🤝｜tech-support

[plain bronze]

?

[ornate elk]

you would need to install xformers 0.0.16 into the sitepackages folder

[plain bronze]

Is there any link for to download that? And is former version xformers can cause bad effect compared to the current state of Comfy of mine?

Lower berformance, some bugs and etc.

[coarse swan]

@ornate elk seems to be working fine now, thank you very much!

[ornate elk]

perfect, no problem!

[pale salmon]

@ornate elki tryed adetailer lol and it crash and i got this

[ornate elk]

can you screenshot me the txt2img settings? steps, resolution, hires fix etc

was tiled vae enabled?

[pale salmon]

its an nsfw prompt idk if i can send it

[pale salmon]

idk really lol

[ornate elk]

lol xD

[ornate elk]

i dont need the prompt

just the rest

[pale salmon]

where can i check if its enabled

[ornate elk]

scroll down

and there you should have a Tiled VAE setting

enable that

[pale salmon]

ok, done

[ornate elk]

leave it like this

but you have to restart now because you did run out of vram. and then a webui restart is required to clear the stuck vram

and then enable tiled vae again

[pale salmon]

k kfine ty

[ornate elk]

nope should work fine

[plain bronze]

Should I delete the folders I copied from A1111?

[ornate elk]

yep

[plain bronze]

Where to download that version, I can't find it.

[ornate elk]

also you use a pony (sdxl) model so you may need to try hires fix upscale by 1.2

if that works you can go higher until it wont work

[plain bronze]

😦

[ornate elk]

You have to run pip install xformers==0.0.16
Inside the site packages

[plain bronze]

How to do that?

And is this the last version of it that I can install?

[pale salmon]

Okay bc my style is like anime screen cap, so I kinda need a good quality either way its ugly lol

[ornate elk]

You can also upscale in img2img Without getting out of vram

Using the SD upscale script

[plain bronze]

How to run that in site packages folder?

Open cmd and write that there?

[ornate elk]

In the folder click into the explorer bar and type cmd and hit enter

Then run the command there

[pale salmon]

Nice idea will try this

[coarse swan]

After rendering one image, I get this whenever I try to generate more images.

the image was forming properly but at like 95% it just disappears along with the loading bar and this pops up. Then the subsequent Generate just goes 'in queue: 1/1' forever

closing cmd and opening again works, but it feels like a bandaid solution since it keeps happening haha

[ornate elk]

You need to have python installed in path

[ornate elk]

Can you share the full cmd Text?

[plain bronze]

And what to do now?

[ornate elk]

Run the Python 3.10.6 installer again.
Click modify, click next, then select add python to environment variables.

[near yew]

hey guys

i dont see img2img tab

what should i do

to see it

and how the heck do i setup my model rpg v5 cause i ve been searching a lot on google and i dont find anything

to make some recommended settings

why it must be so damn hard to make something on it

[ornate elk]

Oh your using InvokeAi

Really rare to see it here

[plain bronze]

I didn't seperately download Python. When I download A1111 and ComfyUI it downloaded with them.

[ornate elk]

Ohhh

Then you have to install python manually.
But then go for python 3.10.11 64bit

[near yew]

someone recommended me

to install it

do you have an idea

[plain bronze]

So, must install 3.10.6 or 3.10.11?

[near yew]

i really want to use img2img to make this https://www.youtube.com/watch?v=9v4lBexN_Mg&t=679s

and i need img2img tab

and also some settings for model: rpg v5

cause i dont see anything on google

and i have no idea what to do

it s so overwhelming

no idea what to do

[coarse swan]

closed the cmd earlier. Works normally this time. 😔

I'll ping ya if I get it again

[near yew]

;(

[ornate elk]

You should install Auto1111 webui

It has img2img

A guide for that is in the pinned messages

[near yew]

ye

im doing this already

i just want to make some cool photos of me

like this video

https://www.youtube.com/watch?v=9v4lBexN_Mg&t=363s

[plain bronze]

@ornate elk

[ornate elk]

3.10.11

[plain bronze]

Customize installation?

[ornate elk]

check add python.exe to path

and then click install now

[plain bronze]

Feeling like hacker.

I upgraded that. Is this recommended? @ornate elk

[ornate elk]

oh i noticed that that is the wrong xformers version. you would need 0.0.23

im sorry

run the command again but change the number

[plain bronze]

Okay.

[plain bronze]

Will version 0.0.16 uninstall?

When I install 23

Or do I have to uninstall it manually?

@ornate elk Again...

[elder mauve]

@ornate elk i tried to install ROCm , but turns out it wrecked my display drivers somehow , and display was all glitchy

so i removed it and now its normal, but the GPU is not getting detected by fooocus

any idea how to install rocm without display drivers?

or like what to do now

[ornate elk]

yes

have you run the pip install xformer==0.0.23 ?

[plain bronze]

Yeah, that's why 0.0.16 uninstalled.

[ornate elk]

cant help much with linux sry

[plain bronze]

Here.

[coarse swan]

happened again. sent the log yer way 👋

[pale salmon]

quick question how do i get the vae section here

[ornate elk]

go into the Settings - User interface - Quicksettings
There add sd_vae
And then hit apply and reload ui.

[pale salmon]

thx u !

[elder mauve]

fixed it

but now foocus gets stuck on " encoding positive"

[plain bronze]

Why it didn't work? What should I do now?

[ornate elk]

what does the cmd shows? full log pls

[elder mauve]

is it one of those times when it only takes very long on first launch?

[plain bronze]

Okay.

Sorry, it changed when I paste it here.

Gotta send ss

[ornate elk]

i dont think so

[ornate elk]

okay looks normal

[plain bronze]

But you said using pytorch cross attention means it doesn't run xformers.

[ornate elk]

yes

but it should automaticly use xformers but it doesnt now. so idk how ti fix it without reinstalling it with a venv

[elder mauve]

My slow ssd 🥲

[plain bronze]

Reinstalling whole ComfyUI with venv?

[ornate elk]

yea but i cant guarantee that it would work

and i dont have the time rn to explain every step

[plain bronze]

https://tenor.com/view/jahrein-jaho-bleach-jahrein-domestos-drink-to-bleach-gif-22001927

[safe inlet]

why are you trying to install it

[plain bronze]

Wanna use xformers.

[safe inlet]

mm

[dense sandal]

@ornate elk I installed the motherboard chipset and now it worked

Although maybe it was just luck this time

Its a new PC and the chipset driver wasnt installed before. Maybe that fixed it?

I dont know if the motherboard needs any other driver besides the chipset

[ornate elk]

interesting

good that it worked

but still strange

[dense sandal]

Also it went so much faster

I had the vid2vid in under 3 minutes and before it was like 10

I will be doing a lot more tests though

[ornate elk]

alright will keep that in mind if someone has the same issue

[elder mauve]

now i get ```
[Fooocus Model Management] Moving model(s) has taken 0.55 seconds
Started worker with PID 23491
`App started successful. Use the app with http://127.0.0.1:7865/ or 127.0.0.1:7865
[Parameters] Adaptive CFG = 7
[Parameters] CLIP Skip = 2
[Parameters] Sharpness = 2
[Parameters] ControlNet Softness = 0.25
[Parameters] ADM Scale = 1.5 : 0.8 : 0.3
[Parameters] Seed = 2425342122506827548
[Parameters] CFG = 4
[Fooocus] Loading control models ...
[Parameters] Sampler = dpmpp_2m_sde_gpu - karras
[Parameters] Steps = 30 - 15
[Fooocus] Initializing ...
[Fooocus] Loading models ...
Refiner unloaded.
[Fooocus] Processing prompts ...
[Fooocus] Preparing Fooocus text #1 ...
Segmentation fault (core dumped)

it does recognize my gpu

Fooocus version: 2.5.3
[Cleanup] Attempting to delete content of temp dir /tmp/fooocus
[Cleanup] Cleanup successful
Total VRAM 16368 MB, total RAM 31851 MB
Set vram state to: NORMAL_VRAM
Always offload VRAM
Device: cuda:0 AMD Radeon RX 7800 XT : native
VAE dtype: torch.float32
Using sub quadratic optimization for cross attention, if you have memory or speed issues try using: --attention-split
Refiner unloaded.

[upbeat atlas]

Hi, a few months back I was told about ZLUDA for AMD gpu windows and after installing it made generation so much better. I was wondering if in the past few months any new/faster ways or optimizations have come out for amd windows, or is zluda still the fastest?

[ornate elk]

nice

[ornate elk]

hey, there is now WSL2 (Windows Subsystem Linux) which can use ROCm nativly. Thats the fastest possible speed you can get with your gpu

but the setup is not that easy

[elder mauve]

its not working , see error at bottom

[upbeat atlas]

I've never heard of that, is there a YouTube guide you'd recommend?

[ornate elk]

oh now i see

[ornate elk]

nope. i would stick to zluda. maybe i will make a guide in the future

[upbeat atlas]

I guess zluda it is then, but do make the guide if you find the time, thanks

[ornate elk]

i got lora training working on zluda today

will make a guide for that too

[upbeat atlas]

Thank you 😄

[elder mauve]

Do u know the issue

[ornate elk]

nope, not an linux expert sry, but if i were you i would try to get auto1111 working and not fooocus

[cosmic furnace]

Is forge good & fine to run on AMD now? or nah?

:<

[ornate elk]

it works with zluda

but i would still prefer auto1111

[cosmic furnace]

y . weird indeed. seems like alot faster when models trained with it can go 4-8 steps. But still it has 10x or even 20x less ppl downloading it, comparing to usual version -_-

[oblique vapor]

Hey @ornate elk!
I was told you're an AMD go too. I have got comfy running well, but you have an AMD cpu CrystalWizard suggested I ping you to get everything set up as best as possible. He almost mentioned I should let you know I have:
32 Gigs of ram
8 Gigs of vram Dedicated GPU memory on my NVIDIA GeForce RTX 2070 SUPER
And an AMD Ryzen 7 3800X with an 8 core Processor.

[dim fossil]

https://huggingface.co/stabilityai/stable-diffusion-3-medium-diffusers/tree/main

anyone know how i can install

[hidden herald]

what can i do about this?

[ruby yoke]

install the version of torch it says it can't find?

[hidden herald]

i installed it and got the same message

[dim fossil]

what version did you install?

[ruby yoke]

how many versions of torch do you have installed on that machine right now?

[hidden herald]

torch-2.1.2+cpu-cp310-cp310-win_amd64.whl

[hidden herald]

1

[ruby yoke]

1. is python and everything else in your sytem path and 2. did you reboot after installing

@ornate elk https://x.com/_akhaliq/status/1821927512823173401

AK (@_akhaliq) on X

sd-webui-comfyui

An extension to integrate ComfyUI workflows into the Webui's pipeline

https://t.co/IlxEwCBhbH

Twitter · Aug 9, 2024, 3:12 PM

[hidden herald]

i didnt reboot, i'll do that now

[charred wagon]

https://stable-diffusion-art.com/video-to-video/#Step_3_Enter_ControlNet_settings
Hi, I'm trying to use a ControlNet-based img2img video creator based on the linked tutorial. One of the settings I am looking for is this:

In my installation, however, it looks like this:

Specifically, there is a lack of places to upload an image

This image uploading place is available on txt2img, but not img2img

[honest tendon]

this goes in Models checkpoints right?

[hidden herald]

alr i rebooted and same msg

[charred wagon]

I am on Win11, GPU is an RTX4060, Python version 3.10

[honest tendon]

neither

[elder mauve]

can someone please link a guide on how to install Automatic1111 on linux?

i am trying to run Automatic1111 and it says torch cannot detect my GPU

[solemn moon]

Check the pinned messages in the channel

[elder mauve]

i dont see any ROCM guides

[uncut fog]

I had a question regarding SDXL image generation, mainly with resolution, I found the resolution cheat sheet to use as a general rule of thumb, but I was wondering if it's better to just use 512 x 512 and then use Hires Fix to change them to 1024 x 1024. It seems about the same-ish speed, but i'm wondering what is generally best for quality

[dim fossil]

from what i understand its trained on higher res pic so it has trouble knowing what anything is at lower res

[elder mauve]

i think my rocm drivers are not working in linux , do u know how to fix?

it says it cant use my GPU

[ornate elk]

It will use the image from img2img at the top as input.
If you still want to upload a different image just check the "upload independent Control image" button there.

[ornate elk]

Thats the good model version. Make sure you also use the workflow without clip Text encoders

[ornate elk]

Can you screenshot the full cmd text?

[ornate elk]

SDXL got trained on 1024x1024 resolution. So you will get really bad outputs when used on 512x512.

[uncut fog]

Yeah, when I use HI res fix it automatically scales it up to 1024x1024, but it seems like people also reccomend 896 x 1152, would they average to about the same thing?

[ornate elk]

Yea but you need to be near 1024x1024 as native resolution

768x768 for example works much better than 512x512 for SDXL

And it dont has to be a square.
840x1240 would work great too

[crystal nacelle]

zluda for the win

[uncut fog]

https://cdn.discordapp.com/attachments/1271014983789969428/1271057000977141820/sdxlres.webp?ex=66b7edd4&is=66b69c54&hm=64efb866a894f8f531cc5c84e044e6e87aa1313ccc7f9219b1aa01388a8be206&

This is what I found when searching for ideal SDXL resolutions, most people saying 1152 x 896 and 832 x 1216 to be ideal, I'm just wondering if those are okay, or if there's something more.. ideal

[ornate elk]

These are fine

It still depends on your usecase

[uncut fog]

At the moment just to learn properly, and adjust my settings i'm just using a generic anime gen

[ornate elk]

If I want to make me WQHD Wallpaper backgrounds. I use 1080x1920 and then upscale by 2 with hires fix

[ornate elk]

I would say just try out and play around with it

There is no perfect resolution

[uncut fog]

Yeah definitely, I was more-so just curious about the range of resolutions I should even be attempting at all, thanks for the help!

[ornate elk]

No problem 🙂

[elder mauve]

@ornate elk can u please help me with setting up SD on linux

i think i broke my rocm drivers

where to start troubleshooting?

[ornate elk]

What happens if you try to install them again?

[elder mauve]

it says its already installed

which command should i use to install them , cuz rn i am using the amggpu-install one

[dim fossil]

i just had the same problem

you need python 3.10

not 3.12

[elder mauve]

on base or venv

[dim fossil]

my bad, was for this guy

[inner dawn]

Hello, I have the following problem. I can't start ComfyUi under SD. I installed extention, restarted SD, there in the ComfyUi tab I read that you should specify the path to ComfyUi. I installed it earlier separately. I tried to write the path in each line on the screenshot one by one (F:\A1111SD\ComfyUI_windows_portable\ComfyUI). It didn't work. I think okay - I'll delete the folder F:\A1111SD\stable-diffusion-webui\extensions\sd-webui-comfyui, and install extention ComfyUi again and from the tab ComfyUi install it with the proposed button. But it was no longer there. Please tell me how to fix the problem.

[tardy cliff]

I was able to get SD to work yesterday. I cannot get it to restart today. I get "error code 1". Can anyone help?

I have Python 3.10.6, too.

[ornate elk]

Hey, you installed it the not recommended way by downloading it as zip.
Better use my install guide from the pinned messages.
You need to install Git, that's what the error is about

[tardy cliff]

I'll give that a try. Thank you!

[ornate elk]

np, your card also needs --xformers --medvram --no-half as commandline args

[honest tendon]

doesnt seem to work

this

[ornate elk]

you have to select the right checkpoint

the 16gb one

i had the same issue

[honest tendon]

but that takes quite a while to load an image

like 10 minutes

at ma

max

[ornate elk]

without clip or with clip?

[honest tendon]

No clip

[ornate elk]

hmm ok

[honest tendon]

with clip it takes a long time too

way longer than 10minutes

@tame yoke mom?

[ornate elk]

ah then there isnt much you can do

[honest tendon]

aww man

[ornate elk]

we need to wait for comfyui improvements on flux

[honest tendon]

whats this file for anyways

[tame yoke]

what

[honest tendon]

You have the same name as her

[tame yoke]

as who 😭😭😭

[ornate elk]

these are the smallest flux files. they go into models/unet

[honest tendon]

my mom

[ornate elk]

but they only wok with clip

[tame yoke]

amazing

[honest tendon]

better than nothing

[tame yoke]

is a1111 as fast as comfyui now?

[honest tendon]

best ask in general

[deft hemlock]

hello, so i was trying to enable share=true but then everything messed up, i reverted my changes but still couldnt get it back to normal

im running the old web ui because the new one is really buggy and slow,

launch.py: error: unrecognized arguments: set

[honest tendon]

Is there like a way to allocate some GPU and CPU power to normal browsing, when i load images it laggs so bad with higher end models

[deft hemlock]

HEAD is now at c1713bfe Merge pull request #14689 from AUTOMATIC1111/fix-nested-manual-cast

M webui-user.bat

M webui.bat

venv "C:\Users\DELL\Desktop\xmattar\Ai\stable-diffusion-webui\venv\Scripts\Python.exe"

Python 3.10.9 (tags/v3.10.9:1dd9be6, Dec 6 2022, 20:01:21) [MSC v.1934 64 bit (AMD64)]

Version: v1.7.0-373-gc1713bfe

Commit hash: c1713bfeac461bc28158b66ef8d956a39e296b94

*** Error running install.py for extension C:\Users\DELL\Desktop\xmattar\Ai\stable-diffusion-webui\extensions\openOutpaint-webUI-extension.

*** Command: "C:\Users\DELL\Desktop\xmattar\Ai\stable-diffusion-webui\venv\Scripts\python.exe" "C:\Users\DELL\Desktop\xmattar\Ai\stable-diffusion-webui\extensions\openOutpaint-webUI-extension\install.py"

*** Error code: 2

*** Error running install.py for extension C:\Users\DELL\Desktop\xmattar\Ai\stable-diffusion-webui\extensions\sd-webui-old-photo-restoration.

*** Command: "C:\Users\DELL\Desktop\xmattar\Ai\stable-diffusion-webui\venv\Scripts\python.exe" "C:\Users\DELL\Desktop\xmattar\Ai\stable-diffusion-webui\extensions\sd-webui-old-photo-restoration\install.py"

i dont wana update my webui

its gonna cause a massive slowdown

im using sd 1.5

a1111

[honest tendon]

wait what new update?

[deft hemlock]

few months ago a1111 released a horrible update which ruined the ui and made it more confusing

i dont want to upgrade to that

[honest tendon]

its comfy not a1111

[deft hemlock]

oh nvm then

i use a1111 for 1.5

[honest tendon]

IT STILL LOAD SO SLOWWWWW

i love God

❤️

[ornate elk]

you can disable gpu acceleration in the browser setting

[elder mauve]

[ornate elk]

lol which one xD there wanst any big ui change

[elder mauve]

CS1o where do i start troubleshooting for my linux installation?

It detects my gpu in rocminfo but doesnt detect in foocus or automatic1111

[ornate elk]

whats your gpu? the new update doesnt cause slowdown

[honest tendon]

Gosh you saved my life

[elder mauve]

help me too

[honest tendon]

what the hecks even the point of GPU acceleration in browsing

[elder mauve]

...
Successfully installed MarkupSafe-2.1.5 filelock-3.15.4 fsspec-2024.6.1 jinja2-3.1.4 networkx-3.3 numpy-2.0.1 nvidia-cublas-cu12-12.1.3.1 nvidia-cudnn-cu12-9.1.0.70 nvidia-cusolver-cu12-11.4.5.107 nvidia-cusparse-cu12-12.1.0.106 torch-2.4.0 torchvision-0.19.0 triton-3.0.0
Traceback (most recent call last):
  File "/home/santa/stable-diffusion-webui/launch.py", line 48, in <module>
    main()
  File "/home/santa/stable-diffusion-webui/launch.py", line 39, in main
    prepare_environment()
  File "/home/santa/stable-diffusion-webui/modules/launch_utils.py", line 387, in prepare_environment
    raise RuntimeError(
RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

i am getting this error

i ran this command
TORCH_COMMAND='pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/rocm5.1.1' python launch.py --precision full --no-half

from here ||https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs||

[tardy cliff]

Clearly, I'm a noob. What am I missing? 🤔

[ornate elk]

You need to install Git 64bit: https://git-scm.com/download/win

[tardy cliff]

oops. >.>

[ornate elk]

loading civitai.com without lags xD

[tardy cliff]

you told me that. sorry, i'm at work & trying to multitask. failing beautifully

[tardy cliff]

Double checking I did this correctly. There's a space after =

Also, bc I'm a billion years old.... In order for SD to work, I need to keep both the folder & the cmd open once the browser launches?

[tame yoke]

just the cmd app

and browser to see duh

[deft hemlock]

intel hd 520

ik it sounds funny but aslong as it generates me images

[deft hemlock]

not sure, but its the one where they added tree folder things

do u know how i can fix it?

[ornate elk]

cmd and browser

[tardy cliff]

Got it up and running. Now going to close it all & try to reopen. It was running yesterday. Couldn't open it today

[tardy cliff]

THANK YOU!!!!!!!!!!!!!!!!!!!! It reopened without issue. Thank you so very much!

[ornate elk]

looks good

[tardy cliff]

Thank you again! 🥰

[real willow]

Hi, when I try to use ReActor extension from extras tab on automatic1111, it does not give any error but it does nothing over image too. It gives the source image as result. Do you have any idea? It's the same on API too.

[hollow heath]

i need help with my account,
i got notification that my card cannot be charged,
when i try to manage my subscription and log in it redirects me to -> https://stability.ai/stable-assistant

[deft hemlock]

what about my issue :(

[ornate elk]

It says no face detected. You need to use a image with more than just a face

[ornate elk]

They reverted that

[ornate elk]

To fix your issue remove the restore old photo restoration extension

Then delete the venv folder

And relaunch the webui-user.bat

[plain bronze]

Can I increase my vram by using RAM?

[ornate elk]

Nope

[plain bronze]

https://tenor.com/view/путин-дождь-путинподдождём-пуджумер-gif-18159717

[fair oxide]

how to fix onnx error everytime i use ip adapter face

[zenith sail]

Comfy ui

[zenith sail]

No but you can use your ram instead of your vram.

[ornate elk]

Do you have reactor installed?

[plain bronze]

How?

[zenith sail]

Change your startup flags.

[fair oxide]

reActor?

no i dont have reactor

[elder mauve]

which version of CUDA does zluda require?

[knotty turret]

Anyone know if ReForge is better than Forge?

[half island]

Hoi, need some theories for why my 3090 is chugging all the memory plus pagefile, while this guy is able to just use 16GB vram on his laptop 4090. #🏞｜general-with-images message

[plain bronze]

And how to do that?

[zenith sail]

Google

[ornate elk]

It works only with cu11.8

[elder mauve]

the whole windows installation is broke i think

[ornate elk]

Oh no wait

It only works with 11.8

[elder mauve]

i installed 12.4 ah

[ornate elk]

Yea to new

[elder mauve]

i should just do a new windows install smh

but linux wasnt working either

[elder mauve]

i was getting this

[ornate elk]

I would try that. Because it should work without problems

[elder mauve]

it should work in a fresh install of linux without problems too right?

[ornate elk]

But harder to setup

[elder mauve]

what did i do wrong?

[ornate elk]

what webui do you use?

[ornate elk]

can you show a full cmd log ?

[half island]

comfyui

Here's the models. All i changed from Mojo's was model to fp8

Or wait, was that why? I didn't use clip loader for fp8 here?

[ornate elk]

you dont need clip for the 16gb or 23gb model only for the 12gb model

[bold field]

Using a bot on my raspberry pi to use the stable diffusion that's on my pc. The pi as on same network as my pc. It works fine when I'm hosting bot on my pc, but on the pi it doesn't work - it can't access to it. Why? and how can I fix that?

Error: Cannot connect to host 127.0.0.1:7860 ssl:default [Connect call failed ('127.0.0.1', 7860)]

[ornate elk]

Have you --listen in the webui-user.bat?

[bold field]

Yes. However now the error is different. I think now I just need to put the right IP in my code.

I'm not sure what ip to put.

[ornate elk]

If you want to access the SD you use on PC from the pi you need to input the PCs IP + the port 7860

[bold field]

Yes. Now I'm tryna figure out what'd be the PC ip

☠️

[ornate elk]

Open up a cmd and type
Ipconfig

[bold field]

Yes

Then

[ornate elk]

Ipv4

[bold field]

Ok I did. now i get this: An unexpected error occurred: [myipv4here]/sdapi/v1/txt2img

[knotty turret]

@ornate elk i can use 2 loras with regional prompt?

[ornate elk]

Yes

[knotty turret]

can you help idk why getting bad

results

its not working well

idk why

have follow the guide but look this

🙃

[ornate elk]

How does your txt2img settings look like ?

[knotty turret]

my prompt?

[bold field]

anyway if anyone can help me with this i'd be glad

[knotty turret]

you mean my prompt? or everything i can send u the pic if u want too

[ornate elk]

Everything

[ornate elk]

Which api do you use?

[knotty turret]

tried to generate 832x1216 still not what i expect

only 1 character too

[ornate elk]

Strange. The resolution sliders of regional prompter are missing

Also Your using old forge

[knotty turret]

forgot to send matrix too here:

no i using ReForge

[ornate elk]

Ahh

[knotty turret]

reloding my stable to see if works just 1 minute

its taking sooo long...

let me just open it again

[knotty turret]

this?

what i can do to fix that?

[ornate elk]

Click visualize and make template.
Then also check "use common prompt"

[knotty turret]

generate now? or change something more

[ornate elk]

Dont check use base prompt

[knotty turret]

ok

[ornate elk]

Only use one of those options

Not both together

[knotty turret]

anything else?

forgot to say sometimes my generations crash/freeze the UI and i have to reload there fix for it?

[ornate elk]

Can't compare to my settings rn. Not at the PC.
But I can check tomorrow

[ornate elk]

Yes probably

[knotty turret]

Ok man

thx

i will make a try now

[knotty turret]

The visual still bad but this time worked better than before

2 charac🙏 at least

just need to find way to better visual now

[half island]

I don't have the 16 and 23GB model. I can't even run 12GB model without running out of vram xD

[ornate elk]

What's your GPU?

[half island]

3090

[gloomy zodiac]

Hello! Has anyone else tried Stable Fast for 3D gen? I asked for help on github but has someone else dealt with this?

[terse walrus]

So I have this nice image here and I was just wondering if anyone knows, like once this is produced in comfyui, how I can maybe select a certain part of it and then rerun the prompt so that it fixes that part? Pretty much this looks perfect except that the legs are kinda fused together? Is this possible in comfyui natively or not?

[fair oxide]

i got rid of that issue (i think) but i have ran into another

[safe inlet]

search for a tutorial on "Inpainting"

[terse walrus]

I did actually, should've deleted that message but I didnt like it

[honest tendon]

https://github.com/nod-ai

@ornate elk

u heard of this shi?

[bold field]

I probably sound dumb but how do I check

Wait I just realized why it didn't work.... 😂

I forgot the "http://"

I'm so dumb bruh lol

[bold field]

Another question. I want to use this extension. Where should I put it?

[slate creek]

theres extension folder on stable diffusin folder, i think theres the place fot that

im trying to use deforum but cant generate the video, how to fix this guys?

[bold field]

well i did that

however it caused problems

[slate creek]

what problem?

[bold field]

with it the generation just.. doesnt work

• this warning appears

[slate creek]

try 'force_download=True'

[bold field]

wheres that at

[slate creek]

text suggest to use 'force_download=True'

[bold field]

yes i mean where do i put that

wdym

[slate creek]

ouh

on the prompt

i guess

[bold field]

u sure

that sounds weird xD

u mean as a param?

hmm

[slate creek]

i not sure :v

[bold field]

also now i see, that theres an error with censor.py

it just takes buncha time to load the error

ssl error tho so its my side's problem xD

[slate creek]

so much error bro

[bold field]

😔

well ik kinda how to fix the ssl error

[slate creek]

im not that good at this ai things, i just follow what prompt say lol

[bold field]

mhm

[slate creek]

can you give solution for my problem bro?

[bold field]

not sure i can xD

[sharp ridge]

what python version does comfy ui use?

[deft hemlock]

Thanks man it worked

[deft hemlock]

Still don't wana risk it

I broke sd many times trying to downgrade my version

[bold field]

Ok so yea if someone can help me with installing NSFW filter I'd be more than happy

[ornate elk]

Yes, but currently its bugged and not worth to use

[bold field]

And I'd like to know how to properly solve this:

C:\stable-diffusion-webui\venv\lib\site-packages\huggingface_hub\file_download.py:1150: FutureWarning: resume_download is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use force_download=True.
warnings.warn(

[ornate elk]

Thats a warning that can be ignored

It doesn't do anything

[bold field]

So I'll just let it be

[bold field]

Ok I did and I think it works

Now if I try to create something with nsfw prompt,
it just doesnt generate anything.

I that how it should work?

O no wait

nvm it does generate it..

[woeful frost]

i tried several time to create an account to acces stable diffusion when the public release was announce i conrtaced support and even submited ticket but i had no response not even a mail where can i find help ???

[bold field]

Ok the filter does seem to work now

@ornate elk btw, with --listen, it doesnt open it on broswer anymore, i do want it to tho

how can i enable that

[ornate elk]

--autolaunch

[bold field]

thanks!

[ornate elk]

3.11

[bold field]

I wonder, is there a way to now turn it into an error the bot can use?

🤔

[ornate elk]

Here you find the info:
#🗣｜artisan-support-feedback message

[sharp ridge]

does lora actually work with schnell

[fair oxide]

hello guys , tell please its possible make text in promt zone more bigger ?

[slate creek]

how about my problem guys? i dont know how to solve this

[slate creek]

did anyone know how to make this kind of style ai ?

[fair oxide]

@ornate elk do you know this error?

[honest tendon]

flux rn takes quite a while

[dense sandal]

Should I have studio or game drivers for nvidia?

[honest tendon]

interesting it takes longer and longer after every image

[ornate elk]

Would need to see the full cmd log

[ornate elk]

Game drivers

[ashen phoenix]

I can never update my a1111 what can I do

[fair oxide]

here

[ornate elk]

Reinstall or reinstall without using Stability Matrix

[ornate elk]

Did you also used the face id lora?

Make also sure your GPU driver is updated.

[main bluff]

am trying to use flux but nothing is showing up

[ornate elk]

Automatic1111 webui doesn't support Flux right now. But mostly with the next big update

[main bluff]

when i try to use comfy i have this error

i know am very far behind , i just got working on AI like yesterday

[fathom rose]

what are the recommendations here?

[livid dock]

Anyone can help me? I already reinstalled animatediff but still, my cmd says 100% but my webui stops at around 90-95% :/

[ornate elk]

That means you used the wrong flux model for this workflow

Oh wait you used an other model

Maybe try Dreamshaper v8 as model

[ornate elk]

You paused the cmd by clicking in it.
Press space to unpause the cmd

[ornate elk]

What's your GPU?

[main bluff]

nividia 4070

[ornate elk]

Okay, very strange that torch wasn't found.

But if your new to ai I would not recommend starting with comfyui

Better use Auto1111

[main bluff]

i got all excited to use flux

ok, will stick to auto1111 for a bit

[livid dock]

still the same, didnt click anything and stop at 95%

[ornate elk]

Have you ffmpeg installed and in the path?

Maybe it struggles making the mp4 version

[ornate elk]

You can try but something with your comfyui is wrong.
Maybe reinstall it and then install the Comfyui manager again too

[main bluff]

i think its because i am using python 312

it needs to be maybe less than that for torch to work

[livid dock]

i just installed ffmpeg again, but i only need to place it into eviroment variables or?

its still not working tho

or do i need to place it into stable diffusion folder somewhere?

[fair oxide]

yeah

[fair oxide]

i think mines already up to date

my laptops pretty old its a latitude e5570 intel i5

[ornate elk]

Yes for comfyui only python 3.11 is supported

Or 3.10.11

For Auto1111 too

[livid dock]

cmd alwas has 1 step to less to get to 100% dont know why

[ornate elk]

can you share the full cmd log and whats your gpu?

[livid dock]

3080

yeah wait

[sharp ridge]

https://github.com/lllyasviel/stable-diffusion-webui-forge/releases/tag/latest
Press any key to continue . . .```
 i got this problem when installing forge, how can i install it?

i have tried another version, yet still the same

[ornate elk]

your running and older webui version and your torch version is outdated too and your using --opt-sdp-attention instead of --xformers. (both have the same speed but xformers uses slightly less vram).
A bigger problem is that you used Context Batch Size 16 for AnimatedDIff which explains the freeze.
A 3080 has only 10gb of vram which would require a context batch size of 10 or 8

[ornate elk]

whats your gpu?

[sharp ridge]

3050 laptop

[ornate elk]

make sure your gpahic cards driver is updated

[livid dock]

thanks alot, i will try to update everything 👏

i totally forgot about the batch size 💀

[fair oxide]

@ornate elk same issue appear after i updated my drivers

[thorny iron]

How do you train for two specific unique characters, then put them in the same picture?

[ornate elk]

have you tried installing the latest version of auto1111 and test if it worked there?
also whats in your wbeui-user.bat? do you run it in cpu mode?

[thorny iron]

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers

call git pull
call webui.bat

I use this maybe it will help him? :?

[fair oxide]

i got latest a1111 lemme check the webui

i have the command arguments as --api --skip-torch-cuda-test --skip-python-version-check --no-half --gradio-allowed-path

[ornate elk]

ahh okay

im not sure if face id can work without an nvidia or amd gpu

[fair oxide]

oh

alright

is there an alternative to face id

that works with intel

[ornate elk]

the normal ip-adapter face plus model

[potent arch]

hey guys im in trouble with control net i can use pose

https://ibb.co/mb8p3N0

[fair oxide]

so no v2

[potent arch]

ony one help me

where am i wrong

[ornate elk]

cant zoom into that image, just paste the screenshot here and not in an image upload thing

[potent arch]

okey

there is

[ornate elk]

your trying to use a sd 1.4 (1.5) controlnet model with an sdxl model

that wont work

[potent arch]

whic one work

[ornate elk]

these ones, there is an t2i adapter for openpose and a normal openpose model at the bottom:
https://huggingface.co/lllyasviel/sd_control_collection/tree/main

[potent arch]

thx mate i will try it

[sharp ridge]

Am i correct? I'm trying for nf4 work flow on comfy

[ornate elk]

looks correct, where did you get the nf4 workflow from?

[sharp ridge]

i created it

but it seems take so much time compared to fp8

[main bluff]

did someone see this before ?

[fair oxide]

@ornate elk i realized that im running on cpu, how do i change that

[ornate elk]

does your laptop has a dedicated gpu from nvidia or amd ?

[fair oxide]

no

it has intel graphics 520 or something

[ornate elk]

then you cant

[fair oxide]

oh

ok thanks

[ornate elk]

sry and no problem ^^

[ornate elk]

you can add --opt-sub-quad-attention to get a bit more speed

[fair oxide]

alright thanks

do i add on to it

or delete something from the original

[ornate elk]

idk why you have --api or --gradio-allow-path in there, but if you dont know too then remove them

[fair oxide]

alright

[ruby yoke]

check to make sure that numpy is installed and if so what verison it is

[potent arch]

yes it works thx mate

[ornate elk]

can you share me that for testing?

[sharp ridge]

sure!

[ornate elk]

thanks!

[sharp ridge]

can you share your testing, i cant even run it, so slow

[hot thorn]

hey im tryna download stable diffusion locally and stuff and rn its stuck on this and keeps staying as this

whats going on with it?

[ornate elk]

i dont have the nf4 node 😮

[sharp ridge]

https://github.com/comfyanonymous/ComfyUI_bitsandbytes_NF4

you can get it from here

[ornate elk]

try click into the cmd and after that hit spacebar

[ornate elk]

awesome thx

[hot thorn]

o tank you!!

sorry im

kinda dum

so like

:3

thank u tho

oki i now leev server since thats all i came for and stoof

[ornate elk]

oh you dont need to leave you can get a lot of helpful information here

[ornate elk]

cant get it to work, maybe because the amd fork of comfyui first needs an update

[sharp ridge]

https://github.com/comfyanonymous/ComfyUI_bitsandbytes_NF4/issues/3#issuecomment-2282837256
check this

[ornate elk]

yea i did that right now before reading this sry

worked

i i still need to update my comfyui

so i have to wait for the amd fork

[knotty turret]

@ornate elk Hey can you tell your regional settings now?

just to compare

just see if there anything i missing

[ornate elk]

oh hey

i almost forget but i found my settings.
check this out: (my regional prompter settings)
#🤝｜tech-support message

[knotty turret]

Ok👍

The quality its good did u use highresfix?

[sharp ridge]

can i know what version of python do you use?

[ornate elk]

3.10.11 64bit

[ornate elk]

yep

and adetailer

[sharp ridge]

after updating dependencies, my rendering became so slow

[sharp ridge]

do you know command that make render faster?

[ornate elk]

in comfyui?

whats your gpu again?

[sharp ridge]

yes, 3050 laptop

[ornate elk]

in comfyui there isnt much you can do

it already uses the best options for performance

in auto1111 there would be a lot of stuff you can change to get even sdxl models working or upscaling

[sharp ridge]

ohh, thank you

[main bluff]

Thank you for your help, it works now , i can sleep feeling good about my self today 😄

thank you CS10 for your help

[ornate elk]

no problem!

[sharp ridge]

for me personally, i don't think nf4 is faster, it takes too long on scnellnf4

[ornate elk]

Full cmd log pls

[stark yacht]

https://pastebin.com/gyw6Nb1h

[ornate elk]

The important part is missing at the start.
But I see your on AMD

[stark yacht]

I really want to know

[ornate elk]

Follow any of my Stable-diffusion Webui AMD guides to get it working.
If your GPU has 8gb vram or more you should use the ZLUDA Guide. If not try the Directml Guide.

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

[stark yacht]

16 gb

[ornate elk]

Yea then definitly go for Zluda

Automatic1111 with ZLUDA

[stark yacht]

i try

[ornate elk]

Let me know if you need help setting it up.
But I'm off now so I can help in 7 hours again.

[stark yacht]

Good thanks

[valid hollow]

Hello. I have a problem that I can't do anything about. Almost every model looks terrible when I generate images. In any quality, in any VAE, with any settings. Disgusting crooked images. Can the quality of the generated images depend on the power of my PC?

[ruby yoke]

change your sampler and scheduler

[long blaze]

How i can train a lora? Anyone have a good tutorial?

[frozen sandal]

https://civitai.com/articles/3921/this-is-how-i-create-and-train-loras pretty good one, you can find others on civitai. they also have an on-site lora trainer you can use with 500 buzz (site credits) you can accrue buzz daily just by being active on the site

[distant plover]

Module not found: torch when I install and try to run live portrait. anyone know about this error?

[long blaze]

he

[long blaze]

the lora train of civitai is as good as a local training with kohya?

[ornate elk]

I don't think so

[ornate elk]

In which software?

[sharp ridge]

help

[valid gorge]

After updating ComfyUI, I lost again the history / image view bar. Can't find a way to get it back. Can someone point me the setting please?

[safe inlet]

on top of queue prompt button the square

[vivid lichen]

im trying to launch the .bat to get onto the website put all it shows on the cmd is this and it doesnt open the tab

[sharp ridge]

any help?

[valid gorge]

I think I'm missing that one?

[safe inlet]

i have the button there, maybe from something else, i dont know

[valid gorge]

I used to have two of those, now both are missing.

[safe inlet]

probably this setting for manager https://github.com/ltdrdata/ComfyUI-Manager?tab=readme-ov-file#security-policy

[sharp ridge]

thanks man

[ornate elk]

What's your GPU?

[vivid lichen]

rtx 3060

[ornate elk]

First make sure your GPU driver is updated.
Then open up a cmd and type
Pip cache purge.

Then Edit your webui-user.bat
At the line commandline_args=
You add: --xformers
Then save.
Then delete the venv folder if you have one.
Then relaunch the webui-user.bat

[vivid lichen]

okay thanks

[worn bane]

How i can make image like artkoikoi it's looks like anime

[stark yacht]

worked!

[ornate elk]

Perfect!

[fierce hollow]

Question, who has worked with flux before ? How to make extremely detailed and realistic images ?
An example of this (As it is reportedly done in flux)

[lost crescent]

Flux error using nf4

[lost crescent]

Use terms to bring in classy lighting. Try this at the start of your prompt - learn how to prompt like this to get awesome lighting - microscopic ::1 closeup ::1 backlit jewel tiffany ::3 intricate needlepoint closeup murmuration ::3 underwater lighting ::2

You can add this for more explosively creative lighting - Neon-Infused Expressionism, Desert Mirage: Surreal desert landscape, Souza's artistic style, neon lights and ethereal glows, blending reality with fantasy, iridescent hues

[fierce hollow]

The point is that flux does not support "DPM++ SDE" But supports only Euler, so the image is not fully realistic.

[lost crescent]

Still see how these prompt examples change the lighting

I am not using euler - I am using uni_pc_bh2

[fierce hollow]

Can I email you and send you some examples ?

[lost crescent]

OK

[fierce hollow]

Hmmm... I'll give it a try

[lost crescent]

Olivio Sarikas uses that uni_pc_bh2/sgm_uniform https://youtu.be/tXO6SJ-6Eb8

[fierce hollow]

Forge. Which folder to load the new samplers into?

[boreal vector]

I tried to install xformers using this guide but im getting this error

https://youtu.be/UP4L_WlMKuE?si=BWpgLSxKZOUImF2S

[ornate elk]

Can you share the complete cmd log?

That error mostly means your running an old torch version. But I want to check the log. Hopefully your not running a webui version from over a year back xD

[boreal vector]

I found this solution on reddit
https://new.reddit.com/r/StableDiffusion/comments/135d9dc/how_to_fix_xformers_error/

so its reinstalling the venv folder now

[ornate elk]

Yea that will mostly fix it. But still check your webui version

[boreal vector]

how do i check that

[ornate elk]

The first few lines of the cmd shows it

Or later at the bottom of the webui in browser

[boreal vector]

Creating venv in directory C:\Users\ashir\stable-diffusion-webui\venv using python "C:\Users\ashir\AppData\Local\Programs\Python\Python310\python.exe"
venv "C:\Users\ashir\stable-diffusion-webui\venv\Scripts\Python.exe"
fatal: detected dubious ownership in repository at 'C:/Users/ashir/stable-diffusion-webui'
'C:/Users/ashir/stable-diffusion-webui' is owned by:
'S-1-5-32-544'
but the current user is:
'S-1-5-21-1342013206-926425775-1834896619-1001'
To add an exception for this directory, call:

    git config --global --add safe.directory C:/Users/ashir/stable-diffusion-webui

fatal: detected dubious ownership in repository at 'C:/Users/ashir/stable-diffusion-webui'
'C:/Users/ashir/stable-diffusion-webui' is owned by:
'S-1-5-32-544'
but the current user is:
'S-1-5-21-1342013206-926425775-1834896619-1001'
To add an exception for this directory, call:

    git config --global --add safe.directory C:/Users/ashir/stable-diffusion-webui

Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: 1.7.0
Commit hash: <none>
Installing torch and torchvision
Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/cu118
Collecting torch==2.0.1

does this help?

[ornate elk]

Yep

Your running webui version 1.7.0

Its from last year June or july

[boreal vector]

how do i update

[ornate elk]

Latest version is 1.10.0

[ornate elk]

The log says commit hash none + dubios ownership detected.
That means you maybe moved your webui.
You better clean install the latest version and then move your models over

You find a quick install guide for Nvidia in the pinned messages of this channel.

[boreal vector]

yeah, I just added the exception in cmd for now, I will do a clean install later

[ornate elk]

My install guide also explains which commandline_args are needed depending on the vram of the GPU

Xformers is always recommended for Nvidia

[fierce hollow]

Forge. Which folder to load the new samplers into?

[boreal vector]

does this work now?

[ornate elk]

Git pull won't work as your webui has no commit hash. Means it has no connection to the original git repo.
When you reinstall you use
Git clone command. With that you can always update later if you don't move the webui around

[lost crescent]

When I d/load the extra-samplers pack in A1111 - in which folder do I put them at all?

[boreal vector]

Exception: ControlNet model control_v11p_sd15_openpose cab727d4 is not compatible with sd model(StableDiffusionVersion.SDXL)

is the in compaitable with the SD or checkpoint?

[tropic arrow]

Hello, I would like to ask where i can cehck my cuda version.

Also thsee stuff on the second image is necessary or not really, i am using Forge version.

nvcc command is not working for me. This mean cuda is not installed on my PC ?

[ornate elk]

Hey, update your GPU driver.
Then delete the venv folder of forge
Then relaunch

[tropic arrow]

Also i think about do worth installing studio driver ?

[tropic arrow]

The first command is solved, it said not recognize the nvcc --versions, because the cuda tool kit was not isntalled, now the nvcc is listing the versions, howevwer the second image still showing the same ( the SD is working fine i just want to know what are they that i highlight with red color.

[ornate elk]

Nope, better use the gaming drivers

[vivid ridge]

Hello, guys I am working with SD upscaler "stabilityai/stable-diffusion-x4-upscaler" and find some unexpected case.
I am trying to upscale different types of images and when I upscale images with large amount of black color model produces some random artifacts, maybe some of you know how to fix it?
I am also providing example of it 1st is basic image and 2nd upscaled image totaly filled with artifacts.

[fair oxide]

i downloaded what people were saying was the default flux workflow but im confused because i have none of these files

do i need to put the flux file in the unet folder instead of checkpoints? for some reason?

what is the DualCLIPLoader and where do i even find those files?

last time i tried to follow a workflow like this i downloaded all the file names and put them in the respective folders and it just didnt work

[sand dawn]

Anyone out there very good at Gradio, that could help me debug an issue with one of my Buttons which works except not in Gradio 4?

[fair oxide]

oh the model also doesnt work normally

everything is just so complicated i feel like i need to go to school for 10 years to know what the basics of any of this is

like the "how to use flux locally" just links you to tons of files and folders where you already have to know how to use all of them to use them

make a guide that doesnt require extensive prerequisite knowledge challenge

how are there people running this on reddit with 3gb when everything ive seen so far says "prerequisite have at least 30gb vram"

im so confused and dumb

[sharp ridge]

which settings produced image faster on forge

[sand dawn]

Its dependent on your hardware. Read OP here https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/981

[valid hollow]

Hi all. I noticed that some YouTubers have some kind of SD parameter that allows you to see the image generation in real time. That is, the picture changes before our eyes and every step is visible, in contrast to quality jerks: 1) noise, 2) image details 3) final result

[fair oxide]

afaik theyre setting up preview image along various steps

one workflow i looked at had a video file rendered and saved of the image previews but that seems really excessive to me

[ornate elk]

Looks normal

Do you get an error?

[fathom rose]

what's this and how do i fix it? happening when i'm trying to start forge using webui-user.bat file but when i'm starting with the run.bat file it's opening without any issue

[sand dawn]

reinstall?

[fathom rose]

full reinstall or venv?

[sand dawn]

I'm not familiar with your file structure... personally, I have seperate directories for each webUI

[fathom rose]

I'm just trying to re install old version and setting file directories in command arg

[sand dawn]

By old version, you mean you are downloading and extracting this package here?
https://github.com/lllyasviel/stable-diffusion-webui-forge/releases/tag/previous

[fathom rose]

i suppose to be so if it's this one

[sand dawn]

The package I linked which is the most recent 'stable' version is f0.0.17v1.8.0rc-1.7.0

[south vigil]

yo im following a tutorial for installing comfyui with flux, comfyui seems to be right instlled but i got stuck when ever i put the schnell pt8 flux version into comy ui ill get only 5-6 windows open instead of around 15
https://www.youtube.com/watch?v=3kljJNosQ1I&t=274s 4:18

[fathom rose]

Do you have it still? Can you show me the version info on the bottom panel?

[distant plover]

Live portrait

[heady pine]

@ornate elk Hey there, sorry for the @, but is there a fix for the batch count when it comes to controlnet? After about a month, I tried making some new images with the batch count, but I came across the same problem I did before. I'm wondering if there is a way to resolve this problem by maybe going into the webui-user.bat and pasting something. I'm not an IT expert, but surely this can't still be a problem, can it?

[fierce hollow]

When I d/load the extra-samplers pack in A1111 - in which folder do I put them at all?

[sharp ridge]

how do i save in jpg on forge?

[ornate elk]

Hey, what exactly is the issue?
Controlnet works with batch count. There was a bug a few months ago but that should be fixed

[heady pine]

Sorry for the late reply. Looking at the old image I sent you and this one, they are both different, but the same problem is happening. Like last time, I have not changed any of my settings compared to when it used to work perfectly for me.

[ornate elk]

whats your gpu?

[heady pine]

My GPU is a GeForce RTX 2080 Ti

[ornate elk]

then you only need --xformers --medvram-sdxl --no-half-vae

in the webui-user.bat

[heady pine]

I wanted to make sure I gave you a definitive response before repling, and unfortunately, after pasting both that into the webui-user file and applying the Upcast cross-attention layer to the float32 setting, I am still facing the same problem.

I have tried applying the Upcast cross attention layer to the float32 setting without what you told me to past, and I have tried pasting what you told me to into the webui-user file while also not applying the Upcast cross attention layer to the float32 setting; both have ended the same way.

I'm unsure what the problem could be; my settings, apart from applying the above, have stayed the same since the time that it worked perfectly.

[ornate elk]

can you show me the whole txt2img stuff you tried?

everything, controlent settings, resolution, model etc

then i can easily see where the issue is

also a full cmd log would help

[heady pine]

Give me a couple minutes and I will get you that info

[coarse swan]

silly question, but where does the tag.txt output goes for the Tagger extension?

[crisp prawn]

anyone use Reactor with Forge or A1111? the dropdown on my face models option has disappeared. anyone know why?

[deft hemlock]

i deleted the venv folder now i got this

i am a silly indiviual

[oblique meadow]

You need 30gb vram if you are running at fp16 precision with no cpu offloading.

I would recommend using the nf4 version of flux, that will save a massive amount of vram and fit in 12gb vram and most likely less. You can use cpu offloading to save even more vram, maybe lower it all the way to 6 or so.

People running it in 3gb vram are probably running most of flux in shared or cpu which will make it extremely slow.

[vapid widget]

how can i know if stabble diffusion used correctly my vram , i think SD use less , i have a rtx4070 super and 32gb ram

[restive sparrow]

Not a mega deal but I noticed my A1111 stopped automatically opening in browser. (sometime after installing Opera GX) anyone know perhaps the cause?
EDIT: fixed just disable and enabled the setting in settings.

[fair oxide]

how do I use this?

[restive sparrow]

I swear im getting slower generation speeds on upscaling today. im scarting my head

[ornate elk]

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

[ornate elk]

you need to do this to fix it:
put albumentations==1.4.3 and pydantic==1.10.15 in the requirements_versions.txt file
Then relaunch or delete the venv

[ornate elk]

Same way like a lora

[ornate elk]

Make sure you have set hires steps to 10

[restive sparrow]

no this was with, i2i upscaling with tileddiff/var

Got my speeds back, didnt do anything, 🤷‍♀️
Perhaps the base resizer in i2i? Ive seen before where it interferes with the Tiled Dif upscale parameter. So i set the deufaly rezier to 0 to make sure its off

[deft hemlock]

it worked now
thanaks man

[boreal vector]

how to use the openpose stick figure images?

i drop them in controlnet image section and they dont load

do i also need to uplad json files?

[tropic arrow]

there is something that i not understand well, I am using forge version, automatic1111

My goal is would be generating different interaction between two character, without any addons like control net and latent couple or regional prompt.

I am used to think is not impossible because sometime i get decent results but most of the time two characters often merging with each other, so why this happening if it does able to generating two characters some time.

Is there some setting that can me help out ?

[ornate elk]

You have to set preprocessor to None when using stick images

[boreal vector]

👍

[final rivet]

Hi all, using automatic1111 I am installing some extentions which install fine and I can see them enabled in the extentions tab however, i dont see them in the dropdowns in webui to use. Any ideas?

[final rivet]

I restarted the server and still the same.

[ornate elk]

Hey, mov2mov is outdated and needs a manual fix to get it work.

Mov2mov can be fixed with that:
https://github.com/Scholar01/sd-webui-mov2mov/issues/146#issuecomment-2054157728

[final rivet]

hey, thank you. its not just mov2mov but any.. Do have any suggestion that you know works 100% with v1.10.1

[ornate elk]

Please show the extensions tab with the extensions you have installed

[final rivet]

I dont have much as I did a clean install again thinking it will fix it. so I am just testing one by one to see if its the eextention or my UI

[ornate elk]

Okay that screenshot doesn't help xD
Try install adetailer and sd-webui-controlnet,
both should work normaly and are good extensions

They should show up after a restart

[final rivet]

ok thank you. adding: https://github.com/Bing-su/adetailer.git

[final rivet]

ocntrol net is only for upscaling or does it do motion/animation also? sorry new to this.

btw, restarting the UI so lets see. 🤞

[final rivet]

yes! both are now showing up.

[ornate elk]

Controlnet is for everything regarding to image manipulation. Like giving a character a specific pose or transform an image into anime line art etc or sketch

[final rivet]

so the issue is the compatibility rather than webui issue.

[coarse swan]

@ornate elk ello, could I ask for:

1. an example file of a caption .txt for SD lora trainers that works with 1.5

2. a caption .txt that uses two types of tags? I want to make a tag to Keep and tag to Exclude.

[ornate elk]

You need to check if a extension is updated. Not every extension in the list is updated so they won't work with the latest Webui version

[ornate elk]

Hey, sry I don't have that

[coarse swan]

alright 😔

[fair oxide]

anyone knows how to gererate eyes only ?

[valid gorge]

Any Forge users here? How can I make Forge to see all my checkpoints loras etc from my A1111 folders?

[ornate elk]

Hey, you have to edit the webui-user.bat of Forge like this:
`@echo off

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

@REM Uncomment following code to reference an existing A1111 checkout.
set A1111_HOME=D:/Programme/AI-Zeug/stable-diffusion-webui-amdgpu
@REM
@REM set VENV_DIR=%A1111_HOME%/venv
set COMMANDLINE_ARGS=%COMMANDLINE_ARGS% ^
--ckpt-dir %A1111_HOME%/models/Stable-diffusion ^
--hypernetwork-dir %A1111_HOME%/models/hypernetworks ^
--embeddings-dir %A1111_HOME%/embeddings ^
--vae-dir %A1111_HOME%/models/vae ^
--controlnet-dir %A1111_HOME%/models/controlnet ^
--lora-dir %A1111_HOME%/models/Lora

call webui.bat`

You only need to change the path to the base Auto1111

[valid gorge]

Perfect, thank you!

[ornate elk]

Copy all and change the line set A1111_HOME=

Only this and the rest will work 🙂

[valid gorge]

Thanks! 🙂

[chrome rampart]

hi all. question how can i change server in comfy ui manager ?

[valid gorge]

Gettin an error for some reason, hmmm

[gleaming mantle]

SD3 are same specs req as 2? I ran on 2 for this entire time i haven't used it in a while so i thought maybe i'll still be fine?

oh no 1.5

[vocal burrow]

--ckpt-dir blablabla - blablabla

General rule of thumb, dont put spaces and funky non latin characters in your folder names

(here you can simply put quotes around it tho)

[vocal burrow]

more like sdxl than sd2

[gleaming mantle]

aaaaah woah

i was just gonna get a new clean 1.5 install but maybe i can run 3

[valid gorge]

Thanks, will give it a try.

[valid gorge]

https://tenor.com/view/love-you-yay-dance-gif-14729481

it's working haha @vocal burrow

[ornate elk]

Show your bat please

Oh its working okay

[valid gorge]

Yes, working now. Got help from Aryetis :), too long folder names with spaces, I placed the folder path in " folder path" and it worked.

I always forget not to use spaces etc in windows when creating folders and files

[coarse swan]

Is there a way to automate the tags you get from Tagger into a caption.txt? Or do I have to do it by hand

[ornate elk]

Normaly the lora or model training tools support auto captioning tools that place a .txt file in the same folder as the images

Khoya and OneTrainer do this

[ornate elk]

Please don't try to install the webui in the downloads folder.
Follow the install guide from the pinned messages

[gleaming mantle]

yeah i realized that once i saw the screenshot, okay thanks its been too long

[raven reef]

anyone using the checkpoint flux for comfyui? i have this same issue here https://github.com/comfyanonymous/ComfyUI/issues/4329 if anyone knows a fix 😭

[safe inlet]

This answer is in that thread, That is not a checkpoint model. You can not load that model via Checkpoint Loader node.
You have to use Load Diffusion Model

[raven reef]

im using the checkpoint model one https://comfyanonymous.github.io/ComfyUI_examples/flux/#simple-to-use-fp8-checkpoint-version

[ornate elk]

Use the 16gb flux dev model

The 8gb one is a unet model

An then use the right workflow

[raven reef]

im using the checkpoint model and the workflow from here https://comfyanonymous.github.io/ComfyUI_examples/flux/#simple-to-use-fp8-checkpoint-version

[raven reef]

updated comfyui and all python dependencies and restarted. works now 🤷‍♂️

[formal trout]

I was trying to run forge using this
https://github.com/continue-revolution/sd-webui-animatediff/blob/forge/master/docs/how-to-use.md#you-have-a1111-and-you-know-git

git remote add forge https://github.com/lllyasviel/stable-diffusion-webui-forge git branch lllyasviel/main git checkout lllyasviel/main git fetch forge git branch -u forge/main git pull

but I got

hint: You have divergent branches and need to specify how to reconcile them. hint: You can do so by running one of the following commands sometime before hint: your next pull: hint: hint: git config pull.rebase false # merge (the default strategy) hint: git config pull.rebase true # rebase hint: git config pull.ff only # fast-forward only hint: hint: You can replace "git config" with "git config --global" to set a default hint: preference for all repositories. You can also pass --rebase, --no-rebase, hint: or --ff-only on the command line to override the configured default per hint: invocation.

or I might install and then activate the venv of Automatic

[fair oxide]

so after i updated dynamic thresholding, the enable button seemed to have vanished and im not sure why, im using SD1.5, AMD GPU, automatic1111 but the directml version. any help would be helpful.

[ornate elk]

Hey, what's your GPU?

[crude trellis]

how do I install extensions in stable diffusion?

[ornate elk]

via the extensions tab

then click on available, then on "load from", then you get a list

[crude trellis]

I just installed one, was watching a video about it but couldnt find this area

this should be a tab where you can change settings about the extensions

[ornate elk]

at the bottom of settings on the left

[crude trellis]

but fsr its not here on mine

[ornate elk]

is your webui updated?

what version are you on?

and what extension did you instelled?

[crude trellis]

https://youtu.be/hXsxBKYMGaA?si=l3VVs8fHmQ4F1ayU

those

[crude trellis]

how can I check this?

[ornate elk]

at the bottom in browser

[crude trellis]

those are the options avaiable for me

[ornate elk]

your one version behind

[crude trellis]

it should have this one

its supposedly where I would configurate the extensions settings

[crude trellis]

how do I update?

[ornate elk]

with git pull

run that in a cmd inside the stable-diffusion-webui folder
or add git pull into the webui-user.bat

[ornate elk]

delete the 1.0.1 from the extensions folder

it was not installed correctly

and you dont need the stage manager probably

[crude trellis]

just to be sure

[crude trellis]

how do I do that?

[fierce hollow]

When I d/load the extra-samplers pack in forge - in which folder do I put them at all?

[fierce hollow]

@ornate elk help..

[crude trellis]

which file is that?

webui-user.bat?

[ornate elk]

Yes

[fierce hollow]

How do I configure ComfyUI to work with flyx models?

[fierce hollow]

When trying to generate - this error, please help me

[daring tulip]

Just got forge to use SDXL and the vae dropdown menu that appears for SD isn't there for SDXL, despite enabling it. Did I do something wrong?

[daring tulip]

Should add this:

[fair oxide]

my GPU is AMD Radeon RX 7800 XT

@ornate elk

and i did just update and it did this. pretty sure it took from the actual SD github and not the direct-ml one

ok turns out the developer of dynamic thresholding made the extension automatically activate instead of having it be manually activated

[daring tulip]

Just to be clear, this was showing I had the quicksetting active but it wasn't working on SDXL

[dense sandal]

How do I uninstall xformers in portable forge?

I don't know how to go to the environment

[deft yarrow]

Hi,
I got this error, anyone know how to fix it?

let me try

nope didn't work 😦

[restive sparrow]

oh no i was just talking out generally soorry

[deft yarrow]

no worries

[deft yarrow]

If anyone knows a good source how to fix this let me know!

[fair oxide]

generation always gets stuck at 98% how fix

[ornate elk]

With that GPU you shouldnt use DirectML.
Instead install SD using my AMD Zluda guide.
It has a much better performance and usage.

[brittle lantern]

Hello hope ur well I heard u can run Flux on SD forge now if I was gonna do that on my 6600 xt do I download the forge tutorial on pinned or is it another version if so is it different steps I have to follow?

[ornate elk]

Hey to use Forge with AMD you would need to follow my AMD Zluda Forge guide.
Its under the first link from the pinned messages

[mortal gale]

can anyone suggest a good hosting for the self-hosted version?

[sharp ridge]

  out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False)```
any help?

flux on comfyui is so much slower compared to forge, idk why

[fair oxide]

i am using zluda but i don't want to reinstall SD because then i'd have to reinstall all of the models im using as im not using XL or Pony since im using SD1.5

[ornate elk]

Remove --opt-split-attention and --disable-nan- check
Then your fine

You don't need to reinstall anything if your already using Zluda 🙂

With your GPU you can even use sdxl and pony models without problems

[rugged latch]

hai I has question

is there a command I need to input in args, to allow stable diff to operate without the need of internet

Bc, just luanching it without internet causes it to throw exceptions

the version of stable in question is like.. well it's not the recent one, but i dunt think that matters too much, quetstion is for both new and old verisons

[brittle lantern]

That supports flux correct?

[ornate elk]

Yes

[brittle lantern]

Ty man appreciate it

[ornate elk]

Np

But it will only support normal flux models. Not the nf4 function or models from Forge

Fp8 models work normaly too

[brittle lantern]

Time to see how long my 8gb 6600 xt will take with Zluda 😭🙏

[ornate elk]

let me know ^^, make sure to use the 16gb flux dev model

also you need 32gb RAM

[brittle lantern]

16gb one is fp8 model correct

[brittle lantern]

Bro 😭 I had to remove 2 of my sticks I'm on 16gb of ram after like 2 years now

[ornate elk]

oh why that 😮

[brittle lantern]

Wouldn't boot randomly with them in

Without them it works fine

[ornate elk]

have you made a bios update ?

[brittle lantern]

Tried using those 2 stick alone after removing the others and it just doesn't boot so I think those sticks are fucked

[brittle lantern]

Yh it's fully updated

[ornate elk]

Ah damn :/

[brittle lantern]

CS1o I can't run it can I 🤣

[ornate elk]

You can try

But you would need to increase your windows pagefile

[brittle lantern]

That or my time gets increased by 60% with 16gb of ram

[ornate elk]

Tested RN without internet it launches normaly

Running version 1.10.1

[rugged latch]

well >-< it did, 3 times, and on the 4th it started fine if I were to guess, I'd say it was prolly trying to update something, and it couldn't contact,, I didn't read most of the log, bc this has happned before and relauching it a few times sometimes fixes it, i was jw if there was sum i could do in args to stop that bc it do be on the annoying side

[brittle lantern]

@ornate elk how would i fix?

[rugged latch]

I'll read log next time, and send ss to see if anything can be figured out, if I don't figure it out and fix it myself

Thx for looking into it tho =3

[rugged latch]

Try the arg it said to use :o.

[ornate elk]

would need to see the full cmd log

[brittle lantern]

i fixed that issue but when i run flux fp8 it gave me this on gen

[ornate elk]

whats the model size of the flux model?

[brittle lantern]

i downloaded the 11.6gb model accidentally

is it possible to send me the 16gb model?

[ornate elk]

that model is a "unet" or diffusers model, it needs a different loading method forge dont support i guess

[ornate elk]

here:
https://huggingface.co/Comfy-Org/flux1-dev/blob/main/flux1-dev-fp8.safetensors

[worn zephyr]

Hi Guys!
3090TI 64gb ram here,
recommend Pytouch and Cuda version?

[south crown]

Do you still need a 24gb vram to run flux all locally?
Anyone know the reqs?

[worn zephyr]

No, run Flux NF4
some one squeeze images out of GTX970

[ornate elk]

nope it runs on 12 and 8gb nvidia cards too

[ornate elk]

you dont need to install cuda manually

the webuis already install the latest compatible pytorch and cuda version

[worn zephyr]

I use ComfyUI xD

[south crown]

I could of sworn you needed a top graphics card even to get good images rivalling Stablediffusion
Ill test it out later 🙂

I have a 12gb vram so hopefully its enough to try it

[rugged latch]

Happend again and I dunno wut to do

I copied the exceptions in the text doc

I may have fixed it, by disabling --share

buh ya, das wut it was doing :3

[ornate elk]

then you can try run
pip install torch 2.3.1 inside the site packages folder to upgrade

[ornate elk]

make sure to use either the 16gb flux dev fp8 model or the nf4 model from forge

[ornate elk]

is that an old webui version? because 1.7.0 had a bug that looks like this

[rugged latch]

ya.. xD I think it is 1.7

[deft yarrow]

what's is_sm80/is_sm90

[lapis raven]

When using with Radeon, is the optimal solution for each command in the bat file "--precision full --no-half --opt-sub-quad-attention --upcast-sampling --disable-nan-check --medvram --skip-torch-cuda-test"?

[finite zenith]

anyone using sd forge

Getting black images from flux nf4 anyone know a fix?

[ornate elk]

whats your gpu? and nope thats not the optimal solution

[ornate elk]

remove the clip and vae, it (the nf4 model) should work without them

[finite zenith]

It worked

Ty

[ornate elk]

np

[lapis raven]

AMD Radeon RX 5700 XT

[ornate elk]

then you should follow my AMD Automatic1111 with ZLUDA guide. with that you get the best performance for that gpu

You can find the Guide here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

[lapis raven]

Thank you I'll try it!

[ornate elk]

no problem, let me know if you need help with any step