🤝｜tech-support

[thorn goblet]

or, that gpu 0

i assume you want gpu 1 which is indeed nvidia

GT 1030

[willow zodiac]

In the upper right corner, there is a thumbtack looking icon. If you click that, probably the second item on the list will be a step by step to install for Nvidia GPUs. Follow that and you will get there.

I'm pretty new here, but there are others that can generally help if you run into something sticky

[thorn goblet]

thanl yopu ever so kiondly

is this an issue

i logged in now should i just rerun it

[willow zodiac]

You are missing the "t" at the end of the string, it should be "git"

[thorn goblet]

i dont mean for you to hold my hand im sorry

so after this- then what?

where do i edit the webui-user.bat

[willow zodiac]

right click and then edit it. You can edit it in Notepad, for example

[thorn goblet]

wheredo i find that file

[willow zodiac]

the folder will be named "stable-diffusion-webui". The file will be in there if the git repository downloaded correctly

I'm checking out. Hopefully someone else will be around. They will almost undoubtably be more knowledgeable than me.

[thorn goblet]

well youve been the only ome to reply, i hope someone else would pick up your burden for my sake\

im sorry to bother you so im just so... gone right now

i cant say sober me would be any better at understanding...

i really appreiate your patience

[thorn goblet]

i think i got it! ill need a little direction as to generating images with prompts and all that but i figure i can grab that within my community and all that

thank you again,i truly appreciate it

[ornate elk]

What's your GPU? And you shouldn't use xformers together with opt-sdp-attention, as the webui can only use one of them.
Also --no-half is only needed for GTX 16xx cards

Hey, do you have a screenshot of the full cmd log with the error?

Because the file name should be .yaml and yours is .yml

[thorn goblet]

it closed and i dunno how to reopen it....

[ornate elk]

Launch the webui-user.bat

[grand fable]

Hi, I'm adding some info - the Macbook has 16GB of RAM, the workflow in ComfyUI is the default one without any changes, the cat on bicycle with sunset and countryside was made in 457 seconds, the cat was done in 372 seconds, both in 1024x1024px, SD XL base model. Attaching the Activity monitor screenshot from the time it was doing the cat.

[honest tangle]

Hi! Can anyone help with an issue? Yesterday all ve been fine/
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.

runnung img2img nothing helps

[honest tangle]

upd: it blowed away after hitting reload ui button in settings

[ornate elk]

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

[hot spindle]

Hi, anyone have any idea if this is serious? On luch get a warning: ControlNet init warning: Unable to install insightface automatically. Please try run pip install insightface manually.

Whenever I try to manually install it get another warning:

DEPRECATION: torchsde 0.2.5 has a non-standard dependency specifier numpy>=1.19.*; python_version >= "3.7". pip 24.0 will enforce this behaviour change. A possible replacement is to upgrade to a newer version of torchsde or contact the author to suggest that they release a version with a conforming dependency specifiers. Discussion can be found at https://github.com/pypa/pip/issues/12063

Somehwere I cant figure out there's a mispelled dependency: numpy>=1.19.* when it should be numpy>=1.19 but cant figure out where to edit it

Solved, there's a 0.2.6 version , forced install that

[hot spindle]

Actually was going the wrong way, using nigthly build of Torch (diferent venv) was just: venv-torch-nightly/bin/pip install insightface

[karmic crown]

do you have this version?

[hot spindle]

1.1.430

[karmic crown]

even better

some older versions had errors on mac (especially with IP adapter)

i hope that will fix your controlnet issues you mentioned before

also, be sure that you use appropriate controlnet model corresponding to your base model (XL with XL base models, 1.5 with 1.5 base models)

[hot spindle]

I don't think I have real issues, but I was wondering if you need an inpaint Controlnet to use for instance on a inpaint on an image while using canny reference from another , meaning: inpaint a canny controlnet from a diferent image

[karmic crown]

ask this in #📝｜prompting-help and you should get an answer

[hot spindle]

I will, too many things to try and work to do, do you happen to know why I cant find the sampling method on that Photoshop SD plug in?

Should be in main panel but cant find:

[karmic crown]

i am actually a relatively new user (about 2 months), but i am reading and experimenting a lot. unfortunately, i haven't try that one yet since i do not have PS installed on M1 at this moment, but i will

[hot spindle]

ok, tks, will try github

helpfull as usual, thanks a lot 👍

BTW: should TurboSDXL use SDXL Controlnet models?

[karmic crown]

i would check the other tabs too

[hot spindle]

I did

Its a viewr, Cotrolnet, Horde (extrenal render) and settings for ConfyUI/A1111

extras is upscale

[karmic crown]

i am almost sure it should be SDXL

but i emphasize ALMOST 🙂

have you tried to scroll down? there is a scroller on the right

i guess you did

as i said, you have a button bellow 🙂

[hot spindle]

hooo,shit,,dint notice the scrolll bar.Dumb. 🤦‍♂️

Tks

[calm cairn]

tried a Samus Aran prompt in SDXL and got this--what is going on here? This is more than the wrong VAE right?

I've had similar errors in all my XL models so I'm hoping someone here knows what's happening

[hot spindle]

wrong size?If you use less than 1024 in SDXL that kind of stuff happens

[calm cairn]

oh wow ty

[rotund lava]

no,thats a vae issue,you probably using a 1.5 vae on sdxl

[calm cairn]

I had no idea there was a specific VAE for XL, will switch that too thanks

[rotund lava]

you could also try to set vae to none because some models already have a baked vae

[karmic crown]

as far as i know, we only have 2 VAEs for SDXL

those two
https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/tree/main
https://civitai.com/models/152040/xlvaec?modelVersionId=264491

[karmic crown]

i agree, but automatic should also work fine for those

[ornate elk]

Automatic uses None if there isn't a VAE with the same name like the model in the models folder

[karmic crown]

exactly, i just didn't want to go into those deails 🙂

[reef patio]

Hi, does anyone know/have working solution on how to batch upscale with esrgan model?

[hot spindle]

ChaiNNer?Easy batch upscale, only downside i you can't choose intermedite upscales

4x models will do 4x...no way to choose 2x

[reef patio]

well, i want to pass my dataset through jpeg remover esrgan model, which does 1x upscale

ah, got it, gott try that one

[hot spindle]

It does support a lot more than just upscale, you'll have to check, the upside is you can convert PyTorch models to specific GPUs like NCNN and the one from NVIDIA I can't remember the name

If it's a "denoiser" it probably can be run from ChaiNNer as well

If it upscales by 1x just use two upscale nodes in tandem

1x isn't upscale though 😁

[reef patio]

ye, i meant process

thanks, gotta look at this one

[calm cairn]

removing VAE to "none" did it, ty for advice, will add in XL VAE tonight, thanks

[versed ridge]

SafetensorError: Error while deserializing header: MetadataIncompleteBuffer

i get this error when trying to geenrate image

[hot spindle]

@reef patio Would be something like this, where the first model is the "denoiser"

[versed ridge]

can someone help

[reef patio]

yup, still installing torch

[zealous cargo]

it's a 1650 card, so i should only go with xformers? and probably -medvram

[hot spindle]

Just becarefull to use orifinals folderdirectory in image file interator windows and set save to another diferrent folder

If its a subfolder disable "recursive" in window

[rotund lava]

sounds like a corrupt model,download the model again,select it and try to generate another img with it

[versed ridge]

so im gonna delete the folder and download again right? the model i downloaded is Aitrepreneurs one click download btw

[ivory bramble]

All my final generations are ending up like this. I did end up installing a bunch of recommended extensions yesterday https://i.imgur.com/z1CT6sP.png

Any reason it might be happening?

[rotund lava]

just redownload that model,then select it again from the dropdown list and try to generate another img

[ornate elk]

Img2img SD upscale script

[versed ridge]

without deleting the folder, I already tried reinstalling bunch of times. I guess I have to uninstall first

[ornate elk]

Your GPU needs --xformers --medvram --no-half

[reef patio]

yup, thanks, works perfectly

[karmic crown]

1.5 VAE with SDXL model or CFG set too high

[ivory bramble]

Oh that might be it, thanks!

[rotund lava]

no need to uninstall just delete the single model file

[karmic crown]

do not mix SDXL and 1.5 resources, easy as that 🙂

[ivory bramble]

Yeah it says "SDXL 1.0"

the model I'm using

[karmic crown]

in that case you can use onlu 2 VAEs

[ivory bramble]

https://civitai.com/models/152525/realism-engine-sdxl

It was this one

[karmic crown]

those two
#🤝｜tech-support message

[ivory bramble]

Nsfw if you scroll

[karmic crown]

and that model already has VAE

so just set VAE to automatic or none

[versed ridge]

These are the files I have im not stable diff expert dont know much what you mean by deleting model? The stable dif webui file?

[ivory bramble]

I'm trying with a different model altogether to see if the final result is still like that

[karmic crown]

find folder with name Models

[ivory bramble]

These are my current VAE settings https://i.imgur.com/4lm96PH.png

[versed ridge]

ok. there are bunch of files

[karmic crown]

and than

and find the one that make problems inside

and just remove it

[ivory bramble]

That's where I put all my tensors

I should be more organized

[karmic crown]

that is your problem

set that to automatic

[ivory bramble]

Really? because everything was working fine yesterday until I installed all these extensions...

[versed ridge]

İ guesd i deletr refiner and go to my one click downlad and install it again

[ivory bramble]

But this one came out ok on a different model

[karmic crown]

that is 1.5 model i guess

[ivory bramble]

I guess it was the model

[karmic crown]

just set that to automatic

go to settings and find this

[ivory bramble]

You sure?

[ivory bramble]

Ok

[zealous cargo]

i see, but does it affect the performance if i change something under the optimization settings in the web gui? for example, increase token merging ratio?

[karmic crown]

then set this

and reload

than you will have those 3

so you will be able to change it anytime you want

[ivory bramble]

Ohhh so you can change it at the top now

Ahh thank you so much viking

What theme are you using Viking? I want a plain black one like that

[versed ridge]

viking the problem solver now may i have your wisdom on the pic I sent?

[ivory bramble]

Sorry didn't mean to hijack your help 😄

[versed ridge]

explain like you explain it to someon who doesnt know a thing

i just want my blender textures dont know anyting abut any of tis

[versed ridge]

.d np bro

[karmic crown]

just a second guys

[versed ridge]

ok

[karmic crown]

remove that one

[versed ridge]

Ok after that? What do I do?

[karmic crown]

i would recommend to install extension called civitai browser+ (ill help you)

[ivory bramble]

civitai browser has been a lifesaver, I started using it yesterday

[karmic crown]

you just put a link from civitai

and it will put everything in a right place for you

so you do not need to guess which folder you need

[versed ridge]

ok im on my way trying it now

[ivory bramble]

Model Preset Manager too

[karmic crown]

i use model organizer

[ivory bramble]

Ohh I should look at that one

[versed ridge]

to do that i open stable diff and download it there right?

[karmic crown]

wait, i will explain

[versed ridge]

ok im waiting

[karmic crown]

go to extensions, than go to available tab, and click this button

[versed ridge]

uhm wait

it doesnt launch stable diff now its doing an installation

cuz we deleted the file

[karmic crown]

than in search type
civitai browser+

it downloading default 1.5 base model

let it install it

you can remove that later, but leave it for now

[versed ridge]

okey

[karmic crown]

when you find extension, click on install

when it finish, go back to installed tab

and click that button

[reef patio]

@hot spindle One more question, do you know, does it support proccessing multiple at once? or only one by one?

[versed ridge]

thank you man im gonna try it after the download finish cuz its lookingkinda big

and if a have a problem can I text you?

[karmic crown]

just ask here, someone will help

i will if i see it

[versed ridge]

okey thx

[karmic crown]

after restarting go to civitai browser tab and pasta link from civitai; for example https://civitai.com/models/84728?modelVersionId=90072

click on magnifying glass

click on picture and it will populate those boxes

i have that model, so i have delete model button, but you will get this one

when you click it will donwload model and put it the right folder

it simplifies the downloading of models a lot

[versed ridge]

so it will fix my bug and make downloading models easier right?

in what way making easier btw

[karmic crown]

as soon the downloading of base model finish, you will be able to generate images

[karmic crown]

in "normal" situation you download model, lora, text inversion... and you need to find a proper folder for each of them, since those goes in different folders

but with civitai browser, you just need to paste link from civitai

it will take care that you do not put lora in model folder, or something similar that might make problems later

[versed ridge]

ok i kinda get it

now just have to wait for this big intall currently on

%25

if it all works out u are a lifesaver i will use this on my blender projects

[karmic crown]

unfortunately, that is relative
from my perspective, anything below 10GB is not big, since i need 2-3 minutes to download that
but i understand that is a big file for many people

[versed ridge]

i use portable wi fi for some reasons and it downloads kinda slow

but works good when gaming 👍

[karmic crown]

than you have a good ping, but not good download speed

[versed ridge]

yeah just like that

[karmic crown]

if you have 2.4 and 5GHz enabled on your router, you can try to disable 2.4 and leave only 5GHz (if your computer supports it). that made a big difference at my home, since i have like 20 other routers around (from my neighbors)

but, lets try to stick with SD issues here 🙂

[versed ridge]

this one doesnt work like a normal wifi. Its more like a phone with hotspot on and it even has sim card like a phone.

[versed ridge]

yepp i need that more than everything now .d

[ornate elk]

Yes token merging can increase the performance

But dont change the optimization in general cause with the args in the webui-user.bat your using xformers.
So don't set it to Dogetexx or opt-sdp

[karmic crown]

can it make my Mac run as fast as MCurto's 🤣
now seriously, where is that option exactly?

[ornate elk]

Its under performance settings. You can also add it to the Quicksettings by adding token merge or token merge ratio there

It can improve the speed but idk how much

[karmic crown]

any advice on ratio?

[hot spindle]

I dont know sry

[zealous cargo]

alright

[zealous cargo]

i've only seen 0.1-0.2 in a optimization video

[ornate elk]

No clue. But here is the Source:
https://github.com/dbolya/tomesd

[zealous cargo]

and i guess it depends which model and settings you use for prompting

[karmic crown]

"Even with more than half of the tokens merged (60%!), ToMe for SD still produces images close to the originals, while being 2x faster and using ~5.7x less memory."

so that is 0.6

but i didnt see any diffrence with SDXL turbo on my iMac, i will try later on m1

[ornate elk]

I need to test it on AMD

Maybe it can reduce the memory usage there

[versed ridge]

@karmic crown download done and I can generate image but this is the quality i get. After the things you told me to do will it be fixed?

[karmic crown]

go to https://civitai.com/models and find some models you like

this one is good for example
https://civitai.com/models/133005/juggernaut-xl?modelVersionId=288982

[versed ridge]

oh so I guess im using default model and if i use models that are created for realistic images or other purposes the result will change right?

and civitai makes model downloaing easier

[karmic crown]

yes

[versed ridge]

ok im understanding now

[karmic crown]

for example with this model

and it will download juggernautXL_v8Rundiffusion.safetensors and put it correct folder for you

[versed ridge]

ok im downloading it

i saw different mothods to download models on internet sites

but is this the method i should be using for downloading models from now?

[karmic crown]

this is the easiest one

[versed ridge]

ok one last question

[karmic crown]

almost forgot one importnat thing

[versed ridge]

on that website there are so much models. actually all of them. if I want to download a model from there what do I do?

[versed ridge]

what is that

[karmic crown]

when you see a image that you like

click on that circled i at the bottom corner

and you will get information what was used to create that image

prompts, cfg, sampler, steps...

[versed ridge]

uhm wher do I see these images with circled i at the bottom?

[karmic crown]

if you click this icon

[versed ridge]

copies

i guess

[karmic crown]

it will copy that for you, yes

and then you can paste that in positive prompt

[versed ridge]

ok ty and where do I see these images tho

[karmic crown]

and when you click on that arrow it will populate all those things for your, and leave only positive prompt there

on civitai

you have million images there

[versed ridge]

ok its helpfull

and

[versed ridge]

what about it

[karmic crown]

if you want for example that one

copy address

and paste into civitai browser

like that

[loud shore]

please heeeelp, NameError: name 'user_comment' is not defined

[versed ridge]

right where i typed jaggurnaut xl right?

[karmic crown]

that is even easier

[versed ridge]

ty so much man u really did a lot

[karmic crown]

it can search for you

[versed ridge]

someone give this man mod

[karmic crown]

at least post a screenshot, or even better text from cmd

[karmic crown]

uh... that is Colab

I am not sure if those work anymore for users without billing

but i might be wrong

[loud shore]

i have a billing

[versed ridge]

@karmic crown im waiting on juggernaut xl download now u rly helped a lot today have a great day king. i hope to see you if I have more questions in the future 👑

[karmic crown]

i am relatively new here (about 2 months), but i haven't seen anyone here who was able to help with Colab errors 😦

CS1o and Aryetis help a lot people with Windows, and I am the one who usually helps Mac users, but unfortunately, none of us use Colab 😦

[karmic crown]

for a questions not directly related to installation and problems with installation or errors, it is best to ask in #📝｜prompting-help
some people there can help with a lot of things, but some of them usually do not look here

[versed ridge]

thx i will look up to that

[karmic crown]

my advice for you is to rise an issue on github page of an author of that colab workbook

i am sorry i can't help more 🤷‍♂️

[kindred tinsel]

Hey guys quick question, Chocolatey for npm required to install python 3.12 for some reason on my system and I needed to create a venv for webui again :c (messed ups omething else wiht faceswaplab) but now I don't know how to create the venv with the 3.10 version of python that I have installed.

Any help on how to select the versions or run the specific exe for python 3.10 would be greatly appreciated

[ornate elk]

You have to edit the webui-user.bat
There is a line set Python=
There you set the path to the python.exe of 3.10.X
You get the path with the command
where python

Then you need to delete the venv

[kindred tinsel]

it worked thanbk you cs !!!!!. God Idek why I have choco in my system, I needed it for one thing, installing dart :c and it installed its own python :c

[kindred tinsel]

lmao csgo 🤤 also yes literally mvp of this chat

[astral comet]

Why can't i see the seed in citivai's online generator?

[umbral plover]

Hi, didn't been up-to-date for two months - what model/sampler is right now a go to for lowend gpus? Did something change with setting up a1111?

[crimson sail]

i dont know why, i didnt change anything i just wantet to boot stable diffusion and i am getting this error every time

[hexed forum]

any way to run SD somewhat smoothly on 4gb vram? i have a gtx1650 with 4gb 12gigs or RAM and i5 10th gen on a laptop and a lot of the prompts i was trying to make crashed so now im wondering if there is any way to run it on an external program or website or something similar?

[distant bear]

Hello friends, imgtoimg results give results that are irrelevant to input. What do you think is the reason?

for example like this

[umbral plover]

cfg scale propably?

[distant bear]

@umbral plover this is my settings

whats wrong idk ?

[umbral plover]

try decreasing denoising strength, or use original seed

[distant bear]

original seed means = 0 ?

[kindred tinsel]

Hi, does anyone know how to fix this issue?

fatal: No names found, cannot describe anything.
*** Error loading script: civitai_gui.py
    Traceback (most recent call last):
      File "C:\Users\santi\Documents\StableDiffusionIDEs\stable-diffusion-webui-directml\modules\scripts.py", line 469, in load_scripts
        script_module = script_loading.load_module(scriptfile.path)
      File "C:\Users\santi\Documents\StableDiffusionIDEs\stable-diffusion-webui-directml\modules\script_loading.py", line 10, in load_module
        module_spec.loader.exec_module(module)
      File "<frozen importlib._bootstrap_external>", line 883, in exec_module
      File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
      File "C:\Users\santi\Documents\StableDiffusionIDEs\stable-diffusion-webui-directml\extensions\sd-civitai-browser-plus\scripts\civitai_gui.py", line 35, in <module>
        ver_bool = version.parse(ver[1:]) >= version.parse("1.7")
      File "C:\Users\santi\Documents\StableDiffusionIDEs\stable-diffusion-webui-directml\venv\lib\site-packages\packaging\version.py", line 54, in parse
        return Version(version)
      File "C:\Users\santi\Documents\StableDiffusionIDEs\stable-diffusion-webui-directml\venv\lib\site-packages\packaging\version.py", line 200, in __init__
        raise InvalidVersion(f"Invalid version: '{version}'")
    packaging.version.InvalidVersion: Invalid version: '.7.0'

---

I am encountering this issue after a fresh install, Im not sure how to update this thing to get it to version 1.7 which I assume it needs for this line
ver_bool = version.parse(ver[1:]) >= version.parse("1.7")

[distant bear]

still no works ?

should i change Sampling method ?

[dark owl]

denoise is too low try between .5 and .8

Also checkpoint model might not support black hair

seed 0 is bad too

-1 is random

[distant bear]

what should i change man ?

[dark owl]

any numbers

5789245294678

[distant bear]

Can you tell me the values I should give? I just installed it, I don't know anything.

[dark owl]

or -1 for random for each gen

the CFG slider increases the prompt strength and denoise slider will increase the blend between img2img

512x512 isn't that great either if you can do 512x768 or 540x968 is better

[distant bear]

what about Sampling method and steps ?

batch count batch size

[dark owl]

the more steps the more details the AI will add

30-70 is good

batch count and size just generates more images at once per queue

[distant bear]

is it good now ?

[dark owl]

swap the width and height

[distant bear]

oh

[dark owl]

then try generating an img

[distant bear]

proccesing now

[ornate elk]

What's your GPU?
Euler a is a good sampler.

[distant bear]

bad result again @dark owl

[kindred tinsel]

boop

[dark owl]

Yeah use Euler A for humans imo

[ornate elk]

You probably need some performance arguments in your webui-user.bat
Check my install guide in the pinned message of this channel

[umbral plover]

1060 6gb, was wondering if sdxl turbo would run on it, right now using 1.5 models and lcm sampler and lcm lora

[ornate elk]

Yes sdxl turbo can run on it

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

[umbral plover]

perfecto, will try it in a moment, thanks

[ornate elk]

Where did you get the extension from? The latest version is available on Civitai

[kindred tinsel]

I got it from the sdwebui list

let me check how to install it from then there ^^

[ornate elk]

https://civitai.com/models/151467/civitai-browser-or-sd-webui-extension

[kindred tinsel]

But in there they say that the automatic installation should work too :c Like in a gif they show how to do it

[ornate elk]

Okay

[kindred tinsel]

Im not sure what it means about packaging.version.InvalidVersion .I checked my packaging module in the venv and it says the version is 23.2 wayyyy higher than the 1.7 :c

[ornate elk]

No clue then. Try a reinstall maybe

The latest version is from 2 days ago

Maybe something is just broken

[kindred tinsel]

Maybe but then how can I install a specific version before said update?:

[distant bear]

@dark owl whats wrong dude ?

[kindred tinsel]

how about, you try inpainting, mask the area for the hair

[ornate elk]

A day ago the creator says to reinstall the extension to remove an test message

So worth a try to delete and redownload.
If not then download an older version from civitai

Then extract it and Copy the extension into the extensions folder

[kindred tinsel]

Okay I downloaded it again, let me test it.

CivitAI Browser+: Aria2 RPC started
fatal: No names found, cannot describe anything.
*** Error loading script: civitai_gui.py
    Traceback (most recent call last):
      File "C:\Users\santi\Documents\StableDiffusionIDEs\stable-diffusion-webui-directml\modules\scripts.py", line 469, in load_scripts
        script_module = script_loading.load_module(scriptfile.path)
      File "C:\Users\santi\Documents\StableDiffusionIDEs\stable-diffusion-webui-directml\modules\script_loading.py", line 10, in load_module
        module_spec.loader.exec_module(module)
      File "<frozen importlib._bootstrap_external>", line 883, in exec_module
      File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
      File "C:\Users\santi\Documents\StableDiffusionIDEs\stable-diffusion-webui-directml\extensions\sd-civitai-browser-plus\scripts\civitai_gui.py", line 35, in <module>
        ver_bool = version.parse(ver[1:]) >= version.parse("1.7")
      File "C:\Users\santi\Documents\StableDiffusionIDEs\stable-diffusion-webui-directml\venv\lib\site-packages\packaging\version.py", line 54, in parse
        return Version(version)
      File "C:\Users\santi\Documents\StableDiffusionIDEs\stable-diffusion-webui-directml\venv\lib\site-packages\packaging\version.py", line 200, in __init__
        raise InvalidVersion(f"Invalid version: '{version}'")
    packaging.version.InvalidVersion: Invalid version: '.7.0'

Here is the error again, Im going to try downloading a previous version or idk maybe uninstalling pacakging and allowing this extension to install it?

[ornate elk]

Make sure to delete the venv too after deleting big extensions like this one or face replacer etc

[kindred tinsel]

no way you want me to install everything again :c god damn

This is really weird, both my normal 3.10 installation, and the venv have packaging version 23.2. Now I am not seeing that error im seeing

CivitAI Browser+: Aria2 RPC started
CivitAI Browser+: Python module 'packaging' has not been imported correctly, please try to restart or install it manually.

[umbral plover]

Takes about 40+ seconds for 1 image - sample steps 4, cfg 2, res 512x512, DPM++ SDE Karras

[dark owl]

the resolution is wrong lol width = 540, height = 968

[analog crown]

anyone that knows well diffusers library ?

[dark owl]

thats not even 540x968 lmao

I dont know whats going on

[dark owl]

yeah but the picture that came out isnt xD

[distant bear]

Could you do it on your own computer and send me a screenshot?

[dark owl]

I am rendering a video right now so I can't

it's half way done, probably 30mins left

[distant bear]

You said something about the checkpoint model. Could it be related to this?

[dark owl]

Yeah

look for a realism one

[distant bear]

wdym

[dark owl]

https://civitai.com/models/4201?modelVersionId=245598

[distant bear]

should i download this ?

[dark owl]

It's what I used to make my pfp on discord here

[kindred tinsel]

Let me try it out mind sending me the image to check if I can get him with black hair

[kindred tinsel]

what's the image size?

[distant bear]

1477x2048

[dark owl]

if you press this it will give you the resolution of the orginal image

[distant bear]

It doesn't have to be this picture. Any picture will do. What I want is to learn the correct settings.

[kindred tinsel]

I want to try it out,. but im having issues withmy webui lmao, the inpaint setting arent even showing up :=====

[glad fiber]

still needing help with this #🤝｜tech-support message

[kindred tinsel]

hmmm training, there is a guy that has great youtube videos, Ive been meaning to watch his long form tutorials, maybe he can help youhttps://www.youtube.com/@SECourses

AFAIK he has some videos on lora training and he has produced some about himself so he knows what he is doing ^^

[glad fiber]

thanks, i will see

[kindred tinsel]

I got some hair using inpaint with a simple prompt and only for that section. But I am trying with more steps to see if I get a better result.

[distant bear]

thank you for your efforts

[thin quarry]

Mask area you want to change, img2img changes whole thing and since you didn't include in prompt what exactly is on the image - it's doing whatever it wants with black hair.

[kindred tinsel]

@distant bear horrible with masked area, I would use ip adapter but I dont have the time for that sorry

[distant bear]

thank you man

could u share your settings ?

[kindred tinsel]

Literally just moving sliders around ^^

[thin quarry]

masked content - should be fill in this case

masked padding higher too

[glad fiber]

looks like his settings are almost exactly like mine

except the number of epochs, which mine was 3

[kindred tinsel]

Well I wouldn't really know I have no idea how to start training loras nor how many images or the like I would need for it to work ^^, but maybe there are other tutorials that can help

[glad fiber]

i will try with 12 epochs 💀

[kindred tinsel]

May I ask where you found the images or how did you tag them?

[glad fiber]

google

I will try to get it to work first, then I will train with more images and I will take them from the actual show

I hope disney wont see it 🙏😭

[crimson sail]

pls

[karmic crown]

have you fixed this?

[dusky wing]

Hi all! A few hours back, my stable diffusion was working just fine, but now suddenly it's really slow, and if I want to upscale with Img2Img, it just freezes, it stays at 0% with 00:00 it/s.. I have no idea how to find out how to fix this, anyone got any advice?

[dusky wing]

I've got 5s/it trying to generate a 512x512 image using a rtx3060ti... Is that a normal number, or should that be faster? Online, I find benchmarks saying it should get 8it/s

[ornate elk]

whats in your webui-user.bat?

[dusky wing]

just --xformers

I actually get around 5it/s if I run a 1.5 model, but as soon as I run a XL model, it goes up to like 5 to 10 s/it

[ornate elk]

yes because for sdxl your card also needs --medvram-sdxl --no-half-vae

that speeds it up

[kindred tinsel]

It works normally and even downloads the models and stuff, idk what that packaging thing does but it isn';t raising any errors yet

[dusky wing]

You're a wizard, it works so much better now hahaha

I' ve now got 1it/s using XL and generating 1024x1024 image

[ornate elk]

thats good 😄

[dusky wing]

It definitely feels good haha

It does freeze whenI try to do img2img on a higher resolution at the last percents though, is that also normal? Like, it says in the console Total progress:100%, but in the web uit, it says 91% ETA :2s and its saying that for a minute now

Is that just having patience or another magic word should be added to webui-user.bat

[ornate elk]

depens what your settings are on img2img

you should not generate at to high resolutions, instead use upscaling

[dusky wing]

It's a 1024x768 file which I try to get to 1536x1024

[ornate elk]

resize by in img2img is not the right way to upscale images (it works, but is slow as you regnerate the whole image in higher res)

better use hiresfix in txt2img and then use sd upscale or ultimate upscale script in img2img

[dusky wing]

Aah I will look into that, thanks

[wraith bolt]

is this fine for a 1650?

[split sorrel]

Hello! I have an AMD 6700 graphics card, and I've installed Stable Diffusion, but it's not utilizing the GPU, it's using the CPU instead. What should I do? Please help.

[ornate elk]

hey, then you have installed the wrong webui. Check the Pinned Messages of this Channel. There you will find my AMD Install Guide

[split sorrel]

No no, I'm sure I installed it properly.

[ornate elk]

okay, then show a screenshot of the cmd when starting the webui-user.bat

[ornate elk]

thats the right webui, but with the wrong startup args

replace all of the args you added with these:
--use-directml --medvram --opt-sub-quad-attention --opt-split-attention-v1 --no-half-vae --upcast-sampling

in the cmd it also shows that it found an nvidia driver. so you also should uninstall that

[split sorrel]

gave this result

There is nothing related to nvidia, I searched but I couldn't find it, do you have any suggestions about what it is?

[ornate elk]

you need to delete the venv folder and relaunch

dont know where some ressts of the driver are found. But some tools like DDU can remove them.

[split sorrel]

The first time I accidentally installed stable diffusion, which is compatible with nvidia, the problem may be due to that, but I couldn't fix it no matter what I did.

[ornate elk]

okay, does it work now?

[split sorrel]

yes yes thank you very much

You can't even imagine how hard I worked 😄 I'm very happy

[ornate elk]

good no problem 🙂
after the installation is done and the webui open in browser.
open up a cmd and type
pip cache purge
that will remove the nvidia stuff from SD

that was downloaded from your first failed install

[split sorrel]

Thank you very much my friend, you are a very good person ❤️

[ornate elk]

🙂 thx no problem

[warm tusk]

When using -share and the gradio.live link I get a "JSON.parse: unexpected character at line 1 column 1 of the JSON data" error if I set the batch count higher than 2. This is on mobile, btw.

[warm tusk]

Nevermind, turns out I had no gradio queue on the command line from past troubleshooting and that breaks it entirely for mobile if generation takes "too long", probably some kind of timeout.

[sharp lichen]

Does it help to put the lora in the front of the prompt?

[glad fiber]

gtx 1650?

if yes, then it is not

I get 1s/it at ComfyUI

~2s/it at a1111

getting almost 8s/it at ComfyUI with a batch size of 8

btw, how many regularisation images are enough to train a person?

[fair oxide]

Hey guys, I need a little help here, my controlnet isn't recognizing the pose. Does anybody know how to solve this ?

[upper linden]

Wanted to ask, for the amd guide to run optimized models, it seems like i cant get it to work. Is it worth doing this process, if i have a amd card?

https://community.amd.com/t5/ai/updated-how-to-running-optimized-automatic1111-stable-diffusion/ba-p/630252

this is the guide

also would more vram lead to better results in general?

[wraith bolt]

1650 super..

what settings do you got

ig it depends on the model you use?

and sampling steps etc

[grand fable]

Hi guys, is there a way how to generate more image previews faster (in some low quality) and then pick one to go for and generate it in higher resolution and work with that in next steps? I need to create some specific model but generating takes me around 7 minutes (1024x1024px 20-25 steps) so it’s quite time consuming to wait for every iteration that long when some of the results are thrash.. usi g ComfyAI on M1 Macbook. Thanks for every help! 🙂

[ornate elk]

The input image is already processed so set preprocessor to "None"

[ornate elk]

Hey, its not worth using this guide from amd.
You would get faster speed, but you then can't use any model you want and you cant change resolutions.
Also not every extension or feature will work.
So for Usability using directml without olive and onnx is better

Do you get out of vram errors ? Then you need the Tiled Diffusion Extension

[thorn goblet]

how do i make generations go faster

current eta is an hour and a half

[ornate elk]

What's your GPU?

[thorn goblet]

NVIDIA GeForce GT 1030

[ornate elk]

And what settings do you used?

[thorn goblet]

negative guidance maxxed, token merging maxxed

[ornate elk]

Okay and the txt2img settings, like resolution, steps etc?

[thorn goblet]

recommended settings from the VAE publisher

[ornate elk]

That looks pretty normal

But over an hour is strange

[ornate elk]

Why did you changed these?

[thorn goblet]

someone suggested it to make it go faster but it didnt work

[ornate elk]

Then set them back to default

Also what's in your webui-user.bat?

[ornate elk]

Okay thats good

You could try using --medvram instead of --lowvram

Should be faster

Then set batch count to 1 to test

Also use Euler a as Sampler

Its faster

And use a model(checkpoint) thats 2gb

[thorn goblet]

i dont have the best hardware

[ornate elk]

Oh okay only 2gb then stay with --lowvram

But do the other changes I mentioned

Sampler Euler a.
And make sure the model you try is 2gb and not bigger

[thorn goblet]

where can i get models

[ornate elk]

From civitai.com

For example get the Dreamshaper model. Its 2gb

https://civitai.com/models/4384/dreamshaper

[thorn goblet]

i dont rember where toput it

[ornate elk]

In the models/stable-diffusion folder

[thorn goblet]

on the page for dream scaper it says it requires LCM Sampler

idk what that is

ok well i cant make anything with the default width and height, it just says not enough memory

images are coming out shit anyway

[ornate elk]

Did you downloaded the v8 ?

That is 2gb in size?

[ornate elk]

No you dont need that If you downloaded the v8 version

[ornate elk]

Restart and try again please

[grand fable]

Hi, is there any optimization settings for using ComfyUI on M1 Macbook? In other words should I look for some settings or something to generate images faster?

[sharp lichen]

Hello how do i properly caption images when training a lora

[karmic crown]

how long does this take for you? (this is a 1.5 model)

1girl
Negative prompt: nsfw
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 1604009253, Size: 512x768, Model hash: ec41bd2a82, Model: photon_v1, VAE hash: 63aeecb90f, VAE: vae-ft-mse-840000-ema-pruned.safetensors, Version: v1.7.0

Time taken: 1 min. 7.4 sec.

A1111 on M1 MacBook Pro 16GB (8 CPU + 8 GPU cores)

[karmic crown]

but i usually use LCM, to reduce number of steps, and make a generation time drastically lower

1girl <lora:LCM_LoRA_Weights_SD15:0.5>
Negative prompt: nsfw
Steps: 8, Sampler: DPM++ 2M Karras, CFG scale: 2, Seed: 1604009253, Size: 512x768, Model hash: ec41bd2a82, Model: photon_v1, VAE hash: 63aeecb90f, VAE: vae-ft-mse-840000-ema-pruned.safetensors, Lora hashes: "LCM_LoRA_Weights_SD15: aaebf6360f7d", Version: v1.7.0

Time taken: 30.4 sec.

"normal" version (20 steps)

LCM version (8 steps)

[sharp lichen]

Is anyone aware of this error?
TypeError: Descriptors cannot be created directly.
If this call came from a _pb2.py file, your generated code is out of date and must be regenerated with protoc >= 3.19.0.If you cannot immediately regenerate your protos, some other possible workarounds are:

1. Downgrade the protobuf package to 3.20.x or lower.
2. Set PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python (but this will use pure-Python parsing and will be much slower).

More information: https://developers.google.com/protocol-buffers/docs/news/2022-05-06#python-updates

The above exception was the direct cause of the following exception:

[thorn goblet]

anyway to make it appear less blurry

[grand fable]

this one took 105 seconds. Don't have the SD 1.5 right now, so I used SDXL instead

so the 105 seconds is fine, but the output is quite bad..

[steep belfry]

I didn't use the program for a while, because I had to change my SSD (2TB), when I did that, it was like installing it on a new PC, but the already installed Stable Diffusion had taken it from my External HD and put it on the SSD, and when I try to open it, give it that and close it.

I had installed Python 3.10.6, I don't know if it has become obsolete

[ornate elk]

You need to set the cfg to 7-12

Everything lower than 5 loses color

[thorn goblet]

2 was recommended i need to stop listening to randos

[ornate elk]

2 is good when your using special models like Turbo or LCM

Normally 7-12 is good

[ornate elk]

Hey, you shouldn't copy your old SD install over.
Better to reinstall the webui and only copy the models over

You find a quick install guide in the Pinned Messages here in this channel

[thorn goblet]

hey so uhhhhh my eta is an hour

[ornate elk]

Can you show your txt2img settings?

[thorn goblet]

these?

[ornate elk]

Thats strange as in your screenshot from before the image took only 2 minutes

[thorn goblet]

ik

[ornate elk]

How big is your prompt?

[thorn goblet]

same exact prompt

[ornate elk]

Anything running In the background?

Wallpaper Engine or Epic games launcher?

[thorn goblet]

i got minecraft running

[ornate elk]

Yea thats it then

It uses GPU too

[ornate elk]

GPU not CPU

[thorn goblet]

yippee

[ornate elk]

It will take longer when you have other programms running in the background

[steep belfry]

ok, I adjusted, taking advantage, my Stable makes an image every 50 seconds, is this the standard or can you make it faster?

[ornate elk]

Did you checked out my install guide?

There are the commands for the best performance depending on gpu

[steep belfry]

ok, I'll look and see.

[weak rune]

Hi! I have a problem with automatic1111. It crashes as soon as I press Generate. Can anyone help me?

[umbral ivy]

I am trying to install model from stability matrix
But unable to do
I am getting package modification Faild

Any solution

[vocal burrow]

sure, but we can t see your gpu in those screenshots

[vocal burrow]

can t help without more details, console logs, etc

[weak rune]

I have a rx 5700 xt

and installed rocm 5.6.1

this is from inside the venv:

$ pip list | grep torch
open-clip-torch           2.20.0
pytorch-lightning         1.9.4
pytorch-triton-rocm       2.1.0
torch                     2.1.2+rocm5.6
torchdiffeq               0.2.3
torchmetrics              1.3.0
torchsde                  0.2.6
torchvision               0.16.2+rocm5.6

[ornate elk]

What's Our GPU?

[ornate elk]

Do you have a screenshot ?

[ornate elk]

Which webui did you installed with it?

[umbral ivy]

ComfyUI

[vocal burrow]

try using --medvram --no-half --opt-sdp-attention command line arguments

[vocal burrow]

and you do have a swap partition/file ?

Otherwise you might want to try using pytorch1 instead of pytorch2

[weak rune]

$ lsblk
NAME   MAJ:MIN RM   SIZE RO TYPE MOUNTPOINTS
sda      8:0    0 931,5G  0 disk 
├─sda1   8:1    0    16M  0 part 
├─sda2   8:2    0    41G  0 part /...
└─sda3   8:3    0 890,5G  0 part /...
sdb      8:16   0 931,5G  0 disk 
├─sdb1   8:17   0 243,2G  0 part /...
└─sdb2   8:18   0 688,3G  0 part /...
sdc      8:32   0 465,8G  0 disk 
├─sdc1   8:33   0   512M  0 part /boot/efi
└─sdc2   8:34   0 465,3G  0 part /
sr0     11:0    1  1024M  0 rom

That means I don't have a swap partition, right?

[weak rune]

I had pytorch1 before (it was installed by automatic1111 by default), but always got gray images

[ornate elk]

Can't help with that. But wouldn't recommend stability matrix for installation
Comfyui is just a zip file to download

[weak rune]

so I figured it must be the case that rocm is not backwards compatible, because pytorch1 uses rocm5.2

[vocal burrow]

looks like you dont have swap indeed

(I dont have much time to deep dive in linux right now, I m still at work)

[weak rune]

ok, so you would advise me to use pytorch1, which comes by default anyway? but with which rocm version? 5.2 isn't up anymore

[vocal burrow]

I d say make a swap partition first

it should be relatively easy with stuff like gpedit

[weak rune]

ok i can try

[vocal burrow]

gparted*

[dim cave]

hello

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

so this is the only line i edit

ok seems python not in the path

[ornate elk]

You can run the python installer again, click on modify, next and then check "add python to system variables"

[dim cave]

ahh i see it !

its running

[kind sapphire]

Hi everyone, does anyone have a guide on how to setup a scalable API for SDXL Turbo?

What I am trying to achieve is real time inference for my app.
So from the client side I would open a socket connection to my nodejs server and that nodejs server should in turn talk to a cloud GPU server that runs SDXL Turbo and exposes it as API endpoint. Is it doable?

[slate shore]

@ornate elk

the 150 is a usb with no antennaes, ultra cheap and the realtek is what im using now and it does have an antennae and its a bit less cheap still quite cheap

but the pci slot one is not there

[ornate elk]

and your mainboard doesnt have one integrated?

[slate shore]

no

[ornate elk]

is the pci one a wifi network adapter?

[slate shore]

its a asus ROG x470 i think

it is wifi

[valid gorge]

Sometime I feel like I get stuck when generating, even I change the promt, seed, scale, steps etc A1111 keeps generating very similar images or a detail I don't want. Anyone else getting this, is it a bug?

[slate shore]

ubit wie4630

[ornate elk]

okay, and you installed the drivers of it?

[slate shore]

yes i have direct from manufacturer website

[ornate elk]

is it made for Win11?

[slate shore]

it didnt work on win 10, when i did the fresh install last night i started with 10, i installed a new water cooler as well so first thing i did was install windows 10 on new drive, then installed icue to monitor temps...then nvidia drivers for new GPU. once that was all working correctly i tried to go online and realized there was an issue... worked on that a bit and tossed a cheap usb dongle in... then proceeded with installing davinic, after effects etc then once it was setup and i had latest versions of everything i upgraded to 11

[ivory bramble]

Quick question, is there supposed to be 2 checkboxes under these settings? https://i.imgur.com/HJtIpsM.png

[honest tangle]

Hi guys i have a problem while running ModuleNotFoundError: No module named 'tqdm.auto' Can someone helpm me to fix it? It all happened afte rinstalling temporal kit from github

it

upd. if someone needs. if you' ve installed temporal-kit extension? and you have the same issue just delete venv folder and run bat. it will reinstall it and start the sd

[dim cave]

ok now i need to learn what the settings do lol, o put many sliders to max, one photo is now taking 15 minutes

any guide to understand the sliders ?

[main axle]

Hello, guys.
I wanna generate a background of product image.
But have no idea about it.
If I inpaint the other area, the edge is affected.
How to avoid this issue?

Does anyone can help me?

For example.

This is the input

And the desired output is this one.

The edge of image does not affected

[dim cave]

what sampling method ?

[ivory bramble]

https://stable-diffusion-art.com/change-background/#Step_1_Install_the_Rembg_extension this might help

I'm not sure, though, I'm not a helper

[karmic crown]

that should work

[karmic crown]

i have no idea who did you managed to have two of them

[ivory bramble]

Me either

[karmic crown]

btw, there is not need to use external service for images

[ivory bramble]

It doesn't affect anything

[karmic crown]

just paste a picture from clipboard here

[ivory bramble]

Yeah, I need to configure ShareX to copy the screen grab

it automatically uploads it to imgur

but doesn't copy it to my clipboard

[ornate elk]

Here:
https://civitai.com/articles/904/settings-recommendations-for-novices-a-guide-for-understanding-the-settings-of-txt2img

[karmic crown]

then ok, I thought you do that manually. i just wanted to tell you that there is not need for that, to save your time

[ivory bramble]

Thx, I'll configure that later

[main axle]

The background removal works well

But after remove background if I just put the product on background, the result is not realistic

And also the edge of prduct is not good

I think inpainting works the best

But affecting edge of the product

[ivory bramble]

What inpainting method are you using? fill, original?

I've been experimenting a lot with both

[dim cave]

to stop progress just close the cmd ?

[main axle]

I think there's not that big different with choice of inpainting method.

Is here any way to do inpainting without affecting the edge?

[karmic crown]

i would use controlnet

Optional: Inpaint with ControlNet

this

[main axle]

I used depth map controlmap for inpainting

[ivory bramble]

https://civitai.com/models/85514/cosmetics?modelVersionId=135830

[spice hill]

i must have done something, i don't remember what, but now every generation with every model looks like this, anone can help? (automatic1111)

[main axle]

But it was affect the edge a little.

[karmic crown]

nice one

[karmic crown]

99,99% you are using sdxl model with 1.5 lora

[spice hill]

i'm not using sdxl model

[main axle]

Seems you need to increase step and decrease noise strenth

[ivory bramble]

Ha! I learned that yesterday and I was going to suggest it but I wasn't sure

[spice hill]

though, when i did, i changed some settings, but i reverted to using no VAE @karmic crown

[karmic crown]

please show us your setting for that image

[spice hill]

i'm going to try, usually even lower steps worked

[karmic crown]

the text below the image

[main axle]

noise strength 0.2, steps 80>

[spice hill]

it looks fine whe generating

[main axle]

Would be help

[spice hill]

it goes awful when it's done

[vocal burrow]

what cfg ?

[spice hill]

i tried multiple models, anitoon/breakdomain, + 20 to 80 sampling steps and every model gives this result

[loud galleon]

when generating video on comfy do you need to press queue twice during the process? after the line art is generated?

[main axle]

adjust the noise strength helps a lot.
cfg - just default works the best

[spice hill]

@vocal burrow so, it worked, until i wanted to try sdxl, then i got sdxl working, now when i revert to any other model it doesn't work

i disabled it's vae

[jade surge]

hello guys .pip install xformers==0.0.16 how can i do that.I dont know any coding

[spice hill]

though it loads by default when i open automatic1111

[karmic crown]

as i said, please copy the info below the picture

[jade surge]

there is a folder pip.exe in scripts folder btw

[spice hill]

yes, 1 sec

[karmic crown]

so we can see what you are actually using

[spice hill]

a bedroom
Steps: 80, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3000954253, Size: 512x512, Model hash: 1870fa10c3, Model: breakdomain_v10, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Version: 1.7.0

Time taken: 11.5 sec.

[vocal burrow]

no but what s the cfg value ? also what model are you using with what settings ? basically screenshot your txt2img interface so we can check

[vocal burrow]

512x512 using sdxl

[karmic crown]

1.5 model with, sdxl vae

[spice hill]

tried various prompts

[karmic crown]

so i was right

[spice hill]

vae is disabled in settings though

that's the point

[karmic crown]

it is used

[vocal burrow]

if you re using sdxl you ve got to prompt for 1024x1024 resolution as that's the resolution of the dataset used to train sdxl

[karmic crown]

as you can see

[karmic crown]

VAE: sdxl_vae.safetensors

[spice hill]

yes i've read, now, the problem is, how do i disable it then

[vocal burrow]

also yes just dont mix and match different base model and base vae

[ivory bramble]

Is there an extension that could compile errors and put them into a A1111 tab?

[spice hill]

...problem is i wasn't trying to 😭

[vocal burrow]

errors should already be printed in the the console log

[ivory bramble]

It's a pain to go through the console log sometimes

[vocal burrow]

np, now you know

VAE quick guide :
1/ Where do I put my VAE ?

• VAE with .vae.pt, .vae.ckpt, .vae.safetensors extensions go into the models\Stable-diffusion folder
• VAE with .pt, .ckpt, .safetensors go into models\VAE
  2/ How do I use my VAE ? Three possibilities :
• Either you name it similar to another one of your model (eg : Anything-V3.0.safetensors + Anything-V3.0.vae.pt), by doing that it should automatically load the VAE when you load the associated model.
• You manually load your VAE by going to Settings -> Stable-Diffusion -> sd_vae and selecting your VAE
• You add an easily accessible VAE dropdown at the top of your page to quickly switch back VAE by adding sd_vae to your Settings -> User Interface -> Quicksettings list

[spice hill]

yes but, even though i disabled it in settings, it still tries using the VAE

[honest tangle]

4090 . actually i've start to put init videos in the route folder of sd and after all is working ou t

[vocal burrow]

cf what I just posted

[jade surge]

guys anyone knows this issue?

[spice hill]

@vocal burrow solved by removing the no half vae flag in script

no, wait

[vocal burrow]

that s not how it works.....

[spice hill]

NansException: A tensor with all NaNs was produced in VAE. This could be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

[ivory bramble]

Also do this

[vocal burrow]

you simply need to pick a VAE compatible with your model

if you re using a sd15 model, use a sd15 vae

[spice hill]

i want no VAE in this case

[vocal burrow]

sdxl model, sdxl vae

[ivory bramble]

Will make your life easier

[spice hill]

it won't show in that menu, let me take a screen

[ivory bramble]

after you figure it out...

[karmic crown]

@vocal burrow i am obviously became blind, i cant find default VAE setting in my interface 🙈

[vocal burrow]

not possible you ll always be using a vae, be it the one built-in the model or one you picked manually or one auto111 picked for you based on your naming scheme

[spice hill]

okay

[ivory bramble]

This is what I see

[vocal burrow]

default should be auto (=> pick the one with the same name as model) or none (=> use one thats built in the model)

[karmic crown]

i know, but i cant find it anymore

[ivory bramble]

Maybe it's your theme extension

[karmic crown]

i do not need it i have quick settings

[vocal burrow]

if you ve put it in the quicksetting list it won t show up in settings page

[spice hill]

btw sorry if i asked dumb questions etc. btw i figured it out, i am dumb, i wasn't applying settings 😭 @vocal burrow@karmic crown

[ivory bramble]

Oh yeah, it doesn't show up in VAE settings anymore

[karmic crown]

oops

[ivory bramble]

but it's on the txt2img page

[vocal burrow]

np glad you figured it out

[karmic crown]

so that's was the problem. so, i am not blind 🙂

tnx man

[ivory bramble]

I still need to figure out the 2 checkbox situation

[vocal burrow]

would need the full log

[karmic crown]

i was really worried i became blind or stupid 🤣

[jade surge]

ok im taking it w8

[karmic crown]

go to your webui folder and show me the result of
git status

[ivory bramble]

<input type="checkbox" id="txt2img_hr-visible-checkbox" class="svelte-1ojmf70 input-accordion-checkbox">

oh ok

[jade surge]

@vocal burrow

[ivory bramble]

should I git restore that file?

[vocal burrow]

there s stuff before that

put the whole log in a .txt and drop the file in there

[ivory bramble]

this is my web-user.bat ```@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --xformers --autolaunch --no-half-vae --lowvram --medvram-sdxl --no-gradio-queue --theme dark

call webui.bat

Oh I just noticed I have lowvram whoops

[karmic crown]

wsl2 is for linux embedded in windows, so that is not really needed in your case, but you can restore it

i do not see any file that was modified that should not be modified

i wanted to see that

[jade surge]

@vocal burrow

[ivory bramble]

Lets see if getting rid of --lowvram fixes it

[vocal burrow]

You are running torch 1.13.1+cu117.
The program is tested to work with torch 2.0.0.
To reinstall the desired version, run with commandline flag --reinstall-torch.

at the very beginning of your log

[karmic crown]

that should not be it

i would disable all extensions and see if there is any difference
i suspect some of extensions causing that issue

[jade surge]

but i dont know any coding ,i dont know where should i run the install what path what folder

[vocal burrow]

also you re using no command line arguments at all so you re probably leaving performance on the table

[vocal burrow]

Command line arguments explained like I'm five :

• First off, golden rule; .bat files are for windows, .sh for linux/mac, do NOT modify webui.bat/webui.sh. The only file you should be modifying ever is webui-user.bat/sh
• What is a command line argument/arg ? Those are words you use when launching a program to specify how it should work/trigger a specific behavior. They can look like this --xxx or --xxx y (y being a specific value associated with the -xxx command line argument)
• How do I set those ? Edit (with notepad or whatever text editor you want) your webui-user.bat/sh file and then :

• for windows users : find the following line set COMMANDLINE_ARGS= and set whatever command line arguments (and its associated value) on the same line after the =. (eg : set COMMANDLINE_ARGS=--xformers --no-half-vae --medvram-sdxl)
• for mac/linux users : find #export COMMANDLINE_ARGS="", remove the # to uncomment it and then do the same as windows users

• Is there a list explaining what those arguments do ? What values I should use with such and such argument ? Check it here https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings#all-command-line-arguments

and what gpu are you using ?

[slate shore]

so i installed python previously and i assume i did not select add to path i uninstalled and reinstalled and it did not go through the same questions as it did the first time. how do i correct this

[vocal burrow]

leave every settings to default except for the "Add to PATH" and eventually install folder if you want to change it.

[slate shore]

how do you change that once its installed?

[grim raft]

So... I trained a lora and I got this when trying to infer images...

[vocal burrow]

change what ?

[slate shore]

add to path

[grim raft]

https://i.imgur.com/BhZOYUb.png These are my settings

[vocal burrow]

either you reinstall or you edit path manually

[slate shore]

i uninstalled and then reinstalled but it did not ask me those options during the 2nd installation it just installed it

[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?

[vocal burrow]

somewhere during the installation be it the first, second or tihrd it surely asked about "add to PATH"

[jade surge]

msi 4060 8gb

[slate shore]

the fist time it did i did not check it i installed before finding this thread

[vocal burrow]

ok then you should probably use --medvram-sdxl --xformers --no-half-vae

[vocal burrow]

ok but like I said it s gonna ask you about "add to PATH" evey time your reinstall

[jade surge]

direct command on windows or a path on sd?

[vocal burrow]

those are command line arguments

so you put them in your webui-user.bat

like mentioned by the tutorial I ve sent above

[jade surge]

its installing xformers now,am i okey ?

[vocal burrow]

yes

[jade surge]

acces point now available

[vocal burrow]

enjoy pretty much doubling your generation speed

[vocal burrow]

log please

[jade surge]

ok ok i got it bro

[grim raft]

Can someone help me with my issue? I'm at a loss...

[jade surge]

tnx a lot

you are archangel

[vocal burrow]

can t really help you with that :/ I dont do much training, maybe you can ask later once everything has calmed down, or move to #🔧｜finetune

[karmic crown]

do you have mcafee or some other 3rd party antivirus active?

problems like those are usually caused by mcafee and friends ☠️

[karmic crown]

here it is https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/5427

see the last post

just remove venv folder and run webui again

it will create new venv for you, and it should work

[jade surge]

i got smt new

[vocal burrow]

You are running torch 1.13.1+cu117.
The program is tested to work with torch 2.0.0.
To reinstall the desired version, run with commandline flag --reinstall-torch.

you still haven t run --reinstall-torch command line arg

remove all your args and run with only --reinstall-torch once

then restore your previous args

[jade surge]

okey

[karmic crown]

it is probably faster and easier just to run

webui.bat --reinstall-torch

without editing webui-user.bat 🙃

[jade surge]

i just learned it but tnx bro 😄

[karmic crown]

but i know why he wrote that

many people just like to click something, and not to go to cmd and type 🙂

[slate shore]

so this thread is filled with probably every answer i will have about this. i was trying to copy and paste into chat gpt to use as a refence point so that i can ask cht gpt questions and esentially have a database of answers... but it wont let me copy ...ctrl a ctrl c doesnt work, nor does right click. is there a way to do that?

[jade surge]

export as text

[slate shore]

i dont see that option

[grim iron]

I get the following error when I try to captoin images in kohya RuntimeError: The size of tensor a (36) must match the size of tensor b (432) at non-singleton dimension 0 can someone helpo?

[slate shore]

how long should this process take? Its been at this point for a while now: emote: Enumerating objects: 277, done.
remote: Counting objects: 100% (165/165), done.
remote: Compressing objects: 100% (30/30), done.
remote: Total 277 (delta 137), reused 136 (delta 135), pack-reused 112
MiB/s
Receiving objects: 100% (277/277), 7.03 MiB | 4.61 MiB/s, done.
Resolving deltas: 100% (152/152), done.
Installing requirements for CodeFormer
Installing requirements

[vocal burrow]

depends of your hardware and connection

make sure that the cmd has focus (click it in it and press space)

[slate shore]

should i not be doing anything else?

[vocal burrow]

depends if your hardware or connection is the bottleneck

[slate shore]

im on a pretty decent machine 128gb ram, 4070, m.2 ssd, with ryzen 7 5900x

processor is runnign at 3% usage

maybe time for nap and well see what it does after

[grim iron]

my 5600x is only at 8% usage while running so tahts normal I think

[valid gorge]

What would be the best option for a GPU under 500 euros / dollars to use for SD?

[grim iron]

would say a 4060ti

[valid gorge]

That's actually the one I'm been looking for.

What about memory, is 8 gb enough or should I get the 16 gb version. Not a big difference in price I think.

[grim iron]

16gb if I undersand vorrectly more vram = better

[grand fable]

I'm kinda not able to run LCM lora, it tooks so much time. Could you help me and tell me what is wrong please?

[valid gorge]

Great, thanks!

[grim iron]

clip for the promts has to come out of the lora node not the load model one

[grand fable]

I did it based on this video and he says it would be degradation. At time 4:00 https://m.youtube.com/watch?v=pw8eNpmhT5o

[steady thistle]

I'm having trouble getting 1024x1024 using dreamshaperXL turbo. I tried 512x512 with hires fix of 2x, 10 steps, .5 denoise, plus tiled vae with encoder tile size 1024 and decoder tile size 112. I had lowvram set then I changed it to medvram and it seems to work now. I don't get why I didn't work with lowvram tho.

[grand fable]

is the "lcm_lora_weights" the right model? I downloaded it from civitai, because don't know how to download it directly from https://huggingface.co/latent-consistency/lcm-lora-sdxl

[steady thistle]

But now it isn't working. What's the best way for me to get 1024x1024 on my Radeon RX 6800M GPU with 12GB GDDR6

[valid gorge]

guys, can you suggest me a YouTube video or a full course about SD in general where basics and beyond would be covered, and also a course for ComfyUI?

[grand fable]

oh thanks 👍🏻 the "pytorch" prefix confused me.. now just put it in the "loras" folder and open ComfyUI?

[slate shore]

Hey guys, I finally got stable diffusion up and running. Quick heads-up: upgrading from a 2060 to a 4070 GPU cut my render time from 13 minutes to just 2.2 seconds. Anyway, I’ve got a specific use case and could use a bit of help. If someone could spare around 30 mins to an hour to guide me through their workflow and process, I'd be happy to pay for your time. What I need to do is take a photo of a client, an electrician for example, and then drop him into various scenes in that industry. The plan is to create about 100 images for their company that can be used on their website, blog content, Google Business profiles, etc. The images need to be bang on though; they can't be all glowy like DALL-E, and they can't have seven fingers like on Midjourney. Also, the bit where you can use a selection tool and only tweak a certain part of the image, I think it’s called inpainting. Anyone down to help me out for a bit sometime today?

[vocal burrow]

it's pretty much the default weight for that kind of things

[grim iron]

?

[hazy oriole]

hello everyone, I just downloaded sd by the guide here for amd
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs
And It is using my cpu not gpu, I have amd vega 64, where can I find a guide for it ?

[steady thistle]

Check out the pins on this channel and follow it closely. I had a similar issue as you.

The 'pip cache purge' fixed it for me.

[hot spindle]

Sry dont have the time but gonna give you a tip that serves me well: If you have access to PS make a rough cutout of person in Photoshop, drop that person on something of other image you want as background, just a rough, dont mind to be neat, if needed free hand draw some lines...generate with Controlnet Canny with a small weight, something like 0.3 strenght and have it stop at about 0.3 or 0.5 then let SD fill the gaps..works wonders with SDXL

[slate shore]

ill give that a shot

[hot spindle]

Works wonders for me to generate backgrounds

[steady thistle]

If you want good backgrounds, this trick worked for me sometimes. Put your background text first and put [all the rest of the prompt:2]. That makes the first step be background only.

[hot spindle]

Small exemple, I had a 700x300 original, erased roughly the person, hand draw some black lines (for Canny to pick up) then let SD "improve" it a bit

Its now 5k width

[grand fable]

I changed the LoRA to the pytorch (which I think is the same as I had before based on the file size), I rewired the Clip from Lora node to the prompts (not as was said in the video https://m.youtube.com/watch?v=pw8eNpmhT5o ) but it still takes a huge amount of time to generate simple image 😕 any ideas how to run LMC lora so I can generete images fast?

[steady thistle]

I'm trying to inpaint with controlnet and I got this error followed by a crash. Ideas? [F D:\a_work\1\s\pytorch-directml-plugin\torch_directml\csrc\engine\dml_util.cc:118] Invalid or unsupported data type.

[steady thistle]

@ornate elk Do you have any ideas on this? Sorry to bug you if you're busy.

[hot spindle]

Not sure, you do need specific models for 1.5 and SXDL and need to choose them accordingly , make sure model is selected, load same image as "independed control image" and hit Preview:

Preview is that "red explosion" button

[steady thistle]

I did that, that's when it crashed, and I'm using 1.5 and only have 1.5 loaded in controlnet. I haven't downloaded the sdxl ones yet

[hot spindle]

I'm not that much into technicalities but try and check if it makes a preview

In terminal sometimes you have a more comprehensive report and suggestion fo fix it (or: enable --no-half or some other thing)

[slate shore]

ok so literally copied the exact settings from this ..i would expect the replicating the settings exactly would produce at least similar results but mine is far from a photo quality. what am i missing?

https://civitai.com/models/122418/drifting-car

[steady thistle]

I have --no-half loaded and it crashes when I select the explosion button

If I forget to do the inpaint_only+lama, and leave it on inpaint, no crash

It's just when I select the +lama

[hot spindle]

Maybe someone more technically oriented can help you, I can get what I want done but don't have extensive knowledge

[steady thistle]

Do you have control_v11p_sd15_inpaint [ebff9138] as the model?

[hot spindle]

Inpaint Lama is usually for outpaint

[steady thistle]

Yeah, I wanted to outpaint

[hot spindle]

I work in SDXL mostly, always need highres

[steady thistle]

Where do you find the SDXL controlnet files?

[hot spindle]

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

[steady thistle]

thank you!

do I need everything? I love sdxl

[hot spindle]

No!

Just the ones you think you might need and some of them have 3 versions

[steady thistle]

ok

I was looking for the one for the inpaint_only+lama for sdxl.

[hot spindle]

I would say depth and canny are the most useful to me

[steady thistle]

I use those often on 1.5

[hot spindle]

Don't think there's a specific controller for jnpaint in SDXL, think there is one experimental for ConfyUI

[steady thistle]

ah

[torn lotus]

Hello. I am using SD Fooocus, was wondering if it's ok it takes >5m to generate an image. 1070TI

[slate shore]

that sounds about right. i just upgraded to 4070 for that reason i was getting very long times with a 2060. but with the 4070 it takes 2-3 seconds

[torn lotus]

very long times like what?

[slate shore]

4-13 minutes

[torn lotus]

oh man

[slate shore]

anyone ever had issue with loras not showing up in the lora tab? it only has this Nothing here. Add some content to the following directories: but the directory has a few in it

[torn lotus]

I'll rather use a cloud gpu at this rate

[slate shore]

a 10gb GPU cloud is like 180 a month, cheaper to just buy a GPU

[ornate elk]

with auto1111 you would take 1:40 for an sdxl image

a 1080 takes 1:30 for a 30 steps 1024x1024 sdxl one

[ornate elk]

switch the models and they will show up

they only appear for 1.5 OR sdxl, when selecting the model

[slate shore]

im on sdxl base 1.0

[ornate elk]

hey, please use my install guide for AMD cards, its in the Pinned Message of this channel

[ornate elk]

then make sure the loras you downloaded are also for sdxl. if not they will only show up when using an 1.5 based model

[slate shore]

ahh that could be it. i will check that , i just downloaded a bunch to try out.

[steady thistle]

@ornate elk Welcome back! I had an error I can't figure out. lol It's this one here. I got around it by simply not using inpaint_only+lama but that's not the best way of doing it.

[ornate elk]

didnt tried inpaint with controlnet

[steady thistle]

Ah, okay, thanks!

[torn lotus]

what is best ? comfyui or auto1111?

[ornate elk]

you didnt have the same settings, as there are hidden settings. you need to click the "copy generation" data button, then paste it into the positive prompt box and click the white/blue arrow to apply them automaticly

[ornate elk]

auto1111 is in general easier to use and has better support

comfyui is for really experienced Node workflow users

[torn lotus]

ok i'll download auto1111

my wife has a 1650 4vram- worth it?

[ornate elk]

will work 🙂

best with auto1111

[ornate elk]

in the pinned Message of this channel you find my install guide with the best settings for each gpu(vram)

[ornate elk]

tested inpaint_only+lama controlnet and got the same error, tried inpaint_global_harmonious and it worked

[torn lotus]

oh thanks, I have followed another guide, but i'll check yours for tweaks

because I tried a couple prompts and I just generated monsters

[ornate elk]

that happens if you set the resolution to high for the model

1.5 models got trained on a resolution of 512x512 for example

[torn lotus]

1152x896

wanted to benchmark vs fooocus

[ornate elk]

for that you need to use an SDXL model

SDXL is trained on 1024x1024

[steady thistle]

Interesting that you got the same error. The videos all say that the +lama is the best, so if you do get yours resolved, let me know how if you don't mind. I doubt I'll need it that much so it's not something I'll fret over.

[ornate elk]

it wont work, i think it relies on stuff used by cuda (nvidia), or its just a directml incompatible thing more likely

do you want to outpaint or inpaint?

[steady thistle]

Both actually. And that makes sense. Curses to nvidia

[blissful nimbus]

Hi, I have an amd GPU but even after i followed the pinned guide it still keeps giving me errors

[ornate elk]

hey, which error?

a screenshot of the cmd log can help

[steady thistle]

I honestly think I'll use inpaint more in the future, but I'm still in the process of trying out the tools just to get familiar.

[blissful nimbus]

it started with the cuda error, after which i used --skip-torch-cuda-test, then it became something like RuntimeError: "LayerNormKernelImpl" not implemented for 'Half' so i had to add --precision full --no-half. now its:

[ornate elk]

then you followed the wrong guide

i have one for nvidia and one for AMD

you have the webui for nvidia installed

[blissful nimbus]

this one right?

[ornate elk]

[NVIDIA]

😄

so no

the other one starts with [AMD]

[blissful nimbus]

oh wait im so dumb

[ornate elk]

xD delete the whole folder, and follow the other guide. Make sure to also do the Important Step i mention there

[blissful nimbus]

the cache purge?

[ornate elk]

exactly

[blissful nimbus]

tysm

[ornate elk]

no problem 🙂 feel free to ask here if you have any questions

[torn lotus]

how do I hand out the command --medvram-sdxl with linux? is it a ./webgui attribute?

[ornate elk]

there you have to edit the webui-user.sh

and outcomment the # before commandline args

[blissful nimbus]

do i need the --medvram with a 7900xtx?

[ornate elk]

yes

xD

only because the directml version is unoptimized

if you would run SD on linux you wouldnt need it (much better performance for AMD)

i have the same GPU btw

[blissful nimbus]

oh ok

[torn lotus]

I guess it's this one

Commandline arguments for webui.py, for example: export COMMANDLINE_ARGS="--medvram --opt-split-attention"

#export COMMANDLINE_ARGS=""

thx

defo stresses the gpu 1024x1024

[kindred tinsel]

@ornate elk
Do you have any idea of what this error means or should I open a new ticket to the directml fork?

[ornate elk]

what did you tried?

[kindred tinsel]

Regional prompting with a new model anyhentai haha and composable lora

deactivated composable lora but it still hapens, a moment ago it could create images with no error :c

Imma do the usual hit the reset button and Ill try again :)

[torn lotus]

how to install more checkpoints?

[ornate elk]

you can download them on civitai.com and then put them into the models/stable-diffusion folder

[torn lotus]

i still get ugly models

ah I see

[ornate elk]

try dreamshaper v8

[torn lotus]

on civitai?

[ornate elk]

yup

[torn lotus]

k

[ornate elk]

https://civitai.com/models/4384/dreamshaper

ist a good starter model, and small too

[torn lotus]

2gbs small?

[ornate elk]

yes, the smallest models you can get are 2gb

[torn lotus]

ah it has many images

[glad fiber]

is 380 images good for regularisation?

[torn lotus]

has improved a lot... still some minor imperfections .. can use refine?

[ornate elk]

you should use upscaling

[glad fiber]

there is another models too

like reliberate, deliberate

you can use a lora

[torn lotus]

what is a lora?

[kindred tinsel]

seems like it was composable lora, now it is back to normal