🤝｜tech-support

[mystic lava]

wait it cooked

why

😭

1040x1040

[ornate elk]

The models aren't trained on that

[mystic lava]

ahh

whats the max resolution u recommend

[ornate elk]

1.5 based models are trained on 512x512
And sdxl/pony models are trained on 1024x1024

You can always upscale to higher resolution.

[mystic lava]

whats the one im running?

True

[ornate elk]

Dreamshaper v8 is 1.5 based

You can always check on civitai in the info box of the models

[mystic lava]

is there content filters?

Like can it draw violence

nsfw

alat

[ornate elk]

Nope

No filters

[mystic lava]

actrually

?

thats dangerous

💀

actually thats cool

i can do the rock oiled up

no way

so how do i make different styles?

[ornate elk]

You can prompt for them

Or use style loras

[mystic lava]

aight

how do i prompt for them

and what or loras

[mystic lava]

can u teach me how to do video gen

[steady thistle]

@ornate elk Would it be best to start with Swarm UI or the regular Comfy?

@ornate elk I tried Swarm and did the launch-windows.bat and got this:

The system cannot find the path specified.
""
""
"WARNING: You did a git pull without building. Will now build for you..."
""
""
The system cannot find the path specified.
The system cannot find the file specified.
The command could not be loaded, possibly because:

• You intended to execute a .NET application:
  The application 'nuget' does not exist.
• You intended to execute a .NET SDK command:
  No .NET SDKs were found.

Download a .NET SDK:
https://aka.ms/dotnet-download

Learn about SDK resolution:
https://aka.ms/dotnet/sdk-not-found
The command could not be loaded, possibly because:

• You intended to execute a .NET application:
  The application 'build' does not exist.
• You intended to execute a .NET SDK command:
  No .NET SDKs were found.

Download a .NET SDK:
https://aka.ms/dotnet-download

Learn about SDK resolution:
https://aka.ms/dotnet/sdk-not-found
The system cannot find the path specified.
The command could not be loaded, possibly because:

• You intended to execute a .NET application:
  The application 'src\bin\live_release\SwarmUI.dll' does not exist.
• You intended to execute a .NET SDK command:
  No .NET SDKs were found.

Download a .NET SDK:
https://aka.ms/dotnet-download

Learn about SDK resolution:
https://aka.ms/dotnet/sdk-not-found
Press any key to continue . . .

[lunar perch]

Wsl

[steady thistle]

I did the preparation and did C:/Ai

[brittle lantern]

I'm checking again after I removed the command I'll wait for gen to be over so I can send u it

[lunar perch]

How do I install stable diffusion on wsl?

I ran the powershell cmd in windows

Install wal

Now shows a penguin on my desktop

Clicking it does nothing

Wat do I do next?

They said I have to run SD on wsl

[brittle lantern]

@ornate elk so now it worked and the gen is fine but it stays on 96% for quite a while before it gets done (using SD1.5 model)

25/25 complete 100% done on CMD but the UI shows 96% interesting

Nothing shows up after the gen is done no errors

[steady thistle]

@ornate elk I got it. I'm running Windows 11, but I installed .Net anyway and that was the issue.

[brittle lantern]

WOW 2min. 53.8 sec cuz it was stuck on 96%

[mystic lava]

Does anyone know how to locally install flux ai

ik it isnt stable if you coukld direct me to where i could download it

[ornate elk]

Can you show the settings you tried?

[mystic lava]

Like locally run it

Ive seen its reallyt good for video generation

or how to use Stable diffusion to do video generation

Im sure that would suffice

How do yuo install comfy and does it run on amd gpus

whats ur personal recomendation

i can see from your youtube you know a lot abt it

nice videos btw

aight

for comfy and stable video?

comfy looks complicated af

[ornate elk]

Yes Video generation is really not easy

[mystic lava]

peak

[ornate elk]

It would require a few hours to explain everything about it xD

But you can try out flux if you install either Comfyui or webui Forge

[mystic lava]

Fire

Which is easier comfyui or webui forge

[ornate elk]

Forge is easier

[mystic lava]

can it usse amd and windows

[ornate elk]

Yes, I made a guide for it

On my github you see all the guides for AMD.

[mystic lava]

https://github.com/comfyanonymous/ComfyUI?tab=readme-ov-file#installing

this?

[ornate elk]

No

The first link from the pinned messages

[mystic lava]

https://github.com/comfyanonymous/ComfyUI#installing

[timber oak]

not comfy

its automatic1111

[ornate elk]

https://github.com/CS1o/Stable-Diffusion-Info

[mystic lava]

Is that for flux?

[timber oak]

i think flux doesnt work on automatic yet

[ornate elk]

Flux is just a model.
And comfyui and forge can use it.
Auto1111 will add support later

[mystic lava]

it dont

is there an eta oin t hat

i assume not

[ornate elk]

Nope but flux is also not really fast for AMD

Will be better over time

[mystic lava]

fire 😭

so stable video it is

hgow hard is that

i was gonna do it ass myt project for college

the consequences of ai on media industry

check kofi

[brittle lantern]

Ok

[mystic lava]

500hz monitor

zamm

[ornate elk]

I didn't used it with AMD

So idk rn

[mystic lava]

i meant learning curve

not performance wise

[ornate elk]

Thank you! 🙂

[mystic lava]

criminally underpaid

for all the effort you put in

icl

[ornate elk]

It depends mostly on prompting, so not hard to learn

[mystic lava]

fire

can u send a link to ur tutorial

if u havce one

😭

or am i on my own for this shi

[ornate elk]

What was your GPU again?

[brittle lantern]

6600 xt

[mystic lava]

whats cfg scale

[ornate elk]

Ahh okay, and I guess you used sdxl models

[brittle lantern]

the one im using is sd 1.5 model

[ornate elk]

Ah okay

Then you should install the tiled diffusion with tiled VAE extension

It can prevent that freezing at the end

Because thats a VAE thing. When the resolution is to large

[brittle lantern]

ahhh i see

[ornate elk]

You can hover over some settings to get a little info about it.
Cfg is like the creativity or how strong the prompt gets followed

[brittle lantern]

I was a bit confused because it's never got stuck on 96% for 2 min before

[mystic lava]

ahhhhhhhhhh

so more cfg

is more creative

[ornate elk]

Yea then the prompt gets more ignored.
But also the quality will get worse.
So keep it between 7-12

[mystic lava]

peak

[brittle lantern]

I use 6 sometimes

[ornate elk]

If you installes the extension then only enable Tiled VAE and it should work

With that you can also upscale easier

[brittle lantern]

Yh if it still takes 2 min with it on does that mean it's just the model?

Cuz for like the whole 2 hrs yesterday I was convinced I had to reinstall 😭

[ornate elk]

It takes longer because you used the DPM++ SDE sampler.
This sampler doubles the generation time

Better use DPM++ 2m or Euler a

[brittle lantern]

Yh the model page said I can use Euler a

Euler a is Euler ancestral

?

I mean ill assume so

[ornate elk]

Yep

[brittle lantern]

Also new update for Auto1111 shows on my CMD should I update or no?

[ornate elk]

Yea you can update

[brittle lantern]

Ok just making sure so for git pull command I put that where?

The bat file at the bottom on a new line?

Or?

[ornate elk]

Before the call webui.bat

So not the bottom

[brittle lantern]

Ok so I just type "git pull" that's it?

Ty I was just asking cuz idk if updating can break AMD support

[ornate elk]

Normaly not because you installed the webui made for AMD

So it has the patches in it to work

[brittle lantern]

Ty one more question if I had a higher end rx 6000 or rx 7000 card would it perform better on Linux with ROCm

I couldn't find a answer online

[ornate elk]

Yea on Linux it would perform a bit better.
For my 7900xtx its 20 it/s on Linux to 16 it/s on zluda windows

But for me it wasn't worth the dual boot hassle

[brittle lantern]

20 it/s is comparable to which Nvidia GPU?

[ornate elk]

3090 ti

[brittle lantern]

Very nice

[ornate elk]

Yea but it/a drop when using more words, higher resolution, upscaling, loras etc

[brittle lantern]

3090 vs 3090 ti how much slower is 3090 in comparison if you know the answer (sorry if I'm asking to many questions just let me know if its to much)

[brittle lantern]

Understandable

[ornate elk]

Here are charts to compare

That was before Zluda was published

[brittle lantern]

With Euler A I get 9.94 it/s

Pretty good

[brittle lantern]

👍

sigh another issue do i do nan check again

[ornate elk]

No dont

Show the settings when you get that error

Its most likely a wrong file or setting used

[brittle lantern]

settings in the settings tab or my gen settings?

[ornate elk]

Gen settings

[brittle lantern]

ill show

[brittle lantern]

i didnt get it again

my pc is dodgy

@ornate elk got it back

[mystic lava]

my img to img isnt rlly working

in stable

NansException: A tensor with NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

[ornate elk]

try upadint your extensions

[mystic lava]

trying to use inpaint

[brittle lantern]

update?

[mystic lava]

ye doesnt work no matter what i do

[ornate elk]

click on extensions, then on Check for updates.
you then also need a relaunch

[ornate elk]

can you show me your img2img settings ?

[mystic lava]

yea

cant show the image for reasons 🗣️

[ornate elk]

your not in the inpaint mode

[mystic lava]

my bnad

inpaint

there

i tried using a prompt one to

none of it works

:(

[ornate elk]

and the model is dreamshaper ?

[mystic lava]

no

lemme switch to it

it was on an experimental one

tghat dfixewd it

fire

try

doi some models not support it

[ornate elk]

the other one might be not a model

you have to always check the info box of the models page

loras for example go into models/lora folder

[mystic lava]

aight

can i use inpaint

then ask the ai to add things to the image

[ornate elk]

yes

for that set the inpaint area to "only masked"

[brittle lantern]

update and restart button

[safe bloom]

Can someone advise ? I am somewhat experienced with stable diffusion (have used it on an Nvidia equipped cloud server), and found out recently it is possible to run it locally on my Radeon Vega 7 equipped laptop. I managed to install it to the point it where it seemingly works. It loads the model and starts generating an image. It finishes the generation in a bit under 2 minutes and shows an image that corresponds to prompt during intermediate steps, but the final result is just gray square. How should I start to troubleshoot this ? I am completely clueless right now.

Correct image during generation

But end result is a gray square...

This is the "lshqqytiger/stable-diffusion-webui-amdgpu" fork if that makes any difference, running using directml

[ornate elk]

whats in your webui-user.bat?

[safe bloom]

@echo off

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

call webui.bat

[ornate elk]

you need --use-directml --lowvram --opt-sub-quad-attention --upcast-sampling
at the line set COMMANDLINE_ARGS=

[safe bloom]

Added those, but the result is still the same. Correct image during generation, but a gray box as result.

Downloading another checkpoint to rule out if the problem is caused by this particular one.

[near edge]

can someone help plz?

[safe bloom]

Had success with another model/checkpoint !

This is pretty cool. I can generate images on a igpu laptop in 1 minute 40 seconds. Not fast but still good enough for hobby use.

[mystic lava]

@ornate elk

am i missing somehting here

💀

[timber oak]

Hey

i got an error when using control net

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

what does that mean?

[ornate elk]

can you try an other model

for example dreamshaper v8

[safe bloom]

Already did, and it worked see above.

Thanks for the help !

[ornate elk]

ohh

nice np

you can delete the other model then

[ornate elk]

that means you used two different things like an sdxl model and an 1.5 lora or an 1.5 model with an sdxl lora

they wont work together

[ornate elk]

thats now how it works, you dont command the ai to change stuff. it cant do that, you have to prompt what it should generate and you need to lower the denois value

[timber oak]

oooh ok thank you

[ornate elk]

whats in your webui-user.bat?

[mystic lava]

aight

[ornate elk]

whats your gpu?

[near edge]

4060 ti

[ornate elk]

then follow my install guide fro mthe pinned messages

the one for nvidia

[near edge]

still not working

[ornate elk]

is D an external drive?

[near edge]

yes

[ornate elk]

ah well

[near edge]

we tried on both

[ornate elk]

then it needs to be reformated to GPT

[near edge]

on c as well

[ornate elk]

on C it should work

but dont copy it over

git clone it into C:AI

[near edge]

its not working on c either

its current path

C:\stable_diffusion\stable-diffusion-webui

(venv deleted)

[ornate elk]

please use cmd and not terminal

[near edge]

It is cmd

[ornate elk]

also whats your python version

[near edge]

3.10

[ornate elk]

3.10.?

3.10.11 is needed

[near edge]

i ahve 3.10.6

it could be the issue

[ornate elk]

nah 3.10.6 should work too

[ornate elk]

this is a terminal not a old cmd

[sleek pewter]

Create me an image with nude or pastel colors, that says "La Real Gracia" and is simple to use as a profile.

[ornate elk]

maybe reinstall python then

[near edge]

its working

thank you very much

[ornate elk]

no problem 🙂

[pale salmon]

sup i download back forge, but its writting "amd" either way im on nvidia, its normal ?

[brittle lantern]

I'm convinced it's might be a diff issue because on a diff SD 1.5 model I'm running 512×512 and it gets stuck again but this time it doesn't even move I've sat there looking at it for 4 minutes at one point idk what's happening

[ornate elk]

yes, its refering to a cpu architecture

so can be ignored

[pale salmon]

alr ty

[brittle lantern]

Should I try reinstalling auto and start again

[ornate elk]

can you show me the cmd?

[brittle lantern]

Alr

[brittle lantern]

never moves

[ornate elk]

looks normal

[brittle lantern]

Yep it does until it doesn't come off 73%

So either it's extremely long for no reason

Or sum else it happening to my pc

Cuz I remember doing this a while back

For the first time and it working fine none of these long wait time issues

But yk now it's just slow

Or broke or both

[ornate elk]

make sure your not running any games or wallpaper engine in the background

[brittle lantern]

I don't use WPE and I have no games on

[ornate elk]

do you have tiled vae enabled?

[brittle lantern]

Yep

Default setting nothing changed when enabled

I am using Adetailer for Eyes and face could that be a issues

It wasn't before but maybe it is now?

[ornate elk]

have you updated all extensions?

[brittle lantern]

I did what u said

And restarted

I even disable both to see if it was them but it had same issue

Both VAE and Adetailer

In extentions but no fix

So I enabled them and updated then still no fix

U think the gen is doing the 40 min gen thing that was meant to happen first time again for some reason?

I can't think if many options that can cause this issue

So I let it complete the gen I was doing after it was stuck and it gave me 3 min 40 seconds

That's quite long no for 1.5?

On my system

My macbook did it in sub 2 min m2 pro on the same gen

wait @ornate elk i got this now

[ornate elk]

that can happen when torch version or amd drivers got updated

[brittle lantern]

finally a error i can go off gave me black gen just now and nanexception

[ornate elk]

try add --upcast-sampling --no-half-vae to the webui-user.bat

[brittle lantern]

I'll try now

[brittle lantern]

I now have got cuda out of memory...

Tried to allocate 8.31 GiB

sigh

[ornate elk]

what are your txt2img settings again

8gb is to much

what yre you trying

[brittle lantern]

Euler a, 25 steps, 1.3× upscaling 0.25 demonising, 512×512 and the Adetailer and tiled VAE on

Rest is off or default

[ornate elk]

make sure you use 10 hires steps when upscaling

[brittle lantern]

I'll try that now

[ornate elk]

also only enable tiled ave and not tiled diffusion

[brittle lantern]

I have VAE only enabled

Not the other

[brittle lantern]

It worked 3 min 27 sec

Idk what the avg time for that type fo gen on my hardware should be

But hey at least it works

(Finally) and I hope it stays working

[ornate elk]

upscaling takes a bit longer

[pale salmon]

i writted this wrong ?

[ornate elk]

whats your gpu?

[pale salmon]

GTX 1080µ

[ornate elk]

its --theme dark --xformers --no-half-vae --medvram-sdxl

[pale salmon]

okay thans

yeah my pc crash mid download,@ornate elklowvram do not exist ? or something better ?

[ornate elk]

what setting or model do you try?

[pale salmon]

pony model

[ornate elk]

ah your using forge

[pale salmon]

with classic

y

[ornate elk]

there --medvram-sdxl wont work and also xformers wont work in forge

better use auto1111

[pale salmon]

alr

will tryt

[ornate elk]

np

[untold hollow]

I've installed sv3d and svd.safetensors from https://huggingface.co/stabilityai/stable-video-diffusion-img2vid and https://huggingface.co/stabilityai/sv3d/tree/main
but when running scripts/sampling/simple_video_sample.py or python scripts/sampling/simple_video_sample.py --input_path <path/to/image.png> --version sv3d_u
as stated in https://github.com/Stability-AI/generative-models?tab=readme-ov-file
im getting
safetensors_rust.SafetensorError: Error while deserializing header: MetadataIncompleteBuffer

im on linux, python 3.10, could i get some help?

[pearl musk]

Hey people, why is a1111's UI so painfully slow compared to forge? Sending an image from the PNG Info tab to txt2img takes like 5 seconds, when on forge/reForge is instant, I have 32GB of RAM, 5600X, 4060TI 16GB and SD is installed in a SSD.

[ornate elk]

whats in your webui-user.bat

[pearl musk]

"--xformers --disable-nan-check --enable-insecure-extension-access --listen --port 7861 --device-id=0"

[ornate elk]

remove --disable-nan-check

its not fixing anything

[pearl musk]

Crap, you're right, must have come from my old webui-user file, from when I had a GTX 1080, lmao

Hmm, still slow, the UI I mean

[sage mantle]

ok i havent touched stable diffusion for a long time after a format i reinstalled the whole gig but things look different and options i had are missing from ui like the search bar under extensions is there any way to perhaps get it added seeing as i dont really fancy searching through thousands of extensions for what i need

[ruby yoke]

how much is your CPU carrying of the load and how much is your GPU carrying

[ornate elk]

make sure to whitelist the webui in any browser adblocker

[ornate elk]

there is a search bar
its the white line below the hide button

[pearl musk]

Oh, let me check that

[sage mantle]

i swear im not seeing it but lemmy double check

argh thank you sorry i simply cannot see that clearly

[ornate elk]

np, the dark mode is better anyway

maybe try that

[sage mantle]

is it built in yet last i remember i had to add it as an extension

ill be honest im basically learning from the start again more or less

[ornate elk]

its build in

if you set your windows to dark mode

or by addint --theme dark

to the webui-user.bat

[sage mantle]

argh yes ty

i bow to your support cheers again

[ornate elk]

np

[pearl musk]

Still slow as hell, wtf

What makes forge's UI so fast? They use the same gradio version and all

[ornate elk]

It could also be caused by --listen

Because with that you host the webui

For yourself

[pearl musk]

I'll try that too, let me see

[ornate elk]

I had slowdowns with it too

If you dont need it remove it and the insecure extension access

[pearl musk]

I use lots of extensions, last time I removed that option I could not install extension lmao

I would move to forge, but lots of extension are not working

[ornate elk]

Because you had --listen in it.
That blocks installing extensions

[pearl musk]

Duh, what an idiot I am

No, still slow, writing prompts makes the prompt counter process for 3 seconds still, and the UI is still sluggish

Maybe it is because I am on Linux?

[ornate elk]

Hmmm

What browser do you use?

[pearl musk]

Tried on Chrome/Vivaldi/Firefox, Chrome has hw accel disabled

[ornate elk]

Okay

[pale salmon]

i guess stable is not for me

but comfyui work fine so idk

[pearl musk]

Are you using your CPU?

[pale salmon]

dont know

lol

[pearl musk]

Do you know your PC specs?

[pale salmon]

yeah

[pearl musk]

What GPU do you have? is it nvidia, intel, or amd?

[pale salmon]

this ?

[pale salmon]

nvidia 1080

[pearl musk]

Should be a lot faster than that

[pale salmon]

it was on gpu already

XD

[pearl musk]

That is the RNG, that option tells Stable Diffusion to use your GPU to generate the random seed (seed is the "pointer" to where to start to generate the image, a number) are you on Windows?

[pale salmon]

yeah windows 10

[pearl musk]

Go to your Stable diffusion folder, and look for a file named "webui-user.bat"

Open it with a text editor, Notepad should work

[pale salmon]

🥲

[pearl musk]

You have it already, and it still is slow, that is not normal, are you drivers up to date?

[pale salmon]

yeah my gpu is last update

[pearl musk]

replace --no-half-vae and --xformers with --upcast-sampling --opt-sub-quad-attention

[ornate elk]

Have you launched the webui-user.bat or the webui.bat?

[ornate elk]

No don't xD

Opt sub is much slower

[pearl musk]

xformers is not going to improve his speed anyway

https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/5581

[ornate elk]

Can you show the whole txt2img settings

[ornate elk]

That bug is from 2022 ...

[pearl musk]

Is not a bug, and facebook removed support for older GPUs

[ornate elk]

Xformers improves speed

[pale salmon]

img2img

txt2img worked fine, slow but not that slow

[ornate elk]

Will test tomorrow with my GTX1080

[pearl musk]

Do try, I would, but my GTX 1080 is long gone 😦

[ornate elk]

What do you wanted to do?

Upscaling?

[pale salmon]

yeah

add detail etc

[pearl musk]

I think I broke it xD

[ornate elk]

Ahh yea thats to be done different

[ornate elk]

Wrong resolution for sdxl mostly

[pale salmon]

can i post nsfw here in spoil?

[ornate elk]

No

[pale salmon]

kk, bc the first pic is nsfw

[pearl musk]

I'm using a finetune, recommended res is 832x1216, it fixed itself after I killed the process

[pale salmon]

tryed to img2img this

[pearl musk]

Is that an image you made? Do you wish to fix the face?

[ornate elk]

He wanted to upscale it in img2img

[pale salmon]

yeah thats the sketch basicaly

[ornate elk]

But without SD upscale script it will take long

[pale salmon]

with txt2img

[ornate elk]

I need to go off now.
Can explain upscaling tomorrow.

[pale salmon]

np,night

[pearl musk]

Try adetailer for faces, shold fix it without taking too much time

You can find the extension in "Extensions>Available>Click on 'Load from:'>Then write adetailer in the search bar just below

[gilded cloud]

sd3 doesn't work properly for me..

Flux is even worse, only a grey square

Is this a known issue?

[ruby yoke]

that looks like you've got the wrong scheduler and sampler chosen

don't use euler with sd3. use heunpp2

[gilded canyon]

When I use Hi-Res Fix, at around 95% my whole computer slows down a LOT and it hangs for a few seconds to even a minute. I'm using Chrome with a 4060ti 16gb and 32gb of ram

[ruby yoke]

which interface? a1111? comfy? something else?

[ornate elk]

Make sure you use 10 hires steps. And also using Tiled VAE extension can prevent the freeze

[gilded canyon]

a1111

[gilded canyon]

How do I download the Tiled VAE extension?

Is it just in the extensions tab

[vocal burrow]

yes, it might be under the "multidiffusion-upscaler-for-automatic1111" name nowadays.

[gilded canyon]

Got it, thanks :)

[molten igloo]

when trying to use flux with comfy i get this error
[F dml_util.cc:118] Invalid or unsupported data type Float8_e4m3fn.

I use flux1-fast-fp8 version
the error sounds like I am missing fp8_e4m3fn weight_dytpe but i am not using them... I think? at least I have no node where it is picked.

[ornate elk]

Hey, that could be a directml related issue

Flux wouldn't work anyway with directmls vram usage

[molten igloo]

(╯°□°）╯︵ ┻━┻

Aye aye I will use your tutorial, best advertisment ever

[ornate elk]

I can't guarantee that it would work with 12gb vram with Zluda too. Because no one has tested it yet.
It works for me with 24gb and someone with a 6800xt with 16gb

[molten igloo]

I will be the tester! in a few days.. when I have time and motivation

[ornate elk]

Alright let me know ^^

[mystic lava]

Hey im looking for how to install flux with comfy on an amd 7900 xt

[ornate elk]

Hey you can do that with my AMD Comfyui Zluda Guide

[thorn badge]

hi there..i am running flux in forge, and i was hoping someone could explain what this means in the cmd prompt as model runs

[Unload] Trying to free 18359.89 MB for cuda:0 with 0 models keep loaded ...
[Unload] Current free memory is 8456.80 MB ...
[Unload] Unload model JointTextEncoder
[Memory Management] Current Free GPU Memory: 8959.56 MB
[Memory Management] Required Model Memory: 6246.84 MB
[Memory Management] Required Inference Memory: 10239.00 MB
[Memory Management] Estimated Remaining GPU Memory: -7526.28 MB

i have 32gb ram, plus 12gb on my 3080RTX.

[ornate elk]

Hey, its just defecting how much vram your GPU has, how much is required for the task and how much is needed

You should use the nf4 flux model for the best performance and lowest vram usage

[hazy wigeon]

using ComfyUI with Krita plugin. I connect to local server but apparently some custom nodes are missing. tried installing from github url but returned these errors

[rapid tendon]

hi frens, im having issues w a1111 :c idk if this is the place to go, but it keeps getting stuck at loading weights, takes like 2 mins to load, if it even loads at all, while using 100% of my ram usage, sometimes completely freezing my computer, and this happens when i change checkpoints, or change any settings and hit apply, or anything like that. ive tried uninstalling and reinstalling, ive tried updating, downgrading. this sux i cant even use it and idk wat to do :c could anyone help

[sturdy ginkgo]

what are your system specs

[rapid tendon]

rtx 2080 super uhhh holdon lemme check the rest lmao

rtx 2080 super, i7 9700kf, 16gb ram

is there anything else

[sturdy ginkgo]

did you install on a SSD or HDD

[rapid tendon]

ssd yis

i tried it on my nvme but it was the same issue then i switched to ssd and still the same issue

[sturdy ginkgo]

try using forge instead as it has way better memory management: https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/981

[rapid tendon]

shore lemme try it

thx u

[sturdy ginkgo]

A1111 has really become a huge memory hog with all the whistles it adds by default

[rapid tendon]

it was so good like a year ago i tried it, and everything worked perfectly but now its all broken n sad. it kept forgetting where python was when i was installing it and cancelled the freakin install

this ones great thank u sm. it fixed it completely LOL set up was quick, didnt eat up ram, didnt take 5 mins, didnt freeze my pc. 10/10 ur the best

[grizzled cypress]

I’m having an issue with stable diffusion where each consecutive image starts taking a longer and longer amount of time (starts at around 6 secs per iteration, then goes up to like 30+ secs per iteration) This seems to happen at least on both forge and comfyui. I have 16gb ram, geforce gtx 1060 6gb vram. The problem starting happening suddenly because it was working fine a while ago.

[ornate elk]

What sort of models do you use ? Sdxl or 1.5 ?

[grizzled cypress]

XL

[ornate elk]

Try enabling FP8 mode for sdxl in the settings

[grizzled cypress]

It doesn’t seem to be an issue with the exact type of model because i experimented with a couple

[grizzled cypress]

Alright, i’ll give it a try in a bit

[ornate elk]

Also make sure you have --xformers --medvram --no-half-vae in the webui-user.bat

[grizzled cypress]

I have xformers, and vae, i’ll add medvram

[spring snow]

cant generate clear images with flux in comfyui. can anyone help?

[ornate elk]

Do you use a VAE?

[spring snow]

yea whatever vae according to the instructions. its called ae.sft in the vae folder

[ornate elk]

and which model do you use?

which flux version

nf4 or fp8 dev?

[spring snow]

fp8

[spring snow]

im folliwing the comfyanonymous.github.io instructions

ive tried flux dev and flux schnell

[ornate elk]

can you show the workflow?

[spring snow]

here are two different workflows provided by the website

[ornate elk]

hmm looks correct

[spring snow]

yeah i was hoping this was a common error that could easily be diagnosed. I followed everything perfectly

[ornate elk]

oh wait

your missing the Flux guideance in the first workflow

try this workflow:

[spring snow]

ill try to translate that code but im not that smart lol

[glass sable]

Windows, Comfy, stallign out on unifiedloader.
Error occurred when executing IPAdapterUnifiedLoader:

'clipvision'

File "C:\Program Files\comfy\ComfyUI\execution.py", line 151, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
File "C:\Program Files\comfy\ComfyUI\execution.py", line 81, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
File "C:\Program Files\comfy\ComfyUI\execution.py", line 74, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
File "C:\Program Files\comfy\ComfyUI\custom_nodes\ComfyUI_IPAdapter_plus\IPAdapterPlus.py", line 532, in load_models
if clipvision_file != pipeline['clipvision']['file']:

[ornate elk]

you just have to drag and drop it into comfyui

[spring snow]

ok cool ive give it a shot

[loud pond]

can someone help me with this error

what is it asking me to do?

please @me if you have a solution

[brazen bramble]

I’m having a really weird issue for some reason whenever I try to use a Lora it crashes but if I don’t use a Lora everything works fine any ideas ?

[ornate elk]

What's your GPU ?

[ornate elk]

What's the model and lora?

[loud pond]

a radeon 480 8gb

[brazen bramble]

the model is reweik_PonyXL https://civitai.com/models/404802/reweikponyxl?modelVersionId=568997 and the Lora is My Hero Academia https://civitai.com/models/626245/my-hero-academia

[ornate elk]

and what was you gpu again?

[ornate elk]

then you should use my directml amd guide from the first link of the pinned messages

[brazen bramble]

It was a gtx 1650

[ornate elk]

does an other lora work?

and what exactly crashes

do you get any error?

[brazen bramble]

I’ll show you now

[loud pond]

i followed the process and i got the same error

i used this one

[ornate elk]

which one of them

the AMD DIrectML one?

whats in your webui-user.bat?

[ornate elk]

oh you followed the nvidia one..

[loud pond]

oh mb

yeah ill re do it with the bottom one thanks

[ornate elk]

np, try the Automatic1111 with DIrectML

[grizzled cypress]

It seems forge does not support medvram

[brazen bramble]

ok so its not doing what it was earlier before i would click generate then the cmd window would have the press any key to continute text and close but now it starts making the image and then i get this error OutOfMemoryError: CUDA out of memory. Tried to allocate 172.00 MiB. GPU 0 has a total capacty of 4.00 GiB of which 0 bytes is free. Of the allocated memory 3.03 GiB is allocated by PyTorch, and 273.55 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

[ornate elk]

oh yea forge doesnt have that. but check if you can enable fp8 mode for sdxl

or nf4

[ornate elk]

oh okay, you may need to change --medvram to --lowvram then

[brazen bramble]

ok ill try that now thank you

[grizzled cypress]

Yeah, it has that

It’s still steadily increasing though

First image took 1:41 minutes, by the 5th i’m already taking 2:30

[loud pond]

thanks so much, its currently downloading everything but its working thank you!

[grizzled cypress]

Could be that gpu is not handling being run for a longer time so it ends up taking longer as time goes on but i have no idea

[ornate elk]

no, its not an hardware issue

[grizzled cypress]

Idk then

[glass sable]

Never got a response on this. Been troubleshooting 12 hours.

[ornate elk]

its comfyui :/ hard to troubleshoot

you can also try ask in #🧣｜comfy-ui

[brazen bramble]

i changed it to --lowvram but im still getting the same error

[ornate elk]

what resolution did you tried?

[brazen bramble]

i usally do 700 , 500

[ornate elk]

hmm okay

have you tiled vae enabled?

or the tiled vae extension installed?

[brazen bramble]

as far as i know i havent installed any extensions

[ornate elk]

ah you could try that extension as it can save a bit vram

[brazen bramble]

ok cool ill try that where can i install it ive never done anything with extensions before ?

[ornate elk]

go into the the extensions tab, there click on Available, then on "load from"

then you get a list

search for tiled

[brazen bramble]

gotcha ty ty ill do that now

[brittle lantern]

tryna run forge on macbook for gguf flux cuz why not and i got this

"TypeError: Cannot convert a MPS Tensor to float64 dtype as the MPS framework doesn't support float64. Please use float32 instead.
Cannot convert a MPS Tensor to float64 dtype as the MPS framework doesn't support float64. Please use float32 instead."

any fix

[ornate elk]

have you installed forge with the script for mac ?

[brittle lantern]

yep

[ornate elk]

hmm

no clue then

[brittle lantern]

rip

[ornate elk]

maybe try the normal nf4 model

[brittle lantern]

back to draw things for mac then

[ornate elk]

not gguf

[brittle lantern]

got link?

[grizzled cypress]

I’ll try reintalling the UI, maybe one of the extensions or something ended up breaking it

[ornate elk]

https://huggingface.co/lllyasviel/flux1-dev-bnb-nf4/blob/main/flux1-dev-bnb-nf4.safetensors

[brittle lantern]

Ty

[grizzled cypress]

tried with smaller resolutions as well but still the same issue, it starts out fine but increases with each image

[grizzled cypress]

even within the same image, the amount of time it takes for each step increases

One thing i realized is now the gpu is only using 4.7/6gb of dedicated memory, whereas before i’m pretty sure it was using more

[grizzled cypress]

The slowdown could be due to temperature throttling right?

[loud pond]

where do i go to install lora's?

[grizzled cypress]

I’m pretty sure this is it, if i wait like 2 minutes between generations it never happens

[glacial tide]

@ornate elk In the amd-automatic1111-with-zluda guide where it says to swap the library folder in C:\Program Files\AMD\ROCm\6.1\bin\rocblas with optimized ROCm libs from a zip, should I go ahead with this if my ROCm is 5.7 instead of 6.1?

Or should I update ROCm

I am using a 6600XT which I downloaded HIP 5.7.1 for, I wonder if that might be related

[ornate elk]

Yes you need to do it for 5.7 too

[glacial tide]

Alright just making sure those optimized library files will work with 5.7

Thx

[ornate elk]

You could also update rocm to 6.1 but then you need other Library files

But I can link them

[glacial tide]

Ill use 5.7 since that's what I have

Less risk

[ornate elk]

Alright

[glacial tide]

Not sure how risky it is tbh as that's a SDK... but it requires a restart so that does give me pause

[ornate elk]

Its not risky at all in this case

[glacial tide]

So just to be clear then, I would swap the library files in C:\Program Files\AMD\ROCm\5.7\bin\rocblas with the optimized ones in Optimised_ROCmLibs_gfx1032.7z

[ornate elk]

Yea make a copy of the Library folder first.

[glacial tide]

Yeah

[ornate elk]

Dont forget to restart after its done

[glacial tide]

Just did

I just wanted to clarify on that because in my experience with sys admin type stuff, mixing and matching versions of anything is liable to cause an issue that will waste a lot of time to google

Glad to know it works regardless

[ornate elk]

My guide links the rocm lib files for 5.7

Mixing them wouldn't work

[glacial tide]

Yeah but the guide says version 6.1

3.3 Go into C:\Program Files\AMD\ROCm\6.1\bin\rocblas folder.

That's what I was referring to

[ornate elk]

Yea thats just a typo in that one line. I will fix it tomorrow

[glacial tide]

Thx

[ornate elk]

No problem, thx for pointing it out

Its late, I'm off now.
For any questions feel free to @ me, then I will answer tomorrow

[grizzled cypress]

I added another fan and lifted have my laptop not touch the desk to let the underside of it cool down and the issue was remedied, and i won an extra 10 seconds in generation lol

[glacial tide]

If you told me stable diffusion would run on a laptop i would have laughed

is it like a gaming laptop?

because back when i had a fairly nice laptop it still struggled to even play fortnite... back in the day

[grizzled cypress]

Lmfao, i get 1:53 minutes generation per image in sdxl models now

Generation is pretty quick on regular 1 or 1.5 ones

[grizzled cypress]

It’s pretty old, just heavily optimised

Gtx 1060 and 16gb ram

[grizzled cypress]

So it “runs” but definitely not as good as most pcs nowadays

[rare summit]

i got the ui, but this keeps coming up

RuntimeError: The NVIDIA driver on your system is too old (found version 11060). Please update your GPU driver by downloading and installing a new version from the URL: http://www.nvidia.com/Download/index.aspx Alternatively, go to: https://pytorch.org to install a PyTorch version that has been compiled with your version of the CUDA driver.

I downloaded like three diff versions of PyTorch, trying to solve the issue anddddd nothing lol.

pls halp

[thorn pendant]

is there;s anything i need to take off, install more?

[waxen lantern]

Genuinely confused how sd just broke. Any checkpoints I use, I just get buildings. It was working 30 minutes ago. Tried to reload UI and restart SD and pc.

I just produced this image earlier and this is the same prompt now minus the hi res fix.

(80's, retro, city pop poster:1.5), (album cover), (masterpiece, best quality, intricate detail), (anime, illustration), (pastel colors:1.3), best photo pose, girl, solo, smile, perfect detail eyes, delicate face, City Scenes, night city, tokyo, high fashion, lora:add_detail:0.5, lora:80'sFusion:0.6 Negative prompt: ng_deepnegative_v1_75t, easynegative, (worst quality:2), (low quality:2), (normal quality:2), lowers, bad anatomy, bad hands, (multiple view), Steps: 27, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 8.5, Seed: 18918579, Size: 512x768, Model hash: b42b09ff12, Model: cetusMix_v4, VAE hash: 735e4c3a44, VAE: vaeFtMse840000EmaPruned_vaeFtMse840k.safetensors, Denoising strength: 0.45, Clip skip: 2, ADetailer model: face_yolov8s.pt, ADetailer prompt: "detailed face, ", ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer model 2nd: hand_yolov8n.pt, ADetailer prompt 2nd: "detailed hand,", ADetailer confidence 2nd: 0.3, ADetailer dilate erode 2nd: 4, ADetailer mask blur 2nd: 4, ADetailer denoising strength 2nd: 0.4, ADetailer inpaint only masked 2nd: True, ADetailer inpaint padding 2nd: 32, ADetailer version: 24.9.0, Hires upscale: 2, Hires steps: 13, Hires upscaler: 4x-UltraSharp, Lora hashes: "add_detail: 7c6bad76eb54, 80'sFusion: b0f47bf00919", Dynamic thresholding enabled: True, Mimic scale: 7, Separate Feature Channels: True, Scaling Startpoint: MEAN, Variability Measure: AD, Interpolate Phi: 1, Threshold percentile: 96.5, Mimic mode: Half Cosine Up, Mimic scale minimum: 4, CFG mode: Cosine Up, CFG scale minimum: 4, Downcast alphas_cumprod: True, Version: 1.10.0

[ornate elk]

Update your GPU driver.
Then delete the venv folder and relaunch the webui-user.bat

[ornate elk]

Lycoris and locon can be removed. They are natively supported like loras

[ornate elk]

Try without dynamic thresholding

[waxen lantern]

no luck. I tried different checkpoints and sort of the same result

random seed as well

[ornate elk]

I see you have Tiling enabled

Thats the issue

Disable it in the settings

Its for making texture images

[waxen lantern]

where is that setting?

[ornate elk]

Search for tiling

In the settings

[waxen lantern]

you are amazing! thank you

working perfectly fine. im not sure how I even toggled the setting

[ornate elk]

No problem

[rare summit]

How?

[ornate elk]

How what ?

[rare summit]

To update the GPU driver?

[ornate elk]

On nvidias driver site, select your GPU model. Then download the driver:
https://www.nvidia.com/en-us/drivers/

And then install it

[rare summit]

uhhhhhhhhhhhh

i will try

I'm lowkey dumb jfnvjsnijf

[ornate elk]

After that restart the PC.
Then delete the venv folder in the stable-diffusion-webui folder.
Then relaunch the webui-user.bat

[rare summit]

I don't think I installed the right one lol

[ornate elk]

Why?

[calm cairn]

hello all--
I have a Pony XL model but every time I use it, the art quality is shockingly bad. I have copied others' prompts, and that improves it, but it still remains very bad.

I now believe something is wrong with the model. I re-downloaded the model from CivitAI but nothing changed. So, how do I know if my model is wrong?

[ornate elk]

Which model?

[calm cairn]

https://civitai.com/models/257749/pony-diffusion-v6-xl

there are multiple tabs, but "start with this one" is the model

[fair oxide]

Why does the docker command fail?

When uncommenting the docker_args i get "something went wrong"

Im trying to create and connect via ssh via python. Creation works fine if docker_args is commented

[ornate elk]

Yea that model is not as good as other pony models

For anime better use anime pony models

And for realism there is pony realism

[calm cairn]

hmm i didnt expect that reply
so, Pony Realism is for realism, can you suggest a good pony anime model?

[ornate elk]

Anime confetti comrade mix

Is a good one

Or Perfect Pony XL

I think that's even better

[calm cairn]

ok I am downloading both of those right now, they look pretty good from the previews. Thanks for the recommendations

[finite zenith]

Is there an extension that shows the filename of the image used for img2img

[mystic lava]

Hey can someone help me trouble shoot comfui

@ornate elk i hate to be taht guy

but idk why its not working 😭

[loud pond]

Could someone tell me if its possible to use Pony lora on stable diffusion 1.5?

[elder mauve]

@ornate elk installed fresh windows

Should i install amd’s app?

[hazy wigeon]

trying to install some custom nodes in comfyui, but i get these errors:

[calm cairn]

ok so i downloaded both of these models
I can tell the problem was in fact the model I had. Neither of these have the problems that I had with the original pony download. And you are right, the Perfect Pony Xl model is better than the confetti one

[ornate elk]

You didn't followed my Comfyui guide

[ornate elk]

AMD adrenalin

[elder mauve]

Can u please send the link again to guide

Since i reinstalled windows

[elder mauve]

Done

[ornate elk]

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

[elder mauve]

Trying

[ornate elk]

you need rocm hip sdk 6.1

and no addtional rocm library files

then it should work

[ornate elk]

not possible

[loud pond]

so what do i have to do to use the pony lora?

[mystic lava]

I got it wokring

but it didnt work when i ran it

Error occurred when executing CheckpointLoaderSimple:

ERROR: Could not detect model type of: C:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\models\checkpoints\flux1-dev.safetensors

File "C:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\execution.py", line 317, in execute
output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\execution.py", line 192, in get_output_data
return_values = _map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\execution.py", line 169, in _map_node_over_list
process_inputs(input_dict, i)
File "C:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\execution.py", line 158, in process_inputs
results.append(getattr(obj, func)(**inputs))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\nodes.py", line 539, in load_checkpoint
out = comfy.sd.load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, embedding_directory=folder_paths.get_folder_paths("embeddings"))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 527, in load_checkpoint_guess_config
raise RuntimeError("ERROR: Could not detect model type of: {}".format(ckpt_path))

[elder mauve]

6.1.2?

[mystic lava]

yea

i used main docs

wait i missedc one thing

lemme try that

then i shall follow ur guide

to the dot

[ornate elk]

yes

[loud pond]

do i just put it in my lora area and then switch to pony diffusion v6?

[ornate elk]

yea or better to any other pony model as the base pony model isnt that good

[mystic lava]

nah didnt work

time to read urs again

😭

[loud pond]

alright

[ornate elk]

you already have set up auto1111 with zluda, so installing comfyui with zluda should go faster as you can skip most of the zluda steps

[mystic lava]

this is what i used

😭

[ornate elk]

thats the one for nvidia

[mystic lava]

which one am i after

[ornate elk]

Comfyui with ZLUDA

you gpu was the 7900xt right?

[mystic lava]

yea

@ornate elk

found guide

im gonna be using flux

what steps do i follow

[ornate elk]

the comfyui zluda install guide

[mystic lava]

git clone https://github.com/LeagueRaINi/ComfyUI.git && cd ComfyUI && git checkout nodes-cudnn-patch

doing this

[elder mauve]

got this

here

[ornate elk]

can you show the full log?

[elder mauve]

this is the full log

[ornate elk]

ah alright

then proceed with the next steps

[elder mauve]

alr thanks

let me try

[ornate elk]

also i recommend installing Auto1111 first instead of comfyui, but both should work

comfyui is just a bit more work

[elder mauve]

this is fooocus

[ornate elk]

ohhh

xD

[elder mauve]

was going to buy a new GPU today

so trying one last time

[ornate elk]

it should definitly work as you resetet windows

[elder mauve]

i hope so

[mystic lava]

starting now

shjall update u in a sec

Error occurred when executing CheckpointLoaderSimple:

ERROR: Could not detect model type of: C:\Comfy\ComfyUI\models\checkpoints\flux1-dev.safetensors

File "C:\Comfy\ComfyUI\execution.py", line 152, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
File "C:\Comfy\ComfyUI\execution.py", line 82, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
File "C:\Comfy\ComfyUI\execution.py", line 75, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
File "C:\Comfy\ComfyUI\nodes.py", line 539, in load_checkpoint
out = comfy.sd.load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, embedding_directory=folder_paths.get_folder_paths("embeddings"))
File "C:\Comfy\ComfyUI\comfy\sd.py", line 513, in load_checkpoint_guess_config
raise RuntimeError("ERROR: Could not detect model type of: {}".format(ckpt_path))

@ornate elk

havent done anything to zluda

[ornate elk]

what model did you downloaeded? you need the flux fp8 dev model which is 16gb

[mystic lava]

link?

I used the fp32 model

cause i got a beast of a gpu

or i thoguht i did 💀

[elder mauve]

when running run.bat

[ornate elk]

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

here

[ornate elk]

you can delete that

[mystic lava]

aight

is it different/

[ornate elk]

nope

[mystic lava]

fire

[ornate elk]

i would need to try forge for myself again, it could be an update that broke it.
can you please try Automatic1111 with ZLuda, ist the easiest setup i can guarantee to work

[elder mauve]

let me try

[ornate elk]

also why have you installed python 3.10.9 when all my guides say to install 3.10.11

[mystic lava]

can this stuff do deepfake images?

[ornate elk]

yes

[mystic lava]

i can make my friends into shrek

peak

[ornate elk]

xD

[elder mauve]

@ornate elk fooocus worked

i had to do
pip install -r req .txt

i will try to generate an image now

[ornate elk]

alright, keep attention to step 6 from the guide, before clicking generate

[elder mauve]

what should i expect in terms of speed?

[elder mauve]

alr done that

[ornate elk]

fooocus uses sdxl models only

[elder mauve]

yeah

what speed do u get on sdxl models?

[ornate elk]

so i guess on a 1024x1024 res you could get 2-3 it/s maybe

[elder mauve]

isnt that too low?

[ornate elk]

would need to do a quick speed check with mine on auto1111

[elder mauve]

oh u said it will take long on first try right

cuz thats how zluda works

[ornate elk]

yea it needs to process zluda/cuda stuff the first time

[elder mauve]

hhope this works

wont have to buy a new gpu

[ornate elk]

you would be the first person i couldnt get it to work then

[elder mauve]

but earlier i got error in last step when i was using windows , old one

so this is good

[mystic lava]

cs1o

u know any good tuts for deepfakes

wait

i got an error

Prompt outputs failed validation
CheckpointLoaderSimple:

• Value not in list: ckpt_name: 'flux1-dev.safetensors' not in ['flux1-dev-fp8.safetensors']

[ornate elk]

3-4 it/s

mostly 3 rn

[elder mauve]

i see

thanks for checking

[ornate elk]

16 it/s for 1.5 models

so you should get 2-3 it/s

[elder mauve]

makes sense

[ornate elk]

maybe 15-20 seconds per image

[elder mauve]

as long as its iterations per second

[ornate elk]

you have to select the model
in the node

[elder mauve]

error

[mystic lava]

im actually gonna die to natural selectionm

i swtg

[ornate elk]

have you done step 5 after you installed the req.txt ?

if not you need to do this

[elder mauve]

yeah

[mystic lava]

firrst iamge take longer im assuming

[ornate elk]

do it again to be sure. and if it still dont work, try auto1111
havents tested forge since 3 weeks

and there was an update

[ornate elk]

yes

[mystic lava]

eta?

also i accidentally hit generate 3 times

that wont messs it up right

[ornate elk]

nope

eta 15-40 mins

you already generated with auto1111 so it should be faster

[mystic lava]

peak

also

can i see some of ur art

u seem very experienced wit it

r u purely technical or do u often use it

[ornate elk]

i used it a lot in the past, but the last months i was just setting up guides for AMD users

i can show some stuff maybe later

[mystic lava]

@ornate elk

[elder mauve]

trying auto1111

[ornate elk]

have you moved the 2 zluda files int othe torch lib folder ?

[mystic lava]

i have not

how would i go abt that

[ornate elk]

Step 5 of the zluda steps from the comfyui guide

Go into the C:\ZLUDA\ Folder.
There copy the cublas.dll and the cusparse.dll and repaste them inside the folder again.
Rename the copys to cublas64_11.dll and cusparse64_11.dll
Copy these two files to ComfyUI\venv\Lib\site-packages\torch\lib and overwrite if asked.

[elder mauve]

@ornate elk

automatic 1111

[mystic lava]

@ornate elk

This goo

[ornate elk]

oh right, runway ml messed up, they removed the base 1.5 model that normaly gets downloaded

you need to download a model and put it into the models/stable-diffusion folder

[elder mauve]

:/ what to do?

i did

[mystic lava]

running now

praying for no errors

[ornate elk]

thats wrong

[mystic lava]

and it crashed 🗣️

ah

[ornate elk]

they need to be renamed

[elder mauve]

stuck here

[mystic lava]

aight

going aqgain

🗣️

[ornate elk]

let it load a bit

[mystic lava]

running

[loud pond]

quick question why do all of my renders of stable diffusion look horrible? I enter a prompt and it gives me a weird distorted mess

[ornate elk]

can you show an example, with settings

[loud pond]

yeah sure lemme re render one

[mystic lava]

hasnt crashed yet

its first one so im expecting a wait lol

[elder mauve]

@ornate elk

[ornate elk]

you got an image ?

[elder mauve]

i guess

[ornate elk]

nice

[elder mauve]

omg

this works

[ornate elk]

your done 😄

GG

[elder mauve]

second image

3 seconds

wtf

[mystic lava]

yea

[ornate elk]

if you used an sdxl model you need to gen 1024x1024

[elder mauve]

ah

it was 500

[ornate elk]

you need a better model. the base 1.5 one is from 2022
get them on civitai.com

[mystic lava]

cs

how do u use

[elder mauve]

brooo this is crazy fast

no way

[mystic lava]

those promopts that are like

[elder mauve]

omg

BRO

[mystic lava]

() <>

ones

[elder mauve]

[loud pond]

oh could you show me where i can find any better models?

[mystic lava]

wit variables

is there docs

[elder mauve]

thank u so much @ornate elk

finally !

itis sooo fast

[ornate elk]

no problem 🙂 i knew we could do it

[elder mauve]

its crazy fast!

[ornate elk]

yep!

[elder mauve]

omg

should i reinstall windows properly now

and copy over my data

[ornate elk]

you can get models from Civitai.com

[elder mauve]

🤔

[loud pond]

alright

[ornate elk]

what do you mean with properly

xD

if you reinstall windows you would need to set it up again

Im afk

[mystic lava]

i HAD to close midway trthu

[echo hill]

hm? Did I see a ping?

[mystic lava]

YEA

Didnt mean to

mb

[echo hill]

np!

[mystic lava]

ngl

flux takes WAYY longer to load

like 5 minutes per image

is this normal

@echo hill

@ornate elk

Mb sunny

I have a rx 7900 xt

gpu

its the first gen

r 5700

[elder mauve]

yeahh it was on a slow nvme ssd

gonna install it on my main ssd

[mystic lava]

stuck on tyhhis

120 seconds

for this

its 1 pass

step sorry

10%|████████▎ | 2/20 [01:56<20:36, 68.70s/it]

am i cookewd

💀

[ornate elk]

which workflow did you used?

[ornate elk]

alright 👍

[mystic lava]

its still runnin

iits running at 124,46s/it

[ornate elk]

is it the 16gb model?

[mystic lava]

yes

[ornate elk]

also make sure you dont run anything in the background

[mystic lava]

im not

😭

[ornate elk]

what was your gpu again?

[mystic lava]

rx 7900 xt

[ornate elk]

and 32gb of ram?

[mystic lava]

yes ddr5

[ornate elk]

you need to try this workflow for flux:

drop that file into the comfyui browser window

[mystic lava]

aight I shall

gimme a minute

this look right?

nah still incredinbly slow

@ornate elk

and the image it generated looks terrible

this was for an astronought holding a moon

wtf is this

[mystic lava]

not this

i ahvent restarted myt pc

sincce starting htis process