🤝｜tech-support

[gusty tide]

im also tryna use zuda

i got it working just this workflow wont import

alr

its expired

I joined

[eternal elbow]

this is a common scam "support desk" tactic, please do not interact with the link the user sent you and leave any server that you've joined from them

[crystal nacelle]

my controlnet causes a connection error when i try to use it. can someone please help? heres the cmd log

[ornate elk]

Delete the DWPose extension.
Its outdated.
Controlnet has the latest version included by default

[crystal nacelle]

okay, deleted. let me try again now thanks

im running dw open pose full and it aint generating the iconic colorful skeleton. this is what my UI looks like

and cmd looks like this

maybe my model should be on none?

[ornate elk]

What's your base model?

Also make sure the extensions are updated

And provide a full cmd log

[junior aspen]

@ornate elk can you please help me

I have the same issues as him

Is ComfyUI the best thing

[ornate elk]

What's the issue exactly?

[junior aspen]

The packages don’t download

I moved it to the c drive

The zuda version of comfyui

But the guy said id have to clean the files up or something and install everything by hand but i dont know what he meant by clean up the files

[ornate elk]

Which install guide did you followed?

[junior aspen]

Comfyui is working fine

Just the diffusers

Won’t install

[ornate elk]

Any specific workflow that triggers the diffusers?

Or why are they needed now?

[junior aspen]

https://github.com/patientx/comfyui-zluda

[junior aspen]

It’s from a YouTube video they told me to come here

Apparently it was because it wasn’t in my root of c drive

But I’m not sure how to fix now. I moved the folder but same issue

It’s when using that extension thing to download the missing extension node things

[ornate elk]

Which YouTube video?

[junior aspen]

https://youtube.com/watch?v=MbQv8zoNEfY

[ornate elk]

Which guy said that?

[junior aspen]

One of the mods from his discord

[ornate elk]

Ah okay

I guess he meant a reinstall. But idk if that workflow works with Comfyui-zluda.
Or if the diffusers are supported.

[ornate elk]

Also this repo is not mine. So best is to open up an issue there.

[mighty scaffold]

Hello @Team,

I’m currently working on generating a Toy Starter Pack-style image using the following API:
"https://api.stability.ai/v2beta/stable-image/generate/sd3"

However, I’m not getting results that align with the attached reference image. I’ve included the prompt in the appropriate section and tested with various "seed" values, but the output still doesn’t meet expectations.

Could you please advise:

1. If this is the correct API to use for achieving this specific style?
2. If there are particular parameters or configurations I should be adjusting to improve the results?

Your guidance would be greatly appreciated.

[opal nacelle]

Just starting out on what I assume is going to be white the long journey. Does anyone have any recommendations for install guides / material for getting started?

[ornate elk]

Hey, sure just check the first link of the pinned messages of this channel for a list of Guides

What's your GPU?

[junior aspen]

Fr?

Stability matrix good?

[potent sun]

is it possible to set up controlnet with flux dill in comfyUI?

[lucid yoke]

I'd avoid it, i recommend following the guide in the pinned messages

For any sd stuff

[junior aspen]

Just comfy ui?

[lusty moss]

im following the installation guide on automatic1111's github and following every step but i still get errors saying invalid syntax

[ornate elk]

Not for AMD

[ornate elk]

Hey, can you show a full cmd log?

[lusty moss]

oh cmd, i thought python thats what it said atleast

but yeah, is this fine?

[ornate elk]

You need to have Git 64bit installed too

[lusty moss]

o i'll do that

[lusty moss]

oh turns out i already had the 64bit one installed, the latest from 38 days ago

[ornate elk]

Then uninstall python and install python 3.10.11 64bit

Also checkout the first link of the pinned messages for a good install guide

[lusty moss]

will do!

[gusty tide]

@ornate elk I re installed and same problem

and why isnt it using my gpu

[lucid yoke]

hmm why do you think its not?

Total VRAM 12461 MB, total RAM 31865 MB
2025-04-24T07:26:23.093609 - pytorch version: 2.3.0+cu118
2025-04-24T07:26:23.093609 - Set vram state to: NORMAL_VRAM
2025-04-24T07:26:23.093609 - Device: cuda:0 AMD Radeon(TM) Graphics [ZLUDA] : native```

[ornate elk]

Radeon TM Graphics is the igpu

[ornate elk]

You have to disable the Radeon TM Graphics in the Device Manager

[lucid yoke]

ahhh mb i saw the 12gb so thats why mb

[slender crescent]

it actually worked!
||for context: "little girl" in the image descriptions are pictures of myself; AI models often detect me to be a woman and also I am 20 years old||

But yeah, ROCm was a success

[red cobalt]

hm weird i just found this place by searching stable diffusion discord

to try and get help

and i got help thx to u lol

btw should i make the sampling steps like 30 for better results

im normally using 20

or would that fuck it up

[ornate elk]

yea 30 is the standard

[red cobalt]

oh its 20 default on this lmao

so i just used that

funny

[ornate elk]

yea i mean what most people use is 30

[red cobalt]

ah

[ornate elk]

default webui values are not the best

[red cobalt]

and what will it take like 20 30% longer thats rly okay for better images

i usually set cfg at like 7.5 or 8

[ornate elk]

should be a few seconds longer

[red cobalt]

do u know if comfy ui can talk like chatgpt

ah wow

should i make it 50 or 100 then lol

[ornate elk]

nope

[red cobalt]

btw is it possible to save this

[ornate elk]

comfyui do has some llm support

but its not like chatgpt

[red cobalt]

ah

[ornate elk]

there are better tools out there for that

like GPT4all or lm-studio

[red cobalt]

bc chatgpt running out of limit after i make it write one script for roblox (im trying to see how much i can hack roblox with chatgpt)

and it turns out upload counts in the limit aswell

i just uploaded 1 txt file

and it ran out

wait this is weird

it just made 2 of the same images

WEIRD

it made 2 of both of the images

one is like 500kbs

and one is 4mbs

@ornate elk would makind 1 2048x2048 image or making a 1024x1024 image and 2xing it on hires fixes be faster

okay actually def not 2048 it started making freaky shit

[ornate elk]

thats normal, as a setting in auto1111 does this when the image is to large

[ornate elk]

no model is trained on 2048 xD

[red cobalt]

ah

okay lmao

[ornate elk]

so it will start to make duplicates and glitches

[red cobalt]

ill just use hires fixes ig right

it made like 7 fingers 2 faces on 1 torso

lol

how many hires steps tho...

20 should be enough hopefully 😄

[ornate elk]

10

is enough

[red cobalt]

fuck

too late

is there a way to force stop

interrupt doesnt seem to work

i forgot

ah also

is there a way to save this

there isnt right :/

[ornate elk]

tripple click interrupt button

[red cobalt]

didnt seem to do anything yet

[ornate elk]

and set denois to 0.5

and upscaler to Resrgan4x+ or anime 6b

[red cobalt]

well u cant save right

tho?

[red cobalt]

done

[red cobalt]

done

i rly need to save bruh is it possible :/

its too much work everytime i open it

ig its the downsides of a1111

[ornate elk]

in the settings

[red cobalt]

ah

do i click apply rn

okay

it worked i think

ty

time to test it out

hopefully i dont get 10 fucking fingers now

on the images

[red cobalt]

are they both the same quality

bc one of em is 4mbs and one is 300kbs

wait i cant understand

they both seem the same

[ornate elk]

?

you mean the images?

one is a .png and one is .jpg (low quality)

you can disable that in the settings too

[azure magnet]

How to fix it: An outdated version(<1.16.9) of the comfyui-frontend-package is installed. It is not compatible with the current version of the Impact Pack.

[granite bough]

RuntimeError: Could not allocate tensor with 268791120 bytes. There is not enough GPU video memory available!

I get this error when I try to use hires or generate a 1232 x 1504 pic

[ornate elk]

whats your gpu?

[granite bough]

rx6950

[ornate elk]

can you screenshot the tst2img settings like resolution steps etc

[ornate elk]

do you use Zluda or DirectML as backend? and which model do you tried`?

[granite bough]

directML

[granite bough]

it happens with all models

[ornate elk]

ohh rip

then you definitly should switch to Zluda, its 2-4 times faster and uses less vram.

with that you wont get these vram problems, and you then can easily upscale and gen at larger resolutions

Follow the Automatic1111 with Zluda Guide from here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

And feel free to ask if any step isnt clear

[granite bough]

Ok ty

[ornate elk]

np

[lyric vector]

Question I'm trying out to train Illustious Lora on Kohya_ss but my 4090 is flling up on VRAM i'm currently training with 2 batch size and 1024 x 1024 size and Bf16

[prisma moss]

well yes, its supposed to fill up innit

[gusty tide]

Whats the best setup for amd users? Just running comfyui straight up or is there any good programs that package other useful things along with it?

[ornate elk]

whats your GPU?
For easy of use Auto1111 or Forge webui is the best
There is also SwarmUI which uses comfyui as backend but with a better UI on top.

[gusty tide]

7900 xtx

I also just ordered a 3090 for my server pc

[crystal nacelle]

okay. i use SD1.5 base models. in this instance it was realcartoonanime. https://civitai.com/models/96629/realcartoon-anime

here is my web user bat thingy:

i got my control pose extensions from https://github.com/Mikubill/sd-webui-controlnet/wiki/Model-download

[ornate elk]

i see no issue here

what exactly doesnt work with controlnet?

[crystal nacelle]

wait, no control net is working in txt2img now. its working fine.
i cant preview the control net, it only comes up o n the final render.
but img2img still fried and not working

this is what i got when i tried to run img2img. eventually i just left it for txt2img instead

[ornate elk]

looks normal

can you screenshot the whole img2img tab?

[crystal nacelle]

okay, let me refresh so that its a clean console + tab

[ornate elk]

hmm looks all normal

you could try a different browser, as opera vpn makes issues and some adblocking extensions too.

[crystal nacelle]

okay, ill play around with my brower, VPN and antivirus thanks!

[crystal nacelle]

can you help me with getting SDXL to run in my a1111? i downloaded the model from https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/sd_xl_base_1.0.safetensors

[crystal nacelle]

however when i try to switch to the sd_xl_base model it loads for abit then goes back to my previous sd1.5 model instead. the cmd shows ALOT like pages of stuff.

but it ends with the "object no attribute is _sdxl line:

[ornate elk]

why is there animatediff mentioned in the error?

do you tried to use it?

[ornate elk]

how much RAM do you have?

[crystal nacelle]

16GB 😭

I remember installing animatediff awhile back, but never used it coz of long loading times. I can try removing it?

[ornate elk]

Yes remove it from the extensions folder

And have you edited your windoes Pagefile?

Thats needed if you only have 16gb of RAM

[frail bobcat]

All of the sudden I get 404's when trying to connect to the ... well here's the message

When restarting I get the same.

I have no idea what is going on.

Wait, I do have an idea. In the web.user.bat it says it can't find the checkpoint and its falling back to another checkpoint. But I guess it can't find the checkpoints anymore??

What is this link, they're asking for a wallet??

I think someone redirected me to a bitcoin helpdesk/support page??

they deleted the link right after I asked the wallet question.

[ornate elk]

Thats a scam

Dont join their discord or provide any personal information

[frail bobcat]

Yeah, figured that out.

Yet my problem is real.

[frail bobcat]

Hi, does anyone know what to do when all of the sudden stable diffusion can't find the models I'm using even though they are still in the folders?

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker

Also disable browser side vpns like opera

[frail bobcat]

would that change all of the sudden, can it suddenly be blacklisted? I was using it before without a problem.

also, using two browsers and on both its not working

[ornate elk]

What does the cmd shows?

And whats inside your webui-user.bat

[fierce mural]

Hello everyone. I'm a complete noob with AI and used to use Fooocus but I feel like it's starting to be a little outdated. I like the idea of running stuff on my computer and loved the user interface of Foocus, so I went and downloaded The Forge Web UI here : https://github.com/lllyasviel/stable-diffusion-webui-forge/tree/sd35

There seem to be A LOT of new stuff and I wondered if anyone could lend me a hand

[frail bobcat]

Can I send you a pm with a screenshot?

nvm, here it is

[vocal burrow]

I d suggest looking at tutorials from stable art diffusion for instance https://stable-diffusion-art.com/beginners-guide/ Most buttons in sdwebui should have a infobox if you hover them with your mouse, explaining what they do.

[fierce mural]

Thank you, the website seems to be clearly very outdated though

I have more specific questions so I'm just going to go ahead

[vocal burrow]

Forge is based upon A1111, the base of their UI did not change for quite a while by now. There s gonna be a few changes here and there but you should be able to understand most of it by looking at the tutorials.

if you have doubt you can reach here for more information.

[fierce mural]

Looking for a checkpoint recommandation, I'm mostly making realistic(ish) photographs of people

[ornate elk]

Looks normal, but remove --disable-safe-unpickle from the webui-user.bat. Thats a security risk

[frail bobcat]

ok, is there a way to recovery the prompt history? I've been using a lot of prompts and crossed them out when not using a prompt. The complete prompt might have grown to big. Is that possible?

[vocal burrow]

by default, prompt and settings should be embedded in pictures's metadatas.

Also it s not possible for a prompt to be too long anymore (for some years by now, it was a limitation at the beginning of sd1.5 days)

But there s a point of disminishing return when it comes to your prompt's length

no need to copy paste the whole mobydick book in here. it will only dillute the strength of each token/word.

[frail bobcat]

Solved it. I'm using https://github.com/Physton/sd-webui-prompt-all-in-one and my txt2img is huge (23mb), there was also a lock file this extention's storage folder. I removed that and the big history file and that solved it.

[vocal burrow]

23mb of text for a prompt ?!?

Did I get something wrong ?

[frail bobcat]

yeah, I cross out prompts I don't use but leave it in the field.

[vocal burrow]

Ok that s for multiple prompts then. It s still quite big tho.

[frail bobcat]

yes, need to organize better

but yeah, I think it was the lock file that caused the problem, and my huge prompt to cause the lock file.

Thanks for the assistance!

[hexed spruce]

Hello.
I'm using https://www.stablediffusionai.ai/ with a paid plan. Unfortunately, I can't generate anything; I keep getting the message "High traffic, you're in queue." For several hours now...

[vocal burrow]

This website is not the official one. Your best chance would be to reach out to them directly..... But from what I remember about it... It s one of the sketchy one...

There are good third party services out there. But I wouldn't recommend this one.

In theory they offer a 3 day refund

"MIRA MUSE LLC" is not stabilityai if you needed any proof.

They used to go by blackmagic.cc beforehand

Hum no. I was thinking about stablediffusionweb.com.

Still... There are many scamy third party websites that are basically paying for one official access to stability s online api and then reselling it's access by forwarding requests to it

Which is obviously against TOS

So I guess that their services go down whenever their official account get terminated.

Hope it helps....

[ornate elk]

with a 7900xtx you can use any webui with Zluda, but i prefer Auto1111 or Forge over Comfyui. Im using Comfyui only for testing of new models

your two 3090s will be really good for any ai tasks

[worthy scroll]

Have you ever found a way to convert a DiscoDiffusion CLIP model into a diffuser or .ckpt file?

[gusty tide]

Awesome ill take a look at them thanks

[ornate elk]

np, checkout the Zluda Guides:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[gusty tide]

Whats the best way u have found for like making consistent characters and stuff? Ive only really seen workflows made for ComfyUI

[ornate elk]

Hmm, ive used controlnet with IP-Adapter to do this

Its not perfect but works okayish

[tiny nexus]

Compilation is in progress. Please wait...
0%| | 0/8 [00:00<?, ?it/s]Compilation is in progress. Please wait...
Compilation is in progress. Please wait...
Compilation is in progress. Please wait...

Using zluda with a 6700xt, it it normal to see no progress on first image generation?

not sure if this can is an issue:

Repository Not Found for url: https:/linkremoved/huggingface.co/None/resolve/main/config.json.
Please make sure you specified the correct repo_id and repo_type.
If you are trying to access a private or gated repo, make sure you are authenticated. For more details, see https:/linkremoved/huggingface.co/docs/huggingface_hub/authentication
Invalid username or password.

nvm seeing progress now

[ornate elk]

looks good, just wait

[barren wraith]

hi everyone

i haven't touched SD in a long time, since early 2023

i just got around to trying it again

followed this guide https://www.youtube.com/watch?v=RXq5lRSwXqo

downloaded a model and just did a test generation

but it's taking a really long time

[lucid yoke]

are you on AMD?

[barren wraith]

AMD CPU

yes

[lucid yoke]

oh cpu only?

[barren wraith]

NVIDIA GPU

[lucid yoke]

ahh gpu nvidia

hmm

[barren wraith]

it shows 100% CUDA

Last time i used SD it only take some ~20 to 30 seconds for one image

[lucid yoke]

with 6gb its trying to offload to ram wich is a lot slower

hmm

you could try forgeUI (pinned messages for a guide) instead since thats better optimized for SDXL

[barren wraith]

pinned where though

[lucid yoke]

ah i meant here:
but ill link it for you
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

i personally dont use Forge but it gives better performance for sdxl models and A1111 has mostly been outdated

[barren wraith]

even so

i used A1111 back in early 2023

and it didn't take 20 minutes per generation, more like 20 seconds

so i think maybe i did something wrong, not installing something properly maybe

ok thanks for trying to help

i'm just retarded

[lucid yoke]

Things update/change hmm thats my recommendation but otherwise

You could post your full cmd log

[barren wraith]

clicked on webui.bat instead of webui-user.bat

[lucid yoke]

Ahh

Yeah that can happen too

[barren wraith]

[lucid yoke]

If you want a UI with nice built-in tips recommend swarmUI

[barren wraith]

i'll try some other UIs, i'm just most used to A1111

[lucid yoke]

Like help tips on every setting (?) you can hover on

Personally i hate this platform but stability matrix is nice for trying out many platforms but it has a few buts like for swarm it has a bug where it doesn't save any settings

[plain pond]

mr@ornate elk sorry for bothering u, i need some help.. so ijust wanna change the checkpoint and show this error and ican't change.. and then i try delete the venv folder and i relaunch the webui and still same... thanks

[slow cedar]

DONT

DONT CLICK THAT

SCAM

SCAM

Yeah that's right get your sus ass link out of here

Ain't no one here gonna send you a LINK to help you

[crude halo]

click what?

[slow cedar]

They deleted it the moment I said scam

[crude halo]

thanks

[slow cedar]

If anyone pings you with a link saying "get help here" don't click

You'll get all the help you need here

Anyone who actually wants to help won't send you elsewhere via link, they'll just tell you what you need to know

[ornate elk]

Whats your GPU?

[lucid yoke]

Scam

[short nymph]

hello

How can I fix this? CivitAI Browser+: Error: ('Connection aborted.', ConnectionResetError(10054, 'Remote host forcibly terminated existing connection', None, 10054, None))

[red cobalt]

@ornate elk is the link on the a1111 webui your ip

[lucid yoke]

try localhost

[young dagger]

i got this error, do i have to change the code ?

i will restart my pc to see if it changes something

[lucid yoke]

What were you trying to do?

Upscale?

[young dagger]

i just kept the same settings i had, but then this error intervier and i cant generate more images

im trying to generate a picture rn i will tell u

it seems that the problem is resolved

if something doesnt work always reboot i guess

[ornate elk]

Its the local IP of the webui. Its the same for every User

[plain pond]

rtx 2060 max q vram 6gb

[ornate elk]

Is the checkpoint 6gb or larger?
Also how much RAM do you have?

[wanton spire]

Can someone give me an invite to comfy org discord?
when I click on it it shows I am not logged in, when I log in the tab forgets I asked to join that room
Cant find it in discover search

[plain pond]

the model photon is just 2gb and my ram is 16gb

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker

[crystal nacelle]

okay thank you. how much should i change it to? its currently set to 24,85GB for my 16GB ram. id like to use SDXL on my PC

what dyu think of this?

[ornate elk]

Make sure its only enables for the C drive.
Disable it for any other drive.
And then set it to 16000min and 24000max
Then reboot the PC

Or is the E a faster drive than C in your PC?

[crystal nacelle]

E and C are both SSD.
C drive only has 20GB of space free.
E drive has 125GB free, it has stable diffusion installed

[ornate elk]

Are both M2 ssds or SATA?

[crystal nacelle]

probably SATA, they use SATA cables.

[ornate elk]

The C too?

[crystal nacelle]

jea both of them use Sata cables

[ornate elk]

Okay

Then you can leave the Pagefile on E but set it to 16000min and 24000max
Then reboot the PC and try load an xl model again

[crystal nacelle]

okay, only E drive. I will leave C drive as it is

thanks

[crystal nacelle]

no dice, after 5 min of tryign to switch checkpoiint to SDXL it went back to my SD1.5 checkpoint. i guess its time for a ram upgrade

[red cobalt]

@ornate elk do you know something like chatgpt but without restrictions?

[pulsar berry]

Hey

[ornate elk]

localy, there is GPT4all or LM-Studio

[ornate elk]

okay, yep more ram will work xD

[crystal nacelle]

with my img2img control net issue, im seeing "OSError: None is not a local folder and is not a valid model identifier listed on 'https://huggingface.co/models'" on my CMD. could this be why my img2img is inconsistent (sometimes works, sometimes doesnt)?

[ornate elk]

That error is normal and has nothing to do with img2img
Its from auto1111 and can be ignored

[halcyon stump]

Hello, can somebody help me with this error? It appears in cmd after installing python (3.10.6), git and auto1111 and opening the webui-user bat-file.

[vocal burrow]

you didn't check the "add to path" checkbox when installing python

[halcyon stump]

The text after stderr translates to python was not found run without arguments to install from the microsoft store

[halcyon stump]

I just reinstalled it before writing here just to click that checkbox and I manually added it in the system something but it didnt change the error

[vocal burrow]

What do where python, python --version and py --list-paths say ? (run those in cmd)

[halcyon stump]

Where python tells me the folder its in

python --version gives the error again

[vocal burrow]

screenshot

[vocal burrow]

all 3 of them

[halcyon stump]

Thats the error again

[vocal burrow]

place the correct python at the top of the list in your PATH

[vocal burrow]

window's shitty app store shortcut has higher priority than the proper real python according to this screenshot

[halcyon stump]

In here right?

[vocal burrow]

there's surpringly few variables in there .... but yes.

[halcyon stump]

Just move Python 3.10 up to the top?

[vocal burrow]

yes

[vocal burrow]

and uninstall anaconda3's python 3.12 if you don't need it, just to avoid any confusion

[halcyon stump]

I can run pycharm with 3.10 too right?

[vocal burrow]

looks like it

[halcyon stump]

Sorry I gotta go thanks for the help so far!!

[halcyon stump]

Are you still here?
Not sure where I can uninstall it but its not that important, right?
I moved the path with python 3.10 at the end to the top, its still showing the same error...

[ornate elk]

delete the venv folder then relaunch

and go into system control panel an uninstall python 3.12 and anaconda / miniconda if there

[halcyon stump]

webui-macos-env?

[halcyon stump]

Or where can I find that?

[ornate elk]

not relevant, but keep it there

[ornate elk]

if there is one its in the same folder as the webui-user.bat

if not then relaunch it and show me the error

[halcyon stump]

translation of the text

[ornate elk]

uninstall everything related to python and then reinstall Python 3.10.11 64bit
and check "add python to path" at the first install box

[halcyon stump]

I had 3.10.6 until now but that cant be the problem right?

[ornate elk]

no its not, but still uninstall it

3.10.11 is recommended

[halcyon stump]

Its installing right now, thanks for helping me :)

[ornate elk]

no problem 🙂

[halcyon stump]

Should I have done anything before running the bat?

[ornate elk]

nope

open up a cmd again and run
python --version
and then
where python

screenshot that

[ornate elk]

not in the folder, open a normal cmd

also it seems your following an old guide as your folder name is called directml at the end

that also means your on AMD, and should follow my latest Zluda guides.

[ornate elk]

ah its still trying got use the microsoft store python

thats bad

can you go back into the system Environment variables on Path?

[ornate elk]

thats user environment path

the system one should be in the bottom box

[halcyon stump]

I might be leaking my whole pc right now and I wouldnt even notice

[halcyon stump]

Thats the right one, right?

[ornate elk]

yep and there is the python path missing

[ornate elk]

add these two there

[ornate elk]

you can also dm me if you want

[halcyon stump]

If thats okay for you?

[ornate elk]

sure

[crystal nacelle]

i found out my AVG antitrack software might be interupting some of my img2img or control net generations (uninstalling it helped). but i still want some kind of antitracking, how can i add an exception for stable diffusion while installing antitrack software?

[ornate elk]

Install ublock origin as browser extension its better than avg

Then add the webui to the whitelist

[ornate elk]

Also funfact, AVG tracks the User too instead of defending him. Thats the report from the mobile AVG app:
https://reports.exodus-privacy.eu.org/de/reports/com.antivirus/latest/

[late sigil]

anybody have any tips for getting better generations with controlnet?
im finally getting the poses i want but everything seems to be a bit blurry not sure if thats because of my settings or smthn else

[ornate elk]

Hey, do you have an example?

[plain pond]

wow.. now its work thankyou very much

[ornate elk]

No problem 🙂

[late sigil]

okay i think i finally got the resolution up but any idea how i can affect the poses more, no matter what i prompt it seems to be very difficult to get the controlnet to change the pose at all

[vocal burrow]

what settings are you using ? screenshot your whole webui page.

[prisma moss]

Is there a solution to clear up/stop the filling up of C drive during (or after it)the use of stable diffusion? When i check tree size, usually related to python/torch stuff. or pagefile (idk why it wont run with this off i have plenty of ram, nor do i overfill)

[vocal burrow]

there s no reason for the c: to fill up during use apart from pagefile use.

you can always try to clear the temporary install files used by pip with pip cache purge

[lucid yoke]

#🌶｜off-topic

[upbeat moth]

Is anyone else having issues with SDXL image-to-image? I have uploaded a product, with white background, I have put in multiple different prompts for it to change the background of the product in the image and not the product itself, but it just doesn’t listen. I am using the api, image strength 0.8, if someone has had similar issues please let me know

@ornate elk do you know why this is happening? Does SDXL not allow white backgrounds or why wouldn’t it want to change image with white background

[ornate ruin]

I need to factory reset my laptop. Any suggestion how to back up my web.ui? Can I just take the whole SD-A1111 folder into a cloud and download it after the reset?

[ornate elk]

Better save it on a external drive

Model files are very big

[upbeat moth]

Thank you for explicitly ignoring my question, lol

[ornate elk]

sry cant help with api stuff

[upbeat moth]

Sdxl is available on regular web ui so it could concern anyone who is trying to do the image-to-image using a white background, not api only, and even if you don’t it’s better to communicate that

[ornate elk]

which regular webui do you mean?

Stability.ai doesnt offer one. Only API access or Stable Assistant or the Artisan Bot.

[upbeat moth]

Regular accounts, paid or free subscriptions, not api only. You can also use it local installations

https://clipdrop.co/stable-diffusion

[ornate elk]

Clipdrop doesnt belong to Stability.ai anymore.
it got sold las year to another Company called Jasper.
https://www.reuters.com/markets/deals/ai-startup-jasper-acquires-image-generator-clipdrop-stability-ai-2024-02-22/

[upbeat moth]

So SDXL is only available in api?

[ornate elk]

no you can use SDXL models localy on your PC too

its usable localy or by other websites that offer image generation

[upbeat moth]

Copy

[ornate elk]

This is just a Community Server for Stable Diffusion.
So if you use Clipdrop or other web services you have to ask their support

[upbeat moth]

I’m aware that this is Stability’s server and I am not using them lol, you said it doesn’t offer one so I sent you that link ^ wasn’t aware that they had sold it to another company last year 🤷‍♀️

[dreamy lodge]

So after I installed Comfyui and updated my drivers and some of the requirements, my A1111 no longer works and I get this console log.

What parts of my A1111 installation do I need to update or reset?

[ornate elk]

Run a git pull in the webui folder and then delete the venv and the .zluda folder.
Then relaunch the webui-user.bat

[woeful knot]

After a pc format I can't for the life of me manage to make img2img work, it keeps giving me a "PIL.UnidentifiedImageError: cannot identify image file <_io.BytesIO object at" which I can't find a solution online for. I'm on reforge, idk if it's something that happens in other webuis too

[late sigil]

how do you properly setup a vpred model on forgeui?

or can i just use it as a checkpoint model?

[feral breach]

I have this issue, can anyone help me with this?

[gentle sundial]

Hi guys , how do I avoid cloned faces in pony? I tried everything

@ornate elk

[ornate elk]

Hey, you can try using the regional prompter extension or sd-forge-couple

If you use adetailer then lower the denois of it

Adetailer makes the faces look the same

[gentle sundial]

thank you for the suggestion! but these too really just for more detailed prompt and more specific things and to separate the prompt aren't they ? I don't see how these can change faces, what I'm looking for is to create variation in faces for each image, like for sd I use dynamic prompt wildcard to do that but for pony and IL nothing is really working

[late sigil]

anybody have a good guide for setting up comfyui on amd with zluda

?

[idle plank]

Hi guys, I've got this error when I want to use LTX video, someone can help me out ? VAE load device: privateuseone:0, offload device: cpu, dtype: torch.float32
no CLIP/text encoder weights in checkpoint, the text encoder model will not be loaded. (It look like it's deconnect my server whenever I want to run the model but I put t5xxl_fp16 right there so I don't understand --> "C:\Users\Documents\Stable-diff\ComfyUI\models\text_encoders\t5xxl_fp16.safetensors" I'm running it with a 7800xt from AMD

[zenith sail]

Using flux I'm getting this error flux [F421 19:52:52.000000000 dml_util.cc:118] Invalid or unsupported data type Float8_e4m3fn. using the dev fp8 on a amd gpu. anyone know the fix? It works on windows.

[lean temple]

Anyone using a Mac Studio and AUTOMATIC1111?

[ornate elk]

First link of the pinned messages

[ornate elk]

ComfyUI with Zluda or DirectML?

[rigid notch]

Hello hello, I'm using ReForge (I know it's not maintained anymore) and I dl-ed an upscaler model trained to deblur images. But when I look for it in the extra tab, it doesn't show (some other I dl-ed on the same site do appear though). Would you know why ?

[ornate elk]

Hey, what's the upscaler?
Have you put it in models/ESRGAN folder?

[rigid notch]

https://openmodeldb.info/models/2x-Pooh-V4

This one. I put it in the esrgan folder. I see every other models in this folder except this one.
It's a .pth file btw, could it be "censored" by ReForge in an aim to protect me from unsercured files ?

[ornate elk]

I dont think so. .pth should work. But check the cmd if you see any warnings about that

[rigid notch]

I'll do. I'll also try to load it with Comfy to check if it comes from ReForge

I'll let your know !

[gusty reef]

Hi! So I recently got Stable Diffusion 3.5 using Comfy. I followed the tutorial below. I am running into a strange problem where it is taking a very long time for 1 image to generate with a very simple prompt. I was wondering what common causes of this are? One thing I noticed is that the memory usage on my pc spikes to 99% when generating an image.

[lucid yoke]

Not strange so far, what gpu do you have?

[gusty reef]

Nvidia Geforce RTX 4070

[lucid yoke]

Ah 12gb vram, yeah makes sense then

You could use a quantized version or sd3.5 medium ngl

For faster speeds

[gusty reef]

Ahhh I see, so is the reason why its taking so long because I am using turbo and there isn't enough v ram?

[lucid yoke]

Yes, they would like to have 24gb but if you don't have that they will offload to cpu

Dramatically increasing generation time

For me with 16gb (5080) it takes a minute

3.5 medium will fit in 9.9gb vram just fine

[gusty reef]

I see, gotcha. Tyvm for the help! I'll give that a shot!

[gaunt lodge]

I have a question. I use Forge for my stable diffusion specifically for XL. It works with illustrious. I downloaded a checkpoint merge. I can pick it in the checkpoint tab but it does not show up at the top. It is blank. I've had this problem before. What do I do?

[ornate elk]

How much RAM do you have?

[gaunt lodge]

32

[ornate elk]

Okay thats good

And does the cmd shows any error?

When you try to load the model

[gaunt lodge]

Its just blank

[ornate elk]

The cmd?

[gaunt lodge]

I checked the others. They loaded

Ive had this same problem before with another checkpoint merge. I removed it

[ornate elk]

Maybe the file is corrupt

What's the checkpoint you tried?

[gaunt lodge]

interesting. It says "PermissionError: [Errno 13] Permission denied

[gaunt lodge]

Okay.....it works now. I did an update for all the extensions

Which I have not done in months

[ornate elk]

Ah okay xD

[potent sun]

How can I increase my training speed? I still have more than enough ram and power usage to increase perfomance

What is a setting I can change

[lucid yoke]

Vram is all that matters

[potent sun]

I still have avaliable vram

[lucid yoke]

Hmm how much are you sitting on? And what ui are you using to train?

[potent sun]

im using onetrainer

I have a NVIDIA GeForce RTX 4090

[lucid yoke]

Ahh hmm i recommend joining the onetrainer discord but with 24gb vram getting 2/it should be good

Iirc

[potent sun]

I did a train test and now I want to train again but with better values

[potent sun]

im already there im asking in multiple sites to get varied answers

[potent sun]

it?

[lucid yoke]

Iterations

Per sec

[potent sun]

you can also use cpu to compensate right?

[lucid yoke]

If you don't mind it being slow yeah

[potent sun]

like, I just want to leave my pc training this exclusively and go out meanwhile

[lucid yoke]

Id just use the gpu for that one ngl

[potent sun]

as long as the model quality stays the same, if I can increase speed

ok

also I did a training test and it seems I can more or less pickpoint the epoch range for the image generation I want

I just decrease the learning rate and generate samples more often right?

[lucid yoke]

Hmm im afraid i cant give your answers for that, i just use a script lol

[potent sun]

where do I change that

[lucid yoke]

You don't, its what being shown in the cmd

[potent sun]

I see dataloader threads but dunno if its that

[lucid yoke]

That's the training speed that's measured

[potent sun]

ah ok but there must be a value ot there that affects it

[lucid yoke]

Imo the base settings are pretty optimized

Could try changing the method from adam to something else but youd need to edit a lot of params

And no i cant help with that

[potent sun]

yeah but I think they are taking into account that you still use the computer

I mean there must be something I could do to increase speed without quality changes

[lucid yoke]

Hmm not sure since mine get 15.4 gb vram use out of 16

[potent sun]

how many dataloader threads do u use

[lucid yoke]

Not behind my pc rn

But the base sdxl template

Should show lol

Granted im using a test version of torch to run it

[potent sun]

anything I could change here?

I could also increase batch size

[ornate elk]

if your vram isnt nearly maxed out when training then this would be the best

[potent sun]

I've read it depends on learning rate

So you have to tak into account the lr, not everything works

But maybe I'm wrong

[ornate elk]

i mean check if you have some room left and than increae the batch size

always keep 1gb free at least

[last flax]

hey

https://github.com/comfyanonymous/ComfyUI#installing
is this the latest comfiui install guide?

im new to this so dont mind me asking but do i also need to install stable diff and a1111 in order to make comfiui work

im looking to try and play around with the hunyuan model in comfiui

[ornate elk]

hey, yes thats the latest guide

and nope you dont need auto1111 for comfyui to work

[last flax]

alr tysm !!

[potent sun]

But aren't lr and batch size correlated among them?

[ornate elk]

i dont think so

[potent sun]

I'm getting mixed answers then because some say you have to adjust batch size to the learning rate and others say they don't go together

[last flax]

am getting an runtime error: no cuda gpu available

oh

i just needed to plug in my charger xd

[idle plank]

Hi guys, I've got this error when I want to use LTX video, someone can help me out ? VAE load device: privateuseone:0, offload device: cpu, dtype: torch.float32
no CLIP/text encoder weights in checkpoint, the text encoder model will not be loaded. (It look like it's deconnect my server whenever I want to run the model but I put t5xxl_fp16 right there so I don't understand --> "C:\Users\Documents\Stable-diff\ComfyUI\models\text_encoders\t5xxl_fp16.safetensors" I'm running it with a 7800xt from AMD

[vocal burrow]

Just in case, if you clicked the linked that was there before, don t follow their instructions.
It was a "support scam".

leave whatever discord they made you join and undo modifications, uninstall program they told you to install, etc

[brittle lantern]

@ornate elk I heard ROCm 6.4 on Linux is pretty good have I tried it? If so how much faster or better is it

No idea what’s wrong with that message

But ok

[gentle sundial]

@ornate elk thank you for the suggestion! but these too really just for more detailed prompt and more specific things and to separate the prompt aren't they ? I don't see how these can change faces, what I'm looking for is to create variation in faces for each image, like for sd I use dynamic prompt wildcard to do that but for pony and IL nothing is really working

[ornate elk]

Hey, I haven't tried it yet. But I doubt that it would make that big of a difference with a 6600xt

[ornate elk]

Wildcards can work with pony and illus too

[gentle sundial]

yes but they don't change realistic faces and they don't work if you put celeb names , only cartoon character stuff

[brittle lantern]

I’m planning to get a 2TB HDD for Linux and AI on it and go cheaper and buy 9070 or 9070 XT u think it will be fine?

And I can use my current drive for windows and gaming

[potent sun]

im getting 1.3, 1.4 s/it

cant I get up to 2?

[warped canyon]

how do you batch inpaint pictures?

[ornate elk]

Why a HDD? It will be slow as OS drive.

[brittle lantern]

Nice I got auto corrected by Apple sorry 😭 I mean SSD NVME NM790

https://uk.pcpartpicker.com/list/FMNwKq I may do this @ornate elk

[gusty reef]

Hi! So I am just downloaded StableDiffusion 3.5 and I am using ComfyUI. I am running into a problem. Is it normal to take a long time to generate images when loading the first message of the session and/or changing the prompt?

When viewing my console, the majority of the time is spent on loading the models and whatnot, but this does not occur if I run the same prompt twice the process will instantly go to the image generation which takes around 30 seconds. But when changing prompts, it spends an additional minute or two on loading the models before starting the image generation.

[ornate elk]

looks okay, but i personally wouldnt go with an Intel CPU. The 13 and 14 gen have massive problems causing bluescreens and dieing randomly. Intel tried to fix it with bios updates.
If you dont need 20 cores, the AMD Ryzen 7 9700X will work great too. or if you get an 7800X3D.
Then you could go with an AM5 Mainboard and 6000Mhz RAM too.

[gritty hawk]

Can someone please ELI5 how to get past the 75 token limit/enable second prompt in Forge? I've tried looking in the extensions, I've tried googling but can't seem to get a complete explanation - but I know you can go well past the 75 limit (I think its like, what, 255?)

Any help would be appreciated.

I know... or I think I know... it has to do with enabling the right text encoder(s)..

I enabled clip_l.safetensors and clip_g.safetensors with Juggernaut (SDXL) and the token counter doesn't turn red anymore, does that mean I did it?

[steady minnow]

Any updated guides on how to run stable diffusion on android ?
Also can my s21+ run it ?

[ornate elk]

there is no app for that currently

there is only an app for llms

[steady minnow]

Even throw termux ?

[steady minnow]

Wich is ?

[ornate elk]

This one:
https://github.com/mlc-ai/mlc-llm

[ornate elk]

idk, i guess some people got it working but its painfully slow

better use your PC GPU and stream to the phone

[steady minnow]

Well I don't have a PC that's why I'm here :p tanks I'll check out Mlc

[ornate elk]

oh okay well xD

[idle plank]

No problem I never click on the link that was way too suspect

[mellow stone]

can i send here pics to help me with i cant generate same pic some details are missing

how can you do refine >?

[gritty hawk]

If im running SD through something like runpod, is there any practical way of not having to use their virtual environment to use cloud based GPU’s?

I figured out how to get the virtual setup running but i gotta fine tune all my settings, extensions, models etc all over again… plus pay for cloud storage… and it really bums me out.

[ornate elk]

dont click on scam links, the discord that got linked is a scam support server

refine is done with a refiner model, works mostly every webui

but its not that good, upscaling is much better

[ornate elk]

you can send example images in the #📝｜prompting-help channel

[prisma moss]

how does 'aesthetic' and 'very aesthetic' affect the image?"

[signal torrent]

Anyone able to get Sage Attention and Triton working on an RTX 5090 in ComfyUI?

[potent sun]

Collecting torch==2.6.0
  Obtaining dependency information for torch==2.6.0 from https://download.pytorch.org/whl/cu124/torch-2.6.0%2Bcu124-cp310-cp310-win_amd64.whl.metadata
  Using cached https://download.pytorch.org/whl/cu124/torch-2.6.0%2Bcu124-cp310-cp310-win_amd64.whl.metadata (28 kB)
Collecting torchvision==0.21.0
  Obtaining dependency information for torchvision==0.21.0 from https://download.pytorch.org/whl/cu124/torchvision-0.21.0%2Bcu124-cp310-cp310-win_amd64.whl.metadata
  Using cached https://download.pytorch.org/whl/cu124/torchvision-0.21.0%2Bcu124-cp310-cp310-win_amd64.whl.metadata (6.3 kB)
Collecting typing-extensions>=4.10.0
  Obtaining dependency information for typing-extensions>=4.10.0 from https://download.pytorch.org/whl/typing_extensions-4.12.2-py3-none-any.whl.metadata
  Using cached https://download.pytorch.org/whl/typing_extensions-4.12.2-py3-none-any.whl.metadata (3.0 kB)
Discarding https://download.pytorch.org/whl/typing_extensions-4.12.2-py3-none-any.whl#sha256=04e5ca0351e0f3f85c6853954072df659d0d13fac324d0072316b67d7794700d (from https://download.pytorch.org/whl/cu124/typing-extensions/): Requested typing-extensions>=4.10.0 from https://download.pytorch.org/whl/typing_extensions-4.12.2-py3-none-any.whl#sha256=04e5ca0351e0f3f85c6853954072df659d0d13fac324d0072316b67d7794700d (from torch==2.6.0) has inconsistent Name: expected 'typing-extensions', but metadata has 'typing_extensions'
ERROR: Could not find a version that satisfies the requirement typing-extensions>=4.10.0 (from torch) (from versions: 4.4.0, 4.8.0, 4.9.0, 4.12.2)
ERROR: No matching distribution found for typing-extensions>=4.10.0```

[ornate elk]

which webui?

[potent sun]

LoRA_Easy_Training_Scripts

[ornate elk]

oh okay

[potent sun]

Im just trying different training UIs because I cant get good results

they dont share the same training settings so some configs cant be adapted to other trainings

[ornate elk]

true

have you tried OneTrainer?

[potent sun]

its a bit frustrating because I ask but they just tell you to just train and train

[potent sun]

im banned from their discord saerver and Im tired of trying to make it work

I dont know what to do with finetuned models in onetrainer, they have good settings for popular models, but dunno what to do with finetuned versions of those

those same settings seem to not work

kohya directly doesnt work

[ornate elk]

hmm, yea its fustrating that there are no perfect training guides for these tools

[potent sun]

I just hate it when you ask and they laugh at you, this isnt something easy to config

[ornate elk]

true

[potent sun]

Im just trying to get some sdxl values

have you been able to train some models with tha one?

finetuned versions of sdxl seem to have different training parameters

[ornate elk]

only loras

but im didnt invested that much time

i followed the lazy lora trainer guide from civitai

[potent sun]

its pk Im trying to do a lora

[potent sun]

do you have a link?

how were the results?

[ornate elk]

Here:
https://civitai.com/articles/4789/lazy-lora-making-with-onetrainer-and-ai-generation

[ornate elk]

the lora was to strong

i tried training on angel mech wings, but the whole character got trained with it

i guess i should have only Masked the wings but thats a lot of work xD

[vast crest]

Hi, I would like to know if in your opinion my 1111 is up to date?

[ornate elk]

Yep its up to date

[vast crest]

ok,thx a lot 🙂

[graceful oar]

hey, im using a1111 with zluda (rx6600), and im getting the 'RuntimeError: Input type (float) and bias type (struct c10::Half) should be the same' error. (even after a reinstall). using the --no-half flag fixes the problem, but massively impacts generation time (like 5x or 10x). can i fix this error without slowing it down?

[graceful oar]

im also getting this error

Failed to load ZLUDA: Could not find module 'C:\SD-Zluda\stable-diffusion-webui-amdgpu\.zluda\nvcuda.dll' (or one of its dependencies). Try using the full path with constructor syntax.
Using CPU-only torch

gpu usage is low, around 20%, cpu usage around 60-70%. (with the --no-half flag, makes it run very slow)

[graceful oar]

i think my gpu isnt involved at all right now since that 20% is some background stuff

[bleak phoenix]

my stable diffusion doesnt have the video bit?

[lucid yoke]

Hi, your question is really generic? can you elaborate a little more?

do you mean the API?

[bleak phoenix]

so i have the one you run through command prompt and it doesnt have the text to video option?

[lucid yoke]

Hmm that's maybe api? Do you have a subscription or do you have it running locally using your gpu?

[ornate elk]

Can you show the full cmd log?

[graceful oar]

im doing a full reinstall rn, a1111, hip sdk, zluda, so i cant currently

but if it still doesnt work after that ill give you the log

[ornate elk]

Alright 👍

[graceful oar]

also i saw that some people had torch 2.2.0+cu118 installed, and ive had previously got 2.7.0. does this change anything? do i need to downgrade?

[ornate elk]

The zluda version should come with 2.6.0

Best is to run pip cache purge in a cmd once before installing the webui. That removes old stuff

[graceful oar]

actually yeah i mightve had 2.6.0

[graceful oar]

alr man, after reinstalling everything works. thanks for help

[potent sun]

is it possible to change the output sample images in easyscripts and kohya?

they are 512 by default

and I see no option to change size or seed in the gui

[potent sun]

I found the solution here https://github.com/bmaltais/kohya_ss?tab=readme-ov-file#sample-image-generation-during-training

[tawdry tide]

Has someone here an solution for rtx 50 cards? I cant use stable diffusion..

[lucid yoke]

Forge UI & swarm natively support it now iirc, just requires a fresh install

[tawdry tide]

Never heard of them.. and there i can normally use models from civitai?

[lucid yoke]

ahh you were using A1111 before then, hmm youd want to try Forge then
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

also yes

SwarmUI is built ontop of comfyui but a useable interface

but forge is just a fork of a1111 one day and " it just works"

theres some posts on how to get forge to work but i personally use swarm

[tawdry tide]

Okay thanks gonna try it out

[lucid yoke]

techsupport too man? 😭 your really spamming this everywhere huh

[fossil dew]

Any idea what can cause stable diffusions python to open and load, but not open up in a webpage, when it has previously?

[lucid yoke]

Assuming you are using a1111 could be something in the .bat launch file but if you navigate to localhost it should just open in your browser

[ornate elk]

does the cmd stay open?

[daring reef]

Can i use 2gpus?

5060 ti 16gb and 3060 12gb

[sterile garnet]

Hello, I was wondering if someone with experience could help me out? I just got a new laptop with a RTX 4070 8GB Vram, and 64 GB Ram. Right now to generate a 512x786 image with 16 steps its taking me around 1000 seconds (16 minutes) for a kind of not so high quality picture. I'm sure however with the right steps and optimization I can crack down the time and get better results. Currently using Flux Schnell on Pinokio. 🙏 lmk

Or otherwise if someone could recommend a better model for my specs

[sterile garnet]

For now I'm switching to RealVisXL V5.0 Lightning Baked VAE but if anyone has some flux tips please lmk

[ornate elk]

Hey which webui do you use?

Flux with 8gb vram isnt that good to use

[tawdry tide]

forge also doesnt work xd

[ornate elk]

You can try these steps from an other 50 series user:
#🤝｜tech-support message

[tawdry tide]

thats just insaneeee, nothing works in 2025 with those bad 50 series cards

[tawdry tide]

will try that then.. xd

[lucid yoke]

Weird it works for me

Ah is comfy backend on "always update (forced)" ?

[tawdry tide]

uh how can i know?

[lucid yoke]

Im on phone rn but under the server tab > backends

Iirc

Click the green pencil to edit the backend and then update to always update (forced)

Then wait till it restarts

You are using SwarmUI and not stableswarmui right?

[tawdry tide]

using: SwarmUI v0.9.6.0

[tawdry tide]

cause youre not using 50 series card..

[lucid yoke]

Good hmm then the backend should work but if it doesn't after those steps

[lucid yoke]

Im on a rtx 5080 thank you very much

[tawdry tide]

didnt you order a 5090??

[lucid yoke]

80, 90 was scalped to 6k here

Got 80 at msrp

[tawdry tide]

damn okay but what am i doing wrong then

[lucid yoke]

But hmm if the backend force didnt fix it hmm i recommend joining the swarm discord

Its possible that the 90 series needs comfy nightly version and you need to use that instead

And im not sure on how to hook that up

[tawdry tide]

mh i posted something in their forum.. will see

[bold jay]

i needed some help

keep getting this error

RuntimeError: CUDA error: no kernel image is available for execution on the device CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.
Time taken: 3.0 sec.

plz help

whats this

[lucid yoke]

Scam if its a discord invite

Your on a 5000 series card?

[bold jay]

yea sm1 was asking for my wallet address 😭

[bold jay]

mhm

[lucid yoke]

What UI are you trying to use?

[bold jay]

a111111111111

[lucid yoke]

A1111 is outdated, forge supports it but needs a manual fix and comfyUI & swarmUI supports it out of the box

I recommend trying out swarmUI

[bold jay]

i saw comfy ui and it looks so complicated icl

[lucid yoke]

swarmUI is all the benefits of comfy without the need to use comfy directly

https://swarmui.net/

[bold jay]

yea this looks complicated asl icl

not the installation

the usage

[lucid yoke]

Hmm its not much difficult then a1111

If anything theres little ? On all settings explaining what its for

[bold jay]

why is there nodes and stuff

[lucid yoke]

Its optional

In swarm

[bold jay]

oh bet

[lucid yoke]

Swarm basically makes the comfy workflow automatically

Change a setting here and it edits in the backend

What gpu do you have? If you have a 5080 or better its almost a shame not to use something like swarm/comfy

[bold jay]

i got a 5080 xd

[lucid yoke]

??

Norton wow that's a long time ive seen that

[bold jay]

i was running the install bat file and it got blocked

[lucid yoke]

Its probably a false flag but @jagged plover

Hes the dev

[bold jay]

xd

[lucid yoke]

its a false flag but if your doubtful you can always dive into the GitHub code

[bold jay]

we dont got time for allat 🗣️

[terse hull]

hey, is there a guide for installing a1111 on linux with amd gpu? i cant find a recent one on the internet

[bold jay]

@lucid yoke how do i add models nd lora

[lucid yoke]

ah the same was as you do in A1111
\SwarmUI\Models

[fossil dew]

Yes

[lucid yoke]

Loras go in the lora folder, any stable-diffusion model (1.5, xl, 3.5) goes in stable-diffusion and flux, video models etc go in diffusion_models

[lucid yoke]

Hi, the first pinned message of this channel has plenty of guides

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[bold jay]

sm1 was saying put em in a111's mod folders

[lucid yoke]

hmm

[terse hull]

i know, but they dont support linux

[lucid yoke]

ah hmm linux i read over that sorry, hmm and you spesifically want a1111?

[terse hull]

or comfyui

i dont care that much

im a bit expierenced with stable diffusion but i recently switched to linux and i still need to learn stuff

[lucid yoke]

https://comfyui-wiki.com/en/install/install-comfyui/install-comfyui-on-linux

[terse hull]

not for amd gpus

[bold jay]

yo @lucid yoke ilysm

thx for the help

[olive topaz]

Hi im fairly new to stable diffusion and have been having fun making images with Automatic1111, but I am trying to use animatdiff to see if I can make a video or GIF in the art style I like but it only generates 1 single image. I have animatediff enabled, 240 frames, 24fps, 0 display loop, temporaldiff-v1-animatediff.safetensors motion module.

[lucid yoke]

no worries!

[terse hull]

that doesnt work for me because i have gfx803

i know theres a guide somewhere out there but i cant find it again

[lucid yoke]

hmm wouldave been handy to know but hmm not sure then

im not too familliar with all the amd cards

but its a whole chellange to get them running

[olive topaz]

can anybody help me? I did get a no motion module detected error message

[open fox]

Sorry no experience with a1111 but it seems you did not download the required motion modules.

https://huggingface.co/conrevo/AnimateDiff-A1111/tree/main

[olive topaz]

I have multiple motion modules in my sd-webui-animatediff/model folder

[potent sun]

if you have to crop something for datasetup for a training, do you cut arms or legs?

[ornate elk]

Can you show the log and the txt2img settings?

[daring reef]

I want everything related to AI (w-okada, wan, stable diffusion, comfy ui, etc) on an external hard drive, will it affect anything? This is the speed using winsat disk -drive in cmd to see the speed

Disk Random 16.0 Read 0.92 MB/s 2.9
Disk Sequential 64.0 Read 113.37 MB/s 6.8
Disk Sequential 64.0 Write 145.12 MB/s 7.1

[lucid yoke]

If you use a HDD it will fastly increase the loading times

Loading a model can go from 15s to like 30min depending on its size

[ornate elk]

AnimateDiff SDXL model doesnt support motion modules

[olive topaz]

so it was my checkpoint? or did I download the SDXL version of animateDiff?

[ornate elk]

oh i see you tried to use the 1.2 animatediff model with an sdxl based checkpoint

so yea you either have to select the sdxl animatediff model without motion modules or select an 1.5 based checkpoint for the 1.2 model

[viscid steppe]

hey peeps, quick random question for you. I'm trying to use adetailer for faces, but it seems to make faces "cartoony" instead of improving quality/realism. Any suggestions?

Using adetailer within a1111, the issue is common across all models I use, even realistic models.

[olive topaz]

I decided to try HOTSHOT because it said that supports SDXL but hit a new issue. I also could not find the animatediff SDXL version of the extension

[fossil dew]

Ill bring this up again when I have pc access

[upper cradle]

I recently upgraded from a 3060 to a 4060, but now it takes way longer to generate anything. I've already updated my drivers, but it didn't seem to help. Is there anything I could do to fix it?

[ornate elk]

Which webui do you use?

[ornate elk]

The issue you got was an out of vram error

Also idk if hotshot works at all cause its outdated since 2 years

[bold jay]

@lucid yoke how do i set resolution to 4k with swampy, it dosent go too high

[lucid yoke]

Make a image within the supported resolutions and then upscale/refine

[upper cradle]

Automatic1111 webui

[ornate elk]

What's in your webui-user.bat?

[upper cradle]

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--listen

call webui.bat

[ornate elk]

Ah that explains why its so slow.
You have to add --xformers --medvram-sdxl
Then save and close the file.
Then delete the venv folder.
And then relaunch the webui-user.bat

[fluid trout]

hi, i followed these instructions but i get many prompt execution errors, im new to comfyui, anyone can help? https://comfyanonymous.github.io/ComfyUI_examples/hidream/

these are the errors

[ornate elk]

Hey, it seems like your missing some models for it to work

Have you downloaded every necessary model, VAE, and clip that hidream needs?

[fluid trout]

i donwloaded everything the instructions mentioned and put them in the folders

[ornate elk]

And you were able to select them in the dropdowns?

[fluid trout]

the instructions included the workflow. i didnt do anything in the workflow but write a simple prompt and hit "run"

[ornate elk]

Ah okay thats the issue

You have to reselect the models and VAE and clip because the one from the workflow can have different names or folder paths

Clicking on the names and then select the correct ones should be enough

[fluid trout]

within the nodes you mean, i guess?

i cant reselect them

it just goes to "undefined"

[ornate elk]

For every model?

[fluid trout]

can i stream my monitor for you?

[ornate elk]

Nope im on phone rn

You can screenshot

[ornate elk]

Okay and if you click on undefined?

Nothing shows up?

[fluid trout]

cant bring it back up without ctrl+z

[ornate elk]

Is the hidream model placed in the models/unet folder of comfyui?

[ornate elk]

You have to move it to the unet folder

What's your gpu btw? The 33gb model won't work on anything smaller than a 5090

[fluid trout]

ok lol i have 12GB on a 4070ti. thought it would just take longer

i also donwloaded the smaller gguf you can see in the screenshot. but i couldnt find a workflow file that is set up to use the gguf

[ornate elk]

The gguf needs the gguf loader nodes

You can install it as custom node via comfyui manager

[fluid trout]

i downloaded that custom node, but i have zero clue how to include it in the workflow correctly tbh

[ornate elk]

Double click on an empty space in ComfyUI and search for gguf

Then there should be a gguf loader

Or checkpoint gguf or something like that

[ornate elk]

Ah yep unet loader gguf

[fluid trout]

ok got it

do i replace the "load diffusion model" node with it?

looks like comfyui doesnt recognize any of the files i put in the folders

[ornate elk]

Yes

[ornate elk]

Have you restarted it?

[fluid trout]

yes

[ornate elk]

Very strange

How did you installed comfy?

[fluid trout]

downloaded it from comfy.org

didnt change any settings during the installation, kept everything as it was preselected

just tried to reinstall comfyui, now it wont even launch and because there are missing python packages. but it wont install them.. what a mess

i get this when trying to reinstall comfyui, but clicking install doesnt resolve it. im going crazy

[ornate elk]

Do you have any third party antivirus that may blocks it?

Or install Comfyui without the exe installer.
I guess that one is still buggy

On the first link of the pinned messages you find install Guides

[fluid trout]

ok the portable version is up and running

this actually shows the files i put in the models folders

but still i dont know how to put the gguf model in the workflow instead of the large model my gpu cant handle

[ornate elk]

Does it show up in the gguf unet loader?

[fluid trout]

just this moment i got my first image

looks kinda blurry/smudgy though

[ornate elk]

Have you selected a VAE?

[fluid trout]

yes, the one that was in the instructions

on futher generations i changed nothing but the prompt and all i get is stuff like this

[ornate elk]

Hmm, no clue why, but you can try ask in the #🧣｜comfy-ui channel and provide the workflow

[fluid trout]

okay, thanks for your time and help!

[glad fox]

Anyone know how to enable live preview of inpainting in AUTOMATIC1111's webui? I completely forgot how to do it.

[upper cradle]

This seems to have partially worked. The actual generation doesn't take long now, but it takes forever at the start or near the end when it's saving the image

I also switched it from --medvram-sdxl to --medvram and it hasn't made much of a difference

[ornate elk]

if you only use sdxl based models it wont

[ornate elk]

do you use upscaling by any chance ?

[upper cradle]

I don't

[ornate elk]

ah okay

how much RAM do you have?

[upper cradle]

16 GB

Looking at my performance metrics my VRAM and RAM are now getting maxed out when I do anything

[ornate elk]

ah okay, then you need to increase the windows pagefile, that speeds up sdxl gen and should prevent the lag at the start and end

Here is how to:
https://www.tomshardware.com/news/how-to-manage-virtual-memory-pagefile-windows-10,36929.html

Make sure its only enabled for the C drive and disabled for any other drive.
Set it to custom for C and set it to 16000min and 24000max
Then hit ok, and reboot the PC

Also make sure to have at least 15-20gb free space on C

[upper cradle]

It's a little better, but it's still taking an extra minute and a half just to save. Would setting the max higher help?

[ornate elk]

Can you show your txt2img settings?

Maybe your initial resolution is just to high

[upper cradle]

I've done a lot of generation with these settings, albeit with 30 steps instead of 20, but I never had an issue before I switched to my new GPU

[ornate elk]

Hmm, the settings seem normal

Have you anything running in the Background like Wallpaper engine Or Games?

[upper cradle]

Just discord

[ornate elk]

Your new GPU has 8gb vram and your last had 12gb right?

[upper cradle]

Both had 8gb, but it's been a while since I looked at the old GPU specs. Let me check

[upper cradle]

Ah, you're right. Old one was 12GB and the new one is 8GB

I still have the old GPU and I think my motherboard supports multiple GPUs, so can I just have both running?

[ornate elk]

Yes that works too

[lucid yoke]

with multi gpu swarm is really nice 👀

[ornate elk]

You can use one webui for each gpu

[lucid yoke]

with swarm you can run both at the same time for 2 gens at once in one uI

once multi gpu comes into play swarm is just peak

[upper cradle]

Unless I can pool my resources for one generation, I'd rather have my 3060 run for generation and then use my 4060 for other stuff

[viscid steppe]

Following up on this to see if anyone has any suggestions 🙂

[ornate elk]

do you have an example?
You could try lower the denois value of adetailer

[viscid steppe]

yeah, lemme find on

[sterile garnet]

Do you have any suggestions for current high performing models for my specs by chance or methods I can use to ramp up my performance

[ornate elk]

you should setup Forge webui and use SDXL or Illustrious models

they should work pretty fast and have a good quality too

You can also use smaller flux models like nf4 ones but they wont be that fast.

A guide to setup Forge is on the first link of the pinned messages here

[sterile garnet]

Bet thankyou so much

is forge webui for flux only or nah

[ornate elk]

forge webui is for 1.5, sdxl based stuff like pony and illustrious, and flux

[sterile garnet]

got it ty

[sterile garnet]

dang bro this is working amazingly on my system, getting high quality generations on flux in around a minute, using my system ram and Vram

[ornate elk]

nice 🙂

[tepid crypt]

im using a google colab i found that works, but i have to download everything each time i want to use it, someone knows if theres a way to maybe save the files at my google drive to not have to download it everytime?
heres the colab: https://colab.research.google.com/drive/1XVwDfDGo5chAGABtwKCCY3ijTZu5azOQ

[sudden pike]

i get a PDHError when i wanna use my other GPU

it runs on GPU 0 but not on GPU 1

no

what is that for a discord

i dont trust that Discord

i dont see anything on this server that mentions this discord

And its sus that you had this with Mandarin letters ready for copy paste

proof to me that its a Stable Diffusion server...

[olive topaz]

where can I find the SDXL animatediff??

[sudden pike]

how can i know?

nothing on here or the website links this server...

just proof to me that its legit

[sudden pike]

ah cool he is gone now lol

[ornate elk]

What GPUs do you have?

And which webui do you want to use?

[supple vessel]

Hello, can someone help me with this. I have depolyed the sd webui in the server and use it remotly. The website works fine in txt2img but it shows error for img2img: Unexpected token '<', "<html> <h"... is not valid JSON

[ornate elk]

Make sure the Webui is whitelisted in any browser adblocker

[supple vessel]

it shows diff error in diff browser. safari for error: The string did not match the expected pattern. Chrome for error: Unexpected token '<', "<html> <h"... is not valid JSON

[silent stream]

I migrated over to SD Matrix instead of independent installs for Forge

Whenever I hit generate, it spikes the CPU usage to 100% and has a long delay before actually generating

This was not the case beforehand, and I assume it was some setting that is not set for the new installation in the Matrix version

Ryzen 5800x, 32GB Ram, 4090

[silent stream]

I seem to have found a solution(?)

Enable graphic acceleration for Chrome

Idk why that fixed it when I never changed that setting before/after switching to Matrix

But it no longer spikes my CPU to 100% with chrome

[gaunt otter]

Hi, I'm new here. Is anyone still using EveryDream2trainer or is it old tech? Found an old videa and want to try it.

[gaunt otter]

can anyone help with that? did every single step from the site. First had the numPy problem ther Triton and now im left with this. Any Ide how to fix it?

[ornate elk]

Hey, if you want to train loras go with OneTrainer or Kohya_ss

Both are the most used tools for that

[regal flame]

My video card is a 5060ti and I recently got an error when running sage attention with comfyui:'Error running sage attention: Triton only support CUDA 10.0 or higher, but got CUDA version: 12.8, using pytorch attention instead.'

[sudden pike]

RX 7800xt and I use a forked that should work for AMD gpus

[ornate elk]

the stable-diffusion-webui-amdgpu fork?

[sudden pike]

Correct

[ornate elk]

then you have to disable the Radeon TM Graphics in the Device Manager or add set HIP_VISIBLE_DEVICES=1 to the webui-user.bat

[sudden pike]

I will try that thx

[ornate elk]

np

[ornate elk]

i hope you installed it with Zluda and not directml

[sudden pike]

Should I reinstall it with "Zluda"?

How

[ornate elk]

you can follow my AMD setup guide

[sudden pike]

Cool can ya send it?

[ornate elk]

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[sudden pike]

Also why no directml

I'm absolutly a dumb mf in this topic xD

[ornate elk]

Its very slow and crashes often due out of vram

zluda fixes that, best performance on your gpu

[sudden pike]

Ahh cool

Thanks

[ornate elk]

np, let me know if you have any question

[sudden pike]

yee i will

[ornate elk]

Go for Automatic1111 with Zluda or Forge with Zluda

Auto1111 is the "stable-diffusion-webui-amdgpu" fork

[sudden pike]

Is that normal?

ok now im generatin a pic

it takes like 10GB of VRam (i have 16GB)

now when i press on Generate it goes into this Compilation loop again..

ah nvm.... now it just generated a image

thanks for you help!

@ornate elk it does Generate now.. Does that look right?

bcs it takes some time (second image)

[vocal burrow]

1/ not gonna work to generate at 2048x2048 without any sort of upscaling. you need to generate at the resolution used during the model training or within +-50% of it in one or both direction.
So for sd1.5 : 512x512, sd2 : 768x768, sdxl/sd3 : 1024x1024.
Otherwise you will get glitched / duplicate artifacts and will eat more vram (possibly causing longer generation) for pretty much nothing cause of the bad output.
2/ Sampling steps at 70 is too much. depending of the model, 20 to 40 is plenty enough. There are even models out there trained for 1 to 3 steps
3/ Euler A is a pretty fast but simple sampler, people nowadays tends to use more complex ones such as DPM++ 2M Karras
4/ CFG Scale 18 is ridiculously high. For usual A1111/forge workflow anything above 10 is exceptionaly high and will result in deep fried output.

[gaunt otter]

Can you recomend solid guide/tutorial for instalation?

[ornate elk]

Yep compilation is normal the first time

Good that it works 🙂

[ornate elk]

Here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Lora-Trainer-Setup-Guides

[sudden pike]

ok now different issue

it stopped generating bcs it asked for 18gig vram

but yeah. i have 16 gig