🤝｜tech-support

[ornate elk]

And if you then have a question feel free to ask

Forge is pretty much just downloading a zip file and running the update bat and then the run bat

[turbid wing]

just one more question

so when i launch stable diffusion(automatic aaaa) and generate image it takes like 5-6 minutes to generate one 1040 by 1040 pixel image on model (perfect illustirious v2) but when i swap models and swap back again to perfect illustrious v2 ( same model) it creates same res image in just 1-2 minutes

why does that happenes ?

[ornate elk]

Thats strange. How much RAM do you have?

And did you used any performance launch args in Auto1111?

[turbid wing]

i have 16 gigs of ram, and what are args ?

[ornate elk]

Arguments for the webui-user.bat

For a better performance

[turbid wing]

it was always like this, even after using xformers it stayed same

[ornate elk]

Ah okay then its probably the RAM issue.
When using sdxl/pony/illus models and you have only 16gb of RAM they can stuck and also slowdown or freeze the PC.
So you have to manually increase the Pagefile size in windows.

I can link a guide for that

[turbid wing]

yes please

[ornate elk]

Here:
https://www.tomshardware.com/news/how-to-manage-virtual-memory-pagefile-windows-10,36929.html
Make sure the pagefile is only enabled for the C drive and disabled for any other drive.
Make sure its set to custom for C at 16000min and 24000max.
Then click OK and reboot the PC.

[turbid wing]

and what is the recomended ram for them ? yes i have seen my ram usage going up to 89 or 98 percent, and also the reason i was reinstalling was because today it was freezing pc so much even after closing the program

[ornate elk]

32Gb is recommended

But with the Pagefile adjustment your fine

Then it won't freeze

[turbid wing]

thanks man, i have downloaded the forge ui and now i will try to install it

[ornate elk]

Alright 👍

[turbid wing]

@ornate elk quick update yes it is working so good, better and even faster than automatic aaaa

thanks for helping me out brother

[ornate elk]

Perfect, no problem 🙂

[ornate elk]

Hey, the video is against the nsfw rule of the server. Maybe you can show an other example and remove the other one.

[dusky shadow]

okay mb!

[ornate elk]

Np, for smoothness of videos checkout interpolation features.

[broken moth]

Is there a way to get the delete button extension to work with --share on

[indigo vigil]

Does anyone know why now in Forge, my extensions tab is now gone? In the terminal it says "*** "Disable all extensions" option was set, will not load any extensions ***"

[ornate elk]

Have you installed forge localy?

[indigo vigil]

Yes

tried looking in the run files but nowhere did it say it was disabled. I just decided to reinstall it and now it works

installed on Pinokio

[ornate elk]

Ah okay

[livid jetty]

could i move my zluda folders and SD-Zluda files to a usb?

[ornate elk]

to copy them over or to launch the webui from there?

[livid jetty]

yes

[ornate elk]

yes is no answer to an OR question xD

but if you want to do boath then you should know that launching from the USB drive wont work

And if you copy the files over to a new PC then you have to delete the venv of the webuis and make sure the PC has python and git installed

[livid jetty]

oops sorry, i moved my zluda and sd zluda file to a usb and i tried launching it from there

[livid jetty]

gotcha

[ornate elk]

it can work but from a USB stick it would be painfully slow or crashes

from an external ssd it could be okay

but your always limited to the USB connection speed of the PC Case or Mainboard

[livid jetty]

alright im moviong them back to my local disk

[ornate elk]

did you moved them over to save space?

because you could also just move the models onto the USB (if its a drive and no stick) and then link the path in the webui

you could cut out the whole models folder and in auto1111/forge/reforge you can then add --models-dir "E:\USBDrive\Path\to\models"

with this method you can share the same model files to all webuis you have installed. Thats what i do. So no duplicated models.

ComfyUi has an own file to edit for the paths.

[livid jetty]

so i was in the middle of moving my items but i stopped becasue i felt overwhelmed. I feel like i may have dont somthing wrong. i now have 2 files called SD-Zluda, The one on my Local Disk has only a Comfyui folder and inside of it it has a bunch of items along with the startcomfyui.bat.
The other one on my usb drive, has 2 folders, a swarmui folder, and a comfyui folder, the comfy ui folder only has Temp, Venv, and Pycache folders in it, is this how its supposed to be?

[ornate elk]

seems like you interrupted the copy process after it was done with swarm it started with comfyui but then got interrupted

so if the comfyui folder on the PC has no venv folder you can move everything back from the other comfyui folder

and then you can move the completet swarm ui back to into SD-Zluda

so that in the end you have only 2 folders in SD-Zluda (SwarmUI, ComfyUI)

[livid jetty]

could you walk me throught that, sorry

but i also remember having other issues last night after you left. i was saving them for now

but ill need to get my models connected first

here is one of the errors i got

All available backends failed to load the model 'C:/SD-Zluda/SwarmUI/Models/Stable-Diffusion/OfficialStableDiffusion/sd_xl_base_1.0.safetensors'.

Possible reason: ComfyUI errored: {
"error": {
"type": "prompt_outputs_failed_validation",
"message": "Prompt outputs failed validation",
"details": "Exception when validating node: validate_inputs() missing 1 required positional argument: 'validated'",
"extra_info": {}
},
"node_errors": {
"101": {
"errors": [
{
"type": "exception_during_validation",
"message": "Exception when validating node",
"details": "validate_inputs() missing 1 required positional argument: 'validated'",
"extra_info": {
"exception_type": "TypeError",
"traceback": [
" File "C:\SD-Zluda\SwarmUI\dlbackend\comfy\ComfyUI\execution.py", line 850, in validate_prompt\n m = validate_inputs(prompt, o, validated)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n"
]
}
}
],
"dependent_outputs": [
"101"
],
"class_type": "SwarmJustLoadTheModelPlease"
}
}
}

it was on swarm

[ornate elk]

Does Comfyui work without swarm?

Thats important.
Then you can just copy Comfyui over into swarm dlbackend

[livid jetty]

Yes

[ornate elk]

Okay thats good.
And the error from above comes after you setup SwarmUI and copied comfyui over?

Its late im off now.
I will recheck the swarm ui setup tomorrow

[livid jetty]

Uh oh, bots

[vocal burrow]

Cleaned

[livid jetty]

i got the same error, and this is what my directory looks like

[ornate elk]

Delete the comfy folder and then rename the ComfyUI folder to comfy then relaunch

[white anchor]

I've been having trouble getting SD to agree with my setup. Rtx 5060Ti with an AMD processor. I keep running into kernal and xformers issues. Is this endeavor hopeless?

[ornate elk]

Can you show the full cmd log?

Which webui do you want to install?

Have you used the recommended Guide from the first link of the pinned messages?
Go for Forge Webui or Auto1111 boath have instructions for RTX50 cards

[white anchor]

Auto1111 is what I've been attempting. I'm hitting issues when trying to integrate DreamBooth for training. I keep having to build locally and still hit CUDA or kernel issues in the UI itself.

[livid jetty]

I got this error now

[Error] Failed to auto-update comfy backend: System.ComponentModel.Win32Exception (267): An error occurred trying to start process 'git' with working directory 'C:\SD-Zluda\SwarmUI\dlbackend\comfy\ComfyUI'. The directory name is invalid.
at System.Diagnostics.Process.StartWithCreateProcess(ProcessStartInfo startInfo)
at System.Diagnostics.Process.Start(ProcessStartInfo startInfo)
at SwarmUI.Utils.Utilities.RunGitProcess(String args, String dir, Boolean canRetry) in D:\SD-Zluda\SwarmUI\src\Utils\Utilities.cs:line 1158
at SwarmUI.Builtin_ComfyUIBackend.ComfyUISelfStartBackend.<>c__DisplayClass24_0.<<Init>b__0>d.MoveNext() in D:\SD-Zluda\SwarmUI\src\BuiltinExtensions\ComfyUIBackend\ComfyUISelfStartBackend.cs:line 351
10:28:29.552 [Error] Failed init of ComfyUI-0 with script target 'dlbackend/comfy/ComfyUI/main.py' because that file does not exist. Please verify your start script location.

[dusky shadow]

Hello guys, when using DWPOSE Estimator, i found that hands motion arent tracked very well, how can i have a precise motion with V2V? Combining with Depth Anything change the final result too much 😒

[ornate elk]

Dreambooth is outdated and breaks the webui.
Delete it from the extensions folder and then delete the venv folder to fix Auto1111.
Then use standalone training tools like OneTrainer or Kohya_ss

[ornate elk]

Don't move comfyui into comfy you have to rename Comfyui to comfy

[livid jetty]

that is what i did i beleive, but i will try again

so in the sdzluda file, i copy the comfyui folder into the same place as the comfy folder correct?

[ornate elk]

You copy the ComfyUI folder and place it inside dlbackends folder of swarm

There rename it to comfy

[livid jetty]

oko ill try again, just wanted to make sure

i got the same error

[ornate elk]

Can you show it?

[livid jetty]

yes sorry

11:09:46.681 [Error] Failed to auto-update comfy backend: System.ComponentModel.Win32Exception (267): An error occurred trying to start process 'git' with working directory 'C:\SD-Zluda\SwarmUI\dlbackend\comfy\ComfyUI'. The directory name is invalid.
at System.Diagnostics.Process.StartWithCreateProcess(ProcessStartInfo startInfo)
at System.Diagnostics.Process.Start(ProcessStartInfo startInfo)
at SwarmUI.Utils.Utilities.RunGitProcess(String args, String dir, Boolean canRetry) in D:\SD-Zluda\SwarmUI\src\Utils\Utilities.cs:line 1158
at SwarmUI.Builtin_ComfyUIBackend.ComfyUISelfStartBackend.<>c__DisplayClass24_0.<<Init>b__0>d.MoveNext() in D:\SD-Zluda\SwarmUI\src\BuiltinExtensions\ComfyUIBackend\ComfyUISelfStartBackend.cs:line 351
11:09:46.734 [Error] Failed init of ComfyUI-0 with script target 'dlbackend/comfy/ComfyUI/main.py' because that file does not exist. Please verify your start script location.

[brave frigate]

Can someone help me here? My SD won't start and idk why:

[bitter pier]

What numpy version are you using?

[ornate elk]

Its still the same problem

Comfyui needs to be one folder path higher

[ornate elk]

I dont see any launch args. Whats in the webui-user.bat?

Add --use-zluda --update-check --skip-ort
Then relaunch

[brave frigate]

This is what i have

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --use-zluda --update-check --skip-ort

git pull

call webui.bat

[ornate elk]

Ah okay

[bitter pier]

This is just the numpy version issue.

[ornate elk]

@brave frigate Then relaunch once, the error should go away

[bitter pier]

Which version are you using?

[brave frigate]

version? like i use zluda

[bitter pier]

You can check it using pip list

[brave frigate]

still have the same issue. I'll delete the venv again and see if that fixes it

[ornate elk]

Yep if not, you may need to reinstall it

[brave frigate]

Version 1.10.11-amd-39

[bitter pier]

You mean numpy 1.10.11?

[ornate elk]

No thats the webui version

[brave frigate]

idk, it just says

[bitter pier]

As a default zluda installs the latest version of numpy that's available.

Please type pip list

On your terminal or command

[bitter pier]

That must be done in your venv

venv\Scripts\activate

And then pip list

[brave frigate]

ok ok I'll do that once venv is installed again xD

[bitter pier]

There is no need to reinstall your venv.

[ornate elk]

Did it worked?

[brave frigate]

no same issue, im looking for that rn

[bitter pier]

So I warnned you. 💢

[brave frigate]

well yes you did but trial and error is good :)

[bitter pier]

Maybe.

[brave frigate]

so which of these do i need to edit?

[bitter pier]

You don't need to edit anythihng.

Just give the below command at your root directory. venv\Scripts\activate

[livid jetty]

im VERY sorry, it might not be clicking but i want to show you what i do

[bitter pier]

Hi, why don't we move onto DM?
I think it's so confused here.

[brave frigate]

sure thing but im on a call rn so can only text if thats fine

[bitter pier]

Sure.

[white anchor]

Wish I could pay for a "tutorial" of sorts. As is I'm attempting to self teach and it's driving me mad.

[ornate elk]

What's inside the Comfy folder now?

[ornate elk]

Hmm looks okay

[ornate elk]

Let me know if its fixed

[random night]

just follow your guide, and i must admit therock is the goat

[brave frigate]

we're on it rn

[ornate elk]

yea right? im so hyped about august when they want to release HIP SDK 6.4 on Windows with probably the First stable TheRock builds

[random night]

yup, i am too very eager to try that out

[brave frigate]

We did all this you can just look at the cmd it has all the steps. u can follow them probably easily.

[livid jetty]

also just to be clear im still geting the error, and comyui works fine

[ornate elk]

And it works now?

[ornate elk]

I will retest the SwarmUI guide. I'll let you know how it goes

[brave frigate]

yerp its compiling on the first generation atm so i can tell in a few min#

[ornate elk]

@livid jetty okay tested and have to correct myself.
rename the comfy folder back to ComfyUI and then create a new empty comfy folder. Move the ComfyUI folder into the comfy folder.
Then relaunch.
But it seems it still wont work and errors out if you try to load a model.
Probably you need to Install ComfyUI-Zluda from PatientX repo and try setup swarm with that.

[dusky shadow]

Hello guys 🙂 Is someone familliar with DW pose ? because i have a good skeleton motion , but in the end result my hands and gesture are not perfectly matching with my source. (especially the end), is there a method to have accurate movement from video source?

[bold jay]

00:10:18.281 [Error] [WebAPI] Error handling API request '/API/DescribeModel' for user 'local': Model not found.
00:10:19.157 [Error] [WebAPI] Error handling API request '/API/DescribeModel' for user 'local': Model not found.
i was installing swarm and got these msgs in cmd

[tall lichen]

Can anyone please explain in simple terms to me how I can get by this error? I am trying a new checkpoint, but everything I generate is getting a blank or black screen:

[ornate elk]

whats your PGU?

[tall lichen]

Geforce RTX 4070 Ti

[ornate elk]

can you show the settings you used?

[ornate elk]

important is the checkpoint an vae

[tall lichen]

I am not at all savvy at this, just learning

[ornate elk]

because there is the issue

these two things

which checkpoint do you use?

[tall lichen]

vixonsOtakuGens_v1

[ornate elk]

how large is that file?

[tall lichen]

I think it was around 6 gigs

[ornate elk]

okay then its probably sdxl based

[tall lichen]

I apologize, but you'll probably have to explain this all to me like I am a 2 year old, very new at it xD

[ornate elk]

no problem

some models need an additional file called VAE

that fixes little details and the colors

[tall lichen]

oh, so thats why I don't have that extra bit that your SS had

[ornate elk]

yea i explain how to get it there:
so first download the SDXL VAE file from here (direct link)
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/blob/main/sdxl.vae.safetensors
Move that file into the models/vae folder.
Then on the webui go into Settings, User Interface/User Interface, Quicksettings on the right.
There add sd_vae to the quicksettings.
Then hit apply and reload ui.
Then you should have that dropdown and can select the vae file.

[tall lichen]

Thank you! I'll go do that now ❤️

[ornate elk]

no problem, let me know if it worked

if the model is 1.5 based then you need a different file

@tall lichen also whats in your webui-user.bat file?

[tall lichen]

Kinda stuck here atm, I don't see "quicksettings" on the right

also not sure what the webui-user.bat is or where I find it, sorry

[ornate elk]

there add sd_vae

dont remove the sd_model_checkpoint

[ornate elk]

oh np, which file do you start to start the webui?

[tall lichen]

this windows batch file thingy is what I use to start up:

[ornate elk]

yea thats the webui-user.bat i meant

can you right click and edit it?

[tall lichen]

yes, it brings up a text box

and yay, the picture generated this time ❤️

[ornate elk]

can you screenshot that? because you can add there some args for the best performance

[ornate elk]

ah okay, at the line set COMMANDLINE_ARGS= you have to add: --xformers
then save and close and relaunch the webui

[tall lichen]

so it should look like this: COMMANDLINE_ARGS=--xformers ?

[tall lichen]

I think that broke something, I restarted my UI, but it gave me this message when I tried to generate a picture:

[ornate elk]

oh then try to delete the venv folder and relaunch the webui-user.bat

probably torch version was not correct

[tall lichen]

That fixxed it, thank you very much ❤️

[ornate elk]

perfect, no problem 🙂

[livid jetty]

Did you comfyui from patients repo? Or is that just a theory?

[ornate elk]

A theory because his comfyui version is more recent

[brave cedar]

Hey, I'm getting this error on Forge UI using ROCm on an all AMD ystem

Does anyone know how to fix this?

I am using Stability Matrix to get onto Forge

I am also using the other OS that isn't Windows or MacOS, but for some reason that word is banned???

[ornate elk]

What's your torch and hip version and GPU?

[velvet cloak]

Hello, wondering if Spar3d is something I can run offline or if it needs internet connection after the installation.
I just got the code from git and ran the gradio_app.py file until it said I need huggingface token.
Wondering if post the token creation and few other initial steps needed, does it still require me to be connected to the internet?
Please direct me to the right channel if this isn't the right one.

[brave cedar]

Thanks, but I am no longer in need of help, I am going to dial boot Windows and run things there as it seems windows doesn’t give so many problems

[random night]

Hi guys. Could you please tell me how to install a working Automatic1111 Stable Diffusion on the new generation of video cards, like the 5060 Ti?

[misty wren]

v2beta/image-to-video is 404, is there anything else i can use instead?

[lime wind]

Has anyone ever made an Extension for A1111 before? I have questions.

[ornate elk]

Yea on windows I have no issues with my 7900xtx.
Best is to follow my Zluda Setup Guides.
First link of the pinned messages here.

[ornate elk]

Hey, you find the guide on the first link of the pinned messages too

[ornate elk]

Hey, the Driver shouldn't affect the output image at all

Can you show an example of the issue?

Because good or better quality is a not much of an info

Also which Webui, model, vae, etc do you use?

Like I said the quality isnt affected by the GPU or driver at all

It comes down to the checkpoints and settings

I can create stunning images even on a CPU powered webui if I would need to.

Maybe just check your images meta data and try to recreate the image on with the new GPU. Then compare them.

[steep jungle]

Hi everyone. Do you know how to make stable diffusion generate images that stick strictly to the prompt, or at least the way DALLE does? I've been generating images using ChatGPT 4o and the images usually stick to the description, but when I generate images in stable diffusion, either parts of the image are indecipherable messes, the image is missing things mentioned in the description, or the image includes things not mentioned in the description. I've been using automatic1111 UI and IllustriousXL as checkpoint for stable diffusion.

I wanna get stable diffusion to work properly because the images it generates feel "less AI" and its styles are better. DALLE generated images are always obvious that they're AI generated

[ornate elk]

Hey, illustrious models are very good for prompting. But important is to install the booru tag autocompletion extension.
With that you know if the model will understand the words and how good it does.

[steep jungle]

Let me try that out and see how it goes

[ornate elk]

Illustrious models got trained on danbooru images and their tag system.
The extension gives you word recommendations based on the words you type

The number behind the words shows how often the tag is in the database of danbooru.
The higher the number the more likely the model will exactly know how to generate it.

[brave frigate]

Turns out after restarting my computer its back to not starting again 😭 plus it couldn't generate an image yesterday so i just left it:

can i maybe factory reset it? might that help?

[ornate elk]

Can you try add --update-all-extensions
To the webui-user.bat
Then relaunch

[brave frigate]

ok I'll try it out

still not starting but it did update some things

hm

maybe I've been cramping this for too long i might just completely delete it and reinstall it

might be necessary its been like 1 year and a half since i first installed it

[steep jungle]

It doesn't help that much sadly. I think I should go and watch a youtube tutorial or something on how to make better prompts

[ornate elk]

yea probably the best idea, what was your gpu again?

You could also setup Forge webui as its more updated. But Auto1111 still works too.

All Guides are on the first link of the pinned messages.

[brave frigate]

I'll still use A1111, im more used to that

[ornate elk]

me too xD

[nimble summit]

I hope there’s someone here who can help me, as I am in need of some serious technical assistance. Automatic1111 won’t install correctly. Not even using —reinstall-torch : for some reason it keeps downloading all the wrong packages, ignoring what I already have installed, ignoring what’s listed in requirements_versions.txt, and grabbing things that aren’t even compatible with each other. I have no idea why it’s doing any of this.

I have python 3.10.6 installed, and CONDA 11.8

I have torch 2.1.2+cu121 installed as well

but when running run.exe or the webui-user.bat file with —reinstall-torch, it downloads and installs torch-1.13.1%2Bcu117-cp310-cp310-win_amd64.whl

i copied the window to a text file:

I have been beating my head against this for two days, and am utterly stuck.

[bitter pier]

I got you.

Could you let me know which GPU you are using?

[nimble summit]

NVIDIA GeForce RTX 3060

[bitter pier]

Then why did you install conda 11.8?

[nimble summit]

because that's the version I've always had before. I did a complete tear-out and reinstall in an attempt to fix this, and made sure I grabbed the same versions I had originally

[bitter pier]

yeap, That's why you got those issue.

Btw, you installed cuda 11.8 or conda 11.8?

[nimble summit]

right, sorry.. CUDA

[bitter pier]

Yeap, excactly torch 2.1.2+ cu12.1 doesn't work on cuda 11.8

And you work on conda?

[nimble summit]

not sure what conda is installed - which cuda should I have?

[bitter pier]

I'm not so active here, why don't we move onto DM?

[nimble summit]

sure thing

[velvet cloak]

Getting error with Spar3d:
File "F:\AI\stable-point-aware-3d-main\gradio_app.py", line 841, in <module>
run_btn.click(
File "F:\AI\stable-point-aware-3d-main.venv\lib\site-packages\gradio\blocks.py", line 2089, in exit
self.config = self.get_config_file()

...........

File "E:\Programs\Python\Python39\lib\typing.py", line 647, in getattr
return getattr(self.origin, attr)
AttributeError: '_SpecialForm' object has no attribute 'replace'

Any help?

seems hacking is present here in discord too. someone tried to scam me with the name Eros by linking me to different channel asking for wallet address.
kicked me off the channel and chat when I told him i dont have a wallet.

[ornate elk]

Yea there are scam bots on discord.
We try to remove them but nowadays they link or DM people directly or here and then delete their messages

[ornate elk]

Best is to open up a cmd and run
Pip cache purge.
Then delete the webui and start over but use the guide from the pinned messages.

[ornate elk]

i would suggest trying python 3.10.11 64bit instead of python 3.9

[nimble summit]

did that. It didn't help.

[ornate elk]

what does the cmd shows?

when you launch it

[nimble summit]

I:\sdwebui>pip cache purge
Files removed: 367 (2428.7 MB)

but I've done that a half dozen times or so already

[ornate elk]

i mean what does the cmd shows after you reinstalled the webui

also have you upgraded to python 3.10.11 64bit ?

[nimble summit]

you mean running webui-user.bat with --reinstall torch?

[ornate elk]

no you have to reinstall the Webui the correct way using this Guide:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides#nvidia-automatic1111-webui-stable-diffusion-webui

[nimble summit]

ok. I will take a run through that guide and see if it helps

[ornate elk]

yea, let me know if you have any questions but it should be working after that

[nimble summit]

thanks. will report back one way or the other.

[velvet cloak]

Thank you! Will give it a try

[nimble summit]

It worked! webui is open and running again. Thank you 🙂

[ornate elk]

perfect, np

[velvet cloak]

File "F:\AI\stable-point-aware-3d-main.venv3_10\lib\site-packages\gradio_client\utils.py", line 863, in get_type
if "const" in schema:
TypeError: argument of type 'bool' is not iterable

• This is where I am at now. Moved to python 3.10
• The local host link shows internal server error
• I had to change this to True from False -> demo.queue().launch(share=True)

[hard hazel]

just came here to say:

If anyone is having trouble with ultralytics giving you an error related to pytorch and weights_only

activate your venv and pip install -U ultralytics

Took me a while to find it, but I should have tried it earlier 😦

[potent sun]

https://github.com/spacepxl/demystifying-sd-finetuning

here you can see there is a loss/train graph

how can you get that

in kohya or derrhian

[velvet cloak]

any help on this please?

[velvet cloak]

So, I was finally able to get the UI working. The requirements.txt is really **ed up.
Now after clicking the run option - I don't get to see any 3d model.

Any help on what needs to be done?

This is all the console says

[dusky shadow]

helllo 🙂 ! so i have this workflow with 2 sampler, one on the top with Depth Anything + DW pose & the one bottom is just cosvid Lora , but at 81 frame it reset the Depth Pose and motion overral in the final video, could someone help me on that ? ty

[quaint saddle]

cancel my subscription

[ornate elk]

https://stability.ai/contact

[haughty pilot]

@ornate elk hey, I followed your guide to setup forge with my rtx 2080 and when launching webui-user.bat I get this error

this font is better i guess

[ornate elk]

Can you show the full cmd log?

[haughty pilot]

https://ctxt.io/2/AAB4vRDjFw

super long so i figured out it will be the best that way

wait, after launching the UI with run.bat am I supposed to later use webui-user.bat or using run.bat all the time?

[ornate elk]

The run.bat if you downloaded forge as zip

[haughty pilot]

yes I did, run.bat launches the UI, but do those commands in webui-user.bat work then?

[ornate elk]

Yep

[haughty pilot]

alright, i guess problem solved

[ornate elk]

Ah nice xD

[haughty pilot]

the error was when I tried launching the UI via webui-user.bat, using run.bat launches it without any errors

[ornate elk]

Ah yep because forge skips the creation of a venv as it has all files already

[granite nexus]

TLDR; want help finding solution to multi-input image generation

hello, can anyone experienced in using flux kontext help me please.

i just have a few simple questions.

i haven’t used image generators on pc in a few months but i’m wondering if flux kontext has support for using many images as input before generation with new memory optimizations (say 8-15 images). i need to merge a set of designs i have into one but i need use all of them in context in one prompt. i see online there are some spaces with support for multiple images and it seems to run quickly. would i be able to spin this up in my local rig the same way? im on rtx 3080 TI with 16gb vram & 64gb ram and 14 core i7. does swarmui or forge or comfyui support this feature with my vram and specs? how?? please anyone guide me

i want to run something like this locally with support for many input pictures, it seems to support many but it says it’s using Kontext Max, is it only a Max feature currently? https://replicate.com/flux-kontext-apps/multi-image-list

but i don’t know exactly where to begin on comfyui, can someone please give me a basic blueprint?

please dm me or tag me so i can receive notification

[sudden pike]

@ornate elk
Hey. Sorry for being annoying again.
idk why but somehow Stable Diffusion needs more VRam lately..
In the past it worked fine with my settings.
now it doesnt work and needs like 2-5 mins for a picture

and i have checked. its not using the CPU

[ornate elk]

hey, which webui do you use? whats your gpu and whats in your webui-user.bat?

[sudden pike]

RX 7800xt

Downloaded used your guide so Automatic1111

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --use-zluda --update-check --skip-ort

call webui.bat

[ornate elk]

looks okay

how much RAM do you have?

[sudden pike]

96gigs

16gig vram

now it takes like 2-5 mins and in the past 10-20 secs for a image

[ornate elk]

can you show me your settings?

[sudden pike]

and no change when i downgrade my drivers

sure

wait

[ornate elk]

right now 25.3.1 is the best and most stable driver for 7000 series

[sudden pike]

this is my usual setting i used always

This for upscaling.
cant do it higher cuz OOM xD

its working with ~2.6s/it

in the past it was like 6-9s/it

[ornate elk]

which driver are you using?

[sudden pike]

25.6.1

[ornate elk]

also make sure you use the sdxl fp16 vae and the tiled vae extension, when upscaling

[sudden pike]

i downgraded to 25.4.1 where it worked in the past easly

[ornate elk]

yea that makes problems

[sudden pike]

but even with 25.4.1 its slow

[sudden pike]

where can i get that?

[ornate elk]

can you show me a full cmd log of the webui?

[sudden pike]

ye wait

[ornate elk]

the vae here:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/blob/main/sdxl.vae.safetensors
The tiled vae extension:
click on extensions tab, click on Available, click on Load from.
Search for Tiled vae and install, then restart the webui

[sudden pike]

What does it do exactly?

[ornate elk]

the vae file uses less vram as its in fp16 format.
The Tiled VAE helps at the last step of upscaling when the vae gets applied. Vram usage is decreased so you wont get an OOM.

But make sure to only enable Tiled VAE and not Tiled Diffusion or DemoFusion.

[sudden pike]

WARNING: ZLUDA works best with SD.Next. Please consider migrating to SD.Next.
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr  5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: v1.10.1-amd-39-g6d85a7dfa
Commit hash: 6d85a7dfa4977ae4eece9cbbab4725123833be14
ROCm: agents=['gfx1101', 'gfx1036']
ROCm: version=6.2, using agent gfx1101
ZLUDA support: experimental
ZLUDA load: path='C:\Users\Kunde\Desktop\AI\stable-diffusion-webui-amdgpu\.zluda' nightly=False
Skipping onnxruntime installation.
You are up to date with the most recent release.
W0727 22:22:50.806000 19312 venv\Lib\site-packages\torch\distributed\elastic\multiprocessing\redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.
C:\Users\Kunde\Desktop\AI\stable-diffusion-webui-amdgpu\venv\lib\site-packages\timm\models\layers\__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers
  warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning)
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
C:\Users\Kunde\Desktop\AI\stable-diffusion-webui-amdgpu\venv\lib\site-packages\pytorch_lightning\utilities\distributed.py:258: LightningDeprecationWarning: `pytorch_lightning.utilities.distributed.rank_zero_only` has been deprecated in v1.8.1 and will be removed in v2.0.0. You can import it from `pytorch_lightning.utilities` instead.
  rank_zero_deprecation(
Launching Web UI with arguments: --use-zluda --update-check --skip-ort
Loading weights [bdb59bac77] from C:\Users\Kunde\Desktop\AI\stable-diffusion-webui-amdgpu\models\Stable-diffusion\waiNSFWIllustrious_v140.safetensors
Creating model from config: C:\Users\Kunde\Desktop\AI\stable-diffusion-webui-amdgpu\repositories\generative-models\configs\inference\sd_xl_base.yaml
Running on local URL:  http://127.0.0.1:7860

To create a public link, set `share=True` in `launch()`.
Startup time: 11.7s (prepare environment: 14.1s, initialize shared: 1.1s, other imports: 0.3s, load scripts: 0.6s, create ui: 0.5s, gradio launch: 0.4s).
Applying attention optimization: Doggettx... done.
Model loaded in 4.9s (create model: 0.7s, apply weights to model: 2.9s, load textual inversion embeddings: 0.2s, calculate empty prompt: 0.8s).
100%|██████████████████████████████████████████████████████████████████████████████████| 30/30 [01:03<00:00,  2.10s/it]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 30/30 [01:08<00:00,  2.29s/it]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 30/30 [01:08<00:00,  2.67s/it]```

hey atleast it takes a min now.
still way slower than in the past xD

[ornate elk]

a min with upscaling?

try without hires fix then it should take seconds

[sudden pike]

no without

[ornate elk]

oh wow thats bad

your webui also needs an update

[sudden pike]

how? it always said "its newset version"

[ornate elk]

ah i see sry

everything looks okay

so make sure your not running games or Wallpaper Engine in the background

[sudden pike]

how do i download this sdxl vae?

[ornate elk]

the link was a direct download link

you should have the file in downloads

[sudden pike]

wallpaper engine is on but with the settings it "stop" everything when i do something

[ornate elk]

you have to exit it completely

stop wont be enough

it slows down SD massivly

[sudden pike]

worked in the past

but i can try it with off WPE

[ornate elk]

yes pls

[sudden pike]

i have a single safetensors

[ornate elk]

yea put that into models/vae

[sudden pike]

ah there
me dumb sht

[ornate elk]

then in the wbeui go to settings, User Interface, QUicksettings, There add sd_vae to the list.
Then hit apply and reload ui

then you can select it in the new vae dropdown menu

[sudden pike]

where exactly?

doesnt show up 😦

[ornate elk]

there add sd_vae

and remove sdxl xD

and then hit apply and reload ui

[sudden pike]

and now?

this is correct right?

[ornate elk]

no dont remove the other thing

[sudden pike]

ok ill change that

[ornate elk]

add this back too

but in the first place

[ornate elk]

now hit apply at the top

[sudden pike]

sorry for my incopetence xD

[ornate elk]

np xD

[sudden pike]

ye now on top i see my Checkpoint and SD VAE with "sdxl vae safetensors"

1 min again

without hires

[ornate elk]

hmmm

[sudden pike]

to upscale it it says 6 mins

[ornate elk]

something is wrong then

[sudden pike]

i can try to downgrade my drivers to 25.3.1

[ornate elk]

i still would try to downgrade drivers as a friend had clock speed problems on 25.6.1

and with 25.3.1 they were gone

[sudden pike]

now it says 14 mins to upscale by 2 xD

[ornate elk]

im also on 25.3.1 and have no issues

[sudden pike]

how long does it take for you?

[ornate elk]

without hires fix 16-20 seconds

[sudden pike]

and with?

[ornate elk]

with hires fix 24-36 seconds

depending on resolution and upscale by

[sudden pike]

https://tenor.com/view/locote-gif-18404685476945988779

[sudden pike]

What if u do 1024x1024 and upscale of 2?

[ornate elk]

i have a 7900xtx and overclocked it + overclocked the vram and undervolted it

[sudden pike]

ah. ok

should i restart my PC too?

after the downgrade?

[ornate elk]

yes

also open up a cmd as admin and run
sfc /scannow
that fixes windows bugs

[sudden pike]

my Windows etc is pretty old

havent done a reset in the past 5 or 6 years

[ornate elk]

would be a good idea to setup the PC fresh when you have to switch to Windows 11 in october anyway

[sudden pike]

idk if i will do that tbh
i have lot of old software that is hooked with my Windows

if i redo Windows its all gone

[ornate elk]

hmm why not save the software

and after 5-6 years it would be time for a fresh install if you want the best performance and less bugs

but for now try the sfc command

if it finds something it gets fixed.
if it errors out your system is not fixable

[sudden pike]

it fixed something

i do the command again

[ornate elk]

not needed

[sudden pike]

like my existence

ye now it found nothing 😄

[ornate elk]

let me know the speed now after killing wallpaper engine before of course

[sudden pike]

i did it before

63 secs with wall paper engine
without 73 sec

[ornate elk]

try run again without xD

thats not possible

[sudden pike]

HEYYY its faster!

like in general faster

30 secs

without hires

[ornate elk]

okay good

[ornate elk]

1 minute

and crashes with oom if tiled vae is not enabled or freezes the PC xD

[sudden pike]

oh wait

[sudden pike]

havent done that yet 😄

where do it do that?

[ornate elk]

do what?

[sudden pike]

dis

[sudden pike]

dis?

and then?

[ornate elk]

and then enable only tiled vae when using hires fix

that prevents OOM

[sudden pike]

idk where i find these check boxes

[ornate elk]

have you restarted?

[sudden pike]

yes

[ornate elk]

in txt2img at the bottom

[sudden pike]

ah there lol

[ornate elk]

also get adetailer if you dont have it

[sudden pike]

i have no "Adetailer"

[ornate elk]

yea install it like the other extension before

[sudden pike]

ok captain

[ornate elk]

its for face fix

and get also the booru tag autocomplete extension

most helpful thing when using illustrious models

[sudden pike]

k

[sudden pike]

now SD is dead

[ornate elk]

relaunch

[sudden pike]

i did

[ornate elk]

relaunch again

[sudden pike]

python is weird

after 4 relaunches it works?

where is the logics

[ornate elk]

yep xD

its a bug

[sudden pike]

"bug"

ok 1024 upscaling by 2 is not possible

my whole system lags xD

[ornate elk]

Yea try 832x1216

[sudden pike]

Compilation is in progress. Please wait...
Compilation is in progress. Please wait...
Compilation is in progress. Please wait...
WARNING NMS time limit 2.050s exceeded
Compilation is in progress. Please wait...
``` ![harold](https://cdn.discordapp.com/emojis/973106562493210624.webp?size=128 "harold")

[ornate elk]

Thats okay

Zluda needs to compile stuff

[sudden pike]

ah now its done

[ornate elk]

Should work much faster and better outputs and then set upscale by to 1.5

[ornate elk]

Also if you want even more speed you can try out my TheRock experimental setup guide.
It uses AMDs TheRock project with doesnt need zluda for the webui and is like 20-30% faster then zluda

[sudden pike]

it works again thanks mate 😄

[ornate elk]

Perfect, no problem 🙂

[fair oxide]

Hey everyone, I’m new to this whole local AI model thing. Is there anyone I can keep asking in DMs? I don’t have any particular questions right now, but I’m sure I will sooner or later

[ornate elk]

Hey, for technical questions its best to ask here

Setup guides are on the first link of the pinned messages

[fair oxide]

Is everything updated to the latest verions? I'm starting as well

Thank you very much in advance, i see a lot of useful stuff

[ornate elk]

Yes updated and tested. Only SwarmUi with ZLUDA is currently not working and I need to try to fix it.

Are you on AMD or Nvidia?

[fair oxide]

Nvidia Geforce RTX 4060

8GB VRAM

[ornate elk]

ah with that card i would go with Forge webui

[fair oxide]

Oh yeah? I just downloaded Automatic1111

What's the difference if i can ask?

[ornate elk]

Forge is a bit more updated than auto and has various extensions already included like controlnet. It also has support of Flux and Chroma. (But with 8gb vram you shouldnt use them)
But they boath have the same Ui.
Auto1111 is a solid choice too. SDXL based models like SDXL/Illustrious/Pony will work good with your gpu.

[fair oxide]

ok got it, i was downloading some models from civitai to put in auto1111

and test it

i think i will continue with it and then change it if i feel it doesnt satisfy me

[ornate elk]

sure, also if you did follow the setup guide from the pinned messages your ready.
If not you need to make sure you have --xformers --medvram-sdxl added to the webui-user.bat

[fair oxide]

i think i did it cause i said it was mandatory if you had a 8-11Gb vram

[ornate elk]

yep

also how much RAM do you have?

[fair oxide]

DDR5 16GB

[ornate elk]

ahh then you also need to do an additional step to prevent slow gen and potential system freezes/stutters when generating.

its pretty easy, you have to adjust your Pagefile:
Here its explained:
https://www.tomshardware.com/news/how-to-manage-virtual-memory-pagefile-windows-10,36929.html
Make sure the pagefile is only enabled for the C drive and disabled for any other drive.
Make sure its set to custom for C at 16000min and 24000max.
Then click OK and reboot the PC.

[fair oxide]

ok thank you so much i take a look rn

[ornate elk]

no problem, feel free to ask if any step isnt clear

[fair oxide]

u r the best man thx a lot

here it says it's for windows 10, is it the same for W11?

[ornate elk]

yep same

[fair oxide]

yeah i see

just did it

i need to reboot the pc now

done

So i should be ready to start now

Do you have any video tutorial that i can look at to take the best from auto1111?

I'm willing to study so i can watch also very detalied tutorials

[ornate elk]

Maybe this is a good start:
https://www.youtube.com/watch?v=nJlHJZo66UA

It covers the basics of txt2img, img2img, controlnet a bit and the sliders and buttons

also a quick tip, Hover over the buttoons and words with the mouse to get additional Info of the Ui

[fair oxide]

perfect ill watch it right now

Thank you again

[ornate elk]

no problem, have fun! and for prompting questions feel free to ask in #📝｜prompting-help

[turbid willow]

Having a technical problem, I think? Not sure if it's clasified as technical or something else.

But my negative prompt field greys out and wont let me further edit the text inside anymore. It does this to me every time now. I've restarted diffusion, restarted my computer, and it happens every time. I add in a negative prompt and then shorty after that, it greys out and locks it in, making it unable to edit.

[ornate elk]

Which webui do you use?

Forge or Auto1111?

[turbid willow]

forge.

Everything else works fine, still generating good images, just can't change the negative prompt field.

[ornate elk]

Do you have a flux model selected or the flux button checked at the top left?

Because flux doesnt use negatives

[turbid willow]

No, I have "all" selected, I always have, though maybe I should switch to "XL" since I'm using Xl >_>

[ornate elk]

Yep switch to xl

[turbid willow]

Ha, that fixed it instantly. Thanks.

Ur too smart.

Oh.. And now it greyed out again.

Oooh, interesting... I switched back and forth, it ungreyed out, and when I specficially changed the CFG to 1, it greyed out.

[ornate elk]

Which model?

[turbid willow]

It gets mad at me for using 1 CFG/

Illustrious

[ornate elk]

Cfg 1 is bad to use

Its only for flux schnell

[turbid willow]

But it makes the images look so much better

I tested and found out yesterday, cus I was using like 6 CFG, but if I drop to 1, the image quality improves dramatically.

[ornate elk]

Can you share an example

Didn't tried such low cfg on illustrous before

5-6 is normal

[turbid willow]

1 CFg on left, 6 on right

This is a low resolution

havn't upscaled

[ornate elk]

Ah okay

[turbid willow]

The image on the right looks more cartoony and janky.

[ornate elk]

True but could be also prompt or model related

Any weighting used?

[turbid willow]

Lots of weights.

Lots of loras

[ornate elk]

Ah well

[turbid willow]

So it could be that. XD

[ornate elk]

To high weights can cause issues with illus and pony

[turbid willow]

I don't use anything over 2

[ornate elk]

Yea anything above 1.3 is bad normaly xD

[turbid willow]

Gotcha, good to know XD

[ornate elk]

On 1.6 and above Its like heavily forcing

No matter how bad it will look

[fair oxide]

question, the models i downloas from citivai and use on auto1111 can be used also on forge?

[turbid willow]

Thanks for the help. Sounds like the issue is the low CFG for some reason. At least now I know how to ungrey it before I switch back to low CFG. Thanks!

[ornate elk]

No problem!

[ornate elk]

Yes

[fair oxide]

perfect thx

[ornate elk]

Can be used on any local webui

And you can link them so you dont need to have to copy them over

[turbid willow]

Bwaha.. I found that if I set CFG to 1.1, the box doesn't grey out. So I get the best of both worlds, low CFG and a negative prompt box that still works. Funny.

[ornate elk]

Nice good to know

[fresh crystal]

I just switched to Linux and wanted to get automatic1111 up and running again. I'm on a Radeon RX 6750 XT, which is apparently one card before the Rocm supported cards. Am I screwed or is there some workaround?

Man we just gonna bury my question with 20 scam pics huh?

[ornate elk]

Here is a linux Guide:
https://github.com/CS1o/Stable-Diffusion-Info/issues/42

Its for ReForge on Arch based EndevaourOS

[fresh crystal]

I'm running Mind, which is Ubuntu. Will is still work?

[ornate elk]

idk, it will differe in the commands

but you could look them up or ask chatgpt to convert them for ubuntu

[fresh crystal]

Got it, thanks. Kinda jumped into Linux on a whim and it can be mad confusing.

[hot thorn]

@ornate elk are you still up by any cgance mate?

or is there anybody that would help me with my prompt? I've been going at an ok pace, but then enter the classic: add this, remove this, try this and so on.

i have a mediocre PC, and it takes me about 6 minutes to create a 1080x1080 pic, upscaled by 1.5 at 30 steps + 10 steps upscaler, along with 10 step adetalier, so I often leave my pc for the night to make some picutres.
(I use dynamic prompts to choose random characters and few other things)
I want to make my prompt properly include everything I want in it and make a great result majority of the time, instead of deleting 30-40% of them, with 10-20% of them being great and the rest being just ok

[fair oxide]

Do someone know why i ask to auto1111 to give me one subject and it always gives me twins? it's annoying af

[vocal burrow]

You re not using the correct base résolution output.
Sd1.5 -> 512x512
Sd2.1 -> 768x768
Sdxl/sd3/etc -> 1024x1024

[ornate elk]

Hey, share an example output with prompt in #📝｜prompting-help

[fair oxide]

Thank you very much I'll try again today

[hot thorn]

i was asleep ;p, your message was at 6:30 my time. and my prompt and pic couldn't really be shared on discord server, we'd need to get into dms

[fair oxide]

it worked thank you very much

[ornate elk]

Alright np

[fair oxide]

but now i have another problem: i tried to generate a picture and i recived this message: OutOfMemoryError: CUDA out of memory. Tried to allocate 2.39 GiB. GPU 0 has a total capacty of 8.00 GiB of which 2.29 GiB is free. Of the allocated memory 3.85 GiB is allocated by PyTorch, and 751.71 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

i used epicrealismXL_vxviiCrystalclear.safetensors [3267fd6443]

and <lora:epiCRealismXL-KiSSEnhancer_Lora

any solution?

I have --xformers --medvram-sdxl --no-half-vae added to the webui-user.bat so i don't know what else to do

[ornate elk]

Which settings did you used?
Resolution, hires fix?

[nimble summit]

I am suddenly getting CUDA out of memory errors, just trying to load a checkpoint.

I have an NVIDIA GeForce RTX 3060 with 12gb Ram. I never got OOM errors like this before.

[hot thorn]

could u help me out in DMs tho?

[south mulch]

Anyone need help with flux or sdxl loras?

[ornate elk]

Can you show the settings you used?

[nimble summit]

nvm. I seem to have fixed it by restarting the computer. There must be a memory leak in the software someplace.

[ornate elk]

Make sure you have --xformers in the webui-user.bat

[fair oxide]

It worked until Steps 28 but I received that message at steps 30

[ornate elk]

Ah the issue here are the hires fix settings

You have to set hires steps to 10 and then upscale by 2

[fair oxide]

ok i'll try again, thank you so much man

[ornate elk]

np, there is also an extension which can prevent out of vram errors

[fair oxide]

oh yeah? how is it called?

[ornate elk]

Tiled VAE

[fair oxide]

ill take a look, really appreciate it

[ornate elk]

np, if you install it then simply only enable Tiled VAE at the bottom of txt2img and it will try to prevent the oom, when using hires fix

[main cloak]

I was Using it to learn pronting and play with diffetent Webui´s, but after having issues trying to install ComfyUI everithing went to s_it. Errors every time I try to intall something. I try uninstalling, re-installinmg everything but it doesnt work. It seems that the program things the packages are already downloaded. It says downloading for a couple of seconds only, and then says; installing, but gives this arror.

[ornate elk]

Hey try open up a cmd and run
Pip cache purge

[main cloak]

Stability Matrix is installed in portable mode. Does that makes eny difference? because Pip cache purge is not recognized by cmd

[ornate elk]

Ah okay, I wouldn't recommend using stability matrix but portable mode should be fine.
Pip cache purge doesn't work as you dont have any python installed then.

So you have to clear the stability matrix temp and cache data manualy

[main cloak]

Thanks. I´ll try that. I use it just because models take so much storage. With comfy, forge and fooocus its like 200gb (maybe more) if I need to use them separately. And it`s a chore copy everithing 2 or 3 times.

[ornate elk]

Ah okay but you dont need to copy them.
Every webui supports custom file paths so my webuis all use the same model folder too.
But its not as plug and play like matrix.

[main cloak]

mmm I didn`t know that

[onyx mist]

Is there an example of using SV4D 1.0 or 2.0 to generate a NeRF?

[brave cedar]

I am getting this error when trying to use Web UI Forge Rocm with a RX 9070 XT. I plugged AMD_SERIALIZE_KERNEL=3 into the terminal and rebooted but the problem persists. Any solution?

I also tried TORCH_USE_HIP_DSA in the terminal but the command was not found

[ornate elk]

Hey are you on Linux or windows?

[brave cedar]

Linux

I'm Stable diffusion through stability matrix

[ornate elk]

If its an arch based one then here is a setup guide:
https://github.com/CS1o/Stable-Diffusion-Info/issues/42

[brave cedar]

Its an Ubuntu based distro

[ornate elk]

Ah I wouldn't recommend using stability matrix with AMD. Better set it up manually

[brave cedar]

Oh okay

[ornate elk]

The User on my Link also had a 9070xt, maybe the commands there can help if you know how to use them on Ubuntu

I'm off for now its very late here.

[sleek escarp]

Can I get some help? I've been using stable diffusion for a long time, but earlier this week I got blue-screened after a few generations and now that seems to be the norm. I can't get more than a handful of images before my computer locks up and crashes, where as before I could leave it almost indefinitely and produce around 1500 images a day. using a 1080ti

[ornate elk]

Hey, whats in your webui-user.bat?

And how much RAM do you have?

[sleek escarp]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=

@REM Uncomment following code to reference an existing A1111 checkout.
@REM set A1111_HOME=Your A1111 checkout dir
@REM
@REM set VENV_DIR=%A1111_HOME%\venv
@REM set COMMANDLINE_ARGS=%COMMANDLINE_ARGS% ^
@REM --ckpt-dir %A1111_HOME%\models\Stable-diffusion ^
@REM --hypernetwork-dir %A1111_HOME%\models\hypernetworks ^
@REM --embeddings-dir %A1111_HOME%\embeddings ^
@REM --lora-dir %A1111_HOME%\models\Lora

call webui.bat

24 gigs. also nothing about either has changed before vs after it started crashing

[ornate elk]

Ah okay your using forge or reforge

[sleek escarp]

yeah

[ornate elk]

Try add --cuda-stream --xformers to the line set COMMANDLINE_ARGS=

Then save and close.
Delete the venv folder and relaunch forge

[sleek escarp]

ok, i'll try it

[fair oxide]

can i run sd with rx6600xt 8gb? i thought it's not possible but i read that vulkan is supported

[dense topaz]

I'm trying to create an image of a demon and three human people but it keeps making every look like the demon. How can I get the 3 people to not look demonic please lol

DM reply please

[ornate elk]

Yes you can use it with Zluda. All AMD guides are on the first Link of the pinned messages.
Go with Auto1111 with Zluda or Forge with Zluda

[sleek escarp]

so it seems to be working, although it's now taking 12 seconds or so per step, when for the last year its been able to work at 3 seconds per step. any idea on how to get closer to that original speed?

[ornate elk]

At which settings?

Model, resolution, steps, sampler etc

[ornate mortar]

Has anyone managed to get ComfyUI working on a 50 series card?

[ornate elk]

SCAM!

[ornate mortar]

lol link to some random support group almost clicked it but saw the clover ahah

[ornate elk]

They are really quick, glad you saw it

[ornate elk]

Should work without issues

[ornate mortar]

just says cuda error needs sm_120 which pytorch doesnt support yet apparently but Automatic 111 works fine

[ornate elk]

Ah maybe then you may need to manually switch the pytorch version of Comfyui

I forgot to add that to the guides

[ornate mortar]

is that PyTorch 2.7 ive been using 2.10

oh jeez im getting pythorch and python mixed lol

[ornate elk]

Yes 2.7.1 cu128 would be good

[ornate mortar]

thank you got it working

[stray flare]

what is best XL base model for very realistic characters(real person)
i want for Lora training ?

[tired hornet]

Help me out please with providing info that i have to fill up here..
I'd like to add a Stable Image Core or SDXL 1.0 atleast, because cannot manage both atm

Sorry, there was an error from the AI: Not Found

Details: {"id":"9710c3d6bdf3f797ed7e99d631c1ec77","message":"Not Found","name":"not_found"}

I keep getting this kind of messages

[tired hornet]

ive tried all that sh*t and like 20 more other variants but still nothing, please help

[hot thorn]

anybody here able to help me with kohya_ss lora training? im stuck ;<

they ask me for some web3 wallet associated with my account?

and they just removed me from the server

wut?

@frank edge explain please?

^ suspicious much?

[ornate elk]

It seems you got messaged by a scammer

[hot thorn]

gotta ban him no?

[stone swift]

Yo guys, I don't know much about fine tuning, but I had a project idea and made my own dataset of images, I want to fine tune the model with that dataset, but have tried lots of things, and got some versions and libraries mismatch, if someone can please let me go through each steps if you have any exp. Thank you so much

[delicate citrus]

@ornate elk can you help me with forge? I'm tryng to run sd3.5 and flux

I neeed a little help with VAE and Clip. I put it but (maybe wrong?) but dont generate a picture and crush my colab

[ornate elk]

cant help with google colab sry, but for flux you need the ae.safetensor, clip l and the t5 text encoder

[delicate citrus]

it is the same thing from local. Later could I show you how I'm built it on Forge to you take a look if looks ok?

I imagine I'm built something wrong

🙂

[stray flare]

help

it showing avr_loss=nan]

im using LoRA_Easy_Training_Scripts

[ornate elk]

Sry can't help with that

[delicate citrus]

@ornate elk I'm using this repo
https://huggingface.co/calcuis/sd3.5-large-gguf/tree/main

I make this built, but show this error message

[ornate elk]

your missing the sd3.5 vae and clip text encoders i guess

[delicate citrus]

sorry I post the wrong screenshot

@ornate elk some idea?

take like 5 minuts and don't generated nothing

[ornate elk]

thats a clip error. maybe that t5 clip model doesnt work with sd3.5

i have mutlple different ones in my clip folder

[delicate citrus]

do youI take all on this repo, it must be work haha

[ornate elk]

oh true

that should work but its still a clip error

oh dont use clip skip 2

[delicate citrus]

wait 🙂

sorry, I already close it. I will try on Comfy now. Maybe today is the day that I will finaly generate some image on COmfy? lol

about clip skip 2, it don't is used on sd3.5? only 1?

[ornate elk]

yep

only 1

same for flux

and for sdxl

[delicate citrus]

idk about it, thanks!

on sdxl I generated with clip skip 2... I'm new ond models out of sd1.5 xD

[ornate elk]

yea on sdxl it doesnt make a difference. But if a merged model was trained on clip skip 2 the creator normaly writes that to the info box on its civitai page.

[delicate citrus]

I understand, clip skip 1 is the default

how can I download models direct to comfy on my foldes? there are something like civit ai browser + here?

[carmine ibex]

How do i get the res_2s sampler and the beta57 scheduler?

[vocal burrow]

It s a scam don't go there

[restive sparrow]

Anyone ever run into the bug in A1111 where your VRAM gets maxed out and it doesn't flush it? This is during standard generations by the way, not upscaling.

no driver change, no new or weird model, no settings changed

[deft wren]

Hi, i am trying to install zluda with sd next, using the tutorial, i think i did everything right, but i am gettting this error

my gpu is a rx 5700xt btw

[ornate elk]

Can't help with sdnext and can't recommend it. But checkout Forge with Zluda or auto1111 with Zluda on the pinned messages first link

[ornate elk]

Nope, feel free to share some more infos when it happens again

[delicate citrus]

@ornate elk could you share your forge sd3.5 repo with my? I'm tryng this one but ins't working =/

[regal island]

https://cdn.discordapp.com/attachments/1080939894626586685/1401258879551017000/image.png?ex=688f9f87&is=688e4e07&hm=fd247ce2b91e07c8f3be8e9487f147f80bbb3d980ba09c4fdb48899518d0e2a8&

anyone know how to fix this, given I already input my civitai api key to stability matrix?

[ornate elk]

nope sry, the repo looks okay

but here is one for sd3.5 medium:
https://huggingface.co/city96/stable-diffusion-3.5-medium-gguf

[haughty pilot]

Trying to use a model based on NoobAI but all I get is just random noise, tried 2 different ones but it's the same. Other models work fine.
I'm using forge ui.
Any ideas?

[haughty pilot]

same thing happens on automatic1111

[vocal burrow]

Can't help much without seeing your settings. But if I had to guess either you re using a wrong VAE or cfg is set to an absurdly high value.

[haughty pilot]

ohh, I think you might be onto something, the previous models I used said it has VAE integrated in it, maybe this one doesn't?

[vocal burrow]

Check the description on the page you got it from. It should indicate of you need a specific VAE.

[haughty pilot]

the author only said that reference is on this page
https://civitai.com/models/833294/noobai-xl-nai-xl
but im not really sure what he meant, I have never used model based on noobai before

[vocal burrow]

And what settings are you using it with ? Screenshot your web UI page.

(just in case.... Keep it sfw please)

[vocal burrow]

The whole thing please

[haughty pilot]

alright, in a few minutes, switching back to forge now that i confirmed the problem persist on automatic as well

[ornate elk]

automatic doesnt support v-pred models

forge does

[haughty pilot]

the thing was happening on forge hence i gave it a try on automatic

[ornate elk]

feel free to show the settings you used

[haughty pilot]

btw those are Illustrious models, I got confused becuase he linked noobai model in the description

copied the setting from the 2nd image, I'm mainly interested in SilvermoonMix03 but tried all of his silvermoonmix models and it always end up like that

[ornate elk]

is your forge webui updated?

what does the cmd shows?

[haughty pilot]

do I just run update.bat to make sure it's up to date?

[ornate elk]

are you on nvidia then yes

[haughty pilot]

alright, I ran it to make sure, will try again when forge launches

[haughty pilot]

well, it seems it works after updating 😂

what exactly are v-pred models?

[modern bone]

Hello, I am seeking some advice, I have been trying for around 2 months to create a system where I can style transfer from a real photo of a pet into a stylized figurine version, ( examples attached)
GPT-Image-1 does this occasionally decently when provided multiple style references but loses the correct proportions (correct muzzle length, limb length etc) and markings so it ends up looking like a different dog, my end goal would be a system that can reliably create output in the correct style that has an extremely good resemblance to the real dog reference

What I have tried so far

1. Different prompts for GPT-Image-1; however, none capture the likeness to my satisfaction
2. Flux Kontext Max using FluxKontextCreator fusion mode; it completely misunderstands my prompt, despite trying many different methods this was a complete failure
3. Build a flux kontext LoRA; this worked but not terribly well, the dataset I used was also built using GPT-Image-1 so it inherited the issue of incorrect proportions and markings.

What I am trying next

1. Train a non paired image LoRA for flux kontext; my hope with this is that kontext will learn the style only and not pick up the proportion and marking issues from GPT-Image-1 allowing it to create more accurate models
2. Create a HiDream E1.1 LoRA; my understanding is that this model is superior and I am hoping I will get a superior output from it
3. Create an ID LoRA or DreamBooth step where I train that step specifically for the likeness of the dog and stack the ID step with the style step

As I am new to AI I am seeking advice from more veteran users, any pointers for where I can go from here would be greatly appreciated 🙂

here are the photos

[steel cypress]

I only have 4 VRAM, which app do you recommend I install for images?

[haughty pilot]

I have a huge problem after updating forge,
before the update:
I have been using "--cuda-malloc --cuda-stream --xformers --medvram" as commandline
everything was working fine, on 1024x1024 I had around 1.6/1.7 it/s, loras loaded pretty quick, pc was not lagging during generation etc
after the update:
it says --medvram is no longer an option, now there's "gpu weights" in the UI (it sets automatically to ~7100MB), the thing is with that setting my PC just freezes completely when I tried to load any loras, I tried lowering it down but:
at 6000 MB I get around the same it/s as before the update, whenever I try to load a lora it takes much longer than before the update and my PC freezes for either 5-15 minutes at the end of generation or just freezes completely
at 5000 MB it goes down to around 1 it/s, loading any lora takes ages, PC lags A LOT for up to a few minutes at the end of generation (~95%)
values below 5000 just kill any performance, takes ages to generate anything or load any lora so it's not even worth running but it did not freeze/lag

Before the update with --medvram everything worked pretty decent, loading loras didnt take that much time and it definitely didn't freeze my PC
I have rtx 2080/8GB VRAM

Arg --medvram is removed in Forge.
Now memory management is fully automatic and you do not need any command flags.
Please just remove this flag.
In extreme cases, if you want to force previous lowvram/medvram behaviors, please use --always-offload-from-vram
Using cudaMallocAsync backend.
Total VRAM 8192 MB, total RAM 16250 MB
pytorch version: 2.3.1+cu121
xformers version: 0.0.27
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 2080 : cudaMallocAsync
VAE dtype preferences: [torch.float32] -> torch.float32
CUDA Using Stream: True
F:\Stable Diffusion\forge\system\python\lib\site-packages\transformers\utils\hub.py:128: FutureWarning: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead.
  warnings.warn(
Using xformers cross attention
Using xformers attention for VAE
ControlNet preprocessor location: F:\Stable Diffusion\forge\webui\models\ControlNetPreprocessor
[ForgeCouple] - INFO - Loaded Adv. Presets...
2025-08-04 06:50:02,676 - ControlNet - INFO - ControlNet UI callback registered.```

this is my cmd log

[ornate elk]

How much RAM do you have?

[haughty pilot]

16 GB, I know it's not much but I close all apps when I do anything in SD, and everything was working alright till I updated the forge today

[ornate elk]

Ah okay but thats the problem, then you have to increase your Windows Pagefile.

Here is the Guide how to increase it:
https://www.tomshardware.com/news/how-to-manage-virtual-memory-pagefile-windows-10,36929.html
Enable it only for C and make sure its not enabled for any other drive. Set to 16000 Min and 24000 Max
Then apply and reboot the PC.

After that it shouldnt freeze or lag anymore

[haughty pilot]

alright, will do that altho im still confused why simply updating the UI made it impossible to do a single thing
gonna do that, test and let you know if that helped

[haughty pilot]

it actually did help, thanks
I had pagefile before but on automatic/hdd, now i dont know whether I should go for 64 or 128 gb in my new pc 😂

also what does that mean? F:\Stable Diffusion\forge\system\python\lib\site-packages\transformers\utils\hub.py:128: FutureWarning: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead. warnings.warn(
do I need to do anything?

[ornate elk]

Pagefile on HDD, slows down the whole PC xD,
64Gb RAM is perfect on a new PC.

[ornate elk]

That warning can be ignored

[orchid geode]

how do i install this

it gives me the error RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this chec when i try and launch

@ornate elk can u help

[deft egret]

im gonna try this again and yeah i already deleted stable diffusion and the a1111 model so i reinstall it so anyways i tried to install stable diffusion which is AUTOMATIC1111/stable-diffusion-webui but it has a problem and it doesnt work. i was following the tutorial on youtube but its not working

so it says stderr: The system cannot find the path specified

and here is another one

and yeah i use the stable diffusion A1111 Model to test it out

sorry if i ask this again so its a little bit different

[ornate elk]

What's your GPU?

[orchid geode]

i already fixed it

why are my generations so low quality btw

[ornate elk]

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

[orchid geode]

rtx 4070

also wheres webui-user

[ornate elk]

The file you use to launch the webui

Webui-user.bat

You need to right click it and add --xformers to the line set COMMANDLINE_ARGS=

then save and relaunch

[ornate elk]

Hey, you should try to delete the venv and relaunch.
Then show the full cmd log if you get errors again

[orchid geode]

u mean run.bat

[ornate elk]

That would mean you installed Forge webui

[orchid geode]

tf did i even install atp

[ornate elk]

What's your folder name of the webui?

And did you followed the updated install guides from the pinned messages first link?

[orchid geode]

wym

[orchid geode]

no i just opened the run.bat and it installed it for me

wtf whyd the art suddenly improve

ts ai frying me

[ornate elk]

Okay, you may want to checkout the install guides as they include steps to get the best performance

[orchid geode]

whys it taking long asl to make images now

[ornate elk]

Quality is defined trough the checkpoints (models) you use.
If you use sdxl/Illustrious checkpoints then make sure you set the button to XL at the top left

[orchid geode]

what button

[ornate elk]

Can you screenshot your ui please?

That would help to figure out what you installed

[orchid geode]

ok

[ornate elk]

Ah okay so you installed Automatic1111 webui but as portable mode

Not the recommended way to set it up

I would suggest to follow the complete setup of Auto1111 or Forge from here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[orchid geode]

bruh ok

lol i should ai generate phonk aura

[orchid geode]

why cant i get openpose working

models dont appear

[deft egret]

i tried but it still doesnt work

is there a way to fix this or redownload the proper stable diffusion because i want to use it to make ai generated art and using anything v3 and v4 model

[ornate elk]

Follow the updated install guide from the first link of the pinned messages

Then it should work

[ornate elk]

You need this openpose model for illustrious bases stuff:
https://civitai.com/models/1359846/illustrious-xl-controlnet-openpose

[delicate citrus]

Forge can run GGUF? I'm having error message with T5, do you have another T5 to recomend me?

[chrome tangle]

Can i gtet some help if its okay to ask. I am having alot of grief and trouble with my adetailer. If its okay to ask

Basically, i am looking for help with itl The optimal settings, optimal prompts. I am having alot of issues with discoloured boxes. Wherte the box doesn't match the colours of the image and it ends up with huge colour mismatches. Also at times, it doens't change much. So if its okay to ask, just the best settings/optimal and stuff. I AM SO SORRY FOR BOTHERING

[ornate elk]

Here a t5 gguf:
https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main

[ornate elk]

Hey can you share an example output and the settings?

[chrome tangle]

of course, thank you fo rhelping!

https://gyazo.com/be849cfd5e304d827438880050febd57 https://gyazo.com/f640e6a2be2d5a52a94693e89d891271

Also, i know i am dumb, i was trying out the multi-diffusion stuff and no matter what i did, i got tile isues.

[chrome tangle]

And once again, thank you for trying to help me, again, i am sorry if i come across as the dumb dumb!

[ornate elk]

yea dont use tiled diffusion. only use tiled vae when upscaling anime/semi realism images

[chrome tangle]

Okays,. so i will drop that feature

[ornate elk]

have you try the default settings of adetailer, they work mostly the best, and if you get wrong face colors then only lower the denois in adetailer to 0.2

[chrome tangle]

i will give it a try, thank you!

[chrome tangle]

Mind if i ask another question. In terms of the prompts. DO you double up weith the face and hand prompts. Or just put them in the adetailer?

[ornate elk]

you can either leave them in the main prompt OR add them to adetailer but dont add them doubled

[haughty pilot]

hey @ornate elk, pagefile fixed my pc freezing issues but it looks like I still can't freely use all the loras, the image generated fine without the lora but when im trying to ad hyperrealistic lora the UI just dies at ~97%
"Press any key to continue" pop ups at the end in the console and it just closes after I press anything
That generated image from first screenshot is actually the one generated without lora so whatever im trying to generate with lora doesn't go through

default cpu weights is at around 7000 MB (works fine without the lora), I tried lowering it down since it crashed when I tried using lora but nothing changed, exactly same thing happend on 5800 and 7000

[ornate elk]

enable low vram mode on controlnet

[haughty pilot]

where can I find it?

[ornate elk]

also you should update your extensions

via check for the extensions tab

[haughty pilot]

yeah, will update them, dont have the option

[fossil saffron]

looking for someone who can help me for $ . cuda not running rtx 3060 12gb , if someone can help please dm me...

raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

Stable diffusion model failed to load

[haughty pilot]

is it supposed to take over 20 minutes? lol

[ornate elk]

no, check the cmd

[ornate elk]

hey follow the install instructions of the first link of the pinned messages

[ornate elk]

okay then relaunch

or try add --update-all-extensions to the launch args

[fossil saffron]

sir i did

since 6 days please :/

[haughty pilot]

now it took 0.1 sec but nothing happened 😂

[haughty pilot]

didn't help for some reason

[ornate elk]

ah maybe forge doesnt have that button

will check

nope it doesnt have that button my bad

[haughty pilot]

alright so (same settings): lora with controlnet works fine, no issues, controlnet without lora works fine without any issues but controlnet+lora dies at 95%-100%
I was able to use both at the same time before I updated forge so it has to be possible

[vale smelt]

I was using A100 GPU on google colab for awhile but it seems to not be working correctly anymore. Specifically this GPU won't connect to xformers. It keeps saying there's conflicting versions of torch even when I uninstall them

[haughty pilot]

would changing pagefile from 16-24 to more help?

[violet swift]

WAN 2.2 users, how do you ensure that hair doesn't blur and appear to move during frames, and that eyes don't become distorted?

[ornate elk]

You could try set Max to 32000

[deft egret]

i tried using COMFYUI So How to Generate Images Faster In Seconds Because the generation is so slow and take 5 minutes to generate and it becomes a bad work so i need help generating images faster with good picture without scribbles so the model i am using is anything model

[deft egret]

and another thing on stablediffusion web ui is that i did follow the instructions and i did add these to it

and i follow the instrctions of installing stable diffusion web ui but the installation still dont work 🙁

[deft egret]

well only comfyui and fooocus Works But Stable Diffusion web ai isnt and i really want to use it so i can play the model which is anythingv3 and v4 and v5

[ornate elk]

Something went wrong at the step where you git cloned the webui it seems

If your from Asia or Russia you may also need to use a vpn for that step.

So try to delete the stable-diffusion-webui folder and git clone it again and then edit the webui-user.bat again and relaunch

[haughty pilot]

I did, didn't help, pc is not even lagging, I even limited gpu weights to 4000 MB and the UI still times out/crashes at the end of generation when im trying to use both controlnet and lora, maybe there's a problem somehwere else? as I said, it worked fine before the update 🤷

[ornate elk]

Which lora, controlnet and txt2img settings do you try?

[haughty pilot]

crashes ("press any key to continue" appears in the console) with pretty much any lora I tried, add-detail-xl, hyperrealistic_illustrious, aesthetic_anime etc, it sometimes goes through but rarely, I increased pagefile to 16-32 GB and even tried capping gpu weights at different values

[ornate elk]

Okay try opening up a cmd and run
Pip cache purge
Then delete the venv folder in forge if there is one

Then relaunch

[deft egret]

i tried to get a better pose but the other file is missing and i cant make better poses...

i tried to do a better pose for comfyui and using anything anime model

[deft egret]

its this one

and its bad

[ornate elk]

Can't help with Comfyui, but you could ask in #🧣｜comfy-ui
If you want to get Automatic1111 working then please relaunch it and share the full cmd log

[delicate citrus]

thnks!

@ornate elk what is this quants? (gguf, fp16, etc)

when I choose a quant, all models (clip, vae, checkpoint) must have the same quant?

[prisma moss]

ive gotten myself a 4090, any args that would help it further? It does fine as is

[delicate citrus]

@ornate elk for diffusion models I can choose a quant gguf as for LLM? For example, Q5 is a good choice?

[stark stratus]

Been trying to run SD on my machine for quite a while now, with no success.
I tried multiple tutorials, guides, everything you can imagine.
I'm on Arch Linux, with a Ryzen 5 5500 and an RX 6600, which is obviously lame but still ||supposedly|| compatible hardware, at least for simple things.
I've been trying to run https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu, but it simply does not work. It complains my GPU isn't supported by Torch and says I can skip the check. Even when skipping the check, it says git hasn't been found, even though I cloned it USING git.
Am I doing the right thing? Is there a better approach?

Any help is appreciated.

Thanks for reading.

[haughty pilot]

did that, seems to fixed the problem but now there isn't a venv folder in forge folder, what happened? 😄 I thought it's going to reappear once I relaunch my UI

[ornate elk]

normaly forge uses its prebuild venv folder which has a different name. so you normaly shouldnt have a venv in forge unless you launched the webui-user.bat in the webui folder by accident or edited the environment bat to set skip_venv to 0

[haughty pilot]

yeah I used webui-user.bat at the start, makes sense

btw I wanted to move my forge on ssd, can I just move the whole folder or will that break something?

[ornate elk]

it shouldnt break anything

but dont move it to an external drive

[ornate elk]

hey, here is an Guide for Arch Linux from 3 weeks ago that should work:
https://github.com/CS1o/Stable-Diffusion-Info/issues/42

[lime pivot]

@ornate elk Hello bro, this is more of a PC question than a Stable Diffusion one, but my PC crashes a lot and I think it's because of RAM. The longer I use Stable Diffusion, the more my 'committed memory' increases. My guess is that when it goes past the limit (65.4+ GB), it crashes. Is this normal? If so, what can I do to prevent it? Am I supposed to regularly restart my WebUI? Because so far, that’s the only thing that seems to fix it

[floral yew]

hey uh I don't know where else to put this but all I get is some fuzzy blue square and then a grey image

[ornate elk]

Which webui, which settings, model?

[ornate elk]

How much normal RAM do you have?

[floral yew]

uh I am not very technical I dont even know how I got this far lol

I using these models it would be easier if someone could just call me and guide me through I just wanted to figure this out for dnd portraits 😂

[ornate elk]

sry the the ram question was for Jason Lee

[floral yew]

oh sorry

[ornate elk]

these are all Loras not models. Made for 1.5

[floral yew]

oh 🙁 I dont know what that means but okay

[ornate elk]

np, can you just screenshot the whole ui with the output?

then maybe i can see the issue

[floral yew]

its uh just giving me people when I need like anime atelier art

atleast its not horror anymore...thankfully I found these negative prompts somewhere in the discord

video told me to put the models here but they are Loras so I don't know what that is or where to put it

[ornate elk]

these files excluding the bottom one are all LORAs, not Checkpoints (models).
Loras are small additional files that got trained on a specific subject, style or Character and they go on top of a model. They cant be used on its own.
Loras go into the models/lora folder

you need a checkpoint other than 1-5 ema pruned to use them with. for example this one:
https://civitai.com/models/23900/anylora-checkpoint

whats your GPU btw?

[floral yew]

Nvidia GeForce RTX 3060

[floral yew]

downloading this thank you

[ornate elk]

ah make sure you have --xformers in the wbeui-user.bat for the best performance

[floral yew]

where is that...is it the notepad for the bat? I was already told to put some stuff in there

[lime pivot]

@ornate elk are you asking about this? i thinks its 64 gb

[floral yew]

also it already looks allot better

[ornate elk]

yea right click the webui-user.bat and edit it

then screenshot it

[floral yew]

is it anywhere in there?

[ornate elk]

can you screenshot it?

the whole thing

[ornate elk]

ah okay already looks good but remove this:

then save

[floral yew]

oh so he already gave okay

yeah I followed a youtube video

[ornate elk]

ah ok

i would suggest you look into illustrious models and loras if you want to got for high quality Anime images

[ornate elk]

yep looks good.
can you check your windows pagefile settings and screenshot them?

[lime pivot]

im not sure, but is this what you mean?

[ornate elk]

Yep looks good too

Open up a cmd as Admin and run:
pip cache purge
And then
SFC /scannow

[lime pivot]

@ornate elk thanks bro hopefully it works now. btw is 445 files removed a lot xD is that bc of the images i saved from sd lol

[ornate elk]

Nope thats just cached stuff from webui installs. No images.

If you get crashes again you should check your Motherboards BIOS Version and mostly consider an BIOS update if yours is older than 1 year.
Also which driver version are you using?

[lime pivot]

is this right?

[ornate elk]

Yes, its from july, check for updates in the nvidia APP.

[lime pivot]

@ornate elk thank you bro!

[ornate elk]

No problem 🙂

[mental vector]

Are you running on CPU exclusively or GPU, and also, what framework are you using? Comfy UI, or Auto1111? I can help you because I've been able to mimic those images based on the parameters that are shared in CivitAI. I'm running on an old non-RTX 8GB GPU.

@floral yew I'd love to help via DM so I don't spam the channel. I've been using ComfyUI but the same concepts should apply to your chosen framework. There are a few things that I learned the past week and you should be able to mimic and modify those images at your whim.

[lavish hull]

I am trying to download SD with my AMD 7000 series 7, but this error pops up

[mental vector]

Seems like the version of Torch you have installed does not know how to reach the DirectML libraries. I don't use Windows, but doing a quick run-through on the error, it seems you may be able to install it: "pip install torch-directml". I would suggest if you're able to setup a virtual python enviroment in Windows, to do so.

[vocal burrow]

You probably don't want to use "²stable-diffusion-webui-amdgpu" or at the very least run it with --use-zluda instead of --use-directml (slow and old). Or better yet, run forge with Zluda, check pinned guide for some guide.

[mental vector]

I'm not a Windows expert, is Zluda a new middleware layer that talks to the GPU on Windows?

[vocal burrow]

More of a "translation layer", stable diffusion (and pretty much any AI thingies) use CUDA "language" (actually an API not a language but let's not get lost). You can use DirectML instead but as mentioned, it's slow. Or you coud still use CUDA if combined with ZLUDA. ZLUDA will "translate" CUDA instructions to AMD compatible ones.

[mental vector]

Feel free to talk to me in CS speak 😸 - I see, I had no idea something like this was even out in the wild. I have friends on both sides of the equation, ROCm purists, and CUDA purists, but I wanted to run CUDA native stuff on my old RX 580 12GB which should be able to handle some of the lower-req models in Comfy UI. AMD has better pricing-per-comparable performance against Nvidia. This is great, I'm gonna try it in my Linux box, assuming I can compile it for Linux too.

[vocal burrow]

No need to compile anything, there's libs for pretty much everything in any serious distribs and AI front is mainly done in python. Just git clone, run the script and it will create a venv with everything for you.

[mental vector]

Sweet, thanks! I'll try to run Comfy under it and run a few workflows.

[vocal burrow]

but yeah even in linux, cuda + amd = no no. You'll want to try ROCM (or wait for "therock" to be usable).

But that's something I'm not familiar with yet. (I'm switching from Nvidia to AMD in the next weeks)

Cs1o is the AMD guy in here (for now :p)

[mental vector]

Noted, I'll read up on ROCm and try to do a setup. I don't need to do programming with the API, just run the Comfy framework, so I think it will be ok to remain somewhat agnostic.

[vocal burrow]

Comfyui is even easier to setup than forge/a1111/sdnext and others. It should automatically detect pretty much the best settings for your setup. (but harder to use IMO)

[mental vector]

It's a bit hard to use, yes, I went through the OpenArt "Comfy Academy" to get started, but it was a few days of workbook-style videos. A1111 is more straightforward for sure.

Are you able to run FLUX and/or variants in your setup? I'd be curious to know what GPU you're using.

Right now I can run any SD1.5 or SDXL, as long as I use --lowvram and upscalers (which IMO are quite fantastic), but I can't even begin to imagine running FLUX.

[vocal burrow]

RTX2070S, and yes but it's really slow as I only have 8gb. Not worth it for my workflow even though it's just a hobby.

[mental vector]

Got it. I'm still torn between getting an Nvidia 5060 ti, or the AMD equivalent. I've read in some sites that FLUX is optimized for CUDA, and although it runs on ROCm, it may have some reduced performance. Not sure how much of that is true.

[vocal burrow]

Boring answer but it all depends of your budget and what you're gonna do with your GPU. Personally I find it really hard to recommend a 5060 (even ti). If you go with green team, I'd say at least pick the 16gb version and not the 8gb one. What's the state of the second hand market where you live ? Are you gonna use it only for AI ? How long ? Maybe renting a gpu is a better option ? Using online generation services ? There's also the "AMD is gonna get better" answer and it is true, they're getting better with every update but it's taking time (cf "therock" exerimental setup).

But yes as of now if you're looking at it only from an AI workload perspective. As long as you get a 16gb+ gpu, the bank for the bucks is in the favor of nvidia. They're the ones behind CUDA after all which is dominating the research field.

[leaden rivet]

Hello, when reading the above discussion, my question seems way too basic 😄 : how to restart a sleeping space via API (gradio), or any other way than logging-in via a browser and pressing the "restart space" button? Many thanks

[mental vector]

So, most of that is easy to answer. I'm going to use for stable difussion (not the training part). I have a 3060 that I use for gaming, and I only play one or two games. Secondhand market here really ssskkksss, with 3000 and 4000 series going through the roof on eBay. Amazon has them at even more ridiculous prices.

and yes, the 16GB is a minimum requirement for me.

[vocal burrow]

I would not qualify API stuff as easy question, especially at 3:00 am. I'm no API expert either but what do you mean by "sleeping space" ?

[mental vector]

I'm also going to try stable audio difussion models with it.

[leaden rivet]

e.g. this space is sleeping right now: https://huggingface.co/spaces/somethingbyai/mnist_counter

[vocal burrow]

ghjkh i woke it up sry

[leaden rivet]

no problem, i just took a random sleeping, public space to show you.

but actually I am interested in waking a private space in any way, except the press-the-button-way. Any Idea?

[vocal burrow]

Ok so if you're not too kin on gaming, nvidia gpu 16gb+ vram is probably the best bet. You can find some "Procyon" benchmarks to see which is the best within your budget or check the sdnext benchmarks database https://vladmandic.github.io/sd-extension-system-info/pages/benchmark.html (but beware that those are community scores so you'll have to factor in the model they use, their environment, etc)

[mental vector]

Understood, at this rate anything is better (i'm using a Quadro M5000 8GB).

[vocal burrow]

tbh I thought that requesting anything would wake it up automatically.

I think you have to pay their premium solution or whatever it's named for it to never go to sleep.

[mental vector]

@vocal burrow Thank you for your help. I'm new to this world of "AI", but it's already addictive. Have a good evening.

[bitter pier]

Hello

How are you doing?

I've something to ask for your help.

May I DM you?

[vocal burrow]

Probably better if you ask in here, more people might be able to help you than in DM.

[bitter pier]

Not something technical but something critical for the life. 🙂

[vocal burrow]

How so ? What's the question ?

[bitter pier]

I recently moved from the US to Portugal, and my US-based company insists that remote developers must live in the US. I'm looking for someone truly exceptional, with a heart of gold, who can help me navigate this situation. In gratitude, I'm happy to share a portion of my income with them.

[lavish hull]

@mental vector @vocal burrow I did what yall both said

But I'm getting this error now

[mental vector]

This is what catches my eye:

At first glance, what that says to me is that ROCm isn't initializing. Can you verify that ROCm is installed? Should be in C:\Program Files\AMD\ROCm

Also, did you install the actual Zluda middleware? https://github.com/vosen/ZLUDA/releases/tag/v3

According to the AI bot (and it correlates with other sites), there is a bit of involvement in the install process for Zluda as the author still states that is "experimental".

Without having an AMD GPU available, I'm unable to reproduce the issue :(, however, what's your running env including GPU model?

I believe @ornate elk is the local AMD Guru; I found some of his excelent guides here: https://github.com/CS1o/Stable-Diffusion-Info/wiki/ Warning: it's a tiny bit technical.

[ornate elk]

Hey what's your GPU?

[lavish hull]

amd ryzen 7 7000 series

[vocal burrow]

That s the CPU not GPU.

[lavish hull]

https://tenor.com/view/drake-drake-meme-rapper-embarrassing-drake-embarrassing-gif-14154814324941847170

[ornate elk]

And your GPU?

[tall pier]

@pure tusk Hi is there is any sort of guide/support to download SD on macOS

I tried to do it with 1 method but an error message shows up

[ornate elk]

have you tried the Mac OS setup script ?

from the pinned messages

[tall pier]

I followed a guide from github

[ornate elk]

which one

[tall pier]

wait let me send you the link

am I allowed to send it?

https://github.com/lllyasviel/stable-diffusion-webui-forge/issues/314#issuecomment-1974804889

this part specifically

the error message in question

[ornate elk]

ah okay yea that guide is not recommended, a Mac OS User made an easy installer script for Auto1111 and Forge.
You can find it here. Read the instructions before starting.
https://github.com/viking1304/a1111-setup

you can define that it should setup forge by using -o as launch arg

[tall pier]

let me try doing it by this guide

I am very rookie with tech so my knowledge is very limited

I am mainly shifting away from web based softwares because of limited points and censorship

[ornate elk]

yea thats a good idea