🤝｜tech-support

[ornate elk]

That chart represents speed without any settings activated like loras, longer prompts, sdxl, or hires fix used

[drowsy pagoda]

so you say that with zluda, amd will be technically higher in the chart?

[ornate elk]

No, as the chart was done with AMD used on Linux with rocm. Where you get the best performance.

I'm using a 7900xtx on Windows with lshqtigers fork and with directml I got like 4-6 it/s
Now with zluda im getting 16 it/s

[drowsy pagoda]

shit

if I get a new pc in the future I will consider linux

too much of a hassle right now

[ornate elk]

You can also use dualboot

But I would suggest installing the zluda version and you won't have any regrets with SD

[drowsy pagoda]

oh shi, I didnt even know that was possible

yea, im too lazy to do all'at

[ornate elk]

Its not that hard

[drowsy pagoda]

im fine waiting a bit more while I am tabbed out waiting on my gen

aight, ima try to install SD again now from the pin

[ornate elk]

Its worth it 🙂

[drowsy pagoda]

do we need to use the medvram command?

is 12gb 'medium vram'

[ornate elk]

--medvram-sdxl

Not --medvram

[drowsy pagoda]

yea, I meant that

[ornate elk]

Yes thats needed

[azure magnet]

Hi i have heard that forge it's a better version of normal automatic1111. Is it true?

[drowsy pagoda]

do I need to do the last step in the pin that says if your gpu is below an rx 6800? 6700xt is close to it

[ornate elk]

Its better if you have a GPU with low vram

[ornate elk]

Yes a 6700xt is below and needs the additional step

[grizzled void]

when a model say that the base model is SD 1.5, what does it mean ? some Lora don't want but i dont understand why

[azure magnet]

I have 12 gb vram. So isn't better?
So the output is worse?

[ornate elk]

The output isnt Affected

Forge is a bit faster than auto1111

But it doesn't support every extension like auto1111

With 12gb vram I would use auto1111

Forge is very good for 6gb vram or less

[azure magnet]

Last question SDXL is better for quality output instead of "normal" SD?

[drowsy pagoda]

@ornate elk Do I install 5.7.zip or ZLUDA.zip?

[ornate elk]

Yes as the base resolution is 1024x1024 instead of 512x512 from 1.5

[ornate elk]

Use the latest stable version 3.6

Scroll down a bit

[ornate elk]

Loras and models are either trained on the base 1.5 model or sdxl model.
You need to make sure both are based on the same.
So model and lora need to be based on 1.5 or both on sdxl

[drowsy pagoda]

[ornate elk]

Yea but you can't destroy it if you follow the steps xD

Or even if you do it wrong

I can take a look when you edited it

[drowsy pagoda]

what do you mean

take a look how?

[ornate elk]

If you screenshot what you have added

[drowsy pagoda]

lol ok, I am just a bit paranoid because I have absolutely no idea what I am downloading and installing and how it all works

but I know that github and AMD is safe

[ornate elk]

Yea understandable

[grizzled void]

that the, im using a lora Mona SD1.5 with a model based on SD1.5, but i have terribl result

[drowsy pagoda]

@ornate elk Should I run clip skip 2?

also, are Pickletensor files safe? or are only Safetensor files safe

[grizzled void]

all my lora are base in SD 1.5 and my model too, but some lora don't work, don't understant why

[ornate elk]

For some anime models its good.
For most semi realistic models its not needed

[ornate elk]

Then lower the lora weight by changing the number at the end to 0.5

[ornate elk]

They are less safe then safetensor

But if you download them from civitai your mostly safe as they scan every model

[grizzled void]

how ;-; ?

[ornate elk]

Change the number at the lora

Lora:name:1 to 0.5

[grizzled void]

in the file ?

namefile *

[ornate elk]

In the prompt

[grizzled void]

aaaaaah

ok ok

if i have this problem i have to put it to 0,5, right ?

[drowsy pagoda]

Just try playing around with different values

[grizzled void]

ok ok thanks ^^

[drowsy pagoda]

make sure to check what value range is recommended to play around with in the description of the lora on Civitai

@ornate elk where is the system tab in the SD UI?

[ornate elk]

System ?

[drowsy pagoda]

https://github.com/vladmandic/automatic/wiki/ZLUDA#compilation-settings-and-first-generation

that's what it says here

[ornate elk]

Dont to these steps

Only the ones for rocm library

[drowsy pagoda]

I got this error trying to do my 1st gen 'RuntimeError: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver from http://www.nvidia.com/Download/index.aspx'

[ornate elk]

Ah okay, thats fixable.
Open up a cmd.
And type
Pip cache purge
Then hit enter.
Then delete the venv folder and relaunch the webui-user.bat

[earnest jackal]

Hello, I'm trying to install dwpose for control net but its giving me an issue

[drowsy pagoda]

what's the issue?

[drowsy pagoda]

I did that and it shows the same error

[ornate elk]

Can you make a screenshot?

[fair oxide]

am inpainting, and wondering how to get rid of this?

working with SDXL pony

the white area

[drowsy pagoda]

try to img2img or hires with the same seed

[fair oxide]

well, in inpainting, i want to get rid of the source of the issue

not a workaround

[drowsy pagoda]

I cant open the web UI

[fair oxide]

it may haft to do with the model itself, but not sure what can be causing the white

increasing mask blue solved my issue, thanks

blur*

[drowsy pagoda]

@ornate elk I removed the 'use-zluda' from the batch file in command args and now the web UI opened but this error pops up as soon as I press to generate an image

'RuntimeError: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver from...'

[ornate elk]

Your missing the --

You need to add --use-zluda

[drowsy pagoda]

oh shit

oh, I still get the same error

'RuntimeError: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver from...'

[ornate elk]

Can you show me the paths you added in the system variables?

[drowsy pagoda]

maybe I downloaded AMD HIP SDK version 5.5.1 instead of 5.7.1

[ornate elk]

The path is wrong

It needs to be C:\ZLUDA

[drowsy pagoda]

omfg

[ornate elk]

Then delete the venv folder again

And relaunch the webui-user.bat

[drowsy pagoda]

ok, I pressed generate and I dont see any progress in the cmd. No new lines of text. I think this is the part where I wait 45mins

[ornate elk]

Yes if the generate button is grayed out and no error is there just wait

[drowsy pagoda]

just says 'Applying attention optimization: Doggettx .... done

jesus, so much time wasted because I went over the guide sloppily

[ornate elk]

Your not the first ^^

I wish more people would read slowly a second time or read something out load

[drowsy pagoda]

just missing the \ and it not working is ridiculous

[ornate elk]

True xD

[drowsy pagoda]

I payed attention to everything else and I forgot about that

[ornate elk]

You still managed to do the extra step with the rocm library without problems

[balmy lily]

HEA FUCKO

sup

[ornate elk]

And also you edit the path in the system variables and not in the user variable like other users.
So you were pretty good

[balmy lily]

i have question

[drowsy pagoda]

bro, aint no way with that name

[balmy lily]

how would most effectively write text to this girls shirt, like a specific sentence (like a few words or symbols) thats written with normal letters. but to not change the rest of the image much

[drowsy pagoda]

getting an image with text to generate is pretty much a no go from what I know

idk if that has changed in the past year

[ornate elk]

You would need to use an sdxl model to generate readable text.
Short words work pretty good.
If you want to add words to an image you can try inpainting but it mostly fails on text

[cosmic furnace]

how do i do this? o.0

[ornate elk]

You would need to run that command in a Powershell console

[cosmic furnace]

Where do i get this powershell console?)

[ornate elk]

Windows has it preinstalled

[cosmic furnace]

same goes for this?

so , basically what to do? )

cmd doesnt run it =\

[ornate elk]

Open powershell

[cosmic furnace]

how?)

[ornate elk]

And copy and paste the commands

[cosmic furnace]

ah i see

[ornate elk]

Click on the windows symbol and type Powershell

[cosmic furnace]

y gotcha)

[random turtle]

Is there a better alternative to GroundingDinoSAM to detect / mask out objects?

[cosmic furnace]

wait it doesnt work with this

[timid ibex]

I tried to install Automatic1111 but it's taking quite some time and won't really install

[cosmic furnace]

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/trumank/uesave-rs/releases/download/v0.3.0/uesave-installer.sh | sh

[timid ibex]

Is there another alternative to that?

[balmy lily]

well i found this concept of using a controlnet to draw the stuff on the shirt, so if i were to make a black and white redition of the image with a controlnet https://huggingface.co/lllyasviel/sd-controlnet-canny, i could manually draw the text and symbols i want onto the black and white image, i could then use that black and white image as an input with the controlnet.

but i cant seem to get it set up

[cosmic furnace]

Install prebuilt binaries via shell script

[ornate elk]

A bit more infos please
What does the CMd shows?

[cosmic furnace]

oh wait w.e its same =\

[timid ibex]

I'll have to retry cause I closed the window

[ornate elk]

Yea pls relaunch and check at which step it hangs

[ornate elk]

Thats just controlnet canny. Do you have the controlnet extension installed?

[timid ibex]

The ETA was taking too long that's all

And won't really budge

[ornate elk]

Its a 2gb file. Maybe you need to stop the VPN for that download

[timid ibex]

I'll try that

[drowsy pagoda]

gyatt, daaaamn 16mins for a 2.7gig file? imagine

[ornate elk]

Its with vpn enabled

[timid ibex]

Meh, not so much of a difference

[balmy lily]

yea, and i have the canny model downloaded, but i keep running into errors that look like this

Error occurred when executing KSampler:

mat1 and mat2 shapes cannot be multiplied (154x2048 and 768x320)

when i use any control net model, i think its related to the controlnets being made for an early version of stable diffusion and not working with modern checkpoints. but idk how to turn my image into a canny image, or how to edit it in the sd ui if you can, or how exactly to use the canny image once i make it and turn it into a similar looking image to the one i showed u, cuz itd require utilizing both an image to image thing and using the canny controlnet to change the image i use as input, so its kind of a complex thing of combining multiple ways of generating images thats a bit too advanced for me right now

[ornate elk]

Okay then you need to wait :/

[timid ibex]

Reminds me of the Windows XP days 💀

[ornate elk]

Yea the controlnet models are either for 1.5 or sdxl

[drowsy pagoda]

@ornate elk holy shiet, it finally works!! thanks

[balmy lily]

what else is there, like is there stuff after sdxl

[ornate elk]

Perfect! No problem 🙂
Have fun

[drowsy pagoda]

now it isnt even using 6gb of vram

[ornate elk]

Yes it uses much less resources and you can even use Sdxl without problems

Also higher resolution without going out of memory

[drowsy pagoda]

what is sdxl? newer models on civitai use that or are based on it?

[balmy lily]

do u mean it only works for those specific checkpoints or stable diffusion versions

[ornate elk]

Yes sdxl is the latest model version. These are trained on a 1024x1024 resolution

[drowsy pagoda]

but I can gen 512x512 for faster gens with no problem, right?

[ornate elk]

Sure

But sdxl models should not be used for 512x512

[drowsy pagoda]

omfg, I can hear my gpu fan going to 100% when generating and then back to 0 lmao

[balmy lily]

will controlnet not work for just any model even ones that are trained on sd1.5 and sdxl, itll only work on the base models with sdxl or sd1.5 at the start of their file name

[ornate elk]

There are different controlnet model versions. Some are for 1.5 and some for sdxl

[balmy lily]

also do u have any comfyui sd images that have done text good

[ornate elk]

Nope I'm not using comfyui

But if you use an sdxl model you can generate good text

[balmy lily]

when i set the latent image to have a batch size above 1, but have the ksample be fixed at one seed, what exactly are scematics of the other images in the batch

[timid ibex]

Finally

[ornate elk]

The seed will increase by 1 I geuss

[balmy lily]

wdym by an sdxl model, like any modern checkpoint, or just the basic official sdxl model called sd_xl base and sd_xl refiner and that everyone says to download

[balmy lily]

actually i tested it and when increased the seed by one and tried it again the image i got wasnt the same as the the second one in the batch i made befor it

also theres a seperate batching feature to incrementally increase the seed by 1 and keep making images

[ornate elk]

Any sdxl model

[balmy lily]

but arnt all models sdxl or sd1.5, what checkpoints could be giving the controlnet problems

[ornate elk]

Using the wrong controlnet model with the wrong model

Like I said you need an 1.5 controlnet model + and 1.5 model OR
And sdxl controlnet model + an sdxl model

Dont mix them

[balmy lily]

how do i know which models are 1.5 and which are sdxl

[ornate elk]

SDXL models right now are always 6gb or larger, so 2gb or 4gb models are 1.5 based

You can also check the models civitai page

[balmy lily]

whats the point of this, it seems to be for converting the pth files into safetensors files or something https://github.com/Mikubill/sd-webui-controlnet/blob/main/extract_controlnet.py
and whats this one
https://github.com/Mikubill/sd-webui-controlnet/blob/main/extract_controlnet_diff.py

[ornate elk]

Idk

[balmy lily]

are ur controlnet models .safetensors or .pth

[ornate elk]

.safetensor

[balmy lily]

did you get them from here

https://civitai.com/models/9251?modelVersionId=11003

[karmic crown]

https://github.com/viking1304/a1111-setup/releases/tag/0.1.0 🎉

[ornate elk]

I use these ones:
https://civitai.com/models/38784/controlnet-11-models

These are for 1.5

[balmy lily]

do they go with the checkpoints or in the controlnet folder

[ornate elk]

They go into models/controlnet

[balmy lily]

pth is pickletensor

btw

i just found that out

[karmic crown]

you will see some tutorials that will say to put controlnet models in extension folder - DO NOT do that, put them where CS1o told you

[balmy lily]

controlnet is preinstalled on comfyui, but no models are

so its not in the extensions folder

[karmic crown]

oops, i thought it was a1111 🙈

[upbeat atlas]

Hi, if I wanted to go from automatic 1111 1.8 to 1.6, can I simply type git pull "the link to 1.6 version" without it messing up the webui?

[karmic crown]

why would you downgrade to 1.6?

it is possible. but I am confused. why would you do that?

[upbeat atlas]

I'm using sdxl model on Zluda and it keeps on crashing, I looked around and apparently there's a bug in the 1.8 version that keeps ramping up the ram usage leading it to crash so I want to try older versions to see if it fixes it

[karmic crown]

to get back to some old commit (I used some random number for reference)

git fetch --all
git reset --hard d80445bc4cf0

to restore to the latest version (this should work)

git pull

if it does not work for some reason, do this

git fetch --all
git reset --hard origin/master

but i would try to fix 1.8.0 if i was on your place to be honest

i hope i didn't mistype something, i am tired

i need to go now

[upbeat atlas]

Thank you 🙏

[ornate elk]

Can you point me to that bug?

I never heard of any zluda users with that problem

[upbeat atlas]

I read it from the comments on this github post

https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/15175

[karmic crown]

good night people, i had a long day
cu some other time

[ornate elk]

Gn

[ornate elk]

How much RAM do you have?

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

And what's your gpu?

[upbeat atlas]

I have 16 GB vram and 16 GB ram
Rn 11.6 GB of ram is being used
My webui is automatic1111 directml running on Zluda
My gpu is AMD RX 6900 XT on windows

[ornate elk]

Okay and what's inside your webui-user.bat?
With 16gb ram you need to increase your windows Pagefile

[upbeat atlas]

The arguments I'm running are:
--use-zluda -medvram-sdxl --update-check --skip-ort

Also, how much do I need to increase my pagefile?

[ornate elk]

Min 16000 and Max 24000 should work.
And make sure to have 15gb free space on C

[upbeat atlas]

Just to make sure, is this correct?

[ornate elk]

You should made that for C drive if its an M2 SSD

And disabled for D drive

[upbeat atlas]

The webui is installed on drive D

Do I still do it for drive c?

[ornate elk]

Yes, pagefile should be placed on the fastest drive

And only use one

Dont enable for both drives

Not even system managed for the other drive

[upbeat atlas]

Is it right now?

[ornate elk]

Yep 👍

[upbeat atlas]

Thanks for the help, ima try and see if it fixes it 🙏

[ornate elk]

Then reboot the PC

No problem 🙂

[vast crest]

Hello,here is one of the errors that google collab gives me: ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behavior is the source of the following dependency conflicts.
tf-keras 2.15.1 requires tensorflow<2.16,>=2.15, but you have tensorflow 2.16.1 which is incompatible.
torchaudio 2.2.1+cu121 requires torch==2.2.1, but you have torch 2.1.0 which is incompatible.
torchtext 0.17.1 requires torch==2.2.1, but you have torch 2.1.0 which is incompatible.

[ERROR] CUID#7 - Download aborted. URI=https://huggingface.co/Linaqruf/stolen/resolve/main/pruned-models/stable_diffusion_1_5-pruned.safetensors
Exception: [AbstractCommand.cc:351] errorCode=24 URI=https://huggingface.co/Linaqruf/stolen/resolve/main/pruned-models/stable_diffusion_1_5-pruned.safetensors
-> [HttpSkipResponseCommand.cc:215] errorCode=24 Authorization failed.

the links don't work. In the end I can't create Lora anymore and I don't know why 😦

[upbeat atlas]

well darn, it didn't fix it, I still get the same crash ):
If you find out anything else to fix it please @ me

[ornate elk]

What sort of crash is it?

Any error message?

[upbeat atlas]

When it gets to final couple steps of generation, it stops for a second, then this pop up comes up and in the cmd it only says tap anywhere to continue and when you do it just closes.

[ornate elk]

What settings did you used?

Resolution steps hires fix etc

[upbeat atlas]

I've tried a lot of different settings
344×512
512×768
768×1024
With and without hires.fix, with and without BREAK, with and without negative prompts
It's always the same, I get that error normally as well, but it takes quite a while before it shows up, but with a lora model the error immediately happens

[ornate elk]

Thats very strange. So you get it with 1.5 models too?

Or only sdxl?

[upbeat atlas]

No no, ONLY sdxl, the 1.5 models work perfect fine

[balmy lily]

do you know whats the deal with all the models on this https://civitai.com/models/4201/realistic-vision-v60-b1, what do the ones with vae vs no vae in the name mean, is the vae included or disincluded or something, or are they vaes, like whats the difference between the models

[ornate elk]

Does it crash at the last step when nearly done?

[upbeat atlas]

Yeah, at the last step of the normal generation, before hires.fix starts

[ornate elk]

Baked VAE = VAE included

[balmy lily]

and (novae) means no vae included, or no vae necessary

[ornate elk]

No VAE included

[ornate elk]

Oka then its possible the VAE step.
You should try out this VAE here for sdxl:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main
The sdxl.vae.safetensor

[balmy lily]

whats different about the models that say inpainting, i see those alot, like i suppose theyre for inpainting, but what actually makes them different

[ornate elk]

The are combined with the Inpaint base model of 1.5

So they are not for generating images

[upbeat atlas]

With the model I'm using the vae is baked in so I don't use any, should I download this and make the model use it to test it out?

[ornate elk]

Yes its worth a try

Also if you can try out an other Browser. Idk how Brave behaves with high load

[upbeat atlas]

Alright then, which one of them should I get tho?

[ornate elk]

The sdxl.vae.safetensor

[upbeat atlas]

What's the recommended browser?

[ornate elk]

Firefox, edge, chrome

[upbeat atlas]

So I used the vae in firefox, same error 💔

[ornate elk]

Can you show the exact settings?

Steps resolution etc

Also please monitor your RAM usage In the taskmanager when generating

[idle cairn]

I've had good results with OperaGX, to be honest, while running Forge and Comfy at the same time

[ornate elk]

We had a lot of users with problems because they had enabled opera vpn

[upbeat atlas]

No hires.fix this time, still crashed at the very last step

[upbeat atlas]

For a few seconds before the start it hit 15.8/15.9, when it started it sat at 13.8/15.9 and in the end it was at 15.9/15.9

[ornate elk]

And you restarted the PC after increasing the pagefile?

[upbeat atlas]

Yes

There was even a pop up after the changes that told me to restart

[ornate elk]

Okay and you also launched the webui-user.bat and not the webui.bat?

[upbeat atlas]

I didn't even know you could launch webui.bat, yeah it's the webui-user.bat xD

[ornate elk]

Okay can you test with a non pony model and lora?

[upbeat atlas]

Ok but it'll take a while, I'll have to download a sdxl model first, I'll test it and get back go you, thanks again

[ornate elk]

Sure, no problem, but I'm off for today/night

[balmy lily]

Error occurred when executing KSamplerAdvanced:

'ModuleList' object has no attribute '1'

[ornate elk]

No clue on comfyui :/ sry

[balmy lily]

turns out its a sdxl/sd1.5 controlnet mismatch problem. is there any way of converting a checkpoint into an xl or 1.5 model, like to make sd think that the model is either 1.5 or xl, but not like actually convert it, or some easy way to actually convert them

@ornate elk https://www.youtube.com/watch?v=PrJFjvz0XhM

[ornate elk]

No you can't convert models to a different base.
They need to be trained on that base.
What you need is to use Sdxl controlnet models from here:
https://huggingface.co/lllyasviel/sd_control_collection/tree/main

They work with sdxl models

[balmy lily]

the opposite is the problem, my controlnet models are all sdxl, and most of my checkpoints are sd1.5

[ornate elk]

Oh then use the controlnet models I linked an hour ago

[idle cairn]

Hello !

I'm having an issue, and Google couldn't help.
I'm trying to launch Forge (Which worked fine 30 mins ago) and I get this error :
Traceback (most recent call last):
File "E:\webui_forge_cu121_torch21\webui\launch.py", line 51, in <module>
main()
File "E:\webui_forge_cu121_torch21\webui\launch.py", line 39, in main
prepare_environment()
File "E:\webui_forge_cu121_torch21\webui\modules\launch_utils.py", line 431, in prepare_environment
raise RuntimeError(
RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

Tried to add the arg, but no success

[ornate elk]

What's your GPU?

Also dont add this arg

[idle cairn]

4090

I was impatient and removed my Forge and A1111 installations.
I'm reinstalling Forge, now

[ornate elk]

Check for any nvidia driver update

[idle cairn]

Yeah, that's the first thing I did, but it didn't help

[ornate elk]

Okay

Maybe an extension broke forge?

[idle cairn]

Yeah...... You see... It happened right after I tried to install the Riffusion extension...

[ornate elk]

Oh okay

Maybe riffusion doesn't work on forge then

And only on auto1111

[idle cairn]

Maybe...

Well, I was just looking to try Music generation, but I guess it'll wait

[ornate elk]

SD audio was announced so hopefully we won't wait long

[idle cairn]

Hopefully

Redid the whole Forge installation and it's working fine again... That's that then

[heady ruin]

ok so i have this work flow that takes a negative then fills it in with whatever u prompt it but I want to add another load image to provide the color scheme but don't know where to connect it

[balmy lily]

Error occurred when executing FacialPartColoringFromPoseKps:

index 9 is out of bounds for axis 0 with size 9

[idle cairn]

ControlNet Recolor ?

[drowsy pagoda]

@ornate elk is it beneficial to use xformers in command args on an amd card?

[balmy lily]

what is fp16

[brittle oak]

Hello everyone, I need your help if is possible, I just recently noticed that using Google Colab Pro, the rate of usage has falling down, now is 1.76 per hour, some days ago was 1.96, does anyone has an idea about this, is possible to ncrease it a little bit....?

[balmy lily]

i cant install these custom nodes https://github.com/sipherxyz/comfyui-art-venture

[dense viper]

https://en.wikipedia.org/wiki/Half-precision_floating-point_format

Apologies if this is not the right place, but does anyone know if there is a way to request specific image size (width, height) using Stability API? (api.stability.ai/v2beta/stable-image/generate/core) The docs doesn't mention it, only aspect ratio.

[heady ruin]

All good I figured it out thk u

[balmy lily]

can u send the workflow

u can also just send the image, that contains the workflow metadata

[ocean basin]

Error occurred when executing KSampler:
Could not allocate tensor with 251658240 bytes. There is not enough GPU video memory available!
hey, can someone help me with this error? i get this error when doing animatediff. i also get this eror when text2image using SDXL checkpoint.
my spec:
cpu AMD Ryzen 5 3400G with Radeon Vega Graphics 3.70 GHz
Gpu amd Rx 6600 8GB Vram
ram 16gb
windows 10
i use ananconda to run comfyui

[ornate elk]

Its for nvidia only

[ornate elk]

Hey, the problem is that your using directml without zluda. So your vram usage isnt optimised in SD.
You would need to install either auto1111 or comfyui with the zluda Backend for best performance and less vram usage

[ocean basin]

I see, ty

[ornate elk]

NP, you find a guide for auto1111 with zluda in the Pinned Messages of this channel

I will make a guide for comfyui too the coming days

[ocean basin]

yes, im following your guide right now

[ornate elk]

Let me know if you have any questions about the steps

[ocean basin]

after i done all the step:
Preparing metadata (pyproject.toml): started
Preparing metadata (pyproject.toml): finished with status 'error'

stderr: error: subprocess-exited-with-error

Preparing metadata (pyproject.toml) did not run successfully.
exit code: 1

[17 lines of output]

• meson setup C:\Users\filip\AppData\Local\Temp\pip-install-6nr0tpxf\scikit-image_42b13d630cb2481c95ecbdaec6f114f6 C:\Users\filip\AppData\Local\Temp\pip-install-6nr0tpxf\scikit-image_42b13d630cb2481c95ecbdaec6f114f6.mesonpy-7nm1wpot -Dbuildtype=release -Db_ndebug=if-release -Db_vscrt=md --native-file=C:\Users\filip\AppData\Local\Temp\pip-install-6nr0tpxf\scikit-image_42b13d630cb2481c95ecbdaec6f114f6.mesonpy-7nm1wpot\meson-python-native-file.ini
  The Meson build system
  Version: 1.4.0
  Source dir: C:\Users\filip\AppData\Local\Temp\pip-install-6nr0tpxf\scikit-image_42b13d630cb2481c95ecbdaec6f114f6
  Build dir: C:\Users\filip\AppData\Local\Temp\pip-install-6nr0tpxf\scikit-image_42b13d630cb2481c95ecbdaec6f114f6.mesonpy-7nm1wpot
  Build type: native build
  Project name: scikit-image
  Project version: 0.21.0

..\meson.build:1:0: ERROR: Unable to detect linker for compiler clang -Wl,--version
stdout:
stderr: clang: warning: unable to find a Visual Studio installation; try running Clang from a developer command prompt [-Wmsvc-not-found]
clang: error: unable to execute command: program not executable
clang: error: linker command failed with exit code 1 (use -v to see invocation)

A full log can be found at C:\Users\filip\AppData\Local\Temp\pip-install-6nr0tpxf\scikit-image_42b13d630cb2481c95ecbdaec6f114f6.mesonpy-7nm1wpot\meson-logs\meson-log.txt
[end of output]

note: This error originates from a subprocess, and is likely not a problem with pip.
error: metadata-generation-failed

Encountered error while generating package metadata.

See above for output.

note: This is an issue with the package mentioned above, not pip.
hint: See above for details.

[notice] A new release of pip is available: 23.2.1 -> 24.0
[notice] To update, run: D:\Ai\SD-Zluda\stable-diffusion-webui-directml\venv\Scripts\python.exe -m pip install --upgrade pip

Press any key to continue . . .

[ornate elk]

What's your python version?

Open up a cmd and run
Pip cache purge

Then delete the venv folder and relaunch the webui-user.bat

[ocean basin]

i have python 3.10.6 installed allready, but its said "This program is tested with 3.10.6 Python, but you have 3.12.1."

[ornate elk]

Then uninstall 3.12.1 if you dont need it

Then install 3.10.11 64bit and check add python to path

[ocean basin]

Its work, ty

[ornate elk]

Perfect, no Problem 🙂

[drowsy pagoda]

I am trying to generate an image with sdxl 1.0 and everytime I get to 95%, I can see the image almost ready to finish and I get this error:

OutOfMemoryError: CUDA out of memory. Tried to allocate 9.00 GiB. GPU 0 has a total capacity of 11.98 GiB of which 0 bytes is free. Of the allocated memory 2.84 GiB is allocated by PyTorch, and 9.52 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)

I am using 1024x1024 resolution for one image gen

could the resolution be too much for my gpu or is this a fixable error?

my gpu is the rx 6700xt and I have zluda installed as the pinned comment says

[ornate elk]

What settings did you used?

Did you enabled hires fix or not?

Normaly that resolution shouldnt be a problem

[drowsy pagoda]

no

[drowsy pagoda]

what do you mean by settings

[ornate elk]

Make sure you close other Programms in the background, especially Wallpaper Engine if you have it

[ornate elk]

Steps, resolution, loras etc

[drowsy pagoda]

no lora, no negative prompt, very simple and short positive prompt, 1024x1024, default steps

[drowsy pagoda]

I do however have 9 tabs on firefox+discord open

[ornate elk]

Hm that shouldnt be an issue

[drowsy pagoda]

does fine in SD 1.5

I just go watch a video on youtube while waiting for my gen

[ornate elk]

And you also didn't checked the refiner?

[drowsy pagoda]

no

didnt touch that\

[ornate elk]

Okay can you screenshot the whole SD page when you get that error again?

Was the model a normal sdxl model or a "pony" model?

[drowsy pagoda]

the original sdxl I downloaded from huggingface

no vae version

[ornate elk]

Okay thats strange.
Have you restarted SD and tried again?

[drowsy pagoda]

yes, same thing yesterday

generation is fine but gets to 95% exactly and stops with that error

[ornate elk]

Ah okay, than thats a VAE issue.
You need the sdxl fp16 VAE Fix

I can link it

[drowsy pagoda]

I just have the base no vae model

so, the model I got is just unusable for some people?

[ornate elk]

No, the VAE just uses to much vram at the end.
That can be fixed in two ways

[drowsy pagoda]

but I have the no vae model

[ornate elk]

First is to download this VAE
(Sdxl.vae.safetensor)
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main

Put it into the models/VAE folder

[drowsy pagoda]

so I download the newer vae model?

or is the base model with the vae fix fine?

ok, it worked now, thanks

this is my first sdxl generation

I am getting a bit over 4it/sec

oh wait, I think it's the opposite, it says 4.2s/it

am I reading that wrong or am I not getting even 1 it/sec with a 6700xt

[ornate elk]

Sdxl is slower, how long in seconds did it take for 20 steps?

[drowsy pagoda]

over 2 mins, not sure

[ornate elk]

Hm could be normal

[drowsy pagoda]

I just did the same prompt as the image above but with a 4 batch count and 1.2 scale highres fix with 20steps and it took 16mins and 39secs with 1024x1024 res

highres didnt seem to affect the image at all

[ornate elk]

Hires should affect it. But you need to set hires steps to 10.
And denois to 0.5

[drowsy pagoda]

hmmmm ok

had 20steps and denoise 0.4

maybe it wasnt high enough I c

[ornate elk]

Denois depends on the upscaler too

With esrgan upscalers you can play around from 0.2 to 0.5

With latent ones stay near 0.5

[drowsy pagoda]

using 4x ultrasharp rn

[hard quiver]

can anyone tell me why image generation is giving out blank pics?

[ornate elk]

Your trying to use a lora as model

That won't work

Loras go into models/lora folder

[hard quiver]

ooh.

[hard quiver]

i pasted it in lora folder but now its not in checkpoint.

where do I access it now

nvm. another guy. explained it. thanks a lot though.

[upbeat atlas]

@ornate elk hi, so I tried none pony sdxl models as well with normal sdxl loras, same problem unfortunately

[sand dawn]

Seeing if maybe anyone is very well versed with SD API formatting...
I have multi-controlnet, layerdiffuse, reactor - all working perfectly fine.
I cannot get s2-forge-couple to work - I get 422 error every time.
The author is offering support but I don't think they quite understand the issue either.
https://github.com/Haoming02/sd-forge-couple/issues/11

Here's the relavent portion of the response from script-info

[fair oxide]

can someone point me to a up to date guide for installing a1111 for amd gpu in linux? i went by this guide https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs , but had a lot of difficulty with the very specific python version

[ornate elk]

Thats really Strange. As someone with a 6700xt doesn't have this issue

What you can try is to enable Tiled VAE extension

Also make sure you use the sdxl fp16 VAE fix

If nothing helps you can try to enable the FP8 mode for sdxl in the auto1111 settings

[plush parrot]

does anyone know why after the last automatic1111 update I have 10x time bigger chekpoints and loras previews? I can't browse it anymore, 1 picture is more than 1 screen size

[simple burrow]

https://huggingface.co/spaces/camenduru-com/webui was producing at the following address. specifically 16/9. But this program is no longer available on this site? Where can I produce?

[fair oxide]

RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

I am getting this error, anyone know how to fix

I am using a amd pc setup

ryzen 5 5600 xt

[fair oxide]

sorry but i do not know what that means

I started a week ago

[ornate elk]

Loras and models need to be on the same base version

1.5 loras only work with 1.5 models And sdxl loras only work with sdxl models

Your tried to use a 1.5 lora with an sdxl model

[vocal burrow]

what did you install ?

A1111 directml ? A1111 zluda ?

did you follow one of the pinned comment as a tutorial ?

[hard quiver]

reloaded stable diffusion while one of these extension thingy were being processed, or installed. am I in trouble?

thought process behind this was that in the meantime I can check other extensions that ive just installed.

[ornate elk]

Try restart the webui

If it fails, delete the Extension and install it again

[hard quiver]

how do u delete an extension

[ornate elk]

Delete its folder in the Extensions folder

[hard quiver]

is there a way to know how much extension is installed so far or something similar.

so u can check if its genuinely stuck or jus taking sometime

so something like this doesn't happen again.

[ornate elk]

You can see all the Extensions in the Extensions folder

[hard quiver]

also. stable diffusion is not starting for me now? xd. should I restart the pc?

[ornate elk]

And when starting the webui check the cmd if any errors appear

[hard quiver]

its stuck at cmd and not going forward

[ornate elk]

Can you screenshot it?

[ornate elk]

Oh you just paused it by clicking into it.
Press space to unpause

[hard quiver]

now is it good?

[ornate elk]

Should go forward now

[hard quiver]

lets see. cuz it doesn't take more than 10 seconds

[ornate elk]

What was the Extension you tried to install?

[hard quiver]

the interrogate one

img to img

[ornate elk]

Also I see an error at the top of the cmd

[hard quiver]

to find out the prompt.

[hard quiver]

that has always been there. afiak. still it opened everytime.

i'll try restarting once I save my data

[ornate elk]

Ok

[vocal burrow]

they're using the zip package

and somehow messed it up I guess

[ornate elk]

Ohh true

[hard quiver]

restarted pc. still not working.

is there a way to repair stable diffusion

what are the chances that its not loading cuz the extension is still being downloaded or something.

it took 10x more time then it normally does for some reason. but its opened now

how do I check if that extension was installed or not

i tried to install 2 interogating extensions. one of them did install. the other one. I need to figure out.

[cinder ermine]

hey anyone know what to use as regularization images when training a style?

[ornate elk]

You only need clip interrogator extension, that should work okayish

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

You may need some additional performance arguments

[hard quiver]

gpu 3080ti and idk what webui-user.bat is

[hard quiver]

u mean extensions?

[hard quiver]

also. Im trying to restore img through imgtoimg. can u guide me any checkpoints that are good for this

[ornate elk]

Oh right you didn't installed the webui the recommended way
So there is no webui-user.bat

[hard quiver]

or to turn real life pics into anime using img to img

or cartoons or stickers

[hard quiver]

whats the recommended way?

where is it

[hard quiver]

is that a problem?

[ornate elk]

My install guide is the recommended way:
#🤝｜tech-support message

It covers also the best performance arguments needed for different gpus

[hard quiver]

is this supposed to take this long?

for a interrogater pic

[ornate elk]

Yes, check the cmd, its probably downloading something

[ornate elk]

For that use an anime model like Divinanimemix or Aam Mix

[hard quiver]

alright. lemme install

[fair oxide]

web ui direct ml

[hard quiver]

might be a stupid question. but I can tinker with setting right? for the next image while one image is currently being generated. I mean. these settings don't apply real time. right?

[ornate elk]

Yep, you can change them and it won't affect it in realtime

[hard quiver]

alright. thanks

[vocal burrow]

ok then what are your command line arguments ? (Also you might want to consider moving to zluda)

[exotic mountain]

https://tenor.com/view/hello-gif-19947459

[quiet swift]

Hello, is there tuto on the net that shows how to install the last updates for Deforum ? (I saw in the extension tab in my stable diffusion window that there was updates for Deforum, I have tried to upload something in Github, but now I've got 2 Deforum tab in my stable diffusion window 🙂 how stupid I am !! I'm bad with computer + I'm french and I don't understand clearly what I must do when I'm on the Github page to upload the latest version.) Thank you if someone has an advise.

[sand dawn]

I figured out my problem… the correct key name for the script included a space in it. I’d never encountered key with a space in it before, didn’t even know it was valid

[quiet swift]

I have nearly figured out my problem of updating Defroum. I understood that I must write "git pull" in my terminal (I'm on MAC). But I must write something before "git pull" and I do not know what. If someone knew, thank you in advance

[short tide]

Hello community, is anyone having problems using stable audio? because from the first stable audio model I want to start using it and it gives me a sign saying that I was blocked and it won't let me enter

[placid igloo]

Hello people! Are there any image making enthusiasts from China? We are having a small contest and need one of the judges to be from China. If you want to be a culler of the best AI art write in private.

[dusk shale]

hoy everyone, anyone having some problem like this, after render the image goes like this

[onyx solar]

You need to use the correct VAE.

[honest tendon]

how do I make 3.12 to 3.10.6, i deleted both of them completely and redownloaded the 3.10.6, idk why its still showing the 3.12 path

venv "C:\ai\stable-diffusion-webui\venv\Scripts\Python.exe"
No Python at '"C:\Users\Johnv\AppData\Local\Programs\Python\Python312\python.exe'

[chrome rampart]

hello all. question. I just wanted to try stable cascade so installed in SD1111forge. When i try to render it takes ages. My gpu is3060 12gb. ANy ideas?

[earnest jackal]

I need help with controlnet, I have it installed but when I check the box and apply and restart ui it unchecks it by itself

[onyx solar]

Did you look at your command window to see what it says?

[earnest jackal]

this is what it said

i'm not that tech savvy so idk much about this

[ornate elk]

your using webui forge, it already has controlnet implemented

forge wont let you install it again

[ornate elk]

python 3.12 is not compatible

[earnest jackal]

oh ok

[frozen pecan]

How to do a SDXL prompt with big sizes? The SD1.5 style "(huge nose)*3" doesn´t work anymore

[subtle quarry]

trying to reinstall a1111 and this came up

i think i know what the issue is.. my s/o has it running on their profile on this pc.

its not running on their profile

what

their profile works just fine

I really dont want to have to use theirs

[ornate elk]

whats your gpu?

[subtle quarry]

3060

AKA: It shouldnt have an issue

[ornate elk]

make sure you have the latest nvidia driver

[subtle quarry]

..good point, one sec

downloading it now

it's weird because it works fine on their profile on the computer

seems to be taking forever to uninstall the graphics driver since i selected to clean install

i guess i just leave it for 20 mins and see if it moves.

It's not been 20 minutes but it hasn't moved

What happens if I force the laptop off

Not laptop

Pc

[ornate elk]

The driver could be broken

Best would be an uninstallation with the DDU tool

[subtle quarry]

..fun

Alright be back in 20

[kind idol]

when it comes to training loras, is it best to use kohya ss or a website like civiai

[ornate elk]

Khoya is for local training. Its Good when you have an nvidia GPU with 6gb (better 8gb or more).

[kind idol]

i have 24gb

but khoya has a different ui from videos on youtube

and is hard to follow guides when making loras

[ornate elk]

Hmm I think khoya is the most used lora training tool

[kind idol]

same

[subtle quarry]

new driver is installed

testing a1111 install now

what's up with this?

[ornate elk]

Seems like a connection issue

[subtle quarry]

well it screwed off for now

alright lets hope this works

alright! got past the step we got stuck on

Thor lives to see another day lmao

okay! Question. Can I set the file path of a1111 to read models from another user on the computer?

that way i'm not duplicating models and using up a lot of storage

figured it out

[honest tendon]

whats this

im using a 6950xt

for my gpu

[vocal burrow]

and what A1111 did you get ?

[honest tendon]

what do you mean?

[vocal burrow]

did you get the default automatic1111 ? the directml fork ? the zluda version ?

[honest tendon]

https://youtu.be/6MeJKnbv1ts?si=fKnkstbvICQUR94E

i just did this video

[vocal burrow]

one more video to add to my bad tutorial list nice

[honest tendon]

i guess A1111

[vocal burrow]

so you got the default A1111

the one not compatible with AMD gpus

[honest tendon]

automatic A111*

is there a tutorial to download that

[vocal burrow]

check the pinned comments and get either the directml one (easier to install but way slower) or go with the zluda version.

[honest tendon]

okeh

ty

how is it on AMD, is it notably slower?

or pretty much the same

[vocal burrow]

if you got directml route it will be probably 2 or 3 times slower when comparing to similar ish nvidia gpus

with zluda it should be pretty much the same, just a bit slower.

[honest tendon]

i guess AMD is truly just for gaming,

[vocal burrow]

but it's still experimental so there might be some issues with extensions and so on

[honest tendon]

still worth a shot

[limpid coyote]

Could somebody help me? I’m new to Stable Diffusion and AI generation.

I’m currently trying to download several models from civitai to use locally in SD A111.

Whenever I input a new model and try to use it, this happens:

Does anyone know what’s wrong and how to fix it?

[meager carbon]

any idea

trying to do zluda

[meager carbon]

i mean it has 50% higher vram than nvidia lmfao

[honest tendon]

is this right

[meager carbon]

yeah not the display driver

[honest tendon]

i think i fucked up something

[honest tendon]

I think its this but idk what this means

what is the file explorer bar

[atomic gorge]

I’m new can anybody tell how subscription works how can I deploy my own instances

[meager carbon]

this part

this is how it should look

[honest tendon]

OH that makes much more sense

thanks

omg finally its working

time to eat my chicken n rice

[fervent lantern]

I followed the instructions for setting up stable diffusion on my Mac M1: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Installation-on-Apple-Silicon

I'm using Stable Diffusion 2.0 (768-v-ema.ckpt) and included the yaml file as well.

When I run ./webui.sh I get errors
loading stable diffusion model: AssertionError
AssertionError: Torch not compiled with CUDA enabled

and I get

RuntimeError: "addmm_impl_cpu_" not implemented for 'Half'

The UI opens. But if I try to generate an image from a prompt I get AssertionError: Torch not compiled with CUDA enabled in the UI.

**I'm confused as to why CUDA is involved here if I used the instructions for Mac. Isn't CUDA for Nvidia? And Mac doesn't have Nvidia.

So what am I doing wrong?**

Update: I updated to Mac OS Sonoma and now I'm still getting this in the terminal when I launch Warning: caught exception 'Torch not compiled with CUDA enabled', memory monitor disabled but using GUI works now to generate images

[subtle quarry]

I don't understand this, it works up until the end of the image making process and then slams its face into the wall

[honest tendon]

does it take this long

nothings changed in like a couple minutes

[meager carbon]

yes

[ancient plover]

Hey, I run Zluda SD at the moment, does anyone know if there would be any issues upgrading from windows 10 pro to 11? Would I need to change or redo anything or should it work without any changes?

[meager carbon]

there shouldn't be any issues, no

[honest tendon]

I think its smooth sailing from here, thank you very much @meager carbon

imma enjoy mah protein shake now

[meager carbon]

cheers

[honest tendon]

is this done

can I close it

[meager carbon]

can you try generating something first

[honest tendon]

it is currently

but its taking a while

[meager carbon]

can you look in dms :3

[meager carbon]

it will take about 15-40 min

just for the first time though

[subtle quarry]

Can someone tell me why my images are coming out so bad??

like all of my images have just come out so bad

the previews look fine, but when they pop out they just...look awful

[meager carbon]

whats wrong wtih it?

[subtle quarry]

Here's where it begins
#📝｜prompting-help message

[meager carbon]

try a different model... im curious

[subtle quarry]

I've been trying different models

[meager carbon]

also i see no color change between the two ima be real

[subtle quarry]

the colors just look...bad

either too dark or way too bright

like this

way too bright

[meager carbon]

is it?

[subtle quarry]

it also just looks bad.

[meager carbon]

but... how???

like

[subtle quarry]

the shapes, the colors,

like everything just. it doesn't look right

like this for example. it looks BAD.

[meager carbon]

i... genuinely don't see what's so bad about it to you

i mean, it just looks like you need to maybe bump your res a bit, or use a dif model

[subtle quarry]

my res is 512x776 before high res fix and after high res fix it's 665x1008...but i'll bump it higher i guess

oh. awesome. thanks eldritch abomination! Very cool!

bah, this is getting on my nerves. everything was fine at first, and then something happened and everything gave up.

like it started FINE.

[honest tendon]

looks nice

[honest tendon]

what is the (red cardigan:1.2)

is that like percentage

12% or 120%?

[subtle quarry]

it's a weight...idk what percent it is.

[honest tendon]

omg

[meager carbon]

oh also, try euler a

[honest tendon]

its taking so long

[meager carbon]

yes, the first time it will take a long time.

[meager carbon]

see here

[honest tendon]

hopefully not 40

;-;

[subtle quarry]

you gotta break in your system or something?

[honest tendon]

idk the science behind it but i guess thats how it works

[meager carbon]

no, just how zluda works

you can look in the terminal

maybe it has more info haha

[honest tendon]

is it this

[meager carbon]

nah

the one that's like

[honest tendon]

is it for the webui file

[meager carbon]

running SD

[honest tendon]

the windows batch file

to be specific

[fair oxide]

lower the denoising strength in hires fix to about .45

[subtle quarry]

lemme finish this batch since this model is actually cooperating with me...unlike the others for whatever reason

[fair oxide]

dont use latent upscaler either

those two things might help

[subtle quarry]

see this is what i expect, like this kind of quality and color

changed it around

[fair oxide]

lmk if its better

[subtle quarry]

hopefully it is, because colors are my enemy today

[fair oxide]

are you using a vae??

[subtle quarry]

nope

[fair oxide]

use one

https://civitai.com/models/22354/clearvae

here

this is a common one

[bright warren]

any news on stable diffusion for non rocm amd gpus?

or on rocm for literally any gpu that's not .001s old

amd moment

[subtle quarry]

batch is almost done

[bright warren]

oh god wrong server, sorry for the meme (or not if you appreciate it)

[subtle quarry]

That's a lot better. Quality is a little funky, but thats fine

gonna try with the vae now

[fair oxide]

[subtle quarry]

It's a lot better than getting colors that were saturated or weirdly composed

or whatever the word is

[fair oxide]

im not good with all the tech stuff but i had the same issue when i first started forge

[meager carbon]

uh, what? p much all amd gpus work wtih SD

[bright warren]

with onnx?

[subtle quarry]

I knew it had to be something 100% wrong with my settings, I just couldn't for the life of me figure out what those settings were

[fair oxide]

i fixed it by using VAEs, keeping my loras at .9 strength or below unless specified, dont use latent upscaler, and keep your denoise at .4-.6

[meager carbon]

?

[bright warren]

hey im making the question

[fair oxide]

hope this helps you quin

[meager carbon]

idk what you mean

[bright warren]

wat

[meager carbon]

like, follow this

[bright warren]

what other way is there then?

[subtle quarry]

ty for the help! they seem to be coming out better

[bright warren]

is this windows only? rip

[fair oxide]

you can find more VAEs on civit with more color

i just gave you a popular example

[meager carbon]

no, i don't think it is

linux too

[subtle quarry]

I used uh... ftmse 840000 or whatever it was

[bright warren]

it looks like directml is tied with dx12

[meager carbon]

ooo

ehm

i mean

there is a run.sh script in there

so idk

[bright warren]

dang im confused now

[fair oxide]

yeah, there are better ones 💀

go digging

[bright warren]

I think ps can run .sh

well, windows qemu it is

[subtle quarry]

ty! Hopefully with the other models it looks great as well! 💜

[fair oxide]

np

[meager carbon]

sh is a shell script that is the equivalent of a batch script

[subtle quarry]

also, sorry senk0 if i seemed snappy earlier. I knew I could do better and it was making me frustrated that i just couldnt for some reason 🫠

[bright warren]

true

i mean you could #!usr/bin/env anything

[meager carbon]

no worries haha

I'm sorry i didn't recognise what was "wrong"

to me it seemed (and still does seem) things look more or less fine

[subtle quarry]

Nah youre good, quality is in the eye of the beholder, and i just have a high standard for myself

[bright warren]

ok this directml bs broke the camels back for me, im switching to proxmox

[subtle quarry]

okay now my poses seem stuck LMAO

Im getting the same poses over and over despite a different seed

time to chuck controlnet at it

[meager carbon]

i just dual boot on another drive haha

500gb nvme drives are like 30$ nowadays

[bright warren]

same but I have windows 10 (iot enterprise ltsc btw) on a 120gb ssd, very restricted all the time

[meager carbon]

120gb is ewaste

[bright warren]

I haven't speedrun a tech job yet

im a neet

gotta make do

[fair oxide]

use wildcards

[subtle quarry]

..oooh wildcards

i kinda forgot about them

[fair oxide]

i just learned about them

theyre really fun for making dynamic stuff

[subtle quarry]

do you know of any for poses?

i'm getting a lot of uh...nsfw ones.

[fair oxide]

https://civitai.com/models/20868/200-wildcards-nsfw-and-sfw

you can just import all of the non nsfw stuff into your forge

[subtle quarry]

ahhh that's epic, ty

[honest tendon]

its been an hour

https://tenor.com/view/death-skelly-move-ball-gif-18688511

[fair oxide]

Is what you generate using stable diffusion private?

[honest tendon]

this is taking too lonmg

its been 2 hours

[meager carbon]

hmm, that is a while...

shouldn't be THAT long.

id wait for cs1o to hop on

[honest tendon]

its doing this

[ornate elk]

Hey, can you show your path settings?

Also what's your GPU?

[ornate elk]

Try to not select the visual studio plugin

And then hopefully it works

[meager carbon]

nope

lets do this tomorrow.

I tried everything imaginable

reinstalled the shit 15 times

and even uninstalled VS

(i use vsc anyways)

[ornate elk]

Then try without the vs plugin next time

[meager carbon]

already did

it didn't even show up

after i uninstalled*

[ornate elk]

Hmm

[meager carbon]

any idea?

maybe some of my prereqs are bad?

[ornate elk]

You have the latest AMD driver?

[meager carbon]

mhm

i think i have the path setup done right

[ornate elk]

Some people also have updated their visual studio code and then the hip install worked

[meager carbon]

the hip install did work

sorry

that ss is outdated

im getting the torch error now

[ornate elk]

Ohhh

[meager carbon]

any idea?

[ornate elk]

The paths look correct

[meager carbon]

cool

[ornate elk]

Which zluda files did you used?

[meager carbon]

uhhhh wdym

[ornate elk]

3.7 got release two days ago

I tested it successful yesterday

[meager carbon]

i installed latest today

[ornate elk]

Okay

[meager carbon]

i can try reinstalling though

[ornate elk]

What's your GPU?

[meager carbon]

7900xtx

[ornate elk]

Okay then open up a CMD
And run
Pip cache purge
Then delete the venv folder and relaunch the webui-user.bat

[meager carbon]

sure

1200 files del

this shouldn't be an issue, correct?

regardless, same issue.

[ornate elk]

Nope

[ornate elk]

What's in your webui-user.bat?

[meager carbon]

any idea?

or is my shit just cursed

[ornate elk]

try this:
go into the C:\ZLUDA\ Folder.
There copy the cublas.dll and the cusparse.dll and paste them inside it again.
Rename the copys to cublas64_11.dll and cusparse64_11.dll
Copy these two files to stable-diffusion-webui\venv\Lib\site-packages\torch\lib, and overwrite if asked.

[meager carbon]

uh

There copy the cublas.dll and the cusparse.dll and paste them inside it again.

what?

[ornate elk]

Make a copy of these two files inside the zluda folder

[meager carbon]

oh okay

[meager carbon]

same issue.

no change

[ornate elk]

Very strange

[meager carbon]

any way to verify zluda is actually even working?

as it currently stands, it seems that it's not even being recognized

even if it's deleted, it has the same reaction

which perhaps indicates it has no idea what to do

[ornate elk]

Is the folder called C:\Zluda?

[meager carbon]

no

C:ZLUDA

[ornate elk]

Ok

Are you on windows 11?

[meager carbon]

mhm

sadly

[ornate elk]

Then try to move the ZLUDA folder into C:\Programms or anywhere else and then edit the path in the system variable

[meager carbon]

sure.

[meager carbon]

nope.

it just has no idea what zluda even is

[ornate elk]

Can you show me the full CMD output?

When starting the webui-user.bat

[meager carbon]

looking through the known issues

this could be a problem

[meager carbon]

sure

[ornate elk]

Hmm

[ornate elk]

You can try to disable the igpu

In the device manager

[meager carbon]

I could... it's already 1am and i have work tomorrow though.

[meager carbon]

how?

nothing shows here

[ornate elk]

Ah okay then its not a problem

[ornate elk]

We can compare our setup when I'm at home.
I have the 7900xtx too and it works with zluda

[meager carbon]

when will you be home?

[ornate elk]

In 8 hours

[meager carbon]

awesome

thats the time i get up at

and sure, I would love to. I really appreciate the help, and id be willing to compensate in some way.

[ornate elk]

I recheck my settings, and maybe I find some differences

[honest tendon]

suree

just a quickie

[meager carbon]

I can completely wipe everything and start from scratch.

[honest tendon]

6950xt

@ornate elk

[meager carbon]

It's there

Just under a different dir

[honest tendon]

really?

[ornate elk]

It's there

Ok

[honest tendon]

is this normal

[ornate elk]

Don't use spaces in the folder name if you can

[meager carbon]

This

[honest tendon]

so I have to rename it to AIDOWNLOAD

or simply just download

[ornate elk]

Yes and change the path accordingly

[honest tendon]

so just path and file folder correct, i dont need to adjust anything else

or I still have to

[meager carbon]

Goodnight cs1o, I really appreciate the help. Have a nice day at work

[ornate elk]

Thanks! Gn and Cya later

[honest tendon]

gn

[ornate elk]

Yes

You may need to delete the venv folder before relaunching the webui-user.bat

[ornate elk]

Your missing \ZLUDA

[honest tendon]

OH mb forgot

there we go

deleting the venv folder

holy thats a lot

looking good so far

this looks promising, it actually stays in "interrupt and skip"

[ornate elk]

That's good

What does the CMD shows?
It should show no errors and just "applying dogetexx..." as last line

Wait 15-40 mins and it should work now

[honest tendon]

WTF IS THATTT

@ornate elk

AJHAHAHHA

how do i fix the faces

is there a mod

[limber pier]

looks ready to post in #🌠｜show-and-tell

[ornate elk]

You need to use an other checkpoint (model)

[sudden hawk]

Hello, i have dl a lot of Loras but they don't appear on my SD, could someone help me pls ?

[ornate elk]

Vae-ft-84000-ema is a vae and not a lora. It goes into models/vae

The other Lora's are only shown when the right base model is selected

Means 1.5 model for 1.5 Lora's
And sdxl model for sdxl loras

[kindred granite]

Hello ! I would like to train an embedding for learning SD how to make a special kind of creature from a licence. If someone may DM me for some help i would love ! The first image is the reference creature, a Sliver, a the second image is my first result. My questions are : May i go further than this result ? Which setting are the best for SD to understand the main features of Slivers ? Best Regards

[sudden hawk]

Ok thank you !

[vapid sluice]

Why are all my images after applying a lora i made black/blank ?

This started happening when i set network dim and alpha to 128

I tried lowering the learning rate but it didn’t work

[fallow arch]

someone can help me?

requests.exceptions.ConnectTimeout: HTTPSConnectionPool(host='huggingface.co', port=443): Max retries exceeded with url: /anyisalin/big-lama/resolve/main/big-lama.safetensors (Caused by ConnectTimeoutError(<urllib3.connection.HTTPSConnection object at 0x000002BA962D26E0>, 'Connection to huggingface.co timed out. (connect timeout=None)'))

The background prompts me this

[hard quiver]

hi. can anyone tell me where am I supposed to put lycoris file?

[honest tendon]

is there a way to like add in multiple image reference photos manually to have one specific style that can edit my already existing photo

[hard quiver]

so it means I can jus paste it in lora folder?

[ornate elk]

Yes

[ornate elk]

Hey, that's a connection issue

Use Ethernet.
And if your from Asia, Russia, Iran, you may need to use a VPN

[fallow arch]

thanks,my bro

[vapid sluice]

What learning rate should i use for network dim 128 lora training?

[ornate elk]

Hey why did you installed Anime4k as an extension in SD?
Its an tool for realtime upscaling video files

[carmine agate]

Hey, forget is working great for me except one thing: often clicking fields on the web you cause it to sort of turn unresponsive for several seconds, then the cursor appears in the clicked field & is editable. Any others have this issue or (better yet) a solution to overcome this? 😆

[hard quiver]

can someone tell me what the problem is?

it was working fine yesterday.

the pic gets generated like normal. however after the generation is done. I get this

[ornate elk]

Wrong VAE file.

First go into settings, User interface, quicksettings, there add sd_vae
Then hit apply and reload ui

[hard quiver]

is it a problem when different checkpoint have same vae files

[ornate elk]

Nope

[hard quiver]

cuz then I only copy one. cuz otherwise u get two same vae files with the second one having 1 in the end. cuz its a copy.

[ornate elk]

But sdxl models don't work with 1.5 VAE files

[hard quiver]

so. should I keep one vae in this particular instance or both of them?

[ornate elk]

You can delete the copy

[hard quiver]

being that they are copies.

[ornate elk]

Keep the ones you like. If you do the steps I described above, you can switch between them easily

[hard quiver]

what do u mean? switch b/w them. isn't it automatic?

[hard quiver]

i don even know what that is yet. xd

[ornate elk]

Different model versions.
The most used are 1.5 based and sdxl based

VAE files are also either 1.5 or sdxl based

And everything only works with the same base

[hard quiver]

no quick settings here

[ornate elk]

User interface

Click on it

Then on the right there should be quick settings

[hard quiver]

user setting itself is unclickable

oh nvm

figured it out

damn im stupid

[spiral lodge]

can anyone tell me how to generate images

[fallow arch]

What happen?somebody know

[hard quiver]

how would I know which vae to use xd.

do I need to remember all of them now.

is the automatic option any good?

[hard quiver]

is something like this possible for loras too? so I don't have to write it in prompt everytime or whatever.

that would be a huge help

[median current]

Is it possible for stable diffusion to crash a gpu? I've only ever had cuda errors due to exceedjng vram limits, but recently started getting white screens and crashes. Trying to determine source and problem component

[hard quiver]

what does weight mean in stable diffusion? are they talking about denoising strength?

[ornate elk]

Sdxl VAE is for sdxl, all others are for 1.5

[ornate elk]

Click on the lora tab and then on the lora to add it to the prompt

[ornate elk]

If the GPU uses to much power and the power supply is to less then yes

[hard quiver]

is there a way to get them to preview?

[meager carbon]

@ornate elk you home yet?

[median current]

That's the thing. I'm running 4090FE on a Corsair SF1000L. Power shouldnt be an issue unless the psu is faulty. No?