support message

[calm wasp]

how do i assign more ram to stable diffusion on my computer?

[valid gorge]

VRAM?

As far I know, you can't assign more system memory unless you run SD completely on your CPU (guessing it will be slower)

[calm wasp]

i dont know im not a tech expert

just wondering how i can make it run faster

[valid gorge]

You could experiment with different boot arguments

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings#all-command-line-arguments

[calm wasp]

i am running it on a 4090ti and it still is really slow despite not even using an xl model

[valid gorge]

I'm on a Mac and use these: --no-half --upcast-sampling --no-gradio-queu

[valid gorge]

Oh, that GPU should be a killer

[calm wasp]

agreed but the generation speed is trash and im incapable of figuring out why

[valid gorge]

How slow is it?

[calm wasp]

3 images take half an hour to generate

i feel like there are some setting i have no idea about

*settings

and my keyboard is fukin broken

@valid gorge what settings should i adjust to increase generation speed but keep the quality at the same time?

[valid gorge]

Your settings look good to me. I personally don't use hires.fix when generating batches of base images I then take to img2img and tweak more

[calm wasp]

the problem is that my gpu usage never goes above 10% and i feel like i dont use its full potential

i might have to look into some tutorials

[valid gorge]

Hope you find the solution. I'm guessing could it be the drivers or GPU settings? (I don't know how Windows works at all sorry)

I usually start with generating base images in lower settings and then tweak them later to add detail and resolution.

[calm wasp]

thanks for helping i will look into it

[valid gorge]

Good luck with it, let me know the solution later when you find it

[thin quarry]

since you are using sdxl - you can gen at 1024x1024, no highres fix and since it's euler a, you can just go somewhere around 25-40 steps

you can upscale later only images you actually like instead of doing highres on everything

[valid gorge]

If you are only able to generate 3 images in an hour, there must be something weird going on. I'm able to generate 1024x1024 base images using SDXL checkpoints with couple LoRAs at 30-40 steps, taking a bit over a minute each.

For example I just generated an image with SDXL together with couple LoRAs and two ADetailers at 836x1024px, took me 1 min 13 seconds

@calm wasp Check this video to get the idea, https://youtu.be/RiN-zrUlneQ?si=0hZLhiAdQ5NN1Tbx

[muted halo]

hi i have a similar question, using my 1660super im only able to generate about 1 picture evry 6-10 minutes is that something i should be worried about?

[calm wasp]

👍

[grand fable]

hi, I tried to use TurboVisionXL and it shows some errors and took quite a lot of time to generate.. Could you help me what to do?

[grand fable]

adding more info - it's on M1 macbook using ComfyUI

[valid gorge]

Runs ok on my Mac Studio M1 Max, not a blazing fast but ok. A1111 feels faster though.

My Macbook Air M1 (the base model with added system memory to 16gb) simply is too slow with A1111 so I did not bother to try Comfy

[lucid niche]

After installing everything i get this..? If someone has a fix please dm me

[high wave]

what I want to know is.... How long before mAI shit breaks and I can't play video games anymore? lofl

[grand fable]

don't you know, what the errors are?

[fair oxide]

hello, I'm using automatic1111 web ui and none of the buttons are working. They literally do nothing. I have installed the web ui in a python virtual environment with version 3.10.6. I've also selected the v1-5-pruned-emaonly.safetensors [6ce0161689] checkpoint.

What else do I have to do? Do I have to install a model? There's already a Stable diffusion folder under the models folder, so I guess i don't have to go to huggingface and get one?

EDIT: I should also add that I've tried restarting the whole virtual environment as well as rerunning the webui-user.bat

[valid gorge]

No sorry, I don't have no idea. I suggest you try reinstalling it from the ground up. Make backups of your models, loras etc. so you can put those back later. You can follow the official guide, first homebrew, python etc and then SD with A1111 / ComfyUI

[astral comet]

Hi guys! My gpu is quite outdated and i'm looking torwards cloud GPU services, any recommendations? I prefer paying monthly than by the hour/image

[high wave]

learning limits....

[hot spindle]

Not sure if Colab is still an option, I know they cracked down on free tiers but I believe it's ok to run on subscription, it's 12eur/monthly that gives you credits where you can selected higher RAM machines, and use T4s or A100, when you deplete the credits you can still use T4 but with a lower 12GB RAM system

I used it often to Render in Blender but canceled my subscription last month, still can render there but smaller project's

[ornate elk]

whats in your webui-user.bat?

also 40 hires steps is to much, you waste time going over 20

[ornate elk]

whats in your webui-user.bat?

[ornate elk]

you installed the webui the wrong way (by downloading it as .zip)
Follow my install guide in the Pinned Message of this Channel

[lucid niche]

thanks, but after installing a bunch of stuff i got it working in my browser

[astral comet]

The tutorials I see online have SD run THROUGH colab, can I use it to "boost" my pc? I like the local automatic111 interface and have already downloaded a bunch of models, loras, etc

[ornate elk]

make sure your webui is whitelisted in any browser adblocker

[ornate elk]

you still are using an old version and you cant update it by the way you installed it

[lucid niche]

oh

Ive been here all day trying to make it work and now i hear this lmao

[ornate elk]

sry for beeing late ^^, but installing it isnt much of work. my guide covers all steps

[lucid niche]

will it work if i already have it working?

[ornate elk]

Google Collab, Hugging Face, Paperspace, Runpod, are the most used

[ornate elk]

yes

[lucid niche]

alr i will give it a try

cuz this is definitely not what i wanted 💀

[ornate elk]

you have to edit your webui-user.bat
at the line Commandline_ARGS=
you have to add: --xformers --medvram --no-half-vae
Then save and relaunch

[ornate elk]

after the installation you also need to download better models

[high wave]

ik https://www.perplexity.ai/search/how-to-set-55WRl5T2Rm.E9rwv0pA3Dw?s=u

[lucid niche]

i tried adding this epicrealism thing but i got it wrong i think

[ornate elk]

no not split size lol

[high wave]

lol

[ornate elk]

the other commands make your gpu use less vram and generate faster at the same time

dont use split size stuff, that wont do much

[high wave]

ty, i may try that

[fair oxide]

i am stuck here and my pc is very slow (amd gpu, directml version)

[ornate elk]

you installed the webui the wrong way (by downloading it as zip file)
Follow my Install Guide in the Pinned Messages of this Channel

[fair oxide]

thanks ima try

[lucid niche]

where am i supposed to click?

[fair oxide]

where the path is

[lucid niche]

yeah?

[fair oxide]

wait

[fair oxide]

yes

there

type cmd and enter

[fair oxide]

you have to delete

what was in there

then type cmd

[ornate elk]

yes only cmd should be in the path

[lucid niche]

alr alr ty guys

[lucid niche]

is this correct if i click on save

[fair oxide]

yes

just follow the instruction

[ornate elk]

you need a space

between --no-half-vae and --medvram

[lucid niche]

thanks!

thank you guys a lot for your help

[fair oxide]

is it normal that this step takes like already 5 mins?

[lucid niche]

yeah bro it took me more than that the first time i did it

i was worried too but everything was okay, just give it time

[fair oxide]

thanks

[zealous cargo]

is it normal that my first generation on epicrealism model takes like 20-30 seconds but the next generation takes like 1 minute or more?

[fair oxide]

did you change the prompt

or any settings

[zealous cargo]

no

[fair oxide]

it can have many reasons but its not a big deal some generations just take longer

as long is it works

[ornate elk]

you have paused the webui (by clicking in it) press space to unpause it

[fair oxide]

oh okay

thanks

[zealous cargo]

yea its just annoying

[ornate elk]

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

[zealous cargo]

1650 super, --xformers --medvram --no-half

[fair oxide]

can i skip the automatic model download i already downloaded some

[ornate elk]

yes if you place a model inside models/stable-diffusion it will be skipped

but you need to relaunch the bat

[fair oxide]

nah then idc

[fair oxide]

nvm solved it. Turns out you cannot have the web ui installed in a path that contains a folder whose name begins with a dot. I had named my virtual environment folder .venv so that messed it up.

[ornate elk]

oh yea that wont work ^^

[fair oxide]

its stuck

[ornate elk]

you paused it again

[fair oxide]

no

nothing happens if i press space

[ornate elk]

okay, then close and restart the webui-user.bat

[fair oxide]

is there a sampler u recommend for amd or in general?

[ornate elk]

euler a is good and fast

also DPM++ SDE or SDE Karras are good

[fair oxide]

thanks

[hot spindle]

I'm biased because I mostly work with photos but my 2cents: Euler a I associate mostly with illustration, strong lines, somewhat blurred midtones, SDE Karras looks like the opposite spectrum

[fair oxide]

nah 💀

[ornate elk]

ERROR: Could not install packages due to an OSError: [WinError 5] Zugriff verweigert: 'D:\stable diffusion\stable-diffusion-webui-directml\venv\Lib\site-packages\cv2\cv2.pyd'
*** Check the permissions.

your Stable-Diffusion Folder need Permissions

[astral comet]

Ok i'm confused, does google colab GPU Computing work across your local pc, so it virtually replaces my physical GPU, or do I need to run everything through the Google Colab interface to use the cloud GPU?

[fair oxide]

how do i give permissions?

[fair oxide]

everything runs in the cloud

[astral comet]

Ok but can I use my local Automatic111 UI/folder/files or do I need to use their UI and start from scratch

[fair oxide]

but the extension is installed now and works

[fair oxide]

you need to do everything from scratch

[ornate elk]

right click the Stable-diffusion-webui-directml folder then on advanced settings, security you have to click edit. then give yourself and System every permission

[fair oxide]

its like a pc you can control in the cloud that has no access to your pc

[fair oxide]

maybe its because my windows user profile is not an admin

[astral comet]

@fair oxide Any alternative? I'd rather subscribe to something that just runs a better gpu on my pc, if that's a possibility

[ornate elk]

yea that could be it

[ornate elk]

there is only your local gpu to generate or using gloud gpu services to generate

[astral comet]

So all cloud gpu services have their own UI, correct?

[fair oxide]

yes

[ornate elk]

no, the most are using auto1111 as ui

but they are not connectet to yours

google collab can use gdrive to upload models

[fair oxide]

google collab uses notebooks

[astral comet]

Aight. Which ones use auto111?

[fair oxide]

there is a plattform called rundiffusion (https://rundiffusion.com) wich offers like every ui there is

but it costs

50 cents/hr

but you only pay when you use it

and you have 30 mins of runtime for free

[astral comet]

Aight

Thanks!

[fair oxide]

they even offer audio generation ai and stuff like that

but i dont if this is the best solution

[fair oxide]

you could look up yourself and compare them

[hot spindle]

Its always remote, but you can use built in Gradio remote in A11111 so you access it in your local browser

Drawing a parralel I use my work much better computer to render SD while using my home computer browser

[fair oxide]

i always get this but it still works ´fatal: No names found, cannot describe anything.´

[ornate elk]

that can be ignored

[umbral plover]

Is it normal that in a1111 between clicking generate and actual start of generation it does something (nothing in console) for 20-30 seconds? Didn't change any thing, rerun same parameters

[ornate elk]

If your models are stored on a HDD at this time it loads them into the vram of the GPU

[umbral plover]

Hmm, i have ssd only... Maybe I could somehow put it into RAM? I think it would be a little faster

[ornate elk]

On SSD it should work fine. If you have more than 16gb of ram you can store multiple models on Ram for fast switch

[umbral plover]

I have 32 so it should be enough for 1-2 models

But I didn't touch model between generations and each run it takes 20-30 seconds thinking about something but I don't see anything in console

[ornate elk]

Hmm thats strange then

30 seconds is pretty long

Any extensions that may do something?

[ornate elk]

make sure to whitelist the webui in any browser adblocker

Idk but you could try disable Lobe Theme and webui-state and lcm extension then reload and test again

[umbral plover]

Will do and come back (Didnt know about adblocker - can you explain why?)

[ornate elk]

because sometimes they block javascript and other stuff on websites, even when the blocker shows 0 blocked sites, as the webui dont have any ads.

the webui sends the settings to the webui cmd and the adblocker is in between, so that often caused problems in img2img tab for some users

using opera gx vpn can cause also problems

[umbral plover]

Oh, maybe I should've mentioned that I use sdxl_turbo model, managed to shorten that stale time to 15-20 secs, disabled preview completly but not sure if it did impact that time or something completly different

[ornate elk]

Preview can impact time

[fair oxide]

is there any way to speed up openpose? a 1 minute generations takes like 35 with openpose

[ornate elk]

You need to use the smaller Controlnet models

https://civitai.com/models/38784/controlnet-11-models

[torn lotus]

I am getting general system crashes while generating with A1111. I got an out of memory error in my terminal.
On deb12

[fair oxide]

okay

[ornate elk]

Amd or nvidia GPU?

[torn lotus]

nvidia 1070ti

I have a 8G swapfile. But I think linux is mismanaging VM

[ornate elk]

Yea on Linux you need more ram or a bigger swapfile

[torn lotus]

more vram?

[ornate elk]

32 would be good. If you only have 16 then you need a 20-30gb swapfile

[fair oxide]

why does it take shorter time when i dont check low vram?

[torn lotus]

I have 32G ram

I'll increase to 20G

[ornate elk]

Because if disabled it dont splits the load it takes on the vram fron controlnet

[fair oxide]

okay. the results are bad too

[ornate elk]

That doesn't depend on lowvram or not

How you define bad?

Do you generate a full body pose in a 512x512 image?

[steady thistle]

@ornate elk I'm having trouble with inpainting again. I am trying to improve the blade of a sword, I got the hilt fixed and now I want the blade etched. But what should I set my denoise, scaling and CFG scale to?

[ornate elk]

I dont know you the perfect settings. You have to try.
But use 512x512 or 768x768 resolution

[steady thistle]

I'm using 1024x1024 on an sdxl turbo model

[ornate elk]

Ah okay

So you inpaint with a Turbo model?

Wouldn't recommend

As it always needs low cfg and steps

[steady thistle]

Okay, that's probably the issue.

I'll use a normal sdxl model for it and see if that fixes it. Thanks yet again for the help. I really appreciate it

[ornate elk]

No problem, hope it works 🙂

[steady thistle]

Probably high denoise tho?

[ornate elk]

The lower the denois the less will change

[steady thistle]

Gotcha.

There's still a blade on it, so I just want to improve it, so maybe low denoise then. I'll try both with the new model

[ornate elk]

Yea lower would be good, try 0.5 and see what it does, then 0.3 etc

[fair oxide]

is there any way to get better resolutions without hires fix, my gpu cant use it

[ornate elk]

What's your GPU?

[fair oxide]

radeon rx580 i guess but ima look

[ornate elk]

Okay that GPU can't use hires fix.
You would need to use the SD upscale script in img2img to upscale.

[fair oxide]

where can i download it?

[ornate elk]

Its already included

In img2img at the bottom under scripts

[fair oxide]

okay

is there any way to automate this?

[ornate elk]

Yes you can click on Batch input

Then select a path to a folder with images you want to upscale

[fair oxide]

and when i generate an image its automatically being upscaled?

[ornate elk]

No

If you generate an image in txt2img you can click on the little painting with frame button. To send the image to img2img

[fair oxide]

its not a big deal either its one button press

uhm https://cdn.discordapp.com/emojis/1100390821435555870.webp?size=48&name=fr&quality=lossless

when i click generate

[ornate elk]

@fair oxide
To upscale in img2img do this:
Put an image into the img2ing tab by drag and drop or from the button in txt2img.
Then use some negative tags.
Set the resolution to 512x512 (thats not the output resolution)
Then set the Denois below 0.4
At the bottom select the SD Upscale script.
Then select an upscaler like Resrgan4x+
Then hit generate.

[steady thistle]

I've worked on the inpainting all this time. I decided to try Inpaint Anything and see if that solves my issue. Crossing fingers as I download models!

oh lord, the models.

so many gigs

[steady thistle]

@ornate elk Welp, how about this error with inpaint anything? 2024-01-13 19:57:43,239 - Inpaint Anything - ERROR - The size of tensor a (0) must match the size of tensor b (256) at non-singleton dimension 1

[solemn moon]

i feel like ive seen this b4, but not ringing a bell

[solemn moon]

if you post a screen shot of the setup i can try help

[steady thistle]

I'm downloading more files on Inpaint Anything, not sure if it will help, but it's supposed to simply be... drop in the image, pick the model and hit Run Segment Anything. But that is where it fails.

Maybe it's the wrong size image. I loaded the large and huge but not base. Just discovered this was a thing. heh

[solemn moon]

Just discovered i use inpainting a lot and still havent tried Inpaint-Anything

😂

[steady thistle]

I was fighting inpaint regular and figured I'd try to work around it. heh

All this is just to learn about it all anyway, so it's useful

[steady thistle]

Okay I said, I can just try to sketch with the inpaint. Great. No problem. Then I get: IndexError: tuple index out of range

What the hell is a tuple and why isn't the index in range of it?

I think they're just making stuff up now. lol

[worldly rain]

guys the result disappears when I generate an image

OutOfMemoryError: CUDA out of memory. Tried to allocate 2.02 GiB (GPU 0; 3.00 GiB total capacity; 6.05 GiB already allocated; 0 bytes free; 8.32 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

it says under it

[warped bane]

Can´t load animateDiff into SD. Could somebody help ?

[vocal burrow]

You're running out of vram/hitting the limit of your GPU. How to fix it (Nvidia edition) :
1/ If you're using a 2000/3000/4000 RTX series cards then you can try to use the --opt-sdp-attention command line argument. Otherwise use --xformers argument. To do so, Edit your webui-user.bat and append it to the set COMMANDLINE_ARGS= line. Please note, if you're using vladmandic's fork then you won't be able to use--xformers.
2/ Also throw --medvram in there if you've got a GPU with less than 8gb of vram, --lowvram if less than 4gb. (each argument must be seperated by a space)
3/ If you're using --precision full, --no-half and don't know what they're doing. Then you probably don't need them and therefore they're increasing your vram usage for nothing. Remove them. Notoriously --no-half is necessary for GTX1600 cards.
4/ Make Batch size lower
5/ Output at a lower resolution (and eventually upscale later on with hires.fix or img2img feature)
6/ Write less complex prompt
7/ Use lighter model (use fp16 pruned variant of your model if available)
8/ If you're using a VAE and your generation crash at the very end. You should try to install multidiffusion-upscaler-for-automatic1111 extension (available in the default extension list) and enable the TiledVAE features (cf screenshot)
https://media.discordapp.net/attachments/1002602742667280404/1111097170150563881/image.png?width=810&height=228

PS : keep in mind that SD1 models are trained with 512x512 images, SD2 models with 768x768 and SDXL 1024x1024. So stick close to those resolutions before doing any upscaling,

[vocal burrow]

there are probably more meaningful errors above in the log

[worldly rain]

hello, thank you so much for the detailed answer, its much appreciated

I have a gtx1060 3gb, I already have --lowvram and --xformers added

I also have --no-half there so I'll remove it now

I use a 1.5 checkpoint model

[vocal burrow]

should be good with those settings. Do you still have issues ?

[umbral plover]

Why I don't get a preview image using sdxl turbo model?

[vocal burrow]

by default the live preview is actualized every 5 steps and you're supposed to use sdxl turbo with 1 step.

So there's no time for the preview to kick in

[umbral plover]

Oh, okay, is it a good idea to try to decrease that refresh time (if its possible)

[vocal burrow]

no, what would be the point of getting a preview at step 1 when you can already get the finalized result ?

Also preview comes at a price, the more you ask for it the slower your generation will be.

(and the more you risk getting out of vram error if you're on the edge)

[umbral plover]

You can get decent results on 1 step? Usually i have 4-6

[vocal burrow]

for sdxl turbo yes, it's made for 1 step gen and sub 1 cfg

Use this extension (read its doc) to use sub1 cfg in auto1111. https://github.com/seshelle/CFG_Rescale_webui. Otherwise setting the cfg at 1 should be enough to get ok results.

[vocal burrow]

clip skip 1

[umbral plover]

Still the same

[vocal burrow]

512X512 ?

[umbral plover]

Yeah, set COMMANDLINE_ARGS= --autolaunch --no-half-vae --theme dark --xformers --medvram the only thing different from yours i think is model

[vocal burrow]

where did you get yours ?

[umbral plover]

https://civitai.com/models/139562/realvisxl-v30-turbo

Tried different samplers and different CFG rescale values and nothing changes

[vocal burrow]

Just tested the model, I got similar results.

So that's not a proper sdxl turbo model and more of a weird merge. I'd have to investigate more to understand what is going on exactly

[warped bane]

this is all i got

[umbral plover]

Okay, im downloading proper sdxl turbo but its slow as turtle so it will take a moment

[ornate elk]

for merged turbo models you need more steps

the creator of yours used 8

[ivory bramble]

Anyone use the Config Presets extension?

by Zyin055

[umbral plover]

Okay, proper one works nicely 10-14s per image. tried also normal sdxl with lcm but my god, 6 minutes for one picture ;D

[vocal burrow]

that's pretty terrible perfs. What's your gpu ?

[umbral plover]

1060 6gb

[vocal burrow]

1/ uninstall the extension
2/ delete your venv
3/ reinstall installation and monitor/save the logs for error during installation

[vocal burrow]

ok and what settings did you use for the generation ?

[umbral plover]

Turbo or SDXL 1 ?

[vocal burrow]

for the sdxl

6 minutes seems like a lot but I'm not used to 1060 so who knows. I just want to check that you're not using over the top settings.

[umbral plover]

cute cat lora:lcm-lora-sdxl:1
Steps: 7, Sampler: LCM, CFG scale: 1, Seed: 688875665, Face restoration: CodeFormer, Size: 1024x1024, Model hash: a928fee35b, Model: albedobaseXL_v20, Lora hashes: "lcm-lora-sdxl: 2fa7e8e56b09", Version: v1.7.0

[vocal burrow]

ok I guess 1060 really sucks regarding sdxl then ^^" sorry

[umbral plover]

Well, vram, from what i read when i upgrade i should aim at 3060 12gb

[umbral plover]

Btw, when using turbo in inpainting should i up cfg scale?

[wind sun]

Yo guys is there any online paid courses for stable diffusion

[stuck knot]

How do I solve this Error with Segment-Anything, I have tried reinstalling Segment-Anything node multiple times, and I have even tried reinstalling ComfyUI Manager to see if it is the problem?

[languid ledge]

Hi i wanna know if its possible to batch face inpaint using roop

cuz stable diffusion doesn't understand the image if i dont focus on the face

[ornate elk]

Does roop still works for you?

[languid ledge]

yep

[ornate elk]

Because most people use Reactor as its more updated

[languid ledge]

reactor?

[ornate elk]

Its a face swap extension like roop

[languid ledge]

can i get a link for that?

if you dont mind

is it easier to use tho compared to roop

[ornate elk]

Yes its very easy

Let me get the install guide

[languid ledge]

if so can you batch face inpaint?

[ornate elk]

Face inpaint?

[languid ledge]

yep

im trying to face swap and image but it doesn't work

only works when i inpaint it

it has some distraction on the side i think

[ornate elk]

Can you show an example?

Here is the site of Reactor:
https://github.com/Gourieff/sd-webui-reactor#features

[languid ledge]

sure lemme go find one

[languid ledge]

i can't find an example but uh just lemme explain so the ai can't detect the faces as the other object/face lookalike that ai thinks is face has a higher priority. so im just inpainting the face to let ai understand which way they should look at

does that make sense

[ornate elk]

hmm idk if i understand it, but inpaint a face and then faceswap is not the right way ^^

[languid ledge]

yk when you inpaint something, ai focuses on that inpaint part right

so i just want ai to focus on that face rather than other objects

[cloud topaz]

hey is it possible to face swap from a custom image with a custom face with foocus?

[languid ledge]

if you get what i mean

[ornate elk]

yes but only if you set inpaint area to "mask only"

[languid ledge]

yea

so is there a way to batch inpaint like that

[ornate elk]

not with fooocus sry

but with auto1111 or comfyui

[cloud topaz]

ohhh okay thanks you saved me lot of time ahah

i dont know if i am allowed to say this here but i am trying to face swap with comfy ai and get consistent realistic results to make an ai influencer
if u can help me i can pay you (sorry if its not alloweed to say here)

[ornate elk]

yes there is the batch option in img2img

[languid ledge]

yea but what if i want to use batch + inpaint

[languid ledge]

what am i suppose to do with that?

do i go inpaint sketch?

[ornate elk]

batch means to use multiple images. so you need multiple input images + multiple inpaint masks, then these would get applied after another

[languid ledge]

do i have to inpaint manually?

[ornate elk]

if you just want to faceswap you should use Reactor

[languid ledge]

okay will do

thanks

[ornate elk]

no problem, roop is also outdated and wont get updates, for reactor its needed to disable or delete roop.
Follow the Install instrucstions from the github link carefully to get it working.

[languid ledge]

okay ill let you know if i have a problem

i will install it tommorow

how do you delete an extension again?

[ornate elk]

sure, you can just delete the sd-webui-roop folder from the extensions folder

[languid ledge]

okay

thanks

[ornate elk]

no problem 🙂

[ornate elk]

cant help with comfyui sry

[astral comet]

ok, so, i'm able to generate good images in low res, then I use high res fix to create my final images, but the results are always hit and miss. extra limbs, fucked up clothing, etc that isn't in the lower res image. I know this is a broad question, but where should I start/what should I be looking at to get better/more consistent results?

[ornate elk]

that depends on your denois strenght and upscaler

the lower the denois the less will be changed

start with denois 0.5 in txt2img or denois 0.3 if you upscale in img2img

[astral comet]

@ornate elk thanks! How would I go about finding the best upscaler? Does it depend on the model or are there universally better ones

[ornate elk]

It depends on the model. For realistic images 4x-UltraSharp is good

For Anime images is Fatal Anime 5000

You can find different upscalers here:
https://openmodeldb.info/

Download the .pth file and put it into models/Esrgan folder

The upscalers dont fix deformations like multiple limbs etc. Thats fixable with a lower Denois

[astral comet]

@ornate elk ok, in other words I need to find the denoize sweetspot

Do "hi res steps" matter in any way?

I always leave it at 0

[ornate elk]

Yes, don't use 0. if you leave it as 0 it will use the same amount as normal steps. So your wasting a lot of time.
Because mostly 10-15 are enough.
If you have a fast GPU you can go up to 20

[steady thistle]

I'm still getting Inpaint Anything - ERROR - Cannot set version_counter for inference tensor

Might be a python issue?

[shy edge]

Does anyone know what this means?
got prompt
ERROR:root:Failed to validate prompt for output 160:
ERROR:root:* (prompt):
ERROR:root: - Required input is missing: images
ERROR:root:* VHS_VideoCombine 160:
ERROR:root: - Required input is missing: images
ERROR:root:Output will be ignored
ERROR:root:Failed to validate prompt for output 488:
ERROR:root:* (prompt):
ERROR:root: - Required input is missing: images
ERROR:root:* VHS_VideoCombine 488:
ERROR:root: - Required input is missing: images
ERROR:root:Output will be ignored
Requested to load SD1ClipModel

It is driving me nuts. I get output, but I see this in the logs and it makes me wonder if something is not quite right.

[fair oxide]

Just throwing this hugging face link here for convenience https://huggingface.co/uwg/upscaler/tree/main as I've noticed downloads at openmodeldb often takes way too long.

[warped bane]

If everything in life would be as easy as this. Deleting the venv folder helped. Thanks a lot👍

[little echo]

hey, after recent update apparently several lora's won't show/no longer usable in A111 webui.
one of which is LCM weights. any suggestions on troubleshooting?

okay, i think in the settings actually it was set to hide if incompatible ver.
is there a possibility they may still work fine?
also is there a way to skip updates/roll back--it seems to force it before launching
i realize these questions may be specific to A1111 build, sorry if it's an inconvenience

[fair nacelle]

helpe

[ornate elk]

What do you mean by not usable?

[ornate elk]

Pls a full screenshot

[ornate elk]

Python version to new

[fair nacelle]

can i have both (old and new)

[ornate elk]

Do you need the new one?

[fair nacelle]

nah

[ornate elk]

Then uninstall and installing the right version would be better

[fair nacelle]

on it

[ornate elk]

Also use my install guide in the Pinned Messages of this Channel

[fair nacelle]

my sd is already installed

[ornate elk]

Ok still check it for the best performance arguments

You also have to delete the venv folder after installing python

[fair nacelle]

right sec

lemme try the run.bat

this technically launcher the webui

but like..everything is loading

im installing torch now

[ornate elk]

idk what guide you followed, but looks like it wasnt a good one

[fair nacelle]

the github one?

[ornate elk]

make sure to update the extension, controlnet has an error

[ornate elk]

the github one doesnt say to use --diable-safe-unpickle

[fair nacelle]

well thats what i tried to avoid the previous error msg

[ornate elk]

also it doesnt say to have a folder called webui

[fair nacelle]

it installed itself

[ornate elk]

i dont think you installed it the right way

[fair nacelle]

its plug and play there aint much for me to choose

click the exe

and it goes

[ornate elk]

yea that means you used an other installation tool rather then the "offical way"

[fair nacelle]

https://github.com/AUTOMATIC1111/stable-diffusion-webui

i use this one

[ornate elk]

i dont see any .exe there

[fair nacelle]

it was rethorical

[ornate elk]

maybe you followed this, but thats still not the recommended way to install it

[fair nacelle]

this

i think

whatever of those two was just click and let it do its thing

[ornate elk]

that would be okay, but then your folder wouldnt have the name webui, instead it should has the name stable-diffusion-webui

[fair nacelle]

now im even more fked

[ornate elk]

i would still recommend following my install guide, as you wont get any problems later with that.
The guide is in the Pinned Messages of this channel.
just follow that and delete the old stuff

[fair nacelle]

i wont

anyway

to add the python to path

do i need to include the .exe as well at the end of it

or just the older

(for some reason the fkin installation didnt add it self)

nvm its there but says it isnt

[ornate elk]

you can edit the webui-user.bat and therer add the path to python.exe

[fair nacelle]

microsoft store python worked instead

fk sake now what

[ornate elk]

You can use --port to define a port in the webui-user.bat COMMANDLINE_ARGS

[fair nacelle]

like this?

Finally

im fking it

thanks..

is there a way to bulk update the models? been a while since i touched them n that

[ornate elk]

Nope

You can either download them manually or use the civitai browser + extension

[fair nacelle]

ty

[ornate elk]

np

[fair oxide]

did they add cross-compatibility between sd 1.5 and xl so 1.5 loras and embeddings can be used with XL models?

x-adapter proposed that idea to offer universal pluggin support, and i just rendered an image with XL model while using 1.5 lora and embeddings

no errors, no warnings.

[fair nacelle]

bruh what are theese

@rustic otter

[rustic otter]

you need to put a space before --

and dont put - - its --

[fair nacelle]

well thats what im asking

[rustic otter]

thats why i said copy paste

[fair nacelle]

running now

[fair nacelle]

ffs now i dont have ports again

[rustic otter]

just remove precision and nohalf then

[fair nacelle]

i dont wnt to

how do i give it 1-20000000 fking port range?

so it doesnt bother me again

[rustic otter]

idk i dont use that command lmao

[rustic otter]

this is all i got

[fair nacelle]

im in

[ornate elk]

What's your GPU?

[fair nacelle]

after adding some of ur args

[ornate elk]

Remove --precision full

And then make sure the model you try to use is larger than 1.98gb

[fair nacelle]

and if it isnt

[ornate elk]

Then its not a model and shouldn't be in the models/Stable-diffusion folder

Loras for example are small and go into models/lora

[fair nacelle]

sec

the quality is nice, but the photo color is kinda washed

im using vae-ft-mse-840000-ema-pruned.ckpt in vae settings

[ornate elk]

Make sure the VAE is in the models/VAE folder

[fair nacelle]

it is obviously othwise i couldnt have selecetd it

[ornate elk]

If the output is washed out. That means the VAE isn't applied

Check the cmd for any errors

[fair nacelle]

working fine

idk maybe im ass at promptin still

[ornate elk]

You can show an example

I can tell if a VAE is used or not

[fair nacelle]

its nsfw

i think it does

[ornate elk]

To compare

[fair nacelle]

i mean, it aint as bad as the first one, but it aint as good as the last one

i can dm u the pic

its not nude, but it is like.suggestive i guess

[ornate elk]

Okay sure

[zealous cargo]

where do you get vae's?

[fair nacelle]

hugginface i uess

[ornate elk]

Huggingface or civitai.com

[fair nacelle]

should loras be divided with , before or after

and should they be placed first or last

[ornate elk]

Yes separate them with ,

Doesn't matter where they are placed

Also adjust the lora strength by changing the :1 to :0.5

If it looks bad for example

[fair nacelle]

yea sure thanks

[proper pike]

Does anyone know how to use this workflow? It wont even start for me when I click queue https://civitai.com/articles/3256/updatedcybearpunk-vid2vid-hotshotxl-ipamultipurpose-lorastack-4cn

[onyx solar]

Look at your console and see what the message is.

[weak ore]

Do we know the price for the stability API

[errant ledge]

Why would a SD 1.5 model have LoRA-v1-CLIP keys? Is there a purposeful reason to do this?

[proper pike]

Where would this message be? Because there's no pop up

[onyx solar]

In the console window, if there is one. If you're only looking in the browser, you're not looking at the console.

The other thing to look for is for any node that is bypassed, muted, or red.

[proper pike]

https://pastebin.pl/view/c1c77172
the command line interface? Yeah nothing. I must be doing something wrong, but what? I did contrl b on all the nodes

[onyx solar]

Without seeing all the nodes and your console, no real way to know what's going on.

[onyx solar]

Well, on the last screenshot above, in the far left side of the screenshot in what is the Inputs (the gray/white group), the Load Checkpoint node in your screenshot doesn't have the MODEL or CLIP connections connecting to the CR Apply LoRA Stack node, so no initial model is being fed in.

[coarse galleon]

should I download the NVIDIA Studio Ready Driver?

[little echo]

there was a default setting to hide loras that aren't compatible versions
so i am guessing an SF update made them hidden
they seem to still work when turning that option off though

another question: are all the usable tags defined by each model?
if a tag is not defined in the SF model--does it get ignored?

[remote agate]

II have encountered an issue with automatic 1111, and I need help. I am unable to recreate images using the same seed and literally the same parameters as the original image. Every time I create a new image, it starts distorting to a point where blotches literally begin to appear. I have tried everything, setting "xformers" and Eta Noise Seed Delta (ENSD) to 31337 temporarily resolves the issue, but then it starts distorting completely again. Please, I would like to know if there is a solution to this, and I would be deeply grateful for any assistance.

[remote agate]

The results are the same if I don't modify the factory script to open stable diffusion on Windows

[ornate elk]

Not needed for SD but you can

[ornate elk]

Yea every lora that isn't compatible with the model will not be shown as 1.5 and sdxl have different loras to use.
Can't say how many words a model understand but it won't ignore words. The output will still change. Worst it can happen is it doesn't know what it is and then it isn't generated in the output.

[ornate elk]

What are your txt2img settings?

[ornate elk]

Remove that path to Python, its wrong

[remote agate]

The only thing I have modified is the webui-user.bat and the ENSD configuration, but other than that, I haven't modified anything regarding txt-img

[remote agate]

Do I leave it empty?

[languid ledge]

Hi guys i fked up when installing reactor

[fair oxide]

i reallly want to organize my loras but whenever i put them in folders loras disappear is there any other way like an extension or smth to organize loras?? or any way to fix them disappearing??

[languid ledge]

tried reinstalling this happened

[woven smelt]

Sometimes when im generating i get timages like this. If i change models ill get a real picture again, but change back to that model and its broken again. Persists through restarting SD entirely only PC retsart fixes. Any ideas whats going on?

[ornate elk]

Yes

[ornate elk]

Is the model 2.1 based or 1.5 based ?

[ornate elk]

The folders should work. Is your webui updated? Also make sure the webui is whitelisted in any browser adblocker

[ornate elk]

If its an older model try Euler a as sampler

[ornate elk]

Delete the reactor folder from the tmp folder

[coarse galleon]

i mean like are there any benefits of getting it?

or tensorRT

[ornate elk]

No benefit with the studio driver.
Nvidia has a TensorRT extension thst increases speed but isn't compatible with every SD function

[woven smelt]

Happened again, black screens this time. changing the sampler yo euler didnt change anything

[languid ledge]

as in ctrl + r and delete?

[woven smelt]

Its only this particular model that is having trouble but I cant understand why a computer restart is required rather than a SD retaart

[languid ledge]

i only got these

*** Error completing request
*** Arguments: ('', 'https://github.com/Gourieff/sd-webui-reactor?tab=readme-ov-file', '') {}
Traceback (most recent call last):
File "C:\Stable Diffusion\stable-diffusion-webui\modules\call_queue.py", line 58, in f
res = list(func(*args, **kwargs))
File "C:\Stable Diffusion\stable-diffusion-webui\modules\ui_extensions.py", line 620, in <lambda>
fn=modules.ui.wrap_gradio_call(lambda *args: [gr.update(), *install_extension_from_url(*args)], extra_outputs=[gr.update(), gr.update()]),
File "C:\Stable Diffusion\stable-diffusion-webui\modules\ui_extensions.py", line 357, in install_extension_from_url
with git.Repo.clone_from(url, tmpdir, filter=['blob:none']) as repo:
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\git\repo\base.py", line 1308, in clone_from
return cls._clone(
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\git\repo\base.py", line 1219, in _clone
finalize_process(proc, stderr=stderr)
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\git\util.py", line 419, in finalize_process
proc.wait(**kwargs)
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\git\cmd.py", line 604, in wait
raise GitCommandError(remove_password_if_present(self.args), status, errstr)
git.exc.GitCommandError: Cmd('git') failed due to: exit code(128)
cmdline: git clone -v --filter=blob:none -- https://github.com/Gourieff/sd-webui-reactor?tab=readme-ov-file C:\Stable Diffusion\stable-diffusion-webui\tmp\sd-webui-reactor?tab=readme-ov-file
stderr: 'fatal: could not create work tree dir 'C:\Stable Diffusion\stable-diffusion-webui\tmp\sd-webui-reactor?tab=readme-ov-file': Invalid argument
'

---

same problem

[ornate elk]

What model is it?

[woven smelt]

hentaiMixXLRoadTo_v40. I think it was some weird combination of tokens. I went back to a previous generation that worked then slowly added the prompts in and it worked, idk something weird in the prompt but seems ok now

[ornate elk]

Okay delete the reactor extension from the extensions folder and then delete the venv folder and relaunch

[ornate elk]

Okay still strange

[woven smelt]

I checked brackets, commas etc and added each prompt one by one until I got to the same prompt and it worked. Maybe some invisble cahracter messing with it

[ornate elk]

Maybe, I've got a similar issue the last days

[grand fable]

Hi, could you give me some tips about what HW is best to buy in order to use SD? I started with this few days ago and running SD on my M1 Macbook, but looking for some desktop with solely purpose for SD. As far as I understand it all depends on GPU, is there anything special like focusing on VRAM or some core frequencies etc.? And what about the other components? CPU, RAM, motherboard? Does it really matter whether I use Intel i3 or i9? How much RAM, 8, 16, 32gb? etc. Thank you a lot in advance for your tips 🙂

[vocal burrow]

The more vram the safer. Avoid AMD gpus as they're the underdog and therefore very few people optimize stuff for it, everything is gonna cost much more vram, some stuff wille be incompatible, etc. So yeah pick an nvidia gpu. And it goes without saying but don't waste money getting anything under 10XX. If you can, avoid 16xx series as they need --no-half => everything will use more of that precious vram.

8gb is the sweet spot if you're on a budget as you can do everything with minimal speed trade-off. You're mostly gonna have slow down on SDXL, heavy upscaling stuff and training.
12gb is the "I don't care spot" where you can do anything without any speed trade-off. But nothing is future proof, so maybe the future SDXXL or whatever will need 16gb to run comfortably.
Unless you plan to go deep in SDXL model training, anything beyond 16gb is probably a waste of money. Instead get a newer generation gpu if you can.

Other than that, cpu isn't used apart from transferring stuff to the GPU or trivial operations. And you'll need at least 16gb to use SDXL without hitting the pagefile / experiencing slowdown during model loading. So if you can, get 32gb anything beyond that is "look at me I have too much money" setup or someone with very specific need.

[royal turret]

hello comfyui users i have an error popping up and well i dont understand what im supposed to do exactly : File "C:\Users\Nishant\Downloads\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\python_embeded\Lib\site-packages\safetensors\torch.py", line 308, in load_file
with safe_open(filename, framework="pt", device=device) as f:
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

[umbral plover]

Hi, in comfyui the lower section of menu with manager etc. disappeard, anyone any idea why? //edit Reinstall of manager helped

[grand fable]

thank you for that extensive answer 🙂 so basicaly nvidia RTX 3xxx or 4xxx with at least 12gb better with 16gb to be on the safe side for future usage. 32gb system RAM. And for the CPU? Is for example some low end i3 ok?
re. the GPU - when I quickly check it, the 3060 ti 12gb is for 355usd and 4060ti 16gb is fo 550usd - is it worthy to go for the 4060 for the extra 200usd? is there any noticeable difference between the 3xxx and 4xxx generations or just the better VRAM?

based on this it looks the 4070 12gb has better score than 4060 Ti 16gb, how is that possible? https://cdn.mos.cms.futurecdn.net/RtAnnCQxaVJNYgA4LbBhuJ.png

[rotund lava]

128 bit bus memory on the 4060 plus more cuda cores

[ornate elk]

Noteworthy is that rtx30 cards have a higher energy consumption than the rtx40 generation.
For CPU go for 4 cores or 6, CPUs aren't that expensive (AMD Ryzen 5 5600X for AM4, or Ryzen 5 7500F for AM5 platform (newer).

AMD CPUs mostly consume less power than Intel ones and with AM5 Mainboard you can upgrade the CPU without changing the Mainboard until 2026.

[vocal burrow]

not much to add on my side

[grand fable]

ok so AMD is the way for CPU. Is it just for the power consumption? Or are there some other differencies?
re. the GPU is it worthy to go for the more expensive 4070 or 4060?

[umbral plover]

Is there any equivalent of styles.csv from a1111 in comfyui?

[woven smelt]

I managed to reproduce it. If I have a prompt of "multicolored hair with highlights" it goes balck. changed to "multicolored hair, hair highlights" and its ok

[ornate elk]

Okay very strange, so Its something with to much words or tokens without separation

[woven smelt]

something like that

[ornate elk]

SD doesn't care about the CPU that much. But its good for the system in general to have a good CPU. And less power consumption is always a big plus if you for example generate images the whole night

[karmic crown]

1660 SUPER is not bad for gaming but is not good for stable diffusion. my son has it in his PC, so I guess it is better to buy 10xx than this (at least for SD). I have two Macs, and they are also slow

but 30xx or 40xx is the safe way

even laptop with something like 3070 should be usable

[languid ledge]

the whole file?

i mean like whole venv file?

[zealous cargo]

yea for my 1650 it takes 30 seconds to 3 minutes or more sometimes

[mighty hare]

hey so ive downloaded A1111, and the images it's pooping out are really. well just not great

[languid ledge]

venv "C:\Stable Diffusion\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.5.1
Commit hash: 68f336bd994bed5442ad95bad6b6ad5564a5409a

Launching Web UI with arguments: --lowvram --xformers --no-half --no-half-vae
Traceback (most recent call last):
File "C:\Stable Diffusion\stable-diffusion-webui\launch.py", line 39, in <module>
main()
File "C:\Stable Diffusion\stable-diffusion-webui\launch.py", line 35, in main
start()
File "C:\Stable Diffusion\stable-diffusion-webui\modules\launch_utils.py", line 390, in start
import webui
File "C:\Stable Diffusion\stable-diffusion-webui\webui.py", line 44, in <module>
import gradio # noqa: F401
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\gradio_init_.py", line 3, in <module>
import gradio.components as components
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\gradio\components.py", line 55, in <module>
from gradio import processing_utils, utils
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\gradio\utils.py", line 339, in <module>
class AsyncRequest:
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\gradio\utils.py", line 358, in AsyncRequest
client = httpx.AsyncClient()
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\httpx_client.py", line 1397, in init
self._transport = self._init_transport(
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\httpx_client.py", line 1445, in _init_transport
return AsyncHTTPTransport(
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\httpx_transports\default.py", line 275, in init
self._pool = httpcore.AsyncConnectionPool(
TypeError: AsyncConnectionPool.init() got an unexpected keyword argument 'socket_options'
Press any key to continue . . .

when i try to open sd again when delete the whole venv folder

[zealous cargo]

why did you delete venv?

[karmic crown]

Version: v1.5.1
your a1111 is outdated

my advice for you, rename your current folder as sd-backup for example

install it from scratch using te guide from pinned messages of this channel

(you have amd and nvidia guides there)

move your models from backup to new installation

[mighty hare]

why do my photos look like they are from dall e mini lmao

they are really poorly generated and i dont know what to do

im using models, im copying prompts from prompt hero and they look like garbage compared to the photos

[languid ledge]

cs10 told me to

[mighty hare]

can someone help please

[languid ledge]

did u try other image?

i mean other prompt

[mighty hare]

yes

i have been trying many prompts

all of them look bad

[languid ledge]

maybe cuz ur model?

try different model

[mighty hare]

i have downloaded many models and all of them produce bad results

i downloaded sdxl 1.0, that looked bad

[languid ledge]

how to update

[solemn moon]

Do you have VAE SDXL? Took me long to realize you need the right VAE too

[mighty hare]

whats that

[karmic crown]

update from 1.5.1 to 1.7.0 can be problematic, so i suggest:

1. backup
2. new install
3. restore models from backup

[solemn moon]

Yea that’s your problem

Download SDXL VAE

[languid ledge]

ah fk.

[mighty hare]

i hate this so much, theres no instructions for this, just have to ask people haha

[solemn moon]

And wait bruh

[karmic crown]

some old extensions can cause problems, so this is usually faster than solving issues with old extensions

but you can try to upgrade if you want, but i would not recommend that path

[mighty hare]

how do i set the vae?

[solemn moon]

Google how to add VAE shortcut to main menu or something

[karmic crown]

go to setting, find user interface, than user inteface again

[solemn moon]

Bro I had this problem for so long lol

[ornate elk]

Yes

[solemn moon]

This is like kicking it to get it started for me 😂

[karmic crown]

and add those

[lost crescent]

Best SDXL Models include CounterfeitXL, DynavisionXL, NightvisionXL and UnstableDiffusersXL

[mighty hare]

i use automatic1111

[solemn moon]

Me too

[solemn moon]

Or this

[solemn moon]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/9703

Maybe that

[ornate elk]

Your webui is outdated

[languid ledge]

do i reinstal

[solemn moon]

Add update command to bat file

@languid ledge

[mighty hare]

where do i find the dropdown?

and where do i put the vae file

[solemn moon]

Restart and left hand side

Argh let me open mine

[mighty hare]

theres nothing on the left hand side

[solemn moon]

1min

Restart bro sheeesh

[languid ledge]

--update?

[mighty hare]

is this right?

[ornate elk]

Git pull

In an empty line

In the webui-user.bat

[languid ledge]

how do i even do that

[solemn moon]

Ahhh git pull

[languid ledge]

it just auto types when i start

[karmic crown]

i am afraid that some extensions will make a problem, and in the end new install will be probably faster

[fair oxide]

My webui should be updated, and I don't use adblock, are there any other reasons this may be happening???

[solemn moon]

git pull 2nd to last

[fair oxide]

I reeeaallly want my stuff organized because I have a lot of loras

[languid ledge]

do you mind uh copy paste to me

:)

[solemn moon]

@echo off

set PYTHON="C:\Users\88\AppData\Local\Programs\Python\Python310\python.exe"
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --xformers --precision full --no-half --lowvram --opt-split-attention --always-batch-cond-uncond --api
git pull
call webui.bat

[mighty hare]

how do i set the vae

[solemn moon]

dont use my command line args

[languid ledge]

its just git pull right

[solemn moon]

dwnld and place in VAE folder and restart '

[solemn moon]

yea

[ornate elk]

Can you show an example of the folder and the webui?

[karmic crown]

if you manage to start webui after update without crash, you will probably be fine, but as soon as it stars go to extensions tab and update all extensions

[mighty hare]

my prompt was "dog, realistic"

[solemn moon]

bro screen shot yr whole setup

[languid ledge]

yea no

[languid ledge]

oh i forgot to type space

[solemn moon]

dont see refiner bar vae

[solemn moon]

1024x1024 brrrooo

[languid ledge]

damn wtf why ur ui like that

[solemn moon]

see on the left "SD VAE"

[mighty hare]

i cant even see that

theres nothing on the left

[solemn moon]

gang shit bruv

[solemn moon]

weird

[languid ledge]

i didn't even press save as

idk why it wanted me to save as

[mighty hare]

there we go, i got it showing up

[solemn moon]

yeeaaa

[languid ledge]

ah i got it

[mighty hare]

its still broken

wtf

[languid ledge]

[solemn moon]

SDXL was trained on minimum 1024 images (i could be corrected)

change width and height

[mighty hare]

THERE WE GO

DOGGYY!!!!!!!!!!!!!!!!

[solemn moon]

also 30 steps for that sampling mathod works best

[ornate elk]

Sdxl models need a 1024x1024 resolution

[solemn moon]

imho

[mighty hare]

why isnt any of this like

made obvious

[solemn moon]

bro

you just into the tribe

[ornate elk]

What's your GPU?

[solemn moon]

1660ti

but

hear me out

[mighty hare]

dogwater dookie

[solemn moon]

everytime a senior has suggested something else it fails

[languid ledge]

yo shesh stable diffusion look so different

[solemn moon]

this worls to point its saved in my dropbox

[ornate elk]

Dont forget to update the extensions

And then restart

[languid ledge]

oke

[fair oxide]

Yeah one sec

[fair oxide]

does this work for SD???

[ornate elk]

Why --precision full and --lowvram and not --medvram

[ornate elk]

Click on lora tab

[fair oxide]

my lora folder

[fair oxide]

and wheni click on a directory, this is my faves

but there should be at least 50 loras in this folder and when i check there it, but it doesnt show up on sd??

[solemn moon]

changed it, didnt work, went by what Aryetis suggested once, and then someone else cant remember but it kept crashing

oh shit

code bro request

[remote agate]

1.5 stable diffusion

[solemn moon]

can Automatic get the same setup as comfy when setting how many pictures you want, increase batch size on Auto and VRAM fails, click generate 20 times on comfuy and you get 20 images.. just something i noticed

[ornate elk]

I see more .json than .safetensor files there

So how much loras in there?

[fair oxide]

i dont want to count all that 💀

at least 100 gigs of loras tho

[languid ledge]

*** Error completing request
*** Arguments: ('', 'https://github.com/Gourieff/sd-webui-reactor?tab=readme-ov-file', '') {}
Traceback (most recent call last):
File "C:\Stable Diffusion\stable-diffusion-webui\modules\call_queue.py", line 57, in f
res = list(func(*args, **kwargs))
File "C:\Stable Diffusion\stable-diffusion-webui\modules\ui_extensions.py", line 641, in <lambda>
fn=modules.ui.wrap_gradio_call(lambda *args: [gr.update(), *install_extension_from_url(*args)], extra_outputs=[gr.update(), gr.update()]),
File "C:\Stable Diffusion\stable-diffusion-webui\modules\ui_extensions.py", line 369, in install_extension_from_url
with git.Repo.clone_from(url, tmpdir, filter=['blob:none']) as repo:
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\git\repo\base.py", line 1327, in clone_from
return cls._clone(
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\git\repo\base.py", line 1236, in _clone
finalize_process(proc, stderr=stderr)
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\git\util.py", line 419, in finalize_process
proc.wait(**kwargs)
File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\git\cmd.py", line 604, in wait
raise GitCommandError(remove_password_if_present(self.args), status, errstr)
git.exc.GitCommandError: Cmd('git') failed due to: exit code(128)
cmdline: git clone -v --filter=blob:none -- https://github.com/Gourieff/sd-webui-reactor?tab=readme-ov-file C:\Stable Diffusion\stable-diffusion-webui\tmp\sd-webui-reactor?tab=readme-ov-file
stderr: 'fatal: could not create work tree dir 'C:\Stable Diffusion\stable-diffusion-webui\tmp\sd-webui-reactor?tab=readme-ov-file': Invalid argument
'

---

still error i guess

[fair oxide]

whats the difference btw

[languid ledge]

did install visual studio incorrectly

[ornate elk]

.json files are not Loras

Only .safetensor

[fair oxide]

hmmmm

so what do i do

[ornate elk]

You sort the folder for filetype and check how many .safetensors are in there

[fair oxide]

ah, 1012 loras in my character folder btw

[fair oxide]

hmmm alright

[solemn moon]

believe it or not, i get 1984x1984 but after like 6hrs lol, dunno, it works, glitch mebe

[fair oxide]

so a majority of them, about 70% seem to be json files

but a lot of those still work like loras

[ornate elk]

Remove --precision full and try --medvram

Should be faster

[fair oxide]

there are safetensor files not popping up and json files popping up

[solemn moon]

ill try that thanks

[fair oxide]

what do i do after that 👁️

[languid ledge]

bump

[solemn moon]

need a screen shot of what you trying, or delete venv folder

[karmic crown]

i was afraid that that will happen. that's why i said it is better to reinstall and move models 🤷🏻‍♂️

[languid ledge]

trying to install reactor

[solemn moon]

or this

hmm

reactor with what model

[karmic crown]

if you had working v1.6.0 or v1.6.1 i would not recommend reinstall, but with 1.5.1... 😦

[solemn moon]

always update wtf brrrroooo

[karmic crown]

i think he is trying to install/update it

[solemn moon]

also check drivers, i learnt that the hard way, why "nvidea driver update" doesnt pop up but "one drive startup" does is beyond me

[solemn moon]

so much fun btw

[fair oxide]

OH HELP ME

how do i update drivers

ive had this computer for like 3 years but like

ive never found the drivers

[languid ledge]

are you talking to me or?

[fair oxide]

im technodumb

[solemn moon]

ya

[solemn moon]

same boat ,\happy to share

[languid ledge]

how do i update too

[fair oxide]

we'll alll learn together 🙏

[solemn moon]

press windows key, type "nvidea" then update drivers

[solemn moon]

wont DM sorry bruv

[fair oxide]

???

[languid ledge]

is aight

[fair oxide]

ah rip, i was actually gonna send a friend request loooool you seem cool

[solemn moon]

shit guys maybe we shouldnt listen to me idk wtf that is lol

[fair oxide]

bro i dont know 😭

[languid ledge]

i think i know

[fair oxide]

this is deadass just my luck with tech

[languid ledge]

but its very troublesome

[solemn moon]

strict sec measures no hard felz

[fair oxide]

i dont understand it at all 😭

[fair oxide]

nah i get it, internet is a wild place man

[languid ledge]

updating my driver actually made it lagger

[fair oxide]

💀

[solemn moon]

this is what i get

[remote agate]

Genuinely, it's quite strange because initially, it generates the images in a very similar way, but when pressing "generate" again, it creates images with very ugly artifacts, and if I keep doing that, stains appear

[fair oxide]

yeah

i opened that

[languid ledge]

guess i can't use reactor

[solemn moon]

go to extentions

look for reactor

download

update

delete venv just incasies \

restert

then see it at the bottom of yr web.ui interface

[languid ledge]

do i have to go available or install from link

[solemn moon]

let me look

i wouldve googled

then wouldve ended up here

https://github.com/Gourieff/sd-webui-reactor

lol

should be at the bottom of yr interface

[languid ledge]

yea but it doesn't even let me install

[solemn moon]

maybe try the extensions/install from

sorry i cant remember how i did it

[ember venture]

I had Auto1111 running on Linux on an old Tesla M40-12GB I got for cheap until last week, now it says it doesn't detect any HIP capable GPUs and I get errors I don't understand. Is this the kind of problem I can ask about here?

[solemn moon]

https://github.com/Gourieff/sd-webui-reactor

[languid ledge]

yea i did that

[solemn moon]

make sure git pull is on, delete venv, maybe check extensions, restart

my extensions shows it

yrs?

[languid ledge]

oh

it worked when i download from available

probably i copy the wrong link

[solemn moon]

loool

LG!!!!!\

[languid ledge]

how do you change ur ui btw

[solemn moon]

oh ya, you have to post any new works created as a result of support by me, just fyi

[languid ledge]

it looks very cool

[solemn moon]

lobe theme

google it now!!!

so dope!!!!

[languid ledge]

its not worth it

trust me

[solemn moon]

i get a thrill everytime, reason i stopped using comfy was cause no lobe theme bruh

lobe for lyf

[languid ledge]

does it slows down ur sd

[solemn moon]

yeaaa, the price we must pay unforts

so worth it

no rush ]

[karmic crown]

if you managed to update webui, and it working, do not forget to update all extensions (if you haven't already)

[languid ledge]

yea i did thank you

thank you all for helping

[karmic crown]

i saw a few cases when updating and fixing extensions actually took too much effort and time, that's the only reason i suggest reinstall 🤷🏻‍♂️

[proper pike]

How come my lora folder doesn't show the images like yours?

[fair oxide]

no idea tbh

im still trying to figure out my own lora issues 😭

[ornate elk]

I can help everyone if I'm at home

[solemn moon]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/7053

maybe

[ornate elk]

Is your webui updated?

[ornate elk]

So reactor works now?

[languid ledge]

didn't try yet I have to go to sleep I'll lyk tmr

[ornate elk]

Alright

[ember venture]

had Auto1111 running on Linux on an old Tesla M40-12GB I got for cheap until last week, now it says it doesn't detect any HIP capable GPUs. All the help I can find for this problem is about AMD systems

[split sorrel]

Hello, what should I do to fix this error? I get it while using hires. fix :--- RuntimeError: Could not allocate tensor with 1073741824 bytes. There is not enough GPU video memory available!

[ember venture]

That probably means the settings you're asking for are demanding more memory than you have. If you know you have more than enough memory there is something else wrong though

[lost crescent]

A1111 used to change model quite rapidly. Now, change the model, and it spends 10 minutes making a yaml file!!!! Why, what's changed at all?

[snow lance]

any linux users here have a preference for which nvidia drivers work best now?

535 still the way to go?

[ember venture]

I think the latest version of Auto1111 broke on my Linux, but I find SDNext is working, sort of. I may need to tune it but at least it's running

[remote agate]

@ornate elk What could be the cause of the problem? I'm not referring to minor differences in the image in question; I mean it undergoes a very aggressive change and starts distorting completely to a point where it looks awful.

[cloud topaz]

hey guys! any idea why the manager isnt showing up? i installed the custom node

[ember venture]

SDNext works for me where Auto1111 fails. Took some dialing in but it's running well now. I guess I don't have a problem anymore

[ornate elk]

What are your txt2img settings?

[ornate elk]

Can you generate an image again with that issue?

[little echo]

you say it may not be generated if its unknown, then in what regard will the output will still change? it will try to guess?

is there a way to extract the safetensors to find out the tags coded into them? (actually i am referring morever to how tags are used by the SF models not specifically loras)

[covert fox]

Need help, How to solve this?

[ornate elk]

make sure your webui is updated

and redownload the loras

[remote agate]

After testing with some scripts, the one that solved the problem was... "--no-half", although I'm not so sure what it does, what is this line of code supposed to do?

[ornate elk]

its a line needed for gtx 16xx gpus, it will load models with half precision, that makes it a lot slower in generation

[remote agate]

It's strange, because my gpu is rtx 3060

[ornate elk]

then you shouldnt use it

better is to use --xformers --no-half-vae

[remote agate]

I'll try it and see if it solves the problem.

[remote agate]

well it actually fixes the problem temporarily, when I reload the image it generates errors.

[ornate elk]

can you show me the errors?

[umbral plover]

Hi, when I'm trying to use face restoration in comfyUI i get following error, not sure why tho (Custom node mav-rik Facerestore CF (Code Former))
//edit when i ungrouped nodes (loader and restorer) it started working (wtf)

[ornate elk]

what does the cmd shows?

when do you get the third image?

[ornate elk]

so the thrid images comes out deformed?

and you changed nothing?

is your webui updated?

[remote agate]

I installed it yesterday, so, yes

[ornate elk]

okay, and you generate 3 images and changed nothing between them?

seed was -1 ?

or what did you tried ?