🤝｜tech-support

[ornate elk]

yea let it run 🙂

[autumn niche]

G Morning,¿excuseme where i write the promtp here to make images whit sd xil 1.0?

[opaque jungle]

Yeah! Got it working!

[bitter zenith]

The bots are down

[autumn niche]

Ok thx

[karmic crown]

OK, we figured out where the problem is. this happens only on macOS 14.4 with Torch 2.2.0 or newer, downgrade to 2.1.2 fixed the problem

[vast crest]

Hi guys, but you can convert Lora SDXL files to the format: SD1.5???

[solemn atlas]

RuntimeError: mat1 and mat2 shapes cannot be multiplied (154x2048 and 768x64)

what does that mean

[karmic crown]

what image size have you tried to make?

[solemn atlas]

1024 x 1024

it worked just before but now idk what happend

[karmic crown]

show me the settings you used to generate image

[vast crest]

now it doesn't give me the loading percentage

[karmic crown]

if you disable hires fix and use 512x512, does it work?

[solemn atlas]

it doesnt work

[karmic crown]

restart ui, put the same prompt, sampler... but do not add hires fix and use 512x512, 512x768 or 768x512 - do not use 1024x1024 with 1.5 models

[solemn atlas]

yes it works now

[karmic crown]

now add hires fix, and it should work

just do not use 1024x1024 as base image size

[solemn atlas]

now it works thank you

[karmic crown]

and any time when you get memory related error it is the best to restart webui before you try to generate something else

[solemn atlas]

I will do it thank you

A question is ADetailer !After Detailer manipulations from th extionsions tool?

[noble belfry]

just as a sidenote, if you have a prompt that length you might want to try increasing the CFG value to maybe 9 or even 10 as its very specific. may or may not help, but just a good rule of thumb

[solemn atlas]

thank you I will do just that

[pale bobcat]

Bump. Can someone please help. Thank you

[indigo ivy]

OutOfMemoryError: CUDA out of memory. Tried to allocate 9.77 GiB. GPU 0 has a total capacity of 11.98 GiB of which 0 bytes is free. Of the allocated memory 3.09 GiB is allocated by PyTorch, and 9.31 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

get this when trying to upscale with hires

My gpu memory usage was around 8 out of 12 so I don't know how its running out of memory

And usually its right when the image is done generating that this error pops out

[noble belfry]

are you on windows?

[indigo ivy]

yes

AMD

[noble belfry]

oh AMD, im not sure. i was going to say nvidia lets you fall back on system memory, im not sure of the AMD equiv

[indigo ivy]

Dang yeah I don't know about amd

[noble belfry]

sorry, i was ready to help you, but i dont know about AMD. hopefully someone else does

[indigo ivy]

Don't worry about it, you're good

[ornate elk]

make sure to set hires steps to 10 and it should work

[indigo ivy]

Tried it but got the same error

Its being upscaled from 832x1216 to 1248x1824, maybe my 6700XT just isn't strong enough?

[ornate elk]

for that you need the tiled vae extension

[indigo ivy]

Do I also need controlnet?

[ornate elk]

that will prevent the out of vram error

[indigo ivy]

Alright will get it

[indigo ivy]

ok installed it, do I just enable it and I'm good to go?

[ornate elk]

yes

but only tiled vae and not tiled diffusion

[indigo ivy]

Alright and what sbout hires?

[ornate elk]

hires steps 10

upscale by 2

[indigo ivy]

alright

[ornate elk]

for sdxl try upscale by 1.5

[indigo ivy]

will do

[indigo ivy]

looks fine?

[indigo ivy]

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.

got this error

[ornate elk]

ah okay, you might need to add --upcast-sampling --no-half-vae to the webui-user.bat

[indigo ivy]

Alright will try that

[indigo ivy]

There was an error with --upcast-sumpling

launch.py: error: unrecognized arguments: --upcast-sumpling

[ornate elk]

oh sry its --upcast-sampling

[indigo ivy]

Oh ok

[indigo ivy]

Dang same error

OutOfMemoryError: CUDA out of memory. Tried to allocate 9.77 GiB. GPU 0 has a total capacity of 11.98 GiB of which 0 bytes is free. Of the allocated memory 4.74 GiB is allocated by PyTorch, and 9.06 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
Time taken: 5 min. 8.8 sec.

[ornate elk]

then use a non sdxl model, or lower resolution

[indigo ivy]

Oh bet, thank you

[vast crest]

Hi, I can't find a solution to this problem other than having the old version of SD. As you can see my preview on the right is really too small. Is there a way to make it bigger?

[errant pelican]

The preview on the right is small in this screenshot because the generation is underway. See where it says 6% ETA: 01:46: that means the progress is ~6%, estimated completion in 1m 46s

[karmic crown]

what happen if you click on that small image?

i am not sure if it work by default, since i am on lobe theme

[vast crest]

I've already tried waiting, even until 100%, but nothing

[karmic crown]

both pictures look the same to me, are you using the fixed seed?

[vast crest]

what happens is that it becomes bigger, but the photo remains fixed. Below I see the sequence of images that continues, while the photo that I enlarged is still

[vast crest]

No,i use: seed: -1

[karmic crown]

its img2img, i see now, but anyway, should not be the same, unless it is upscale, than it should be very similar

[vast crest]

this and version 18 of SD.MA in version 17 I didn't have this problem. The preview of the photos was much larger. I don't know how to fix it. If it's bigger, I can notice first if errors appear and then stop everything. But such a small preview, I struggle 😦

[arctic abyss]

Hey guys, I write an illustrate a sci fi comic series called system earth (found at my website tomgardenart.com). I’d love to be able to use stable diffusion to ‘ink’ my pencils using my drafts as its guide and my 100 + pages of my own work as a style guide. Could anyone give me advice on the best plug ins etc to make this happen? Thanks in advance

Here’s an image of my comic (human made) whilst also showing levels of drafts that I would like to feed SD so I can speed up my inking process.

[errant pelican]

I’d love to be able to use stable diffusion to ‘ink’ my pencils using my drafts as its guide
You mean coloring in the lines? I haven't done this myself, but this sounds like a job for ControlNet: maybe the lineart version. If you search "controlnet lineart" that should pull up some tutorials if that's what you're looking for

[karmic crown]

https://www.youtube.com/watch?v=Wl0JK5Ser28 <- start from here

[errant pelican]

If you want consistent coloring—again, just guessing here—maybe you could try loading a page into img2img, with half the page colored by you the human, and the other half pencil sketch. Then mask the pencil sketch for inpainting

[arctic abyss]

tbh i dont really want them 'coloured'. these days i just go greyscale. To me, its all about SD copying my style

[vast crest]

you should create a Lora Style with your drawing style

[arctic abyss]

Lora style?

Here's a clearer description of what I am talking about 🙂

[noble belfry]

oh it'll do the shading etc well, but it will be very inconsistent in color, even if you tell it a green/brown pallet one image the background guy is wearing brown, the next green

[vast crest]

https://civitai.com/models/2032?modelVersionId=2179

this is a Lora Style so to speak

you can recreate the style of other Masters etc

[noble belfry]

techniques to control colours only really work on fairly minimalist images consistently. what you might be able to do is use an SD1.5 inpainting model (the SDXL inpainting models barely exist).

[arctic abyss]

ok. So if i give it more of a suggestion of values in the draft, would that help its consistency? ei:

[noble belfry]

they are their own master 🙂 its about inking their own work 🙂

[arctic abyss]

I'm happy to keep it greyscale

[noble belfry]

oh then yeah SD can absolutely be used. i'd recommend controlnet canny with anime lines and inverting the image, but at a low level. its something you have to just play around with and from piece to piece you'll have to adjust

also, props for being an artist using SD as a workhorse and not a copycat 😄

[arctic abyss]

Haha thanks. yeah I like comics but i really like the storytelling element and if this can help speed things up so im telling more stories faster, im all for it. I feel if i can just add enough detail to the draft, hopefully SD can get something similar to my final inkings. Here's another example

[noble belfry]

yep thats totally doable

[arctic abyss]

ok thanks guys. And i can feed controlnet canny all my images to copy my style?

[noble belfry]

yep. you might need to tinker with its strength but you'll get the feel of it. i'd recommend doing it over panels rather than whole pages, it loses attention and tries to do too many things otherwise and you get kinda washy blurry faces etc.

[arctic abyss]

ok good idea

[noble belfry]

maybe add text after the fact though, its notoriously bad at things like that, ends up writing that language the sims talk 😄

[arctic abyss]

oh god, of course. It wouldnt be writing the script

[noble belfry]

oh i mean BLAM!?! action text

im not sure just how well it would work all the time and might cause awful results, but controlnet also has a reference feature and if you input an existing shaded image it should try to keep to that kinda "feel" of it too. you can have more than one controlnet. so you could try even adding that in too

[arctic abyss]

haha yeah. best to add that in. But yes that would be a slight issue cos good onomatopeas in comics are ones where its built into the composition of the image, not just put on top

[noble belfry]

that said, if you have the text thick lined already over it, if you put that into image2image it will trace over them

so if sketch A already had !! woah!! in it, then it would know when generating to the render

[arctic abyss]

Yeah it should just count it as a line like all the rest of the lines ae

mk. ill install controlnet and let ya know how it goes...

Do i need to install stable diffusion THEN controlnet canny? or can i just go straight for controlnet canny. sorry, newb here

[noble belfry]

i would install webui forge, personally, its a version of a1111 made by the controlnet developer

https://github.com/lllyasviel/stable-diffusion-webui-forge

[arctic abyss]

thanks. ok, ill install this first then 🙂

[noble belfry]

it comes with controlnet installed you just need to download the models for it. https://huggingface.co/lllyasviel/sd_control_collection/tree/main you'd primarily want the SDXL models with 'canny' in the filename. the other ones are fun and useful just not for your task

but focus on those after, installing a SD client the first time can take a lot of effort and headscratching

comfyui might be a better idea long term for setting up a digital inking sweatshop on your computer, but if you havent used SD before start here 🙂

[arctic abyss]

yeah there isnt a simple .exe after extraction. i might have to find a youtube video 😄

[noble belfry]

yeah theres a whole environment needs set up. i suggest olivio sarikas or sebastian kamph for handhold install videos

whats your system btw? what kind of graphics card do you have?

[arctic abyss]

nvidia geforce gtx 1050 ti

[karmic crown]

ouch

[noble belfry]

hmm. im not sure if thats enough to run it. you might be able to get away with forge and using SD1.5 models. since you're not making new images from scratch that should be fine. but it will be slow

[arctic abyss]

oh dam

[karmic crown]

for a decent speed, especially for a lot of images, you need at least something like nvidia 30xx

[noble belfry]

yeah you scrape through with SD1.5

[karmic crown]

with at least 16GB

[noble belfry]

he only wants to img2img shade/ink things

[arctic abyss]

i got 32 gb of ram?

[karmic crown]

vram

on graphics

[noble belfry]

try it and see. the speeds might be monumentally slow but if youre just using it for touch up work it should be tolerable

theres lots of little tricks that let you squeeze all the toothpaste out of the tube

[arctic abyss]

ill give it a try 🙂

I need to install python?

even tho its a language

[karmic crown]

being able to run something and use something for work are two different things

[arctic abyss]

if it works with intended purpose, id consider a pc upgrade

[noble belfry]

oh it 100% will

[arctic abyss]

maybe 😉

[noble belfry]

just....maybe not on your pc. itll be like childbirth each generation

[karmic crown]

i know how frustrating it is to wait 3 minutes just to realize that the image is not what you want and to do a little tweak and wait again... in the and you spend 1 hour for a image that you could do in a 5 minutes with decent hardware

[karmic crown]

make sense

[arctic abyss]

Yeah i have reservations if this process will even work. It might work, but might take more time cleaning the image up than me just drawing it. But also, i am due for a PC upgrade

[arctic abyss]

hmm. after installing python, when I open 'webui.bat' I am getting text saying 'couldnt launch python' .

this is my python install folder.

Apparently it should look like this:

[karmic crown]

#🤝｜tech-support message <- please follow this tutorial

at least for python and git

since you are with nvidia on windows, and you are installing forge, you can also try this one
https://github.com/lllyasviel/stable-diffusion-webui-forge/releases/download/latest/webui_forge_cu121_torch21.7z

that's >>> Click Here to Download One-Click Package<<<
from https://github.com/lllyasviel/stable-diffusion-webui-forge

[arctic abyss]

i dont think i checked 'add python to path' when installing python

[arctic abyss]

ive already installed so i dont have that option anymore. plus python cant detect an installation file aso cant uninstall properly

[karmic crown]

run installer again

[arctic abyss]

ok

[karmic crown]

i am searching for images on the internet, since i am on mac

but that is the deal, you need to check that littel box at the bottom

and be sure that is 64 bit version of python 3.10.11

[arctic abyss]

ok so run installer again to unintalled? cos 'modify' option in the set up doesnt have the tickbox

[karmic crown]

if modify doesnt work, uninstall than install

just clikc that box this time

[karmic crown]

should fix it, but if you already uninstalled, just install it

[arctic abyss]

mk ticking that didnt work(run webui.bat) so will reinstall and tick box

might reset comp too

[karmic crown]

restart cmd

do not need to restat the whole windows (i hope)

in the last few years, i used Windows only a few times, so I forgot many things

[arctic abyss]

the folder that python installer installed is still there after installation..

[karmic crown]

if you installed that for all user it should be in program files, or in your home folder (somewhere) if you installed only for yourself

[arctic abyss]

i was expecting it to delete all files cleanly after running uninstall

not leave main file with its .exes etc

i can just delete that folder, reset comp and reinstall

[karmic crown]

if you are reinstalling at the same palce, no need to delete anything

[arctic abyss]

ok

[karmic crown]

but i am little confused what it left

[arctic abyss]

tthis

its like it didnt uninstall anything, but anyway. ill run install again

[karmic crown]

open cmd and type python -version

does it work?

also try where python

[arctic abyss]

some weird directory...

[karmic crown]

that is fine

it should be there

that is default location for "use for myslef"

or whate that is called

if you type
python --version
does it work now?

also try
pip --version

if those two works, you installed python correctly

[karmic crown]

ok, 3.10.6, that should work, even 3.10.11 is recommended

and what about pip --version

[arctic abyss]

awesome. it opened properly

[karmic crown]

looks fine

leave it like that

it take some time to download and install everything, depending on your computer and internet speed

i need to go, i am tired and need to work in the morning
but someone else will help if you need anything else

its 1am here

[arctic abyss]

thanks so much for your help 🙂

[karmic crown]

come to the dark side and buy a Mac. I made a script for this - just one command and everything magically works 🙈

[golden arch]

Hi, thanks to @glad fiber and @ornate elk I was finally able to download SD. I did it very late, closed the folder and went to sleep and now I don't know how to open SD! can you please help me?

[onyx solar]

Find the web ui batch file.

[golden arch]

Sorry, I'm in the folder but find nothing...
C:\IA\stable-diffusion-webui

[onyx solar]

what do you see?

[late cove]

AttributeError: module diffusers has no attribute StableCascadeUNet

wtf is this error

[safe ferry]

I've always wondered why when you generate two pictures with exactly the same parameters they are not absolutely identical to each other? They look almost identical to a human, but some small details are always slightly different. Where does this randomness come from?

[onyx solar]

Depends. If using a GPU vs a CPU, that can cause differences. Could also be that a parameter you're not seeing is causing the changes. If you truly have all the same parameters across the board, including hardware, then it generally will be identical.

[soft fiber]

the randomness is part of how SD works, it's trained to create the image from noise

[safe ferry]

interesting, didn't know.

[safe ferry]

I think it is on every setup is so. You just need to look really close to notice difference in pixel noise and very small details.

[onyx solar]

You can do a bit by bit comparison to prove that an image with all the same parameters ran on the same hardware is the same.

[safe ferry]

Hmm, I see I have param Random number generator source set to GPU. I wounder how much will it be different between different options here.

yeah, if I use CPU it actually creates two identical images, but my GPUs RNG gives some randomness between them.

Does it make no difference which generator to use? It only makes a difference in the starting noise and has no effect on the final quality?

[viscid cobalt]

Hello everyone, I saw this function on the official website, but I didn’t see this model in the interface, and the help document I clicked on was also a 404 page. How do I use this model?

[torn sigil]

Gm

[vernal cargo]

CUDNN installed but won't show up in the kohya folder for some reason

[proper glacier]

I'm trying to train SD so that I can eventually make a video with certain faces, and I'm getting this error when I click to train... I'm pretty new to this stuff and the tutorials have options that arent on my panel. There's no "preprocess images" tab like the tutorial states. It says it's up to date, but do I have an out of date SD?

[unique hornet]

i amcoming back with the same problem now when i install new lora they wont show up now

[clever flume]

1. Install Python 3. 10. 6 (newer versions of Python do not support torch), and select "Add Python to PATH".
2. Install git.
3. Download stable-diffusion-webui 4. If your computer's video memory is less than or equal to 8G, please edit the webui--user.bat file and add the following parameters after set COMMANDLINE_ARGS=: -medvram (video memory is less than 8G ) or lowvram, (video memory is less than 4G)

這個怎麼使用

[viscid cobalt]

这个好像不用改，默认就可以

[vocal burrow]

yes you do need them as stated by the vram requirements. And english please.

[dapper basin]

trying to install stable diffusion from this tutorial: https://youtu.be/mcqREz9xE74?si=uKPy-S95X3ZF76ER
It shows an error in mine tho

[vocal burrow]

it s telling you everything you need to know

invalid python

use any python from 3.10.6 to 3.10.11

[dapper basin]

I tried using that version
then it just says no files found in python 3.12 folder

[vocal burrow]

(if you have a venv folder in your stable-diffusion-master, delete it after downgrading to a compatible version)

[vocal burrow]

probably because you didn't uninstall python 3.12 and/or didnt check the "add to path" checkbox during the install of the compatible python

https://cdn.discordapp.com/attachments/1002602742667280404/1075947280307331132/image-49.webp

[dapper basin]

ok
let me try all this

[ornate elk]

Wow what a bad tutorial

[dapper basin]

can you suggest me a better one?

[ornate elk]

Please consider doing a proper installation by using the guide of the pinned messages of this channel

[dapper basin]

ok

[ornate elk]

Chose the one matching your GPU type. Nvidia or AMD

[dapper basin]

this one right?

i have nvidia

[ornate elk]

Also make sure to uninstall python 3.12

Its not compatible

[dapper basin]

yes
just did that

[dapper basin]

After running the command it says following:

G:\Stable diffusion 2>git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
'git' is not recognized as an internal or external command,
operable program or batch file.

[ornate elk]

You have to install git

[dapper basin]

https://git-scm.com/download/win

installing from here

clicked a lot of 'next' buttons

[ornate elk]

Yea everything can be default

[dapper basin]

its installing now ig

[ornate elk]

Its for downloading SD, (not as zip) and later to update SD in the future

[dapper basin]

oh

is this correct

I have 4 gb dedicated memory

[ornate elk]

Looks good

[clever flume]

-lowvram

[fair oxide]

is it normal that my gpu isn't being used?

unless "memory" also counts vram

[ornate elk]

What's in your webui-user.bat and what's your GPU?

[fair oxide]

3060ti and I got --xformers --no-half-vae

[ornate elk]

Thats good, but for a ti you should add --medvram-sdxl

Also check the GPU vram usage when generating

In taskmanager under performance

Then on gpu

[fair oxide]

7.2/23.9GB

i guess it does

[ornate elk]

24gb is not the actual vram, thats with shared memory

Make also sure to always start the webui-user.bat and not the webui.bat

And yea add --medvram-sdxl for a speedup in generation when using sdxl models

[fair oxide]

k thx

[ornate elk]

Np 🙂

[ember crystal]

Hello, I'm expecting my new computer shortly. It's not very expensive, but it's better than the one I had. I've installed a 16GB VRAM graphics card; the previous one had 4GB. Now, let's get to the point. When creating images with SDXL, it can take up to 15 minutes with the following settings: 1134x833 upscale x2, hires steps 15, denoise 0.3, cfg scale 7, steps 30. How much do you believe the time will decrease with the 16GB card? For me, if it goes down to one minute, I'll be very satisfied. I know it might sound like a lot, but 15 minutes to 1 is a huge difference.

[ornate elk]

Hey, what are the GPUs?

[ember crystal]

4070ti super

[ornate elk]

That will be really good for sdxl models. You should generate a 1024x1024 in seconds and with upscaling by 2 maybe 1 minute

Make sure to use --xformers

[ember crystal]

I'm looking forward to trying it out

[fair oxide]

Grok went open source with public release of their weights and codes

around 318gb downloadable size

[grim iron]

how do I fix this? CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

I have tried restarting but I got the error again

using a1111 on a 3080

[valid flame]

y this error can anyone help me plz

[grim iron]

ur using the wrong vea

[valid flame]

tnx bro i will download sdxl so i can fix right

[vernal cargo]

anyone know why its saying I dont have python when I have python????

[soft fiber]

probably didnt set the path when installing

[vernal cargo]

wdym

I did though

and it says 311 when I have 3.10

I tried installing 3.11 but I can't have both 311 and 310 installed but I need 310 to be installed so I can use kohya but I can't use SD without 311 but I also can't use kohya with 311

do I need to just completely reinstall SD???

[soft fiber]

delete your venv folder wherever you installed SD

itl recreate one when you start up the webui-user.bat and that will resolve your issue

pretty common issue with that additional information, happens all the time

[vernal cargo]

okay thanks I will try that

[soft fiber]

oh wait i misread your message

SD does not support 3.11

youll want 3.10

[vernal cargo]

well it's what I was using lmao but it won't let me use kohya with it so I installed 3.10

[soft fiber]

oh okay i probably did read the message correctly then lmao

yeah so youll want to uninstall 3.11 then do the venv deletion step above

[vernal cargo]

so the kohya gui is what I use to train models right

[vernal cargo]

yeah I'm now reinstalling torch

[soft fiber]

looks like youre good

[vernal cargo]

thank you also after I install the kohya gui do I use that for making my own model?

[soft fiber]

yep thats right

[stuck wraith]

Hello, I have recently just started using automatic1111 and it has been working fine for the past couple of days but today, it has been using all of my memory and crashing my pc just from starting up. What can I do to fix this? I have 8GB of VRAM

[ornate elk]

hey, are you using an nvidia gpu?

if so you need some performance arguments in your webui-user.bat
--xformers --medvram-sdxl --no-half-vae
Add them to the line COMMANDLINE_ARGS=
Then relaunch the webui-user.bat

[vernal cargo]

trying to get it setup but have an issue where cudnn only installs to torch but not as a folder so I don't get the 'upgrade.ps1' file

is there any other way for me to install cudnn or do I have to use kohya > cmd > .\setup.bat

[soft fiber]

hopefully cs1o or someone else knows, im unfamiliar with kohya setup

[sacred rampart]

how do we write a propt? and which channel

[karmic crown]

this can happen for 2 reasons - wrong vae (as someone already wrote) or too big CFG (might happen if you are using lcm/lightning/turbo... those require very low CFG (usually between 1 and 2)

[placid pier]

Hey guys,

I wanted to use this upscale method. Is there a way to set the settings? or does it automatically use premade settings?

[stuck wraith]

Yes I am, I will give it a try and see how it goes

[ornate elk]

it uses the hires fix settings you can select

[placid pier]

The settings are different from my img2img settings that I usually use for upscaling.

[ornate elk]

not that much, you have a denois, some upscalers, set the hires steps to 10. and upscale by X depending on the resolution.

[glad fiber]

is it strictly needed that the training images for a lora be the same size?

[fair oxide]

moved this here since I believe it's the right place

the tool for PT to SAFETENSORS

has some screwed coding errors

I did set gdrive on my pc and put the right path

allowed it

installed all the requirements

pfew

so uh ...all in all can you give me some troubleshooting advice ? :))

[cursive flame]

Hello, can anyone recomend a good setup for someone getting started making video work with stable diffusion 1.5? Approximately $2500 budget, whether it's a laptop, desktop, or using service like RNDR. I'm a beginner, would appreciate any advice. Or if anyone can point me to a better place to ask these questions.

[grim iron]

would say a 4080 super

but where u from

because that changes with prices

[karmic crown]

4080 by itself is over 2000 euros here, for example

and he said 2500 for the whole computer, not just gpu

[fair oxide]

hello

does someone know what this error might be ?

I'm trying to fix it around

but it

it's just not going anywhere , like I am not into coding

[grim iron]

for me its 1300

so thast why I ask where he's from

https://pcpartpicker.com/list/

thisis for the us amazing

[stuck wraith]

I did this and it’s still freezing my pc. I also see in my task manager that’s it’s using 100% of my memory every time now.

[grim iron]

nvm its 1100 for me in the us its 1300

[karmic crown]

i agree with you that price depends on the local market

[fair oxide]

do you think that'd work /

also what path exactly ?

hmm

[ornate elk]

How much memory do you have?

[fair oxide]

like

uh

someone told me to use forward slashes

instead of backslashes

and I think the coding language it's python

is the writing in red generally a bed deal , like it announces an error ?

[unique hornet]

sorry to repost i am having the same problem again now all the new lora are now not showing up even when i refresh

[grim iron]

are they for sdxl? while using a 1.5 model?

[unique hornet]

both

[ornate elk]

change the model, loras are either made for sdxl OR 1.5

[stuck wraith]

16 GB

[late cove]

Should have gone into swap file

If it was using 100%

[ornate elk]

you may need to increase your windows pagefile then

[grim iron]

when ysing a 1.5 model u can only see 1.5 lora and vice versa

[stuck wraith]

@late cove @ornate elk how do I do either of these? I'm sorry, i don't know much about software.

[late cove]

Means it should have either started using system ram or storage to make up for the ram usage being 100

Which is why it slows down to a crawl

[unique hornet]

i would agree with you but some of my 1.5 arent showing

[stuck wraith]

oh I see what you mean. Yeah it does that but it takes 30 minutes to actually go to normal speed. Is that normal for it to take that long?

[ornate elk]

you can do that by following this:
https://www.thewindowsclub.com/increase-page-file-size-virtual-memory-windows
set min to 16000 and max to 24000
make sure the pagefile is set to C and disabled for other drives
make also sure to have 15-20gb free space on C drive
and restart the PC after that.

[late cove]

Doesn't go up to normal speed

Uses system storage as ram

And painfully slowly generates the image

[stuck wraith]

Okay, I will see how this goes

[unique hornet]

how can i just fix this

[late cove]

This my setup : 💀

[stuck wraith]

Why is it doing this all of a sudden though? It was working fine but now it’s so laggy and I have to wait so long. Before, it only took like 5 minutes for a big generated piece.

[late cove]

You got to many loras

Causing overflow of ram probably

Base sdxl is very easy to run

[ornate elk]

make sure to also disable Wallpaper Engine before using SD, as WPE causes freezes

[unique hornet]

even my emdedding arent showing up\

[stuck wraith]

Where do I find wallpaper engine at?

[grim iron]

hp r dell?

[late cove]

Dell

[grim iron]

thats where it goes wrong

[late cove]

Swapped psu

Upgraded CPU

Works decent

[ornate elk]

If you dont know, you mostly dont have it

Make sure to close heavy programms in the background

[grim iron]

I might have a hate against dell

[fair oxide]

Eh ?

[golden arch]

Hi, it's me again.
I'm trying to open SD with webui-user but this is the result...

[grim iron]

it has to be --xformers not xformers

[noble belfry]

i dont know how well xformers works with v1.8, or more specifically the torch and cuda versions it uses

[golden arch]

Thanks, so I have to open webui-user every time I want to use SD, isn't there a more direct interface?

[ornate elk]

You can make a shortcut to it

[golden arch]

How?

[ornate elk]

Right click the webui-user.bat and press Send To, Desktop

[little echo]

curious is there a reason why img2img tab does not support different upscalers too (r-ersgan, swinir, etc)

[ornate elk]

It does

You have to select SD upscale script to use them

[little echo]

not in forge then?

[noble belfry]

it does, open the scripts dropdown menu 🙂

its at the very bottom of img2img tab

just be sure to set your denoising level low, otherwise it will wrestle with you to make a new image over the current one 🙂

[little echo]

ohhhh

thereotically you can upscale twice then?

[noble belfry]

thats what i do for best results. i normally only scale factor 1.5 or lower each time, it keeps things consistent. a bit more work but on photographic prompts it adds in crazy detailing for pores, eyebrows, skin creases

[little echo]

i'm gonna go all out just to see what the detail is
2048x2048 r-esrgan 16x

i wonder if 4090 can handle it

[noble belfry]

it works by breaking the image into tiles and works on them as individual jobs so it can, itll just be a longer job.

[little echo]

so it'll be 32k wow

[noble belfry]

and you wont be able to use your computer for half an hour 😄

[little echo]

already 2048 x4 is taking over an hour

[noble belfry]

as i say though, low denoising is the key

[little echo]

i am doing de 0.35

hires 150 i guess is default

on a dp++2mk sa 60

[noble belfry]

60 steps is kind of pointless for an upscale

its not creating anything, just filling in and making things bigger, you could easily half that, but go hog wild and feel it out. i'd recommend youtube tuts on it too, theres a bit of a feel for it you'll get watching someone else use it

[little echo]

why? there can be significant changes to the image between 20, 60, or 150 steps

really just depends on if the output is good

are you saying higher sampling is also like a denoise?

[noble belfry]

yes and if youre upscaling what you want are LESS significant changes and only subtle ones.

[little echo]

the problem is getting perfect arms/hands on people before upscaling

[noble belfry]

steps cfg and denoise settings are all something you'll have to play around with. theres no specific scale

[little echo]

i knew this would happen but i am testing anyhow

[noble belfry]

yeah work on that first 🙂

[little echo]

if i use a hand lora would it be good for distance shots

any good guides on getting perfect anatomy? may have to check civit
esp. when the scene contains many people

think it's pretty achievable reliably?

[noble belfry]

the more people the vaguer it gets, and inpainting is a bit crap on SDXL

[little echo]

yikes

[noble belfry]

theres an extension called Adetailer which is useful

id honestly suggest walking before running, a busy tableaux type image upscaled to the size you could put it on a buildings wall is achievable, but you're best getting there in steps.

[hasty orbit]

Is there a way to make the preview image closer to the final image? I'll have an image that looks perfect in the preview, and then at the very end everything gets messed up and the final image is an abomination.

[noble belfry]

are you using latent upscaler by any chance?

because that can trash a perfectly good image.

[hasty orbit]

Yes

Which one should I use instead?

[noble belfry]

try one of the other upscalers instead, its utter rubbish

what kind of image is it?

[hasty orbit]

Oh my gosh!

That REALLY fixed it up!

[noble belfry]

my spider senses tingled, it happens to me once in a while when i forget to set the upscaler and i get so mad 😄

[hasty orbit]

What's a good all-round upscaler?

[noble belfry]

these are my ones, i only have 2 you need to download manually

4x-ultrasharp is really good for photorealism and esrgan_4x i find best with vectors and illustrations. but you might just find your own favorites

there used to be an upscaler model wiki for downloads but they've moved it or something

[little echo]

yea, is there a guide how to install it

[noble belfry]

oh you just pop it into a folder in your models folder

[little echo]

when i checked the site for it seemed to involve editing a py script

[little echo]

image isn't loading

[noble belfry]

its just an image of forge\webui\models\ESRGAN where you can put the models in

https://upscale.wiki/w/index.php?title=Model_Database&oldid=1571 heres an old version of the upscaler database

[little echo]

nice

shows when i click it

[noble belfry]

discord is being weird these days

anyway, 4x-ultrasharp is my fav all rounder model

[nova ermine]

What would you do in order to better this face?

[noble belfry]

bit of a broad question. do you mean with settings or prompts?

[nova ermine]

Anything you can consider if that betters the face, since the rest of the image is pretty nice

However, I've already included lots of prompts for the face in here

It is understable considering this is the reference image (controlnet included) and the face is just a drawing witch is not blended so well. But it should be a way to better results

[noble belfry]

is this an SD1.5 or SDXL model you're using?

i would chop so many of those prompts out if its SDXL

https://discord.com/channels/1002292111942635562/1011743094309396631 this is a good channel for promptcraft btw

here is primarily "why computer no go brrr?" type questions 😉

[nova ermine]

So let's go there?

[nova ermine]

https://civitai.com/models/118086/3d-animation-diffusion My checkpoint

[noble belfry]

if you're using controlnet, you could also experiment with cutting its ending step down to 0.8

[little echo]

I don't understand, why does it go under esrgan/

then where are all the other upscaler models

[nova ermine]

Are you meaning I set this down?

[noble belfry]

yep, just to see. also things like reduce or increase your CFG as that is an enormous prompt

if you drop the ending control step down then it lets the model try to generate the last bits on its own and add in little things

[nova ermine]

Yeah it may work!

[noble belfry]

might make it worse too 🙂 its like letting go of the handlebars

[nova ermine]

Talk about this point

in #📝｜prompting-help if you want, but I'm interested on that

[noble belfry]

sure

[golden arch]

Hi guys, now I have this problem
SafetensorError: Error while deserializing header: MetadataIncompleteBuffer

[ornate elk]

The upscalers you can download on openmodel db are all based on the Esrgan technique, thats why they go into that Folder

[ornate elk]

When do you get this?

A bit more info would help

[golden arch]

Ok, finally SD is running, I wrote some prompts and this is the result.

[ornate elk]

Maybe you can show your settings or the full cmd log

[golden arch]

Ok.

[rain stump]

Hello, can someone please help my dumb ass make a lora? I already make the dataset and all, I was trying to do it in kohya and I was following a yt tutorial, but I can't do the training part
thank you

[ornate elk]

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

Also what model did you tried?

[ornate elk]

Where is the issue?
But remove the space in the image folder name

[golden arch]

set PYTHON=C:\Python\Python310\python.exe
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --no-half-vae

call webui.bat

NVIDIA GeForce RTX 2050
I don't know the model, just followed your installation guide step by step.

[ornate elk]

You need to add --medvram

[solemn atlas]

Hey guys I have a big problem. I asked someone and he kindly helped me out to understand SD better and how to generate just like him. My SD did not do the same results as his. I used 1:1 the same parameters, vae, prompt, neg prompt, everything. Our results are very difficult. The outcome is very difficult. My version is 1.8 and his 1.6. That is the only diffrence

So I delated SD completly, download it again, and it still does not work

this is my out come of the image with the same lora etc

and this is his. We tried everything but I still get it so diffrently as you can see above

[ornate elk]

1.8 had a change that is "seed breaking" I'm not at PC rn to tell you which. But its listed on the github changelogs.
Also you both need the same GPU type or it won't work. Also you need to both have RNG CPU selectet or it would be still different. Also xformers makes minor detail changes

[solemn atlas]

As I know we have the same VRAM, is that important too?

[ornate elk]

Nope

[golden arch]

In COMMANDLINE_ARGS?

[solemn atlas]

so that means I first need the same Version which is 1.6 . But how is it possible to get that version back again?

So it's not possible to generate like his image at all even though the lora was created by him and the outcome should look like that

How do I or we select the same RNG CPU

[ornate elk]

No it wouldn't bother at this point. 1.8.0 is the better version.
You won't get 100% the same image but the lora should work the same for you as for him

[ornate elk]

You can select the CPU RNG under settings and then search for Seed

[solemn atlas]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-NVidia-GPUs I downloaded it from here

[solemn atlas]

And that is what i also wonderd. Why is my image not even near to his? I understand if its not 1:1 but at least the character should be regognizable

[ornate elk]

Do you have the meta data of his image?

And yours?

I could compare it

Nvm its included

[solemn atlas]

Meta data means from the black window

I only have mine but I can ask him tmr

[ornate elk]

No its in the image

I found it

[solemn atlas]

Okay (I dont know what that is but I'm happy you found it

We tried to find out what it could be for at least 4 hours or more

[ornate elk]

Okay so it seems you didn't used the same settings as him. Even the seed is a different one

[solemn atlas]

We really used the same settings. Same Prompt, same neg. prompt, same vae, model etc.

But what is the diffrence in these two images

I did everything he wrote so we have the same settings

[ornate elk]

No your seed is different also the denois strength is different

And more

[solemn atlas]

What should I change so I have the same as his. It's weird because I copy pasted everything he send me. I was sure I did everything just as the txt.file

[ornate elk]

@solemn atlas To check for yourself do the following:
Drop his image into the PNG-Info tab in auto1111.
Then you see all the Settings he used.
Screenshot it.
Then drop your image in and compare

[noble belfry]

seeds dont work on different computers. i think theres one old sampler where it does

but you will not get identical images.

its not the same as random seeds for map generation on say a video game

[ornate elk]

Seeds work if the RNG is set to cpu

[noble belfry]

yes on both sides

[karmic crown]

@solemn atlas if one of you have nvidia and the other do not, you wil NEVER get the same image

unless you switch to CPU generator

[solemn atlas]

I put it into the PNG Info and then just "Send to txt2img" and used his parameters but it's still as mine

[karmic crown]

it is in here

[noble belfry]

ugh im an idiot i thought that was specific to one of the samplers

[ornate elk]

Then it should be a different image to your last one

Because it would have used an other seed and denois

[solemn atlas]

I just looked it up. I have Nvidia but it is on GPU

[karmic crown]

if you both have nvida and both on gpu it is fine

[fleet obsidian]

Is there a text channel for developer based question for Automatic1111’s webui repo?

[karmic crown]

i learned that lesson the hard way

[noble belfry]

theres github

[solemn atlas]

Ohh I understand. I never heard of that. I'll try to change to NV and look if maybe that was the problem all along

[karmic crown]

i am on mac and i was desperately trying to recreate image from civita (made with nvidia) for 2 days 🙈

[noble belfry]

i do think though that seeds across clients should be a thing for the future as a standard. i like the idea of collaboration

[karmic crown]

then i tried on my sons pc (with nvidia) and got the 99.5% identical image from the first time

[solemn atlas]

I have Nvidia but my settings were on CPU I guess maybe that was the problem. So how did you find it out and could you generate them just like on civitai?

It still reloads. I didn't know it takes so long

[noble belfry]

civitai prompts are weird. so many photos of a beautiful unicorn have crazy prompts and you could tell the creator was going for a horror scene but missed

[solemn atlas]

can't wait to try it out

[fleet obsidian]

Maybe my question is generic enough for here. I’m curious about tile based Upscaling with img2img and ControlNet through an API. Mostly want to know what the flow should be, e.g Generate image > ESRGAN upscale > img2img ControlNet > ESRGAN upscale > img2img ControlNet
Maybe I’m wrong in how I’m thinking about it

[ornate elk]

If it was on CPU and his setting not, that is a big difference yes

[golden arch]

I added and the result is

[ornate elk]

Can you go into the stable-diffusion-webui/models/Stable-diffusion folder?

There check the file size of the model pls.

[solemn atlas]

😦

It did not worked at all

[solemn atlas]

does any of these other settings help

[ornate elk]

@golden arch I guess the model could be corrupted.
Best is to download this one from here:
https://civitai.com/models/4384/dreamshaper
And put it into the models/Stable-diffusion folder.
Then select it in the checkpoint dropdown and try again

[karmic crown]

you need to press both of those

[worldly grove]

hello does it really take 10+ minutes to generate an image on 3080ti?

https://i.imgur.com/dLUoLto.png

[ornate elk]

Hey, what webui do you use?

[jaunty seal]

I dunno if this question goes here but...
How can I link the models from my Automatic1111 webUI folder to the Stable Swarm UI (or a custom folder located on another drive)

[worldly grove]

@ornate elkpinokio installer

[ornate elk]

For auto1111?

[worldly grove]

https://i.imgur.com/E1rLbz0.png

[golden arch]

Ok, I think there's no model...

[rain stump]

Thanks for answering
I don't really know where is the problem, I got stuck in this part from the video
I left the space because the guy from the tutorial also left it (2nd image)

Something I have to highlight is that I reached min 46:17, saved the configuration, and closed Kohya, logged in again and loaded the lora configuration as it was.
I don't know if my mistake is that Kohya no longer detects where I put my image dataset???

[ornate elk]

Oh stable cascade. Thats a special new thing. Thats not the normal Stable-diffusion webui.
So yes it could be normal that it takes so long

[ornate elk]

Its in there but the file size is not correct. Please download the one I linked you

[worldly grove]

@ornate elkcan u recommend anything else with faster renders?

[golden arch]

Coming

[jaunty seal]

comfyui or stable swarm could work

[ornate elk]

Sure, if you want to use the normal stable diffusion webui most people use then follow my guide:
#🤝｜tech-support message

Its much faster

And easier to use

[jaunty seal]

@ornate elk do you think you can help me with this problem?

[worldly grove]

okay will try it now thanks

[jaunty seal]

since I am having difficulties with configuring it

[ornate elk]

I dont think stableswarm has args for adding paths. So your only option is to use the good old Symlinks

[jaunty seal]

it does have this tho

doesn't this work?

since I tried placing windows paths but it didn't detect them

[ornate elk]

Oh indeed that could work

Make sure the path ends with /

[jaunty seal]

E:/ai/stable-diffusion-webui/models/Stable-diffusion/
This doesn't seem to work

[solemn atlas]

I really don't know. So i changed from RV to GPU and to CPU and the outcome is still the same. Shouldnt that at least look at bit diffrent? But it looks the same

[jaunty seal]

I even restarted the backend

[ornate elk]

Look near the bottom it locked your RNG settings

You have to click the X

To remove the locked setting

Or it won't change

[jaunty seal]

wait no it worked

it seems that it requires the entire UI to be reset

[solemn atlas]

where is the X ? Sorry i cant find it

[karmic crown]

see the picture i posted

here

[jaunty seal]

yeah I just needed to set the root path for models and everything else was picked up automatically

[solemn atlas]

Ohh okay, should i remove both?

[ornate elk]

Good to know

Its like comfyui then (as its the same Backend)

[solemn atlas]

It's still the same. Didn't changed a bit

[ornate elk]

Need to go off now gn 😴

[solemn atlas]

I'm really confused

[jaunty seal]

Stable Swarm seems to be missing the comfyui manager

[karmic crown]

i am off too its 1am, and i am awake from 5am

[solemn atlas]

good night

[golden arch]

It worked! Thanks!

[vernal cargo]

trying to get kohya setup but have an issue where cudnn only installs to torch but not as a folder so I don't get the 'upgrade.ps1' file
is there any other way for me to install cudnn or do I have to use kohya > cmd > .\setup.bat

[arctic abyss]

Hey guys, so I've successfully run the webui.bat using python. Im unsure what to do next 😄 does that mean its installed?

Oh maybe it didnt work given what its saying in the text above

[ornate elk]

Hey, what's your GPU?

[arctic abyss]

Hi, @ornate elk

[ornate elk]

Okay, make sure you update your nvidia driver to the latest version, and restart the PC

Then relaunch the webui-user.bat

[dense delta]

Any local open source tools describing images and generating texts?

[arctic abyss]

Nice! I have webui forge going (web browser based, didnt know that), now i am tring to add some models for it. Noob question, how do I add them from here? Im used to github and adding its url, but this, unsure.

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

[arctic abyss]

referencing this message from Wit 🙂

[grim iron]

how do I put a max on the amount of vram sd uses

[ornate elk]

You can try add --medvram to the webui-user.bat

[grim iron]

I already have that

[ornate elk]

These models have to be downloaded manually and place inside models/Controlnet

[grim iron]

but I mean like a slider or something

[vocal burrow]

nope

[grim iron]

becaus normally I can just use everything, its jsut that I now need to have my wallpaper enginge on because clock

[ornate elk]

Wallpaper engine causes stutters and freezes and even PC crashs when used at the same time while SD generates images.

[solemn atlas]

Good morning guys, about my problem yesterday that we can't generate the same images with the same settings
These are his version: version: v1.6.0-2-g4afaaf8a  •  python: 3.10.6  •  torch: 2.0.1+cu118  •  xformers: 0.0.20  •  gradio: 3.41.2  •  checkpoint: 354b8c571d

and these are mine: version: v1.8.0  •  python: 3.10.6  •  torch: 2.1.2+cu121  •  xformers: N/A  •  gradio: 3.41.2  •  checkpoint: 354b8c571d

[grim iron]

yeah ik

thats why I ask for the slider

[vocal burrow]

Why do I get different results from civit.ai/my friend/alien lifeform examples :

1/ --xformers (and some other flags but mainly this one) will make stable-diffusion non deterministic. Meaning for the same prompt+settings running on different hardwares you'll get different results. The output should still correspond to whatever the prompt is.... It might just be slightly to very different if you're unlucky. So yeah, some command line arguments can alter the outputs of your prompts and your hardware might also do the difference.
2/ Do you have all the loras, embeddings mentioned in the prompt ?
3/ Are you really using the same settings ? Copy pasting manually what's inside the "Generation Data" boxes is not enough. Civitai is hiding a bunch of data there. If you want to get the exact same settings you'll have to click the "Copy Generation Data" button at the bottom of the image page, paste its content into Auto1111's prompt field and then click the blue arrow under generate to automatically apply each value to the correct field. Hidden values can be; clip_skip, ENSD, token merging ratio, etc
4/ Did you have any overrides set beforehand ? They should show up at the very bottom of the page (some will probably show up too if you use data from civitai like I mentioned in 3/)
5/ Maybe it's using some extensions that does not record its settings in the metadata
6/ Maybe you're using different versions of some extensions/auto1111/lora/models/etc that yield different results
7/ Are you using the same random number generator ? GPU / CPU / NV (Settings->Stable Diffusion->"Random number generator source")
8/ Maybe you're using a newer version than OP. One that includes a "seed breaking change", cf https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Seed-breaking-changes

tldr don t waste too much time trying to replicate images exactly

[solemn atlas]

the thing is, the Lora is the character I want to create. He created a Lora for that specific character. I don't want to copy 1:1 his image and pose but at least get the same hair, expression etc which is recognizable for that character but when i use the lora it looks completly diffrent. I only want to generate her as she is (with a few diffrences ofc but to get "her" into the image" And I don't know why it doenst happen

Is there a way to adjust xformers?

[vocal burrow]

you set it on. or off

that s it

[ornate elk]

There is no slider. What you can do is to get yourself Rainmeter and a simple clock widget for it. And then disable WPE

[grim iron]

ok thanks

[vocal burrow]

(or look at your watch, smartphone, clock in the start bar, open a browser page with a clock on it, etc)

[solemn atlas]

where can I do that? I looked at SD but I don't know where to try it out

oh xformers are just the NV GPU and stuff I understand

[ornate elk]

Xformers Is on when you added --xformers to the webui-user.bat

[solemn atlas]

is this on settings?

[vocal burrow]

set COMMANDLINE_ARGS=--xformers --whateverIsAlreadyThere

[solemn atlas]

So I go to webui-user.bat and open cmd

these things are in my black window thing when i open sd

there are a lot of errors

[vocal burrow]

what s in your webui-user.bat ?

[vocal burrow]

that s your log, not what s inside your webui-user.bat

[solemn atlas]

Oh I#m sorry it's this

[vocal burrow]

so yup you have litterally no arguments whatsoever ....

what s your gpu ?

[solemn atlas]

I'm not sure but GPU is this right

[vocal burrow]

set COMMANDLINE_ARGS=--xformers --no-half-vae
then

[solemn atlas]

okay I did and then open it again

[solemn atlas]

Okay I opend it. Should I now put this too and remove the other one

[vocal burrow]

use this one .... set COMMANDLINE_ARGS=--xformers --no-half-vae

[solemn atlas]

I did

[vocal burrow]

then there you go

xformers is on

(and whatever error you had at startup because of some extension is probably still there)

[solemn atlas]

The outcome is still the same and yes the errors are still there. It says that " Unable to load ESRGAN model" and this "RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory"

[wary wind]

hi, i cannot find any # dreambot in left hand side, may i know it is normal?

[solemn atlas]

I even re-installed SD and still have problems. I don't understand what is wrong. Even after downloading Comfyui there is not bat.file. Something is off with my whole system

[ornate elk]

Can you screenshot the whole cmd with that error?

[solemn atlas]

it's this

[ornate elk]

Does the webui open in browser?

Do you have access to the Extensions tab?

[solemn atlas]

Yes it opens and I also have acces

[ornate elk]

Then click check for updates, let it load and then restart the webui-user.bat

[vocal burrow]

click apply update before restaring the webui

[jaunty seal]

In Stable Swarm UI I have a problem where my controlnet preprocessors aren't being showed I am just stuck on having canny as an option.

[solemn atlas]

I did

It's still these problems

Unable to load ESRGAN model
RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory

I think i know where the problem comes with the ESRGAN model

I copy pasted these folders before delating SD and when putting all files back there was not ESRGAN folder and i just created a folder and named it that way and put all upscalers in there

[ornate elk]

You should have had this folder in models/ folder
When installing sd

Otherwise something is broke

[solemn atlas]

It was not there, I also wonderd thats why i created it

[ornate elk]

Maybe you should do a clean install of the webui

[solemn atlas]

So I need to delate SD again?

[ornate elk]

What you can do first to try to fix your webui:
Rename the Extensions folder to _extensions
Then delete the venv folder.
And restart the webui-user.bat

And then check if you get any errors

[gusty geyser]

anyone manage to get stable diffusion run on intel core ultra integrated gpu?

[ornate elk]

You mean Intel UHD 620 ?

That won't work

SD can only work with Intel Arc or maybe IrisXE GPUs

[gusty geyser]

no

[gusty geyser]

*it should be called intel arc graphics

i tried using the accelerate with openvino script but it isnt using the gpu at all

[solemn atlas]

So I did and the error is still there

[split robin]

I got CUDA runtime error
and it's like, felling all my RAM and VRAM...
I found out this issue when I searched about the commandline too, any fix for that?
RuntimeError: CUDA error: out of memory
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with **TORCH_USE_CUDA_DSA** to enable device-side assertions.

[gritty compass]

When will the bot be fixed?

[ornate elk]

Okay, then for Intel arc you need to use the SD.next webui

[ornate elk]

Then you need a reinstall

[ornate elk]

What settings did you used?

[split robin]

which settings?

[solemn atlas]

from which page? I did that yesterday too

[ornate elk]

Here is my install guide:
#🤝｜tech-support message

[split robin]

I'll start it later, I'm updating a driver xDD

[solemn atlas]

Or let's do it on weekend. I will message again.
I forgot I have an exam I need to learn for

[ornate elk]

Resolution, model, etc

[split robin]

oh, it was 768^2 with 2x upscale

idk, it's pretty normal settings tho

[ornate elk]

Hmm

Try a restart and test again

Make sure wallpaper engine is not running in the back

[split robin]

my wallpaper is literally a 768^2 image I generated xDD

can it maybe because I install a plugin?

[ornate elk]

Yes, which extension?

Some of them can break stuff

[split robin]

well, this is a potato gpu for image generator afterall.. gotta deal with it xDD

[ornate elk]

Do you have any errors in the cmd?

[split robin]

I only see this

[ornate elk]

For the Photoshop plugin you would also need to add --API to the webui-user.bat

[split robin]

Ooooo

okok

[ornate elk]

Easy photo extension won't work on your gpu

[split robin]

ah, sadge

[ornate elk]

Or wait how much vram does it have?

[split robin]

RTX 3050 Laptop 4GB

[ornate elk]

Okay nope

[split robin]

lmao, sadge

[ornate elk]

You can delete it from the Extensions folder

[split robin]

done

the webui prompt all in one is helpful

[ornate elk]

Boorutag autocomplete is also a good extension

[split robin]

btw, how do I know what version of SD do I use in AT1111?

[ornate elk]

And you should get the tiled vae extension

[split robin]

or it can automatically use any SD version?

[split robin]

interesting, what it do?

[ornate elk]

With the auto1111 webui you can use any SD model version (1.5, 2.1, sdxl)

[ornate elk]

It helps when using Hiresfix upscale to not run out of vram

Making it easier to upscale higher

[split robin]

oh ok

this one?

[gusty geyser]

but it is intel arc on intel core ultra cpu? can it still use ipex?

[split robin]

you're in core ultra now? how's the iGPU performance?

[ornate elk]

Yes it supports Openvino and ipex

I can't guarantee that it will work, but its your best chance

[ornate elk]

Yes

[gusty geyser]

alright. i will try it with sd.next

[ornate elk]

Best is to read on their github for the Intel instructions

[gusty geyser]

still haven't got it to actually be used in sd

[split robin]

oh ok

[gusty geyser]

i couldnt find documentation made for it, those intel arc guides out there are made for their discrete gpus

[ornate elk]

On SD.next it says igpus are not supported with Ipex. Use openvino instead

[split robin]

do I need to restart the AT1111 or ReloadUI is enough?

[ornate elk]

Best is to restart

[split robin]

hmm, okok

oh nuts, apparently I have to turn this on xDD

yes, the extension really useful, thanks

[ornate elk]

Yes enabling it is enough. Also you dont need to enabled tiled diffusion

[split robin]

I don't even know what it does, so I'm not enabling it xDD

[sage lava]

Hey I've got an issue with the fetch_uri for my queued images

if I call it with curl I get

{"status":"error","message":"Invalid API Request","tip":"1. Make sure you are passing Content-Type: application/json in header. 2. Make sure you are doing POST request with valid JSON. 3. Make sure your JSON does not have error, use jsonlint to validate json array"}%

going a bit mad with this one, the api key is definitely correct and is used earlier to generate the images in the same package

[vernal cargo]

hey so I trained a lora model using kohya but its a JSON file instead of a 'safetensors' file

[gusty geyser]

I got the following error every time I want to generate a slightly higher resolution image. I'm using openvino acceleration

BackendCompilerFailed: backend='openvino_fx' raised: InvalidCxxCompiler: No working C++ compiler found in torch._inductor.config.cpp.cxx: (None, 'g++') Set TORCH_LOGS="+dynamo" and TORCHDYNAMO_VERBOSE=1 for more information You can suppress this exception and fall back to eager by setting: import torch._dynamo torch._dynamo.config.suppress_errors = True

here is the full error msg

[glad crescent]

hi team, hope you are doing good

can someone help on this issue?

[vapid sphinx]

So there are so many versions of ControlNet canny, depth, lineart etc. for SDXL 1.0 and from like 10 canny models, only 2 work good out of the box. I tested them all. Why is that and why the controlnets for 1.5 models were better?

[jaunty seal]

Stable Swarm UI problem here. It isn't detecting any of the controlnet preprocessors only canny no matter what other model is chosen.

models used: https://huggingface.co/comfyanonymous/ControlNet-v1-1_fp16_safetensors

[dense tapir]

Any tips guys?

https://github.com/danielgatis/rembg

This is supposed to show up at the Extras tab, however it simply doesn't.

[vocal burrow]

use this one instead https://github.com/AUTOMATIC1111/stable-diffusion-webui-rembg

[glad fiber]

if I use the lora and the wb17myself in the prompt, the output image is nothing like what i want

how can i fix this?

[noble belfry]

dont know that model so don't know what triggers it best or the best strengths to use. look up wherever you got it and see how examples use it i guess

for stable diffusion theres a nice little civit model helper, but im not sure if it works for downloading anymore as the site changed things around and requires auth. but anyway what it did was fill out your lora with information like how to trigger it

[glad fiber]

this another version I have the lora is applied

[noble belfry]

wait, is this a lora you trained yourself? just looked at your name

[glad fiber]

yes

[noble belfry]

oh right. damn there could be a dozen different things causing that which happened during your training

[glad fiber]

i guess it is related to my prompt or the captioning

[noble belfry]

olivio sarikas' discord has a pretty good training channel, i could send you an invite if you'd like to ask there, they're good with in depth things like that

[glad fiber]

send it

[cursive beacon]

Can I ask about those online models which generate text exceptionally well, how do they do it?
compared to SDXL models

[noble belfry]

are you asking how the models work?

if you want to make them locally i could link you a really good video that dropped today, but its comfyui workflow

[terse mulch]

How can I train pix2pix in SD to apply a style based on training data that consists of paired images, where A is a piece of artwork, and B is an image based on that artwork that is potentially very different. e.g. the 'style' image is outlines traced on the artwork. or the 'style' image is normal maps that were manually painted to match the source image.

i.e. I have a piece of software that animates images, and uses multiple textures based on that image. I want to train SD to be able to generate each of the different texture types, based on those which I have already done manually.

[inland sierra]

hi guys, anyway to get more credits for stability . ai? I think will need more to test the app

[noble belfry]

controlnet has a good feature, 'reference', you might want to read up on that. its very good at translating the feel or style of an image over to another. https://github.com/Mikubill/sd-webui-controlnet/discussions/1280

its essentially pix2pix

[mellow wagon]

Hello, I keep getting an omegaconf error regarding a missing key for ddconfig. I'm trying to get SVD running on a M1 Pro Macbook

Here is the traceback, I'm using the stable-diffusion-webui

[gusty geyser]

any help on this? thx

[rapid gale]

whenever I use embeddings such as "easynegative" in the negative prompt I get the "RuntimeError: expected scalar type Half but found Float" error.

[terse mulch]

how do I actually use this. It just seems to be a discussion about that feature? Is it just in SD if I update to 1.1.171?

[noble belfry]

controlnet works in a1111 and comfyui

both have seperate extensions, that was just a quick overview with pictures in case it looked like something you'd wanT*

lol i accidentally typed want with another character and got blocked by the server!!

[terse mulch]

So then to have SD take an existing artwork, and generate the different versions of it, I would train a style that is all of the texture types I have manually generated. Then use a new image I want transformed in the same way as controlnet image?
I feel like something where training accepts both 'original artwork' and "texture based on that artwork" is more likely to get the results I want "X artwork changed to Y texture in this way". So that I just give a new X artwork, and it changes it in the same way. But I"m not sure what functionality in SD trains "image A turned into image B".

[gusty geyser]

whenever i run lora script, this also happens. any clues why?

[errant edge]

hey just starting out using fooocus and im currently trying to recreate a generated picture on civit.ai
i tried copying the models, prompt, negative prompt, the seed but my image is entirely different, not even in the same artstyle
anyone care to help me out figuring out what the problem is?

[noble belfry]

you wont get the exact same image, the generation from someone elses random seed is different to if you run it on your own. also a lot of civit generation data isnt easy to copy over, they might have used other upscalers or inpainting.

[errant edge]

oh yeah that would make sense

but that big of a difference? LOL

[noble belfry]

oh hell no, thats a completely different model lol!

[errant edge]

yeah im confused because as i understand it, im not allowed to change the basemodel to the "vixon's fantasy mix" checkpoint

so i tried to understand how refining works

i assume theres the problem, but no idea

[noble belfry]

the one on the left seems to be using an anime trained model, you wont get anything even a mile close to that. the two model flavours interperet the prompt differently too. anime models wouldnt know what to do with a photorealistic prompt and vice versa

[errant edge]

huh, isnt the "resources used" category telling me what models they're using?

[noble belfry]

by "not allowed to change" do you mean its running on a server?

[errant edge]

no it gave me the error "You have selected base model other than SDXL. This is not supported yet." when i tried to

after googling i came across this:

so i went onto the vixon's fantasy mix model's page and it says "base model SD 1.5", due to the look i assumed it counts as "anime"

so i figured i put 0.667 on the refiner

[noble belfry]

what client are you using? is it local or running off someone elses computer?

[errant edge]

its local

[noble belfry]

if its local you can literally just download that model

[errant edge]

thats what i did

[noble belfry]

wait, which client are you running? a1111?

[errant edge]

fooocus

[noble belfry]

ooooh right

[errant edge]

so as i understand it, it doesn't support changing the basemodel or something

im clueless

im just trying to recreate similar images 😦

also unsure if i should change something here?

or whether i should use a style

[noble belfry]

its a bit undersupported, i dont know many people that use it that could talk you through settings mate

[errant edge]

so im better off looking for SDXL checkpoints?

at least i assumed using the SD 1.5 as refiner it would at least get me closer to the result i want

[noble belfry]

no i mean rooned. not many people use it so dont come across problems and can help you. you could bump this again tomorrow maybe during the day

i say during the day, i mean its late in US and europe

[errant edge]

maybe thats why

[errant edge]

yeah using lora based on sdxl 1.0 it works much better

[grim iron]

u could join the sd sub reddit discord the developer of foocus is in there

[errant edge]

with the help of @noble belfry i just switched to forge and having great results so far

[grim iron]

that also works XD

[inland sierra]

anyone successfully build https://github.com/Stability-AI/generative-models on Apple M1 pro? keep getting this error and related issue (but closed) https://github.com/Stability-AI/generative-models/issues/125

ModuleNotFoundError: No module named 'torch'

[split robin]

isn't torch python module?

[grim iron]

try pip install torch

[ornate elk]

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

[rapid gale]

my gpu is a gtx 1660 ti and this is inside my webui @echo off

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

[ornate elk]

Okay for your GPU you need to add
--xformers --medvram --no-half
To the line COMMANDLINE_ARGS=
Then save and relaunch the webui-user.bat

[rapid gale]

still dosnt work

[ornate elk]

Thats strange, it should work.
Does the cmd shows any error?

Make sure you always launch the webui-user.bat and not the webui.bat

[rapid gale]

Hoooooolyyyyyy shhhiiiiiiiit im stupid 😂 thank you so much its fixed now.

https://tenor.com/view/crying-meme-black-guy-cries-sad-man-thank-god-for-my-reefer-hood-news-gif-1799983865423589252

[ornate elk]

No problem 🙂

[errant jewel]

Hellow
I have a mainboard with 2x pcie 16 slots
I have a rtx 4080 running in it
can I use a rtx 4000 ada together with it?
Does someone run two different gpus in their system?

[vocal burrow]

only one gpu can be used at a time.

(but you can run one instance on each of your gpu, stable swarm UI might help you with this)

[errant jewel]

Oh ok thanks
I wanna do dreambooth training and maybe I can do that on two gpus simultaneously

[vocal burrow]

I don't think so either

same problem, you can't unify their vram

[late tree]

I'm getting an error when trying to generate, it goes through the whole thing but then fails at the end saying:
Cannot set version_counter for inference tensor

[inland zephyr]

Ay I see this in my cmd prompt startup, any of them worth using?

[karmic nova]

does anyone know of a dataset that has just one person's face in a lot of different poses and angles?

[vocal burrow]

cf https://github.com/lllyasviel/stable-diffusion-webui-forge?tab=readme-ov-file#forge-backend

[inland zephyr]

ty

[jade finch]

is there a torrent for automatic1111, installing the portable version always fails and trying to use git fails as well

[gusty geyser]

if you are still wondering, i got about 1.3it/s on openvino and 1-2s/it on ipex

[sudden hawk]

hello !

I need help regarding Stable Diffusion, I have installed Python and stable diffusion, i follow all the instructions but now i still can't launch it. and the first time it launched automatically I got an error message and I couldn't generate.

If someone can help me pls 🥹🥹

[jade finch]

what webui did you install?

[ornate elk]

hey, whats the error? and whats your gpu?

[ornate elk]

Here is a full guide to install auto1111:
#🤝｜tech-support message

you would need to install python onto the external drive aswell as SD

[jade finch]

my problem is that all standalone portable versions of automatic always crash for me, my internet connection over VPN is not stable enough and keeps crashing

that's why i am hoping for a torrent that doesn't restart every time your internet connection gets lost

[sudden hawk]

it’s a red error msg up to the right in the screen
my GPU is nvidia RTX 3070

[ornate elk]

did you followed my install guide?:
#🤝｜tech-support message

you can also screenshot the whole cmd output for me to check

[split robin]

Ooh, wow ok

[formal trout]

I'm getting this error with Ultimate SD Upscaler. Only with "ultramix_balanced", for example not with "4x-UltraSharp"

[gritty ridge]

Any link to install Kohya_ss on .MAC if possible at all… if not any other tip for training model through dreambooth / SD?

[dense tapir]

Thank you so much! 😄

[scarlet igloo]

“CIDAS/clipseg-rd64-refined” Please help me, should I download the entire folder or which part of the file in the hugging face, and which directory should I place the downloaded file in the local folder

[sudden hawk]

Ok ! i succeded to install SD in the good way I think cause now I can lauch it, but, when I try to genere i have this grey screen, i don't know what to do

[gritty ridge]

Any advice for changing in stable dif webui adapted for latest M3 Pro configured to max 16 C CPU / 40 C GPU with ANE 16C

[ornate elk]

Is the model you try to use 6gb ? (Sdxl models are 6gb, 1.5 models are 2gb or larger)
Everything smaller is not a model

My guess is your trying to use a model as Lora.
I think old anime style xl v3 is a Lora for sdxl and not a model itself

If its a lora it goes into models/lora folder

As model try out aamMix

[sudden hawk]

i'm sorry to ask, i searched on the web and didnt find. Where can I dl it pls ?

[ornate elk]

You can get all the models and loras from here:
https://civitai.com/models
You can choose between 1.5 and sdxl models and loras.
1.5 is trained on a 512x512 resolution while sdxl is trained on 1024x1024 resolution.

For example this is AAM Mix
https://civitai.com/models/269232?modelVersionId=303526

An sdxl model for Anime

[sudden hawk]

Oh thank you so much, now i understand better (i think myself so dumb now too 😅 )

[ornate elk]

no problem ^^

in img2img you can use the sd upscale script

at the bottom

set denois below 0.3
set the resolution to square
select the upscale script and an upscaler

hires fix is still better, but it uses the same technique

in img2img upscale you cant set the denois as high as with hiresfix

so best is to generate the image with hires fix and then inpaint and upscale it with the sd upscale script in img2img

[frail verge]

anyone mess with adetailer before?

trying to run adetailer over this image to fix the face but it just messes it up

[ornate elk]

are you on AMD?

whats your GPU?

[frail verge]

i am yes and i'm using a 6700 xt rn

on rocm/linux

[ornate elk]

ahh okay

then you may need to add --no-half --upcast-sampling
to the webui-user.bat

[frail verge]

alright ill give that a try

also: any idea why i have to reload the ui after starting it in order to do anything in it?

like the first startup is always unresponsive; won't generate, load extensions page or even PNG info

until i do "reload ui"

[ornate elk]

thats strange, maybe a gradio issue, make sure to whitelist the webui in any adblocker

[frail verge]

oh wow lol i'll try that, i didn't even consider that it could be my adblocker

[ornate elk]

yea, we had such issues here before, if its visual (and no console error), then its mostly the adblocker

[frail verge]

erm, so before i could add these flags to my launcher i did a complete fresh install of the webui to make sure there wasnt any funny business going on (also to fix that weird reloading issue because at the time i thought maybe a fresh install could solve it (it didnt ofc))

and for some reason now adetailer is working fine?

i did a lot of shit that was almost 100% unrelated, like installing rocm sdk and not installing the insightface module so i really have no clue why it's working normally now

[ornate elk]

hmm, yea normaly it should have worked on rocm linux without issues before

good that its back to normal now ^^

[frail verge]

it is!! but i'm also scared because it's just randomly working now xD

ok. i disabled my ublock origin completely and lowered firefox's security settings and the first launch is still buggy for some reason

web console errors from the ui

might be related to the agent-scheduler extension..? either that or it's a symptom. im going to try disabling it and see if that helps.

[fair oxide]

@ornate elk any thoughts on this? you see lcm lora works with 2-8steps and 1-2 cfg, and within those same parameters lcm lora has both sd and sdxl versions, while sdxl lightning lora too works with 2-4-8 steps wit cfg 1-2, what exactly is the key difference to them?

[ornate elk]

there is a lightning lora?

[fair oxide]

yes

[ornate elk]

can you share a link?

[fair oxide]

https://huggingface.co/ByteDance/SDXL-Lightning/tree/main

i've used it a lot

[ornate elk]

ah interesting

[fair oxide]

also have a sdxl model merged with it

[ornate elk]

i guess the difference is thats its locked to steps?

[fair oxide]

no

you can go from 2-4-8 with either

hmm

technically you could just work with lcm for both sd and sdxl for decent output

after - detailer

why do you need adetailer to work before hires fix?

after detailer kicks in after the image has processed

its a way of correcting the image

you cant assess correction method before image is processed

dunno

doesnt sound logical what you are asking

hires fix does some correcting too as far image quality goes

hires fix sorts out image depth, adetailer works on image content quality specific to faces / hand attributes

they are entirely different in their function

or think of it this way, adetailer is sort of inpainting, it does its work when you have an image fully rendered

no not possible

due to their functional criterea

np once you understand their ussage concept its easy to understand how they are applied

[frail verge]

what's the difference between tiledvae and ultimate upscaler? which one is better for upscaling higher than you normally can?

and is ultimate upscaler still a good/viable option for sdxl?

[ornate elk]

Tiled VAE is for getting higher resolution

SD upscale script is for upscaling

If you get out of vram errors enabled tiled vae

[frail verge]

gotcha ty

[little echo]

does anyone ever use non SD images in img2img tab? would it be of any use?

[soft fiber]

it can be any image in img2img, you dont need a generated picture for it

[frail verge]

@ornate elk sorry to bother you but i figured you'd want to know this. i think i figured out why adetailer was failing.. and it seems like it has to do with styles not being applied by adetailer

i was using a style for my negative prompt which worked fine for the base image, but not adetailer

after manually adding the negative prompt, adetailer worked normally again

[ornate elk]

oh good to know thx

strange behaviour still

[little echo]

i tried to upscale an image sized 24576x10240 and it seems to just fail

is there an upper limit?