🤝｜tech-support

[honest tendon]

do I need a special workflow for this @ornate elk

[ruby yoke]

flux, right? did you update comfy with the update.py file in comfyUI/update or did you try to use the manager

they should be in the images with the rest of your workflow

[rich garden]

you mean the text that appears under the generated image?

[misty marten]

Yes, I just did the check

[ruby yoke]

when you create an image with comfy, do you just do preview or do you use a save node

[rich garden]

oh uhh.. i don't think i have comfy, i think i'm using auto1111 whatever it's called (?)

[ruby yoke]

ah. then i don't know, you'll have to ask @ornate elk

[rich garden]

ok

also i keep getting this message when generating an image with img2img impaint which is weird cuz a couple of days ago i didn't encounter this issue

[honest tendon]

i will cry

thats so much storage just gone

[thorny iron]

I keep getting this issue and it doesn't progress.

What can I do

[solemn moon]

This might sound funny but it worked for me
In a new tab, open sd, hit enter a few times, close it (leaving the original 98% window open), repeat, repeat.. wait a lil

[honest tendon]

Uhmm

[drowsy creek]

is it at all possible to get rocm working with an AMD Radeon 5600 XT?

[honest tendon]

didnt work

[ornate elk]

You have to git pull first

The support for Flux was added a few hours ago to the zluda comfyui fork

Then it should work. But I will test it later

[ornate elk]

Can you show me the content of your rocm library folder and what's your GPU again

[ornate elk]

Use the fp8 model of flux

[ornate elk]

Yes should be possible with Zluda

[drowsy creek]

is zluda a drop-in replacement for cuda?

eg, will it also work with ollama and koboldcpp?

[ornate elk]

Ollama works natively with rocm

Kobold too

You still may need to replace some rocm files manually

[drowsy creek]

no, I mean like, I cannot use rocm since my AMD GPU is from the 5600 series

[ornate elk]

Its supported by zluda

So it should work. And ollama should work then too I guess.

Follow my Automatic1111 Zluda guide from here;
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

If you replaced the rocm files then ollama should detect the GPU too

[honest tendon]

How do i git pull do I just type "git pull" in the comfyui directory?

its already the latest model i think

[ornate elk]

Looks good

[honest tendon]

https://comfyanonymous.github.io/ComfyUI_examples/flux

i followed this guide

is there an AMD version so far its just a black output

so it doesnt really work unless my workflow sucks

[ornate elk]

use clip l for clip 1 model

[ornate elk]

in 20-230 mins i can try myself

[grim tapir]

anyone got SD Forge? how do i revert the img2img ui from pic 1 to pic 2?

[honest tendon]

im using the guide so i only have this model for the clips

[ornate elk]

But the guide says you need clip l too

[honest tendon]

oh fk i missed that

i clicked all the blue stuff without reading the rest lol

[honest tendon]

Sickkk

[willow river]

not, rtx2070

[ornate elk]

okay, can you show an example of a blurry inpaint?

[azure ivy]

hey guys where can i find a solotion for this . i use stable diffusion and new version of controlnet stoped working after i update it , how can revert this and install the last version ?

[ruby yoke]

you could just delete the files...

[azure ivy]

that does not worl

k

[misty marten]

I have one RX 6600, there are 654 files in the folder

[ornate elk]

looks right

[misty marten]

yep :\ but it still gives error 0x0000005

[ornate elk]

are you on AMD Adrenalins 24.7.1 ?
can you show me full cmd log again?

[plain bronze]

Can't run Juggernaut X model, why?

[ornate elk]

how much RAM do you have?

[plain bronze]

8

I can run others.

Must upgrade ram for Juggernaut?

[ornate elk]

yea that or increase the windows pagefile

SDXL models need a lot more RAM

you can still use Juggernaut Reborn

[plain bronze]

How to do that?

[ornate elk]

Here:
https://mcci.com/support/guides/how-to-change-the-windows-pagefile-size/

[plain bronze]

I've mistaken. I have 16gb ram. Is 32gb enough for that?

[ornate elk]

make sure its only enabled for C and not any other drives

set the pagefile to 16000 Min and 24000 Max and it will work

Also restart the PC after the change

[plain bronze]

What are the side effects?

Side effects for laptop after doing that.

[ornate elk]

faster performance on some programms and SD.
more usage on the SSD (read and write cycles)

[misty marten]

yes, 24.7.1

[plain bronze]

I'll buy 16gb ram and it will be 32gb total in a few days. Should I wait and not change windows pagefile?

[ornate elk]

download this ZLUDA folder and replace your existing ZLUDA files in C:\ZLUDA wih these:
https://github.com/lshqqytiger/ZLUDA/releases/download/rel.86cdab3b14b556e95eafe370b8e8a1a80e8d093b/ZLUDA-windows-amd64.zip

[ruby yoke]

yeah. wait

[ornate elk]

yea then you can just wait for the new RAM

[plain bronze]

Thank you.

[plain bronze]

Can I get high quality results with 8gb vram?

[ornate elk]

yes

[plain bronze]

Is A1111 good for that or should I use ComfyUI?

[misty marten]

done, but it gave the same error

[ornate elk]

if you use auto1111 make sure to add --xformers --medvram-sdxl --no-half-vae to the webui-user.bat at the line Commandline_ARGS=

[plain bronze]

I don't know how to do that.

Is there any guide link?

[ornate elk]

right click the webui-user.bat and edit it

[ornate elk]

can you recheck the environment variable paths?

[ornate elk]

go into the webui folder

there is the webui-user.bat

[plain bronze]

Right click and click properties?

[ornate elk]

edit

on windows 11 its hidden behind "more options"

[ornate elk]

is that the systemvariables or user variables window?

the top or bottom one

[misty marten]

system variables

[plain bronze]

Like that or must click space between them?

[ornate elk]

nope its fine

[plain bronze]

Thank you, what did I do?

What changed?

[ornate elk]

okay very strange. then there is only one thing you can try:
uninstall HIP SDK 6.1 and install 5.7 and then restart the PC.
After that you need to again replace the Zluda files with these ones here:
(they are made for 5.7)
https://github.com/lshqqytiger/ZLUDA/releases/download/rel.11cc5844514f93161e0e74387f04e2c537705a82/ZLUDA-windows-amd64.zip
Then you have to add %HIP_PATH%bin again tot he system variables.
And then you have to replace the rocm library files again with the one from my guide for your card
then restart the PC again.
Then delete the ven folder and the .zluda folder inside the stable-diffusion-webui-amdgpu folder.
Finally launch the webui-user.bat

[ornate elk]

it increaes the generation performance a lot and needs less vram for that. allowing you to generate faster and higher resolution without any negative impact.

[plain bronze]

How?

That sounds fantastic but how...

[ornate elk]

xformers is a performance booster and --medvram-sdxl loads sdxl better with 8gb vram

[plain bronze]

But why it wasn't like that by default?

[ornate elk]

because its an nvidia thing. and if people would use the webui on AMD on linux it wouldnt work

[misty marten]

okay, i'll do it now

[plain bronze]

That's great! What settings changed in my Nvidia gpu for that exact file?

[ornate elk]

nothing

its only for SD

[plain bronze]

Why Nvidia didn't do that by default. I can't just get it, sorry.

[ornate elk]

nvidia has nothing to do with stable diffusion webui project

its the devs that decide how the webui works ^^

[plain bronze]

Okay but I used sd before with Nvidia, right. And now, something must changed because I will get higher performance.

And what's that, what changed?

[ornate elk]

yea the programm can now use something (xformers library) that can work better with nvidia gpus. using scripts and stuff to increase performance and using less vram

[plain bronze]

Thank you. Helped a lot. You're just... perfect with that kind of things I think.

[ornate elk]

No problem ^^ yea im using stable diffusion since the beginning

[misty marten]

don't install, right?

[ornate elk]

yep

thats important

to not install pro drivers

[misty marten]

do i delete the files from the zluda folder and put these new ones in or do i just put them inside and replace the existing ones?

[ornate elk]

delete the others

[misty marten]

Now in the library, do I create the old_library folder, delete the library files and put these new ones in or just put them inside and replace the existing ones?

[ornate elk]

you make a copy of the library folder and rename the copy to old_library

in the other library folder you input the files and overwrite the existing ones

[misty marten]

like this, right? the library folder has 654 files

[signal sigil]

Guys.. How do i get ComfyUi to offload to my CPU?
Is it done automatically if the GPU doesn't have enough Vram or something?

[plain bronze]

After that edit

I can't run SD and there's an error:

And Kaspersky said: Webui User bat wanna access your camera by using browser...

[plain bronze]

@ornate elk

[ornate elk]

can you show me the full cmd log

[misty marten]

Now the UI has opened, I'm trying to generate an image to see if it worked

[plain bronze]

Okay but what about cam access?

[plain bronze]

NotImplementedError: No operator found for memory_efficient_attention_forward with inputs:
query : shape=(2, 4096, 8, 40) (torch.float16)
key : shape=(2, 4096, 8, 40) (torch.float16)
value : shape=(2, 4096, 8, 40) (torch.float16)
attn_bias : <class 'NoneType'>
p : 0.0
decoderF is not supported because:
xFormers wasn't build with CUDA support
attn_bias type is <class 'NoneType'>
operator wasn't built - see python -m xformers.info for more info
flshattF@0.0.0 is not supported because:
xFormers wasn't build with CUDA support
operator wasn't built - see python -m xformers.info for more info
tritonflashattF is not supported because:
xFormers wasn't build with CUDA support
operator wasn't built - see python -m xformers.info for more info
triton is not available
cutlassF is not supported because:
xFormers wasn't build with CUDA support
operator wasn't built - see python -m xformers.info for more info
smallkF is not supported because:
max(query.shape[-1] != value.shape[-1]) > 32
xFormers wasn't build with CUDA support
dtype=torch.float16 (supported: {torch.float32})
operator wasn't built - see python -m xformers.info for more info
unsupported embed per head: 40

---

[ornate elk]

thats not normal

but most likely a false positive

[ornate elk]

the whole cmd log pls

[plain bronze]

Can you explain?

[plain bronze]

Okay.

[ornate elk]

what browser do you use?

[plain bronze]

Brave

[ornate elk]

remember that the first image can take up to 40 mins

so if the generate button is grayed out and the cmd doesnt show any error its working, let it load

[misty marten]

It's exactly like this, gray button and nothing in the cmd, I'm using another model too, not the default one that comes with stable diffusion

[plain bronze]

And?

[ornate elk]

thats fine, just let it load and wait

[ornate elk]

your torch version is to old for xformers

you have to delete the venv folder and relaunch the webui-user.bat

and your webui is located in a OneDrive synced folder. That can cause massive issues

[plain bronze]

It must located in drive?

I installed it after I delete OneDrive, how can this happen?

[ornate elk]

if you deletet onedrive your fine

then only delete the venv inside the webui folder and relaunch

[plain bronze]

Where is venv folder?

And after I delete it, can I use xformers?

[plain bronze]

These are all folders inside the webui folder.

[ornate elk]

Only delete the venv folder

[plain bronze]

Where is venv folder?

[ornate elk]

Oh wait

[plain bronze]

I'm just searching it for about 10 minutes now.

[misty marten]

@ornate elk IT FINALLY WORKED

[ornate elk]

Perfect!

[ornate elk]

Oh because you installed the portable version its not there I guess

What folders are in your main SD.webui folder ?

[ornate elk]

What's in system?

[ornate elk]

In python or transformers cache

[plain bronze]

Python

Transformers

[ornate elk]

Is there a torch folder and a torch 1.12 folder ?

Inside lib

Site packages

[plain bronze]

Yes there is site packages folder inside lib

[ornate elk]

And inside that

Are there multiple torch folders?

[plain bronze]

All torch folders

[ornate elk]

Okay delete torch and torch 1.13.1.cu17

Then relaunch the webui

That should do it

[plain bronze]

It is what I should expect, right?

Just downloading newer version of torch?

Wouldn't update.bat do this?

[plain bronze]

@ornate elk And what about that?

[ornate elk]

Nope

[ornate elk]

Should work

[plain bronze]

With xformers, right?

[ornate elk]

Yep

[plain bronze]

Why cam access request before?

Is Kaspersky just being schizophrenic?

I can now generate higher resolution images!

[ruby yoke]

why are you running kasperksy?

[obsidian carbon]

Allo, I followed this tutorial to the bone to set up an upscale method and now I am having issues. I am using RGTHREE custome nodes and the node seems to have an issue. I get this error.

D:\ComfyUI_windows_portable\ComfyUI\comfy\ldm\modules\attention.py:407: UserWarning: 1Torch was not compiled with flash attention. (Triggered internally at ..\aten\src\ATen\native\transformers\cuda\sdp_utils.cpp:455.)
out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False)

[plain bronze]

Because it's good.

[ruby yoke]

maybe, that's debateable, but you don't need it running while using the generator, you're not doing anything that it's going to be useful for

[plain bronze]

It's just running in background.

Sometimes I turn it off manually and sometimes not.

[ruby yoke]

i know. suggest that you turn it off, it doesn't need to be running at all, while all you're doing is using the generator

[knotty turret]

someone know how to fix?

[oblique meadow]

This is diffusion discord not embeddings.

But anyway it seems your embeddings are a list but it expects it to be a tensor.

Try doing embeddings = torch.tensor(embeddings)

[knotty turret]

can i dm you?

[ruby yoke]

you should really post this either on the banodoco discord or the L2 discord

[knotty turret]

can you send me the link?

[ruby yoke]

you're ON banodoco

[knotty turret]

yeah i find it on reddit

what chat should i post it?

[ruby yoke]

yeah, those are the serious programmers. post there. i'll DM you the L2 link

[ruby yoke]

probably dev-chatter

[knotty turret]

fine

[oblique meadow]

Yeah

[torpid osprey]

im back

wait

no

i can try on my own

[knotty turret]

send u msg

[honest tendon]

works but takes a while

[torpid osprey]

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

  [21 lines of output]
  + meson setup C:\Users\catsa\AppData\Local\Temp\pip-install-yvnbat_8\scikit-image_5ec85cb028c2465ba745b7839a472a5e C:\Users\catsa\AppData\Local\Temp\pip-install-yvnbat_8\scikit-image_5ec85cb028c2465ba745b7839a472a5e\.mesonpy-c1ag78ig -Dbuildtype=release -Db_ndebug=if-release -Db_vscrt=md --native-file=C:\Users\catsa\AppData\Local\Temp\pip-install-yvnbat_8\scikit-image_5ec85cb028c2465ba745b7839a472a5e\.mesonpy-c1ag78ig\meson-python-native-file.ini
  The Meson build system
  Version: 1.5.1
  Source dir: C:\Users\catsa\AppData\Local\Temp\pip-install-yvnbat_8\scikit-image_5ec85cb028c2465ba745b7839a472a5e
  Build dir: C:\Users\catsa\AppData\Local\Temp\pip-install-yvnbat_8\scikit-image_5ec85cb028c2465ba745b7839a472a5e\.mesonpy-c1ag78ig
  Build type: native build
  Project name: scikit-image
  Project version: 0.21.0
  WARNING: Failed to activate VS environment: Could not parse vswhere.exe output

  ..\meson.build:1:0: ERROR: Unknown compiler(s): [['icl'], ['cl'], ['cc'], ['gcc'], ['clang'], ['clang-cl'], ['pgcc']]
  The following exception(s) were encountered:
  Running `icl ""` gave "[WinError 2] The system cannot find the file specified"
  Running `cl /?` gave "[WinError 2] The system cannot find the file specified"
  Running `cc --version` gave "[WinError 2] The system cannot find the file specified"
  Running `gcc --version` gave "[WinError 2] The system cannot find the file specified"
  Running `clang --version` gave "[WinError 2] The system cannot find the file specified"
  Running `clang-cl /?` gave "[WinError 2] The system cannot find the file specified"
  Running `pgcc --version` gave "[WinError 2] The system cannot find the file specified"

  A full log can be found at C:\Users\catsa\AppData\Local\Temp\pip-install-yvnbat_8\scikit-image_5ec85cb028c2465ba745b7839a472a5e\.mesonpy-c1ag78ig\meson-logs\meson-log.txt
  [end of output]

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

Encountered error while generating package metadata.

See above for output.

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

what am i supposed to do

here

assuming its vs?

[honest tendon]

it works but took like 500 seconds to generate

omgg

[ruby yoke]

flux is huge.

[honest tendon]

works great tho

Time tho not so good...

whats like a cheap GPU for image generation thats not a gaming card?

[ruby yoke]

um how long per generation do you want to wait? and do you want to use your machine to do other things with at the same time?

[honest tendon]

I just want to be noticably faster than my 6950xt which in the list is fairly slow, im thinking of a secondary GPU to kind of be used mainly for image generation

since its scoring 7

i just need a cheap GPU

[ruby yoke]

do yourself a favor if you're buying something new. get the best nvidia card and machine you can - so that's a good gaming machine. the reason that gaming machines are used, is because the card they use is what you need for this

[honest tendon]

i already have a card thats very good for gaming i just need a secondary card thats cheap and faster for ONLY image generation or running LLMs

[ruby yoke]

you need an nvidia card. you need to run python, you want to run cuda, and cuda runs only on nvidia

and you really don't want to waste your money

[honest tendon]

I know so i want cheap non gaming card like one of those A1000 if itheres a significantly cheaper alternative preferable 500 bucks

is there like no dedicated AI gpu thats not for gaming?

[ruby yoke]

or maybe you do want to waste your money. i dunno.

[ruby yoke]

no. because the gaming machines use the GPUS you need for this. so the cheapest way to get a good machine is to buy a gaming machine.

[honest tendon]

well that sucks

[ruby yoke]

most of the games you probably play use pre-rendered graphics or they stream low res images rendered in real time. what you're doing with this requires a lot more power than your games do

[honest tendon]

4070ti seems to be the best

[ruby yoke]

something like this https://www.newegg.com/abs-aqa14700kf4060ti16g-stratos-aqua/p/N82E16883360436

intel processor, not AMD

[honest tendon]

nty im pretty sure CPUs dont matter as much.

also i have a pretty good system already

i just need the gpu

12700f

[ruby yoke]

they matter

[ruby yoke]

shop around for parts, i think newegg sells jsut the gpus

[brittle lantern]

some1 help me fix this i did the install right

using 6600 xt and zluda

[ruby yoke]

did you follow @ornate elk zluda guide? click on the push pin icon at the top right of discord and read his guide

[brittle lantern]

yes

[gaunt juniper]

hello guys, I have a question does anybody use Comfyui? and does anybody have a Server I could join? I'm a noob and I really need some help and experience hahaha

[ruby yoke]

most of us do, and there's a #🧣｜comfy-ui channel here

[gaunt juniper]

yeah i see it now hahahahah thx hahaha

[ornate elk]

Can you try these steps?:
#🤝｜tech-support message

[knotty merlin]

Trying to run Forge, but I keep getting this error. I've installed it from the "one-click" installer, and trying to install the correct version of Torch mentioned here https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/241 I get ERROR: No matching distribution found for torch==2.1.2.

I've also tried uninstalling and reinstalling Forge twice, including one time with not running update.bat and just running run.bat immediately as suggested later in the linked Github issue.

[ornate elk]

Seems like your trying to install an old forge version

Checkout the link to my guides in the pinned messages.
There you find the latest version of forge and the version before the big changes.

[knotty merlin]

I have an Intel GPU, which of the guides should I follow?

[ornate elk]

What's the exact GPU model?

[knotty merlin]

The GPU just lists "Intel(R) Graphics"

Here's the hardware IDs in Device Manager

[ornate elk]

Okay then your only chance is to use SD.Next with Openvino.
Here is a link how to get it work:
https://github.com/vladmandic/automatic/wiki/OpenVINO

[knotty merlin]

Alright, I'm trying to install that and it seemingly wants me to both have Python 12 and not have Python 12, somehow.

First one is with Python 3.12.3 installed, second one is with Python 3.11.9

[ruby yoke]

you need 3.11

you're going to need to downlevel

[knotty merlin]

As stated above, I had 3.11.9. Do you mean that I need 3.11.0 specifically?

[ruby yoke]

you have 3.12 installed some how, what do you get when you run the python version command?

[knotty merlin]

This is two separate runs, one with 3.12.3 installed, and one with 3.12.3 uninstalled and 3.11.9 installed in its place.

[ruby yoke]

just go back to base 3.11 and let pip tell you if it needs to upgrade

[knotty merlin]

I'm a bit of a novice, so how do I use pip to check if an upgrade is necessary?

[ruby yoke]

you'll see errors in the logs if you try to start up the server

[knotty merlin]

Alright, I tried running it again with 3.11.0, and I get the same error that seems to imply it's looking for a 3.12 installation.

[ruby yoke]

always fix the first error that showed up, the rest might be caused by that one

[ruby yoke]

are you specifying 3.12 in your .bat file?

[knotty merlin]

When I try to search in Notepad, the number 12 doesn't appear anywhere in the file, so I don't think I am.

[ornate elk]

Uninstall python 3.12
Then install python 3.10.11 or 3.11
Then delete the venv folder and relaunch

[ruby yoke]

and reboot before you do that, please

[knotty merlin]

Alright, I've already done the first two steps but I'll try restarting my laptop and deleting the venv folder.

[honest quail]

hey guys so I tried downloading some models but when I use them I just get a gray image

[ornate elk]

Hey, make sure the files are actually models and not loras

2gb-6gb = models
Everything below 2gb mostly not a model

[honest quail]

I got them from civitai

@ornate elk is it ok if I screenshare?

I just got stable diffusion again so I'm assuming I messed up somewhere

[ornate elk]

Make a screenshot

[honest quail]

shift windows s right?

[ornate elk]

Windows shift S

[honest quail]

it's not working for some reason

[ornate elk]

You can also use the Snipping tool

[honest quail]

got it

had to manually open the app lol

[honest quail]

I think those are models

[ornate elk]

Only one file there is an model. The others are loras

[honest quail]

oh

what's a lora

and how do I tell the difference

[ornate elk]

A lora is a small file trained on a specific character or style to go on top of a model to help it generate in that style or that character

[honest quail]

oh ok

where do those go

[ornate elk]

On civitai check the info box on the right when your on a model or lora page

You can also use filters

[ornate elk]

Models/lora

[honest quail]

so I make a new folder in the models folder?

[ornate elk]

There should be a lora folder already

But if not create one

[honest quail]

just "lora" right?

[ornate elk]

Lora

With L instead of l

[honest quail]

it's saying that they use the pony model

[ornate elk]

There are 3 commonly used model types today.
1.5
SDXL
Pony
Loras are also based on one of these three types.
They have to be the same type to work together.
Pony is based on SDXL, so some SDXL loras can work with Pony but Pony loras won't work good with SDXL.

[honest quail]

how do I get those

[ornate elk]

You can use the filter on civitai.

Select pony as model type then search filter for most downloaded

Always check the little info box every model has on the right

[honest quail]

I'm seeing checkpoints

[ornate elk]

Checkpoint's = models

[honest quail]

ah ok sweet

the download options are saying pruned in vae

*vae

whoops

[ornate elk]

Always go for the smaller version

[honest quail]

double whoops

*and not in

[ornate elk]

With VAE is good

[knotty merlin]

https://pastebin.com/58EqEiPG
Okay, I tried that and still got some errors that I don't really understand this time

[honest quail]

that was from this checkpoint

wait it says something about a vae folder

[knotty merlin]

Okay, here's the mentioned log file
It appears to be a dubious ownership problem, so I'll try adding an exception for the directory

[ornate elk]

Did you moved the webui folder onto an other drive after git cloning it?

[honest tendon]

@ornate elk u tried Flux yet?

[ornate elk]

Yes

[honest tendon]

does it also take like 350sec for it to work?

[knotty merlin]

No, it's still on C:

[honest tendon]

so far for me the avg is 350 seconds

first one took 500

[ornate elk]

The VAE is not a model.
VAE files go into models/vae

[honest quail]

so how do I use a vae

[honest tendon]

but the images look fking amazing

[ornate elk]

On my RX7900XTX it takes 60 seconds for 20 steps at 1024x1024
Getting 1.9-2 s/it

[honest tendon]

Didnt follow Batman facing off darth vader but it works

[ornate elk]

Move it into the models/VAE folder.
Then in the webui go into Settings - User interface - Quicksettings.
There add sd_vae
Then hit apply and reload ui.
Then in the new VAE dropdown select the vae

[honest tendon]

do you know a way to make it faster

other than buying a new gpu?

[ornate elk]

Nope we have to wait for comfyui updates

Or a smaller model

@honest tendon oh did you used the 12gb or the 24 GB model?

Because the 12gb model is faster

[honest tendon]

16gb 6950xt

[ornate elk]

I mean the flux model

[honest tendon]

Oh

[ornate elk]

You have to use the 12gb variant fp8

[honest quail]

I'm getting a connection timeout

[honest tendon]

WAIT i didnt know there was a smaller model

[ornate elk]

You need a link?

[knotty merlin]

Alright, I ran the listed git command but it's still giving me dubious ownership errors

[ornate elk]

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

[honest quail]

I got a 4070

[ornate elk]

@honest tendon Here is the smaller Flux model for Comfyui:
FP8 12gb.
https://huggingface.co/Kijai/flux-fp8/tree/main

[honest quail]

I'mma restart my pc real quick

[ornate elk]

Then you need to add --xformers --no-half-vae to the Commandline_args=

[knotty merlin]

Do I have to restart for safe directory additions to take effect or smth?

[honest tendon]

thanks man

[ornate elk]

Did you moved the webui after git cloning it?

[knotty merlin]

Again, no

[ornate elk]

Okay. Thats strange

[honest quail]

my dumbass also just realized that I was putting everything in the stable diffusion folder

[ornate elk]

@knotty merlin But I guess the problem is that you got cloned it into a system folder "Documents".
You have to create an own folder call it Ai and git clone it in there.

[knotty merlin]

Alright, I'll try that
Thank you

[ornate elk]

NP

[knotty merlin]

Should it be under C:/Users/Kaz or directly in C:?

[ornate elk]

Directly C is good

[knotty merlin]

Alr

[ornate elk]

You used the 1.5 model with and sdxl vae

That wont work

Only sdxl models or pony models work with an sdxl vae

If your using 1.5 models set the VAE to Automatic

[honest quail]

I'm assuming that means that I should download this and put it in the stable diffusion folder

[ornate elk]

Yep 💯

[stone orchid]

Hi im new here ive built a python script to generate NFTs with STABLE but i cant seem tro get the quality of the image correct I'm running 8gb NvidiA RTX3070

Any coders here able to help with getting the most out the quality

[knotty merlin]

Alright, I managed to get it up and running but halfway into generating the first image I got this

[tame yoke]

is it possible to run sdxl at 512x512 with no issues

[ornate elk]

The issue would be the quality of the output

[ornate elk]

Hmm I don't see any information on what failed

[ornate elk]

Why using a self made script and not a normal webui?

[stone orchid]

i need to creat metadata with each image and genarate 5k collection for

the script does this for me

Unless there is some open source web ui that i can alter

[stone orchid]

thi sis the quality im getting

[ornate elk]

There is Automatic1111 webui which also creates meta data

It also has an API

[honest quail]

@ornate elk I got the feeling that something was off with my generations so I copied as much of an image as I could to see how mine would generate and I got this

[ornate elk]

Quality is define by Checkpoint, resolution and prompt. And can hmbe enhanced by using a better checkpoint and upscaling

[honest quail]

also it keeps telling me that I'm out of memory. I never had that issue last time I put stable diffusion on my computer

[ornate elk]

1.5 models are trained on a 512x512 resolution.
SDXL and Pony on 1024x1024
So you have to change the resolution

[stone orchid]

could i share the script with you to take a look at

here is the script

[ornate elk]

I'm on mobile rn can't check the code sry

[ornate elk]

Sdxl and pony models using a lot more resources

But with your GPU they should work too without problems

[honest quail]

I fixed it for a sec, but now I'm getting connection errors all of a sudden

[honest quail]

maybe this will help

[ornate elk]

Looks normal

[honest quail]

my pc killed itself lmao

[plain bronze]

You are right 👍

[ornate elk]

What settings did you used?

[honest quail]

Everything should have been default

Except the stuff you said to change

And I also added a Lora thing to the ui

[ornate elk]

I mean the txt2img settings like resolution, steps etc and model

[honest quail]

Oh

30 steps, 1024×1024, euler a, and some models I found

The models are working fine now

Btw I remember their being a fix faces option

On the previous version I used

[ornate elk]

Install the Adetailer extension

Its for face fix

Its really good

[honest quail]

Oh yeah it's awesome

[plain bronze]

Is there a big difference between 16+16 and 16+32 RAM for SD?

[ornate elk]

Only when using sdxl and pony models. As with 32gb ram you don't need to increase your pagefile to use them at best performance

[plain bronze]

So, 16+16 is ideal?

Or you mean 16+32 by saying 32?

Which one should I prefer?

[elfin root]

Has anyone experienced SD3 taking a reasonable amount of time doing inference steps (1-2it/s) then after take several minutes to finish?

but then follow up image generation is reasonably quick?

[plush spire]

How do I get preview images in the Lora tab? At the moment they all say No Preview. Automatic1111

[ornate elk]

Oh, 16+16 is ideal. Dual channel is always faster

[plain bronze]

Why this channel vanish when I go to another channel?

[dull frost]

Heya, i have a specific task, which i dont know how to approach looking for ideas/solutions, i have a collection of cartoon images like objects, food, people, etc in personal art style, and goal is to utilize them to generate new images with similar style but different objects

[plush spire]

Sounds like you might want to look into LoRa training

[dull frost]

i was looking at fine tuning vs lora, also what is the reccomended dataset? or throw all in:D

[bronze plinth]

well

[plush spire]

Someone with more experience would need to advise you on that. I'm just floating the idea that it sounds like what you need can be accomplished via a Lora trained on your images

[terse lichen]

so is there a error or something while updating through the manager or ?

[brittle lantern]

I get now failed to load Zluda: could not find module then using CPU only torch

[ornate elk]

You need to recheck the hip path in system variables

[brittle lantern]

It's still in there I removed it and re put it

is it normal to see this

[ornate elk]

Yes

[brittle lantern]

Ok it seems to work now after I deleted the zluda folder in the stable diffusion folder and the open the bat and it remade it

I do have one more question with rx 6600 xt and zluda on windows what's the speed equivalent like to a nvidia card?

[ornate elk]

Idk you have to test the speed of a
512x512 resolution 1.5 (2gb) model.
With 30 steps and Euler a sampler.
Prompt: Cat
Neg prompt: Blurry

Then check the cmd for the it/s you get

Then I can compare

[brittle lantern]

Got u

[plush spire]

How do I set a preview image for a model in Automatic1111?

[brittle lantern]

this is the speed

idk where to find gpu comparison

[ornate elk]

A GTX1080 gets 1.8 it/s

[ornate elk]

put an image (.jpg or .png) into the folder with the model and rename it to the same name

[brittle lantern]

do i look at first or 2nd bar?

so my gpu performs like a 1080?

[ornate elk]

the first

[ornate elk]

you have 3.2 it/s

so its like nearly double the speed

[plush spire]

thanks 🤝

[misty marten]

Is there any way to use kohya_ss with AMD on Windows?

[ornate elk]

Nope. But OneTrainer should work with Zluda.

[misty marten]

I've never heard of this one, is it used for the same purpose as kohya?

[ornate elk]

Exactly

Training loras, embeddings and models

[misty marten]

Perfect, can I install it normally using the github guide, or is there something different I need to do because it's AMD?

[ornate elk]

Currently they investigate why its not working for me with HIP SDK 6.1
But with 5.7 it should work

There is a USE_ZLUDA flag but idk how to use it.
So if you don't get a .zluda folder in the OneTrainer folder after the install you have to manually run the
Zludainstall.py to get it.

[misty marten]

I understand, I'll try here

I installed it but I'm confused, where do I put the images to generate Lora?

[ornate elk]

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

[ornate elk]

Under Concepts you add a path to a image folder and a prompt

[brittle lantern]

@ornate elk sorry to disturb i also did comfyUI but gave me this on generation

[ornate elk]

Have you copied the zluda files into lib site packages torch folder?

[brittle lantern]

lost me at that last part😅

[tepid crypt]

now im having a problem with ip adapter, im trying to install it this way: https://www.youtube.com/watch?v=CSEnlkXHoIA
(btw, im using a colab, sagemaker) there said that i need to install insightface for the version of python i use (which is 3.10), im following each step but when trying to install it using the same command as the video appears this:

[karmic crown]

you are using collab, but you are typing windows specific commands

whoever made that guide should be tortured

just do
python -m pip install insightface==0.7.3
or eve
python -m pip install insightface

[tepid crypt]

well, there said to use the name of the file i downloaded, i uploaded to the colab and its name is: insightface-0.7.3-cp310-cp310-win_amd64.whl, if i do that command would be the same as installing this file?

[karmic crown]

win_amd64 - this is file for windows

you cannot use that one

[tepid crypt]

oh, ok going to try!

[karmic crown]

that is a package prebuilt for windows

and that is your problem

those people expect that everyone will use windows

i hate turorials like that one

that's why i said that guy should be tortured

[tepid crypt]

also btw, how can you know that version would be for python 3.10? it would install that version automatically?

[karmic crown]

python --version

[brittle lantern]

dw i feel the same about any 1 min long tut that show basically nothing

[karmic crown]

you cannot install package for wrong python
it will install the package for your version of python, whatever that is

[tepid crypt]

oh got it, thanks!

[ornate elk]

From my guide ?

You need to do Step 5 of getting zluda to run

[brittle lantern]

omg im stupid i didnt see that

[brittle lantern]

works now ty u think flux might work on my 8gb card the fp8 models

[ornate elk]

Hmm I don't think so

[loud stream]

Hey! not sure if its here im meant to post issues. New to this and even my simple prompts look wierd... for example 'apple'?

[brittle lantern]

ill assume we need some extreme optimised variant?

[ornate elk]

Yea we would have to wait for more improvements on comfys side or a smaller model variant from the flux devs or community

[ornate elk]

Hex it seems like your using an sdxl model.
If so then you need to change the resolution to 1024x1024
Because that's what sdxl got trained on.
1.5 based models are trained on 512x512

[loud stream]

sweet thank you, how are you able to tell im using sdxl?

[vapid widget]

hi

i have a problem

lora doesn't works

[tepid crypt]

where are your loras? what folder?

[karmic crown]

by the model name
https://civitai.com/models/425222/cinenaut-xl-a-true-cinematic-diffusion

[vapid widget]

in this folder

[ornate elk]

The model name has an XL in it. Thats mostly an indicator.
But to verify you would have to check the model info box on civitai.com or the model file size.
If a model is 6gb its mostly sdxl. Everything fron 2-5gb is an 1.5 based model

[karmic crown]

i just posted model link 🙂

[ornate elk]

Is the lora sdxl or 1.5 based or pony?

And the checkpoint too?

Loras only show up when a compatible model Is selected

[loud stream]

awsome thx for the heads up

[vapid widget]

i download the lora in this url

[vapid widget]

and also download the model pony in the same page , afterthat i put the model in the model carpet

[ornate elk]

The lora is 1.5 based
So it won't work with pony

[vapid widget]

but the interface doesn't recognize lora

[vapid widget]

which model do you recommend?

[ornate elk]

You either need to download a Pony compatible lora of aqua or download an 1.5 model

For 1.5 try Divineanimemix or koji

[vapid widget]

thanks

[ornate elk]

Np

Switch the model and then reload the loras on the right and it should appear

[tepid crypt]

dont know why ip adapter doesnt work as good as the video, is it becouse im not using the right preprocessor? in the video it shows its only using three and those ones i dont have

[brittle lantern]

Hmmm what would u think the next best option for a gpu upgrade would be basing on lower than 550 budget

[barren monolith]

any one know how to fix this ? when i click open webui-user it gives me this error ?

it said RuntimeError: Couldn't install torch.
Command: "C:\Users\buate\Desktop\tutorial\stable-diffusion-webui-amdgpu\venv\Scripts\python.exe" -m pip install torch==2.3.1 torchvision torch-directml

[barren monolith]

i had no idea . how to check that ?

its show like this

then how to installed that ? i just follow tutorial on youtube to install stable diffusion and i already downlaod python . git . and clone git

[foggy horizon]

what python version do i download?

[barren monolith]

im following this tutorial https://www.youtube.com/watch?v=eO88i8o-BoY&t=544s&ab_channel=Northbound

its for windows gpu amd . seems like its auto1111 cause i clone git from here https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu

[foggy horizon]

@ruby yoke

[ruby yoke]

stick with 3.11.0

pip'll update it if it needs to when you launch or at least tell you if you need to

[foggy horizon]

ok downloading it rn

[ruby yoke]

amd - i'm going to toss this over to @ornate elk

[barren monolith]

ahh okay hopes he online soon

[ruby yoke]

he's here every day, not sure what time of day or night it is for him, but please be patient. he's the AMD expert

[barren monolith]

glad to know that . thankss . i wait him to help me then .

[ornate elk]

Hey, what's your GPU?

[barren monolith]

Hi . its rx 6800 M

[ornate elk]

Okay then you should follow my AMD Automatic1111 with ZLUDA Guide for the best performance.
The guide is in the first link if the pinned messages

[barren monolith]

this one right?

okay will chat again later

[ornate elk]

Sure, let me know if any step isnt clear

[foggy horizon]

where do i put files i download from civitai?

@ornate elk

[ornate elk]

Depends on what they are. If the file size is 2-6gb they are models and go into models/stable-diffusion

[foggy horizon]

oh ok thanks i figured it out

[foggy horizon]

If I press this is it supposed to show new ones?

or do i need to restart the whole thing

[barren monolith]

hey i still get the same error like this

[ornate elk]

It should show new ones

[ornate elk]

Do you have any antivirus Programm?

Because they most likely broke the Python installation

So uninstall Python.
Disable the Antivirus Programm
Install Python 3.10.11 64bit
Delete the venv folder.
Then launch the webui-user.bat
And enable the Antivirus Programm

[barren monolith]

i did but now im already disable it but still give error

[ornate elk]

You have to delete the venv folder too

Before trying again

[barren monolith]

already delete venv folder and this shows up after click webui-user

and then venv folder exist again

then when i open webui-user its show like this again

[ornate elk]

Reinstall Python again while the AV programm is disabled. Then restart the PC.
Then delete the venv again and relaunch the webui-user.bat

[ornate elk]

This means the Antivirus broke that python installation again

[barren monolith]

oke i try reinstall phyton again

[ornate elk]

If you use Avast, Avira, AVG, Norton360, McAffe, or 360 Security, you can also uninstall that. They are crap.

[ruby yoke]

yeah. just use windows live defender - but most of those are boot sector and hard to uninstall once you have them

[barren monolith]

i alredy uninstall all AV and turn off windows defender all off

already uninstall phyton

and reinstall

i open webui-user now?

[ornate elk]

Reboot the PC.
Then delete the venv
Then launch the webui-user.bat

[ruby yoke]

no. you reboot now. shut down, turn off machine. leave off for a full minute. then turn it on and tell us when that's done

let's get stuff out of memory

[barren monolith]

i delete venv folder

wish me luck

now like this ? wtf had improvement

[ornate elk]

Looks good

[barren monolith]

ty so much turn off av makes this work bro

[ornate elk]

Also turn on windows defender

[ruby yoke]

don't instally any more anti-virus. all they do is take your money and cause problems.

[barren monolith]

okay let me update when its completely done

[barren monolith]

yea its free comes when i bought this laptop 😦

[ornate elk]

Yea McAffe and Microsofts Laptop Deal....

[ruby yoke]

it's hardly free, look how much trouble it just caused you.

at least it wasnt' crowdstrike

[barren monolith]

after download many things and install many thing at the end its show like this ?

[lethal kiln]

Hello experts, I would like to ask a question. I want to put the material of a product in the same scene and integrate it perfectly without using the function of distorting the perspective of PS

Is current technology capable of doing this?

[ornate elk]

Have you done the step 4 of the getting zluda to work?

[ruby yoke]

can you explain what you mean by "the perspective of ps"

[lethal kiln]

Distort and Prespective

[ruby yoke]

so you have a scene, you want to add an image and have it baseically warp around to look like it' spart of the scene?

[barren monolith]

i skip that because i had rx 6800 ?

[lethal kiln]

yes！

[ruby yoke]

yeah - no. that would take img2img inpainting and i'm not aware of anyone doing that right now. however, photoshop has extremely good tools to do exactly that

[lethal kiln]

I tried the img2img inpainting, but the results were highly random. This morning, I also tried mimicBrush, but it was difficult to achieve the desired effect.+

[ornate elk]

I thought you had an 6800M

Thats below a 6800

[lethal kiln]

I wonder if this is because too few people use this function. If I don't use SD, which features in Photoshop can achieve the same result (quickly, accurately, and in a user-friendly way)?

[barren monolith]

yes 6800M . ahh okay i will try download

[ruby yoke]

the technology just isnt' there right now. so with photoshop you'll want to watch some youtube videos on the subject. start with piximperfect's channel

[lethal kiln]

okay, thank you！

[barren monolith]

what should i download for rx6800M ? i worried wrong download

[ornate elk]

The optimised rocm for GFX1031

Same as for 6700xt

[barren monolith]

now like this

i extract gfx1030 and drop all files to library

in above

[ornate elk]

When do you get that error?

Also you need to restart the PC after moving the files into library

[ornate elk]

I said GFX1031 not 1030

[barren monolith]

after drag and drop 1031 all files to library

[ornate elk]

Restart the PC

Then relaunch

[plain oak]

Is there any particular set of steps I need to take to uninstall Fooocus, or do I just delete the folder?

[ornate elk]

Deleting the folder is enough. And then open up a cmd and run pip cache purge to remove old installer files

[honest quail]

hey @ornate elk I noticed that there's some failed stuff in here, might that be the reason that my stable diffusion is so unstable?(no pun intended)

[ornate elk]

Hey, what extensions have you installed?

[ornate elk]

Did you tried to use adetailer hand fix?

[honest quail]

nope

this has been a thing even before I got the extensions

[ornate elk]

Okay close everything. then open up a cmd and type
Pip cache purge
Then delete the venv folder and relaunch the webui-user.bat

[honest quail]

btw the version I got looks a bit different from the old one I used with its file structure

the webui-user.bat doesn't even launch

[ornate elk]

Yea your folder path names are also not correct

[honest quail]

makes me use that run thing instead

[ornate elk]

Best would be a clean install using my guide

[honest quail]

probably yeah

do you have a video

[ornate elk]

No but in the pinned messages you find a step by step guide. Or on the first link too

Its pretty easy and takes 5mins

[honest quail]

alright I'll do that in a sec

[honest quail]

I'mma reinstall python and make sure it's added to path

dang same issue

[ornate elk]

No module named pip means that an third party antivirus Programm broke the Python installation

[honest quail]

what on earth

I don't remember getting one

[ornate elk]

If you use Avast, Avira, AVG, Norton360, McAffe, or 360 Security, you can also uninstall that. They are crap.

Any of these can break the python install by blocking it

[honest quail]

I couldn't find them in my search bar, but I see a mcafee webadvisor thing in my task manager

alright I think I uninstalled it

wait it's still showing in my task manager

should I delete the folder?

I deleted it

now gonna reinstall python and restart the guide

[ornate elk]

Uninstall it via settings/apps

If its there

[honest quail]

I did and the program was still running so I used bug spray

lmao jk I ended the task and deleted the folder it was in

[ornate elk]

Thats not good. You always need to uninstall Programms

Deleting only breaks more

Check the system control.
There under "uninstall Programms"

It there is an entry for it

[honest quail]

yeah I did that

I still saw it in my task manager and files though afterwards

[ornate elk]

But you uninstalled it via system control before?

[honest quail]

yeah

[ornate elk]

Okay good

Then yea delete the rest if there is anything left and also restart the PC

Then Python needs to be reinstalled

[honest quail]

copy that

should I also redownload the folder for stable diffusion?

still getting the same issue @ornate elk

wtf is reason cybersecurity

[ornate elk]

Thats some stuff from a third party AV tool

[honest quail]

yeah I found it in my control panel

uninstallling it

[ornate elk]

Perfect

[honest quail]

I'm gonna have to restart again aren't I

[ornate elk]

Yep

[honest quail]

[ornate elk]

Googled it and it comes bundled with stuff like
RAV Endpoint protection. RAV web filter and RAV VPN

Can alle be uninstalled

[honest quail]

yeah I did that

thanks btw for all of this

[ornate elk]

No problem ^^

[honest quail]

holy hell finally I see pip

let's gooooooooooooooooooooooooo

last time I used stable diffusion I was able to make a shortcut for it that I pinned to my taskbar

I forgot how to do that though since the video was abscure

[ornate elk]

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

Then you have a shortcut. In the taskbar you can also right click it and click add to taskbar

[honest quail]

I don't see that option

also it still crashes a decent amount 😭

[ornate elk]

Crashes? Where when?

[honest quail]

whenever I change models I get a connection error

The only way I've been able to switch is by relaunching and changing before I generate something

just now

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument index in method wrapper_CUDA__index_select)

also it keeps saying that I'm out of memory

I got 32gb

idk what's happening

[ornate elk]

What's inside your webui-user.bat?

[honest quail]

one of my monitors just black screened after both of them flashed white

I gotta restart my pc

this has been an issue ever since I got that 4070

a part of me feels tempted to put my 3070 back

[ornate elk]

Make sure your GPU driver is up to date

Or reinstall it with DDU

[honest quail]

I'll ask more about that when I restart one sec

alright I'm back

[honest quail]

I watched task manager while it generated an image and it for some reason used 100% of the gpu

[ornate elk]

Did you editet the webui-user.bat like my guide suggests? Whats in it?

[honest quail]

you mentioned to put the extra stuff depending on your ram

I got 32gb

[ornate elk]

Depending on vram

Not ram

Vram = GPU memory

But a 4070 should only need --xformers

[honest quail]

for the dedicated it says 12

is that what I'm supposed to be looking at?

[ornate elk]

Right click the webui-user.bat and edit it

There add --xformers to the Commandline_args=

Then save and relaunch

[honest quail]

erase the no half part?

[ornate elk]

--no-half-vae is okay

[honest quail]

it's already there

I did that during the setup

[ornate elk]

Then show me the txt2img settings you used when it crashes

[honest quail]

changing my model fucks things up too though

[ornate elk]

Have you set your pagefile back to system managed for C ?

[honest quail]

my what?

[ornate elk]

Oh nvm you didnt changed it

You waited for the ram upgrade

[ornate elk]

Can you try again switching it

[honest quail]

it's once I've already generated something

[ornate elk]

Okay

Show the me cmd when it crashes again

[honest quail]

just finished updating my gpu so I'm gonna restart real quick

[honest quail]

alright was generating images to get it ready

now gonna switch models and see what happens

yeah connection error

[ornate elk]

What does the cmd shows?

[honest quail]

nothing changes except that it says press any key to continue

[ornate elk]

What's the model you try to switch to?

[honest quail]

it does it with any model

[ornate elk]

What's the browser, any adblocker or AV Browser extensions?

[honest quail]

actually

I do have an adblock

[ornate elk]

If its not ublock origin, you can remove it and replace it

[honest quail]

it is ublock

these are the extensions I have

[ornate elk]

You can remove the
"online security" one
And disable or remove the Free VPN "Ultra secure.....🙃"

Then restart the browser and try again

VPN extension can cause issues with the webui

[honest quail]

alright generated a few images

now to try it

yep

it loads then gets an error

[ornate elk]

Can share the whole cmd output?

[honest quail]

venv "C:\Ai\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: v1.10.1
Commit hash: 82a973c04367123ae98bd9abdf80d9eda9b910e2
Launching Web UI with arguments: --xformers --no-half-vae
[-] ADetailer initialized. version: 24.8.0, num models: 10
Loading weights [439085d766] from C:\Ai\stable-diffusion-webui\models\Stable-diffusion\goodfitPony3D_v41MoreRealistic.safetensors
Running on local URL: http://127.0.0.1:7860

To create a public link, set share=True in launch().
Creating model from config: C:\Ai\stable-diffusion-webui\repositories\generative-models\configs\inference\sd_xl_base.yaml
C:\Ai\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(
Startup time: 11.5s (prepare environment: 2.4s, import torch: 3.3s, import gradio: 0.9s, setup paths: 1.1s, initialize shared: 0.2s, other imports: 0.6s, load scripts: 1.7s, create ui: 0.6s, gradio launch: 0.5s).
Applying attention optimization: xformers... done.
Model loaded in 5.2s (load weights from disk: 1.0s, create model: 0.5s, apply weights to model: 3.2s, move model to device: 0.1s, calculate empty prompt: 0.2s).
100%|██████████████████████████████████████████████████████████████████████████████████| 35/35 [00:14<00:00, 2.47it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 35/35 [00:15<00:00, 2.21it/s]
100%|██████████████████████████████████████████████████████████████████████████████████| 35/35 [00:12<00:00, 2.75it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 35/35 [00:12<00:00, 3.05it/s]
0: 640x640 1 face, 7.0ms
Speed: 4.0ms preprocess, 7.0ms inference, 15.0ms postprocess per image at shape (1, 3, 640, 640)
100%|██████████████████████████████████████████████████████████████████████████████████| 15/15 [00:04<00:00, 3.13it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 35/35 [00:26<00:00, 1.32it/s]
100%|██████████████████████████████████████████████████████████████████████████████████| 35/35 [00:11<00:00, 3.04it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 35/35 [00:15<00:00, 2.25it/s]
100%|██████████████████████████████████████████████████████████████████████████████████| 35/35 [00:11<00:00, 3.07it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 35/35 [00:18<00:00, 1.86it/s]
100%|██████████████████████████████████████████████████████████████████████████████████| 35/35 [00:13<00:00, 2.58it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 35/35 [00:17<00:00, 1.98it/s]
100%|██████████████████████████████████████████████████████████████████████████████████| 35/35 [00:11<00:00, 3.07it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 35/35 [00:15<00:00, 2.23it/s]
100%|██████████████████████████████████████████████████████████████████████████████████| 35/35 [00:11<00:00, 3.08it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 35/35 [00:15<00:00, 2.23it/s]
100%|██████████████████████████████████████████████████████████████████████████████████| 35/35 [00:11<00:00, 3.02it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 35/35 [00:16<00:00, 2.17it/s]
Press any key to continue . . . ███████████████████████████████████████████████████████| 35/35 [00:16<00:00, 2.82it/s]

[ornate elk]

Do you have any 1.5 model you can try switch to?

[honest quail]

I could try switching to the regular model

[honest quail]

still an error

[ornate elk]

And your not running anything other than the webui and discord?

[honest quail]

I guess steam and snip & sketch are also on

[ornate elk]

Thats okay

Then please check your Windows Pagefile.
Make sure its enabled for the C drive only and disabled for any other drive.
And it should be set to system managed and not custom sized.
https://www.tomshardware.com/news/how-to-manage-virtual-memory-pagefile-windows-10,36929.html

[honest quail]

I've been on this before

I think it's when I had different specs

[honest quail]

I imagine that's not what it's supposed to be

[ornate elk]

Nope

Set it to system managed.
Then restart the PC

[honest quail]

alright I'm back

let's test this

oh shit

it might have worked

it would be wild if that's also what's been causing my gpu to act up since I got it months ago

[ornate elk]

Maybe, or the GPU driver updated fixes it

[honest quail]

nah this was definitely it

I've updated my driver plenty since I got it

[ornate elk]

I mean the GPU part not the "can't switch models" issue

[honest quail]

well my screens would randomly bug out and programs would randomly crash

none of that is happening now

[ornate elk]

Ahh well yea that could be related to a fixed pagefile that smal

[honest quail]

I was about to buy a new gpu 😭

[ornate elk]

Also you can open up a cmd and type
SFC /Scannow
To let windows fix itself

Always a good idea after big windows updates

[honest quail]

for some reason it needs admin even though this is the only account on the pc

[ornate elk]

Yea you have to open the cmd as admin

With right click start as admin

[honest quail]

now I gotta learn how to train custom models and stuff

oh shit it's 9am

[ornate elk]

For training I would suggest training loras. That can be done with seperate tools like Kohya_ss or OneTrainer

[honest quail]

completely off topic but is there a way to use it on my laptop while hosting on this pc

[ornate elk]

Yes thats easy.
Just edit the webui-user.bat again and add --listen
To the Commandline_args.
Then relaunch it.
Then you can access it by typing in the IP address of your PC + the port of SD in the browser of the Laptop or smartphone

[honest quail]

the port of the sd?

[ornate elk]

:7860

[honest quail]

I'm bad at this

yay I did it

[honest quail]

I assume I've got to be connected to the same network right

[ornate elk]

Yes

[severe saffron]

Hello. I have an issue with forge ui.
I used to use forge ui with no issues. Removing it from my pc and reinstalling 2 days ago, I can't use it anymore.
My system specs are:
1650 (4gb)
8gb ram
amd ryzen 7 3700

the error message is attached.

I tried all those --no half and shit and nothing works. I tried everything I saw on the internet. Interestingly VAE is set to float32 and nothing changes it. I tried reinstalling, updating pytorch and everything but still the issue persists.

[strange patrol]

even when i try to use pip install torch torchvision torchaudio -f https://download.pytorch.org/whl/cu111/torch_stable.html the cuda version that will install is 12.1(different of my system), any solution?

[ornate elk]

Hey, how does your webui-user.bat looks like?

[severe saffron]

Its the default. I tried many different arguments but nothing worked so I reverted back

[ornate elk]

Can you add --no-half to the Commandline_args= ?

And then relaunch

[severe saffron]

I did it didnt work 😭

The hell is that emoji bro i ain crying

When I try to generate something nothing is generated and I get this error. runtimeerror: "layernormkernelimpl" not implemented for 'half' "layernormkernelimpl" not implemented for 'half'

I am about to give up really cuz nothing seems to work

[ornate elk]

You can still try install the old forge version

Because the new one is an experimental build

[severe saffron]

How do I get that? I installed by copying the git repository

Yk from the terminal and I didnt see a link for the old versions

[ornate elk]

Here is the old version:
https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/849

[severe saffron]

Thanks so much bro you are the best

[ornate elk]

No problem

[obsidian bluff]

yo is it normal for my 3090 to take like 2 hours to generate a 2048 image with 40 sampling steps

[ornate elk]

Nope.
What webui do you use?

[obsidian bluff]

mb btw i meant batch count 1 batch size 8

i use comfyui and the regular stable diffusion thing

either one

it takes the same time

[ornate elk]

Oh okay

[obsidian bluff]

is that normal still

[ornate elk]

Try batch count instead of batch size
Or reduce batch size

[whole widget]

So look there is about 5 mentions of this error in this discord and 0 of them have been solved from what I can gather. No one posted a solution at least.

Does anyone know how to fix this error? I have done 2 fresh installs back to back and got it both times. Everything else is a+

` File "F:_cozyi2\ComfyUI_windows_portable\python_embeded\Lib\site-packages\PIL\Image.py", line 3498, in open
raise UnidentifiedImageError(msg)
PIL.UnidentifiedImageError: cannot identify image file 'E:\_input\n-suite\1 (1).mp4'
got prompt
Failed to validate prompt for output 14:

• LoadVideo [n-suite] 45:
  • Value not in list: video: '1 (1).mp4' not in []
    Output will be ignored
    Failed to validate prompt for output 16:
    Output will be ignored`

[obsidian bluff]

still takes the same amount of time

[ornate elk]

Okay do you use xformers in Auto1111?

[obsidian bluff]

no when i launch it it says "xformers not found laucnhing without it"

[severe saffron]

i downloaded this and it didnt work eithher

giving up

[ornate elk]

Oh then you need to edit the webui-user.bat and at the line commandline_args=
You need to add --xformers
Then save and relaunch

[ornate elk]

Then try my nvidia Automatic1111 guide from the pinned messages of this channel.
It will work 100%

[severe saffron]

i hate A1111. when I used to use it it would generate a 512x512 photo in 2-5 minutes. Forge allowed me to do the same in 30 seconds usually and 1 minute at most

I dont know whats wrong really. I used to use forge with no issues uninstalled and reinstalled and now it doesnt work.

[obsidian bluff]

like this?

[severe saffron]

what even is a kwarg i swear to god boutta throw my pc out of the window

[tight sinew]

I was trying to update comfy nodes because I couldn't get flux to work and now my stable diffusion is all messed up. I ensured python was 3.10 and that the venv was deleted and reinstalled and I get "requests.exceptions.ReadTimeout: HTTPConnectionPool(host='127.0.0.1', port=7861): Read timed out. (read timeout=3)"

I tried changing the port and it still didn't work, and I allowed the port through the firewall and still isn't working. I'm regretting trying to get comfy to work now that I can't even get a1111 working anymore

[tight sinew]

I would really appreciate any help I've been trying to fix this for almost 3 hours now

[lost stump]

Hey guys 👋!
Is there any way we can set alert for stability api credit exhaustion.
??
Help needed

[tight sinew]

okay updating with the manager was probably the issue causer because I did attempt that

at the moment im uninstalling everything

got it

[ornate elk]

Yes

[tight sinew]

yeah I did do that

[obsidian bluff]

alr bet thx

[tight sinew]

I appreciate your help btw

[ornate elk]

Can you show the full cmd log?

[tight sinew]

hardly anything stresses me as much as fixing stable diffusion

[plain bronze]

Hey, how to update my sd1.5 to sdxl?

[severe saffron]

I ended up trying it to see if it has the same issue. Just like i said, it takes 10 minutes to start the process and then another 5 minutes to generate one image. Infinite times more than what it would have with forge.

A1111 isnt worth trying unless you spend like a houses worth on a pc.

[woven fjord]

/画图

[plain bronze]

How can I do that? Deleting this model and download sdxl?

From Civitai, I select sdxl basemodel filter and download one of them?

Oh right, I totally misunderstood the system :)

What is sdxl hyper?

SDXL lightning, turbo...

What is the difference?

[obsidian bluff]

is it normal for just one image generation to take like a minute

w the same settings i told you

[ornate elk]

Screenshot your txt2img settings

[boreal iris]

why does inpainting draw random gibberish in the masked area instead of following the prompt?

like here

i tried to inpaint a cat

[ornate elk]

Why do you even generate on 2048?
Sdxl is trained 1024x1024
And going to high over that causes duplicates

[ornate elk]

Set inpaint are to "mask only"
Then set the resolution to 768x768

[thorn junco]

hello

I have a question

[ornate elk]

Hey

[thorn junco]

Hello Cs1o

[obsidian bluff]

idk cause someone told me its better quality

is it normal tho for those times

[ornate elk]

Normaly you generate at a lower resolution and then upscale to get much better outputs

[obsidian bluff]

oh i got sd upscaler or sum but idk how that works

[ornate elk]

You can use hires fix for upscaling.
Set the resolution to 1024x1024
Then enable hires fix.
Set hires steps to 10 or 15
Denois to 0.5 or 0.4
And as Upscaler try Esrgan4x or Latent bicubic
Upscale by "2"
Then you get a 2048x2048 output

[boreal iris]

@ornate elk this happens

[ornate elk]

Try a bigger mask

Or try set the resolution to 1024x1024

[boreal iris]

its just doing this now

@ornate elk

[ornate elk]

strange

[boreal iris]

by the way how do i update? it says to do "git pull" but i have no idea where

[ornate elk]

go into the stable-diffusion-webui folder
click into the explorer bar (not searchbar)
then type cmd and hit enter
then type git pull

[thorn junco]

hello someone can give me a hand please? You do not have a valid subscription. Please login with your Discord account while signing up to https://stability.ai/stable-artisan#choose-stable-artisan-plan.
Important: Make sure you click 'Continue with Discord' at the login screen!

[ruby yoke]

wrong channel for this. post this in #🗣｜artisan-support-feedback

[plain bronze]

How to know which SDXL I must download?

[ruby yoke]

if you're on civit, you're probably downloading checkpoints and loras. if you want the base model, you might want to get it from stability.ai

[plain bronze]

I want checkpoints and loras but I don't know which base model checkpoints I must download.

[plain bronze]

Among these.

[ruby yoke]

read the pages they are on to find out what they do

[plain bronze]

I need DPM+++ SDE Karras here but I don't have. What is the problem?

I don't have other Karras sampling methods too.

[ornate elk]

scheduler has karras

[plain bronze]

So this is DPM++ SDE Karras?

Thank you!

[ornate elk]

note that DPM++ SDE takes double the time than using Euler a or DPM++ 2M

[barren monolith]

@ornate elk hi i get this error when open webui-user

[plain bronze]

Why?

[ornate elk]

hey, try remove --medvram-sdxl from the webui-user.bat and then relaunch

[ornate elk]

its a different technique

[barren monolith]

so it will be like this ? set COMMANDLINE_ARGS=--use-zluda --update-check --skip-ort

?

or delete whole line?

[plain bronze]

I know, but I don't know what is different.

[ornate elk]

no only --medvram-sdxl

[ornate elk]

Sampler overview: https://stable-diffusion-art.com/samplers/#Samplers_overview

[plain bronze]

I donwloaded 4x ultrasharp and dropped it into ESRGAN folder but it didn't work.

[barren monolith]

its still same error

[ornate elk]

okay very strange. then there is only one thing you can try:
uninstall HIP SDK 6.1 and install HIP SDK 5.7 and then restart the PC.
After that you need to again replace the Zluda files with these ones here:
(they are made for 5.7)
https://github.com/lshqqytiger/ZLUDA/releases/download/rel.11cc5844514f93161e0e74387f04e2c537705a82/ZLUDA-windows-amd64.zip
Then you have to add %HIP_PATH%bin again to the system variables.
And then you have to replace the rocm library files again with the one from my guide for your card. (gfx1031)
then restart the PC again.
Then delete the ven folder and the .zluda folder inside the stable-diffusion-webui-amdgpu folder.
Finally launch the webui-user.bat

[plain bronze]

What should I do?

[barren monolith]

okay thanks will try

[ornate elk]

models/esrgan ?

there shouldnt be a arch file in there

[plain bronze]

There is no esrgan folder in models folder.

[ornate elk]

because you didnt used an upscaler in the webui before

then create a folder in models called ESRGAN

[plain bronze]

Is RealESRGAN folder okay in models folder?

[ornate elk]

upscaler go int ESRGAN only

[plain bronze]

Okay.

It didn't work @ornate elk

[obsidian bluff]

thanks

[ornate elk]

are you in txt2img or img2img?

[plain bronze]

img2img

[ornate elk]

okay go into the txt2img tab and generate an image using hires fix

thene relaunch the webui

[plain bronze]

It worked!

Thanks :)

[ornate elk]

no problem 🙂

[barren monolith]

sir this step i confuse how to do it

Then you have to add %HIP_PATH%bin again to the system variables.

[obsidian bluff]

what does a refiner do

[ornate elk]

then click on New

and there add the path to the zluda folder and second entry for
%HIP_PATH%bin

[ornate elk]

it was used to enhance the quality of the base SDXL 1.0 model. now its not used or needed anymore because custom models + upscaling have a better quality

[barren monolith]

like this sir?

[obsidian bluff]

alright bet thanks

[ornate elk]

yes

that could be the issue why it wasnt working for you

so maybe restart the PC and try again before uninstalling hip 5.7

[barren monolith]

okay

[mellow sequoia]

How do I generate images? I can't find the channels to type

[ornate elk]

Its not free anymore. Channel is #1237459938901491852

[barren monolith]

is this already work now sir ?

[ornate elk]

you now have to select a checkpoint at the top right. then prompt Cat and then hit generate

then check the cmd

the generate button should stay Gray when pressed

first image will take longer if everything is working

[barren monolith]

its like this

[ornate elk]

can you show me the full cmd log?

[barren monolith]

this one since install until open and generate im not closing cmd log

[ornate elk]

restart the PC and then delete the venv and .zluda folder from the webui
then check the webui-user.bat that you have --use-zluda in there

[astral oracle]

Hey chat

How can one install auto1111 on a AMD cpu?

[barren monolith]

okay i try