🤝｜tech-support

[ornate elk]

thx, thats why i wrote those guides

because there are so many outdated and wrong guides for AMD out there

[primal pendant]

you went through alot of hell to then

[ornate elk]

got every webui working in the end ^^

and also Lora training

[primal pendant]

oh sounds nice power hungry gpu though no ?

i assumed undervolted

[ornate elk]

yea pretty power hungry

but great card

its a beast in gaming

[primal pendant]

btw what your best ui you prefer as you tried them all

[ornate elk]

Automatic1111

best ui, and biggest extension support

[primal pendant]

ok cool so i choose right then

this thing gonnna take over an hour i think it's barely using anything

[ornate elk]

does it still shows compiling?

[primal pendant]

yeah 0%

[ornate elk]

ok wait 7 mins and then close and relaunch

[primal pendant]

think it might of crashed

[ornate elk]

crashed?

if the cmd is open its fine

[primal pendant]

oh it did something

[ornate elk]

works

you got an image?

[primal pendant]

nope #

[ornate elk]

okay then close and relaunch

the webui-user.bat

[primal pendant]

now i do

[ornate elk]

nice

[primal pendant]

so it wi;ll be good now ?

[ornate elk]

your ready now

[primal pendant]

no settings to alter ?

[ornate elk]

nothing speciifc for AMD

but there is stuff you can adjust

[primal pendant]

ok cool thanbks very much now need to study ai lol

yeah i need to find cheat sheet for 6700 xt

[ornate elk]

no problem 🙂 have fun

[azure magnet]

where i check how many patreon version i have?

[ornate elk]

open up a cmd and type
where python

[azure magnet]

i see patreon this C:\Users\x>where python
C:\Users\x\AppData\Local\Programs\Python\Python310\python.exe
C:\Users\x\AppData\Local\Microsoft\WindowsApps\python.exe
C:\msys64\ucrt64\bin\python.exe

so have i python 3.10?

[ornate elk]

yes and an unkown python version from the Micrsoft store

[azure magnet]

how could i check what version is?

[ornate elk]

open up the ms store and search for python. then you see all and some wil show "installed"

[azure magnet]

coulb they be a link because they are only 0 KB both

and no one python version is installed

[ornate elk]

where are they?

[azure magnet]

https://tenor.com/view/idk-i-don't-know-unsure-spongebob-friendiversary-gif-9486917737230537182

[azure magnet]

But why when i run comfy it say me python version 3.11.8?

[ornate elk]

maybe you installed python 3.11 with the ms store

[azure magnet]

No just checked

Any idea

[ornate elk]

Check under installed programms for the pythons

[azure magnet]

find only 3.10.11

[ornate elk]

Hmmm

[ornate elk]

I guess comfy comes with python 3.11 already

[shy charm]

for some reason DWM is using more than half of my VRAM so i'm unable to change models

starting DWM seemed to fix it

[oak jay]

Can I please get some recommendations for settings for Dynamic Thresholding with Flux models on SD Forge? Everything I create is either wired, blurry, washed out, burned or nothing like what I want. TIA X

[proper pike]

does anyone know why I get this error?

[fair oxide]

How do I fix the saturation being really high in image generation?

My cfg scale is at 1.5 and it helped a bit but it still looks wrong

I am also putting: "low saturation" and "dull colors" in the prompt but it totally ignores it

[steady thistle]

RuntimeError: D:\a_work\1\s\onnxruntime\python\onnxruntime_pybind_state.cc:636 onnxruntime::python::CreateExecutionProviderInstance CUDA_PATH is set but CUDA wasn't able to be loaded. Please install the correct version of CUDA and cuDNN as mentioned in the GPU requirements page (https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements), make sure they're in the PATH, and that your GPU is supported.

I hadn't had this error before and I didn't touch CUDA, though I installed several new nodes. I haven't changed the CUDA_PATH certainly.

(venv) D:\Ai\ComfyUI_windows_portable\ComfyUI>nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2024 NVIDIA Corporation
Built on Wed_Aug_14_10:26:51_Pacific_Daylight_Time_2024
Cuda compilation tools, release 12.6, V12.6.68
Build cuda_12.6.r12.6/compiler.34714021_0

This is the version I'm running.

Also, that was in PATH.

[median geode]

hey there, anybody know how to handle this?\

[ornate elk]

You need to use a custom vae

[ornate elk]

Do other models load?

Also make sure the webui is whitelisted in any browser adblocker

[median geode]

none of the models can load

just like this

[glossy pumice]

Hey I need help

I need help cloning the git hub thing it cuts off at 43%

Err 8 of something

[snow breach]

err 8 and your description kind of sounds like you are on really bad internet or wifi or something.

[glossy pumice]

Is that the reason

I tried it several times and it says the same getting stuck on 43

What's the best way to install it if I'm doing anything wrong in the first place

[snow breach]

git clone repoaddress

make sure you have enough space on your drive and your internet is good. if there are already files there, delete the folder of the repo name and run git clone again.

[glossy pumice]

How do I check that

[median geode]

you can follow the pinned guides

[glossy pumice]

Where do I check for those

[glossy pumice]

I think I did do a video but bumping into issues

[median geode]

you mean you have downloaded it successfully and made a video?

[glossy pumice]

No

I just downloaded git and python 3.10 following a video

Next step is cloning git thing but it's getting stuck

[median geode]

oh just follow the pinned message

once i followed a video and i failed too

[glossy pumice]

Ok I'll check those too

Ohh

I'll come back trying everything

[median geode]

or can you screenshot your cmd?

[ornate elk]

What does the cmd shows?

[ornate elk]

Have you checked the adblocker settings?

Seems like the browser interface can't connect to the cmd

[median geode]

idk how 🥲 where is it? i just try on edge and google and none of them works

[loud pond]

hello! i want to generate bigger images but i dont have enough vram, how can i make the image less intensive so i can use less vram?

[median geode]

sincerely. could you please give me a screenshot

[median geode]

how big

[loud pond]

704*1200

[median geode]

You can first scale it down to create a small image.

[loud pond]

is there anything else i can do apart from decreasing the size?

[median geode]

Then drag the out put image to the img2img, upscale it

[median geode]

idk, maybe it will work

[ornate elk]

What's your GPU again?

[loud pond]

amd radeon 480

[loud pond]

ok ill try it thanks

[ornate elk]

Get the tiled VAE extension. And enable only tiled VAE.
Then also generate at lower res then upscale with hires fix

Or in img2img with the SD upscale script

[loud pond]

ok ill try that too thanks!

[median geode]

you mean like this? doesn't help

[wraith valley]

try without the webui arguments

sometimes when i try out new arguments and i write them wrong it gives me error too

xformers medram nohalf

[median geode]

u mean delete the arguements?

[wraith valley]

yes

[median geode]

i try try

[ornate elk]

That won't help

Dont remove them xD

[ornate elk]

Try delete the config.json the ui-config.json and the venv folder and relaunch

[wraith valley]

my bad sorry just tried to help haha

[median geode]

never mind xD but thank you anyways.

[median geode]

i finally figure out what's going on. it's my vpn. i closed it and everything just went well

[ornate elk]

Ah nice

[steady thistle]

Does running Comfy on SSD work much better than running off a HDD?

I swapped to a HDD when I had an error and had to redo venv anyway, and it's terribly slow and stuck on loading a Flux model.

[ornate elk]

Yes

But the model should also be on an SSD preferable

Also the windows pagefile needs to be on an ssd

[steady thistle]

Thanks, yeah. I noticed a vast difference but wanted to make sure it wasn't a separate issue.

My generations are reasonably slow, but I attribute that to my using much larger models and larger images than I previously had. Good to know, thanks.

[ornate elk]

The generation speed isnt affected by HDD or ssd.
Only when you go over the vram or ram limit

Then it will use the pagefile.
And if the pagefile is on hdd, rip

[steady thistle]

I had a CUDA issue with ReActor, I don't know if I need to delete my venv or if I should remove ReActor and restart it.

[steady thistle]

The model loading was the rough bit.

This series of images took 540 seconds first time through and once everything was in memory, it took 229 seconds.

[ornate elk]

Yea model loading time depends on the drive. SSD vs HDD.
And for flux, best is to leave It on ssd

[steady thistle]

To be fair, that was a 1024x1024 with an LLM prompt, Flux model (no lora), batch size of 4, and then 3 different ReActor run-throughs.

[ornate elk]

Batch size? And not batch count?

[steady thistle]

Umm, I meant I ran 4 images at once.

It says "batch size" tho

This is the error I got on my SSD tho. I'm guessing there's a conflict involving Reactor. I was about to try to reinstall reactor but I wasn't sure if that's the best route

[primal pendant]

hey you know after than long rendering yesterday it's only suppose to do that once right ?

it does it each time i use it

[ornate elk]

@steady thistle Check your vram when using batch size

Because that can overflow your vram

[ornate elk]

It shouldn't do that

[primal pendant]

that's what i thought to

[ornate elk]

@primal pendant Can you show me your txt2img settings and the cmd log when it happens?

[steady thistle]

oooh, okay. So if it is overflowing, I need to do 1 at a time? or is there another method?

[ornate elk]

Batch count for example makes the images one after one

Or set batch size to 1 when using flux

[steady thistle]

Okay, thank you! I'll look for that.

I'm still getting that runtime error tho.: RuntimeError: D:\a_work\1\s\onnxruntime\python\onnxruntime_pybind_state.cc:636 onnxruntime::python::CreateExecutionProviderInstance CUDA_PATH is set but CUDA wasn't able to be loaded. Please install the correct version of CUDA and cuDNN as mentioned in the GPU requirements page (https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements), make sure they're in the PATH, and that your GPU is supported.

[ornate elk]

You can try to install onnxruntime-gpu into comfyui

[steady thistle]

The weird part is that when I ran it on the HDD it worked fine (but slow of course) after running a fix on Reactor.

But I'll try that.

[primal pendant]

weird it failed after all that gonna reboot pc

now it won't even start

[steady thistle]

Do I go into venv\Scripts\activate then do install onnxruntime-gpu ?

[primal pendant]

delete venv?

[steady thistle]

I tried pip install onnxruntime-gpu but it was already installed.

[ornate elk]

Are you on windows 10 or windows 7 ?

Because it says installed on 32bit system 😮

[primal pendant]

10 64

[ornate elk]

Check in the system info

But also install Python 3.10.11 64bit and then relaunch the webui-user.bat

[steady thistle]

I guess the next step is to rebuild my venv.

[primal pendant]

it was new install only few days ago

only version of python on here

but you say i need 3.10.11?

i thought it don't work with any other version ?

so do i leave 3.10.6 on or remove first ?

[ornate elk]

You can overwrite it

[primal pendant]

ok if your sure

[ornate elk]

We can't see if its 32 or 64 bit here

[ornate elk]

OK good

[primal pendant]

32gb ram

so only upgrade no custom install ?

jsut wanna make sure i don't mess anything up

for instance if i go custom

that py launcher not ticked

[ornate elk]

Upgrade

[primal pendant]

ok so launcher not needed?

lol

[ornate elk]

Okay uninstall everything from python

Then install 3.10.11

[primal pendant]

nice

i can't remove it and can't upgrade it what the hell

[ornate elk]

Uninstall via control panel

[primal pendant]

it won't remove

how is it so god dam broken

need to find way to force it off

getting somethign called revo uninstaller after quick google

how does somethign egt that broken that it won't even remove without another program lol in only a few days

do i need to do this you think ?#

will it break anything you think ?

i didn't even know what python was till few days ago

[steady thistle]

Thanks again for your help. I dunno if there was a conflict or what, I'll probably find out soon enough, but it's running again.

[primal pendant]

i give admin right ?

[ornate elk]

Yep looks okay

[primal pendant]

so no need to admin ?

or is that prefered?

like i want to do things the way that will bring less issues

[ornate elk]

Not needed but doesn't Hurt

Leave everything at default is the way to go

[primal pendant]

ok

i can't even install it now

i guess admin then

[ornate elk]

Also open up a CMD as admin and type
SFC /scannow
That will fix windows issues

But very strange that you have these issues

[primal pendant]

yeah i'm cursed

all good

oh

ffs i swear

how is 3.10.6 newer ?

only other python had on here

this is turning into such a ballache

any ideas how i can install ?

stupid python

[ornate elk]

Check the system control

For python launcher

And uninstall it

Or uninstall via revo

[primal pendant]

nothing

[ornate elk]

And in the Windows control panel?

[primal pendant]

nope

it's not here and yet tells me i have newer version installed already

which it clearly is not

amazing software

👏

so now i no lonegr have python on pc and yet tells me i do

i ain't doing clean windows install just for this

well thanks for helping i guess it's just not possible even though i did one picture and had running yesterday

just not possible to install python

right after deleting god dam everything in registery it's installed

so what now ?

remove all start over ? from yesterday

@ornate elk

this crap so stressful

less than a week ago python wasn't even known to me now it's known to me as worse crappy software with headaches

[glossy pumice]

Sorry I was very tired I needed to rest

I'm trying again

It says RPC failed curl 56 openssl sslread ssl_error_syscall, errno 0

If anyone is here can you please help me clone git to my computer

@ornate elk can you come here I have some questions please

[ornate elk]

Its really strange that is says installed on win32

[ornate elk]

Delete the venv folder and relaunch the webui-user.bat

No need for a reinstall

[ornate elk]

hi

[glossy pumice]

Hi

Help me please I can't clone git stable diffusion thing

[ornate elk]

What error do you get?

Make sure you have no VPN running

[glossy pumice]

Early eof
Invalid index pack output

[glossy pumice]

I have nothing

Just plane wifi

I deleted python 12 and got 10.6 even

Git downloaded with all default settings

[ornate elk]

Any third party antivirus?

[glossy pumice]

Cloning dies over a few while

[ornate elk]

Also try over ethernet

[glossy pumice]

Nope window antivirus only

I don't have ethernet

[ornate elk]

Oh

[glossy pumice]

[ornate elk]

Can you share the cmd log with the error?

[glossy pumice]

Hold on

[primal pendant]

oh your to late i started over lol

jst did driver did the rename library to old dropped in the new and dll and rebooted part

[ornate elk]

Ah okay

[primal pendant]

yeah so afr so good

[glossy sorrel]

Question, is sharing my comfyui workflow to other people safe? i.e., does it expose me to any sort of security risks or anything at all?

[glossy pumice]

@ornate elk

[primal pendant]

going really well

[glossy pumice]

How are downloading it

@ornate elk

Sorry if It's annoying but I kinda need it today

[ornate elk]

Check that your PC time and date is correct.
If your from Asia you may need to use a VPN

Or try change your DNS server

If your internet is really slow. You would need Ethernet

[ornate elk]

Nope

Its safe

[glossy pumice]

How do you install a VPN on a laptop

How to change my DNS

[glossy sorrel]

thank you CS1o! i appreciate the reassurance 🙂

[primal pendant]

what is all this hugging face thing ?

i downloaded model from https://civitai.com/

the dream one you said before

that's only one in the folder

[ornate elk]

Thats a bug that will be fixed in the next update

But can be ignores

[primal pendant]

now do i dare close it

why does my gpu never go up ? stays very low

surely it should go like 50% or more no ?

not 1 or 2%

[warped canyon]

hey, sd forge only sounds notification when im on the tab. If im on another tab/other stuff on pc, it doesnt sound off the alert

[cosmic furnace]

so theres almost no cons of forge comparing to auto?)

[fair oxide]

How to I move stable diffusion out of C?

I need to reinstall windows but I don't want to nuke stable diffusion with it

[ornate elk]

Check the vram usage.

The processing usage is always pretty low

[ornate elk]

You can move the whole folder.

[ornate elk]

Extension support is the biggest con

But other than that nope

[warped canyon]

well, i just mentioned one right before ur message :/

[fair oxide]

Ok thx

[primal pendant]

i knew this would happen how do i stop this doing like the very install run all the time ?

compiling like this as from what i understood it's suppose to only be this long once?

not every time u run it

seems really slow

and now it's doing this

won't even let me create image yet

something is wrong surely you don't have to leave open constantly otherwise compiling for ages on each run

is there some kind of setting i have to do in the ui ?

as that finally popped up but still compiling

😭

[ornate elk]

Let it run it can take very long

But it should be only on time

[primal pendant]

every time i turn on pc ?

[ornate elk]

No

Maybe it wasn't finished before

[primal pendant]

well i let it do long one earlier today after that python nightmare

even created pictures

[wraith valley]

how can i benchmark on a1111

[primal pendant]

but now after closing pc and relaunching it's acting like it's very first install run again

[ornate elk]

If its done once. Try a full restart. And test again

[primal pendant]

it compiles

[wraith valley]

i saw this one in yt video

[primal pendant]

every run

[ornate elk]

Thats an extension

[ornate elk]

Wouldn't make sense

[wraith valley]

yeah its under system info what i have installed

but there is no benchmark

[primal pendant]

so how does one stop it compling every launch for an hour before can use?

[ornate elk]

Dont have the time right now for troubleshooting

[primal pendant]

i googled it can't find any information about it either

it's like you say only suppose to happen once then jsut work

[ornate elk]

Yep thats like it should be

[primal pendant]

but it's not lettting me lol

[ornate elk]

Because in the end you have a zluda folder in appdata with around 2gb compiled files

[primal pendant]

each check point i add messes with it ?

[wraith valley]

ah lol i found the benchmark tool

[ornate elk]

Compiling also can restart if you Update GPU drivers or the torch version

[ornate elk]

Did you tried an sdxl model?

[primal pendant]

all i did was add few checkpoints of things i wanted to try while it was still working i thought great we all done now did the kitty pictures ect alll was fine so shut down my pc for a bit startedback up tried to start stable diffusion now it;s doing long one again

no idea what that is

[ornate elk]

Hmm okay

I'm off now. Check if it happens later or tomorrow again. Then we can do some tests

[primal pendant]

ok

i might try another version of ui

[wraith valley]

damn i think thats not that much

[livid lark]

Ok so its experimental and no one knows for sure anything yet but what are your recommendations for lora training in general? is say 5 steps and 5 epochs and 25 images or so good enough? when is it over trained and when is it undertrained?

also can i merge LORAS? like how we merge SD15 and sdxl models...

[tepid crypt]

is it possible to use Pony loras on a flux checkpoint?

[obsidian carbon]

ComfyUI is slower than usual now :// 1080 ti here and its been like 15 mins and SDXL is like... bruh

heres an example

Requested to load ControlNet
Requested to load SDXL
Loading 2 new models
loaded completely 0.0 2386.120147705078 True
loaded partially 2953.1062858581545 2953.1054763793945 0
40%|██▊ | 12/30 [14:15<22:07, 73.73s/it]

[warped canyon]

hey, sd forge only sounds notification when im on the tab. If im on another tab/other stuff on pc, it doesnt sound off the alert

[raw parrot]

Hey everybody ! I am currently installing but my thing gets stuck here always... If you have any idea on what I should do, please ping me ! Thank you.

[vague vigil]

I'm using automatic1111 webui v1.7.0, because I mainly use it on my phone and it's the last version to have a stacked UI on the generation tab. Newer versions split the screen vertically, with the settings on the left and the preview image on the right. The phone is too small to have a split view like this. Is it possible to have the old vertically stacked view on the latest version? Thanks.

[raw parrot]

well if anyone has any ideas about my.. predicament, i'd appreciate it.

[raw parrot]

Thanks !

[ornate elk]

If I use it on the latest version on my Phone in default Vertical mode. Its all stacked vertically

Didn't tested horizontally

[ornate elk]

How much RAM do you have?
And what's your GPU and what's inside your webui-user.bat?

[ornate elk]

Do you have --xformers in your webui-user.bat?

[snow breach]

Just thought I'd ask while I've got the time to do. I don't have any examples but I can get a few later. I'm using Forge UI; If I produce an image of anything like a cat or a person, everything is fine without upscaling it. As soon as I upscale it starts to stretch them right out height wise. 1024x1024 and upscale by 2 was selected. If I reduce the upscale, I reduce the issue. So if I set it to upscale by 1.1, the issue is slightly noticable, while at 2 the person or cat becomes a very stretchy alien person or cat! Any ideas what has taken place? This is affecting SD15 and SDXL both across multiple different model checkpoints & VAEs.

[raw parrot]

16.0 GB (15.8 GB usable)

so uh

[ornate elk]

You may need to increase your Windows Pagefile then

[raw parrot]

ah

[ornate elk]

If you upscale with hiresfix, make sure the denois is 0.5 or below and hires steps on 10

[ornate elk]

The first installation takes a lot of resources so you either have to wait very long or increase the pagefile.
Then it should work faster and your also ready to use sdxl models (depending on the GPU)

[raw parrot]

nvm then

i have one other question however

if you dont mind

?

is civitai as good or at least up there with diffusion'

[ornate elk]

You mean their image generation?

[raw parrot]

precisely

[ornate elk]

Yea its okay

But you have more flexibility with a local install

[raw parrot]

flexibility ?

wdym

?]

[ornate elk]

More freedom. You can use more extensions, custom upscaler etc

Have more control over the stuff you want to do

[fair oxide]

Can someone come help me?

my discord bot i created, image generator based on cyber realism, (which is hella realistic), produces such such low quality images, no matter how i prompt.

Also, im trying to get faceswapping a thing, using reactor, i have it coded, it just doesnt work, at all

but yea, my image quality is about 50% maybe 40% that of what i make on my PC.. same sampler, same steps, same negative prompt, just so bad

Can someone come help me?

Also, im trying to get faceswapping a thing, using reactor, i have it coded, it just doesnt work, at all in my discord bot image generator writtin in PY

Using ReActor api:

I tried like this
1: generate image, save image
2: send image
3: ask user for a selection of face models to choose from
4: swap face out
5: send new photo
6: delete all images
but it just errors

[glossy pumice]

@ornate elk

@vocal burrow can you come here for bit

[vocal burrow]

someone spamming ? Did I miss it ?

[glossy pumice]

No I need help I installed the git hub repitory what's next

If you know what to do

[vocal burrow]

which git repo ?

[glossy pumice]

Automatic1111

[glossy pumice]

Sorry I'm trying to find what model to get

1.5 right

What's the best place to download models

[vocal burrow]

it should get 1.5 automatically for you.

stable diffusion models go in models/stable-diffusion

[glossy pumice]

Ohhh

[vocal burrow]

1.5 is ancient at this point, would probably recommend browsing civitai and get something more recent ( make sure at first that you re downloading a sd model and not a lora )

[glossy pumice]

It says put stable diffusion checkpoints here

[vocal burrow]

dont know what you re talking about. I don t know which page/readme/tutorial you re reading.

[glossy pumice]

Give me a minute

I checked git status and it says no more commits needed

[fair oxide]

this is where you place your models

[glossy pumice]

Yes and I have non right

I should download 1.5

Right

[ornate elk]

Download any model from Civitai.com

For example Dreamshaper v8

[glossy pumice]

Oh

I'm downloading 1.5 rn should I stop

@ornate elk

[vocal burrow]

Depends of your connection. It s not necessarily a bad idea to take a look at how 1.5 handle things. But again it s outdated by current models.

it s weird that it didn t download 1.5 for you tho.

[glossy pumice]

It is fine Poseidon has cursed me by birth

I think I want to try a model and make my own if I can somehow

Lot of anime in civitai

[vocal burrow]

you can filter stuff out

[glossy pumice]

I have a question

Stable diffusion is the best image generator right

[ornate elk]

Runway deleted the 1.5 model of huggingface

Next webui update will fix it by using a reupload

[glossy pumice]

So I'm too early

[ornate elk]

No the 1.5 base model is outdated anyway ^^

Better get Dreamshaper v8 to start

[glossy pumice]

I see

Let me see if I can find it

Also what's the difference between Lora and checkpoint

[vocal burrow]

aah yeah true.

[vocal burrow]

checkpoint == model == complete stable diffusion model that can work on its own
lora == patch for a specific model architecture (sd 1.5, sd 2, sdxl, etc)

[glossy pumice]

Ok I think I want a anime or a cartoon one before I get a more realistic model

[glossy pumice]

So I should only get checkpoints

I'm seeing a dreamshaper but no v8

Oh shit it says v∞

Guys I need help navigating civitai

I'm downloading a model

I'll come back then

Checkpoint trained

[gaunt crystal]

Every time I open pinokio to launch comfyui it starts installing something? I don’t understand what it’s doing, I was just generating images and then I closed it and now it has to reinstall every time I open it …?

[ornate elk]

Strange but wouldn't use pinkio for it.
Better install comfyui the normal way, (as its only downloading a zip file + clicking a bat)

[gaunt crystal]

I guess so, do I have to uninstall everything then?

[ornate elk]

Yep

[gaunt crystal]

Anyone know how to limit vram usage with flux? Cannot figure it out

[ornate elk]

In Comfyui or Forge?

Have you reinstalled comfyui now?

[hardy void]

Hi, I want to make my LoRa for a specific hairstyle only.

But how can I make sure that I only train it to understand hairstyle, and doesn't take anything else.

It is so funny, I cut only face picture with several people on the same hairstyle. and then ask Lora to train it. It come out so well at first. I thought It work.

I try to say this hair name and then use it with realisticVision. and it turn out everyone is having that hair.

But then it seem every photo are zoom in close-up to the hair.... so it turn out It is a failure because it seem to limit photo style to be the same

[hallow latch]

I was running my models and doing some differents prompts etc and from nowhere I get error connection and inside the CMD it says "Press any key to continue . . . " but I did not even touch it. Why does it happen ?

[vocal burrow]

tldr : just use pictures of some characters with your specific hairstyle and label them properly. Don't try to cut the pictures so only the hairstyle is shown. Otherwise the model won t have any idea of the context behind your hairstyle. Training with very croped picture of hairstyle it will be hard to distinguish hairstyle from just a random crop field or whatever. Context is important. If there is "a picture of a woman with azerty hairstyle" and "picture of a man with azerty hairstyle", the ai is """"""intelligent"""""" enough to isolate the differences between man and woman, and the common points of azerty hairstyle. Althewhile learning that "azerty is an hairstyle, hairstyle describe the fluffy stuff above human head, etc"

for more details, ask in #🔧｜finetune

[vocal burrow]

hard to tell, would need more details about what you were trying to do, what settings, etc.... But please... keep it tasteful.

[hardy void]

about this kind of context. The problem is that I use tags style. Should I try it with sentense style instead?

[vague vigil]

@ornate elk Thanks for the reply. I figured out the difference. I use accordian mode to hide the settings so I don't change them accidentally when I'm scrolling. If I disable accordian mode, then the UI is stacked. @last heath It would be great to have an option for a stacked UI with accordian mode. Thanks.

[primal pendant]

hey back again so about waht we last discussed last night any ideas how i can stop it doing that each run it does it for ages not like 5-10 mins it's over an hour each run

it takes about 5 mins for the web ui to launch

but still can;t create due to compiling

[primal pendant]

@ornate elk

web ui now just popped up

still can't create anything

[signal sigil]

Hey guys. This is possibly a very obvious qustion, but i just want to make sure. Is it possible to change the meaning of a tag that is already in the dataset of a model you're training a LoRa on? Or should it not be attempted?

[neon orchid]

Hey guys, sorry for the sudden question but...
I've been having trouble installing my AMD stable difussion for days now.
And I haven't found a guide or a video that explains why I get so many errors, even when I follow the guides/videos to the letter.
Is there a guide or something that can guide me? 😦

[ornate elk]

Hey, there is my Guide for AMD linked in the pinned messages

With that you shouldn't have any problems. But let me know if you have any questions

[ornate elk]

Its so strange. No one had this issue before

[primal pendant]

so no idea then i think i will try something else

trying focus amd

[ornate elk]

A windows clean install would be my last idea

[primal pendant]

actually maybe not that one

[ornate elk]

You could try forge or fooocus

[primal pendant]

yeah forge

focus one put me off a bit trying to find and add to text document i don't have numbers so finding the line seemed hell

[ornate elk]

Ah okay

Yea forge would be better

[neon orchid]

oh, okay ty!

[primal pendant]

i can run from any drive right don't have to be c?

[ornate elk]

Yes

[primal pendant]

ok good jsut checking as i will leave that other one on there for now this is more of a test

i really don't think my windows install broken

like it was only few months back i did new install

because i upgraded to m.2

from sata 3 ssd

[ornate elk]

Hmmm okay, but your python install issues where strange too

[primal pendant]

well yeah that was odd python is all new to me

so it only had the ones from the guide installed

this seemed quicker

hmmm

[neon orchid]

is this for AMD?

Sorry, my English isn't the best hehe

[ornate elk]

Check the first link of the top pinned messages

[neon orchid]

okay 🙂

[ornate elk]

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

If your gpu has 8gb vram or more.
Go with the Automatic1111 with ZLUDA Guide

[neon orchid]

my GPU is 8gb

[ornate elk]

What's the model?

[neon orchid]

AMD radeon RX 580 2048SP

The strange thing is that before formatting the PC, stable diffusion did work for me.

😦

[ornate elk]

Okay, that can work with ZLUDA, but is a bit more tricky to get it work.
But Directml shouldn't be a problem.
Note that directml is much slower

[neon orchid]

okay, i'll try

[primal pendant]

so far so good

now do i dare close it

[neon orchid]

I hope to have the same luck as you, my friend 😦

[primal pendant]

luck i gone through hell

😭

[neon orchid]

I tried without exaggeration... for 10 hours since yesterday/today

without success, this is my last chance 😦

[ornate elk]

Should work ^^

[neon orchid]

ummm... here I always have doubts, should I close or should I give that option?

It's just that no guide mentions it.

😦

[primal pendant]

oh no i rebooted pc after few tests and closing the ui now it's doing long thing again

it's doing exactly what the other did

[ornate elk]

You can do it

But isn't required

[neon orchid]

okay 🙂

So I follow the Guide with Zluda, right?

[ornate elk]

Its to late to here for me to check. But tomorrow I can send you the Path where the compiled zluda stuff should be located

[ornate elk]

Yes, but its harder to get it work in the last steps. So you might need my help tomorrow

[neon orchid]

Oh, I see.
But those last few difficult steps are due to the capabilities of my graphics card?

or is it because of the guide?

[ornate elk]

Because of the rx580

[neon orchid]

oh and there is no other minor alternative?

only provisionally or temporarily

[ornate elk]

The directml guide

Thats easier to install but slower

[neon orchid]

Ok, I'm going to try to install DirectML and when you have time tomorrow or another time, give me a hand if it's not too much trouble 😦

[ornate elk]

Yes tomorrow I have more time

[neon orchid]

okay, thanks! ^^

[ornate elk]

The directml version shouldnt be a problem to setup for now

[neon orchid]

This part confuses me a little

So should I add --no-half now or is it already implemented?

[ornate elk]

You need to add it

[neon orchid]

ohhh oka

[primal pendant]

sometimes pip version can cause issues apparently

it won't break anythign if i upgrade it right ?

[ornate elk]

Nope

[ornate elk]

You can first try with --medvram before using --lowvram

[neon orchid]

oops

fuck

i change fast

second try

XD

[neon orchid]

so...

I should try first with

--use-directml --medvram --opt-sub-quad-attention --opt-split-attention --no-half-vae --upcast-sampling --no-half

right?

or --use-directml --medvram --opt-sub-quad-attention --opt-split-attention --no-half-vae --upcast-sampling

[ornate elk]

Try the second and if you get an no half (precision) error add --no-half

[neon orchid]

okay 🙂

Thanks and sorry if I ask very obvious things, I'm just someone who doubts a lot xP

[ornate elk]

No problem at all

[primal pendant]

To create a public link, set share=True in launch().
Startup time: 154.7s (prepare environment: 1.2s, import torch: 8.0s, initialize shared: 141.2s, load scripts: 1.4s, create ui: 1.8s, gradio launch: 1.0s).
Environment vars changed: {'stream': False, 'inference_memory': 1024.0, 'pin_shared_memory': False}
[GPU Setting] You will use 91.66% GPU memory (11248.00 MB) to load weights, and use 8.34% GPU memory (1024.00 MB) to do matrix computation.

[ornate elk]

Looks normal

[primal pendant]

then this

now will compile for an hour or more

[neon orchid]

I got this message 😦

and this

[neon orchid]

I think I already fixed it

but I remember that in my old stable difussion I used --xformers, can it be used here too?

NotImplementedError: Cannot copy out of meta tensor; no data! Please use torch.nn.Module.to_empty() instead of torch.nn.Module.to() when moving module from meta to a different device.

fuck again

😦

[wary tendon]

Can anyone help me with this using Facefusion

[wary tendon]

Uninstalled the old runtime and installed DirectML, it's fine now 👍

[neon orchid]

well

I managed to fix it

however if i have 8vram will it allow me to use --medvram?

[wary tendon]

Is DirectML the best for 40XX cards?

[solemn moon]

https://tenor.com/view/oliver-twist-1948-please-sir-i-want-some-more-please-sir-i-want-some-more-gif-2228167917865608284

looking for that that auto flux canny

[hardy void]

I have issue when I try to train Lora using Flux as checkpoint.
I try both Flux_Civitai and Flux_unchain, and none of them work. it has this early that it says "training completed" quickly after, but no result.

can anyone help? From Civitai, those two version of Flux are largest and by far most popular one already. If I can't use Flux, is there any alternative that work well with caption as "description sentence"? I don't like to use tags, because it is really hard to work with very specific stuffs I want to give it a naming.

(like I want to name hairstyle, and a person)

[ornate elk]

Xformers only works for nvidia currently

[neon orchid]

ohhh okay

In the end it gave me errors, what I fixed didn't last long

😦

[ornate elk]

Try to load a different model

[neon orchid]

I did it, I used another model but the message always appears

[ornate elk]

For Nvidia? Nope

[neon orchid]

even if i don't use it

[ornate elk]

Try add --no-half to the webui-user.bat

[neon orchid]

only

-no-half

?

[ornate elk]

--no-half to the other stuff

[neon orchid]

kk

[primal pendant]

wow u start early

@ornate elkso do you have any ideas why i have constant compiling every run after closing it if it was only few mins i wouldn't mind but it goes on for over 1 hour

you mentioned something about zluda

[ornate elk]

Can look into it later

[primal pendant]

ok cool thanks alot

i did some research i can't find anything

there is no setting in amd driver that can mess with anything right ?

gonna try something called stability matrix found that in my research all in one installer

[ornate elk]

That doesn't support zluda

[primal pendant]

yeah just noticed that

removed it lol

[ornate elk]

Nope but you could try reinstall the AMD driver.
Or you can try removing the Library DLL from the rocm library folder

[primal pendant]

hmmm

which driver version ?

the dll i put in from the other folder ?

i will try 5.7.1

now it's back to this lol

as the 5.7.1 don't ahve dll

now it's really broken

i'm removing everything

[neon orchid]

I have this problem

[hardy void]

Can anyone help this?

[vocal burrow]

You re flying too close to the sun. turn down some settings/enable some optimizations, your gpu does not have vram for whatever you re doing

[ornate elk]

You could try again with 5.7.1 and the default guide

[warped canyon]

i cant seem to change default sampler to dpm 2m karras

it always opens on euler a

anyone knows why?

[primal pendant]

why is this shit so stressful i think i'm callign it quits nothing works i thought ai woiuld be fun but it's nothign but stress

[warped canyon]

wdym

[primal pendant]

it don't matter what i try it just don't work as it should and love to complie after every time i close and open again later if i wait 1 hour i can create images but then if i close and reopen it then it compiles for 1 hour before i can create again

[rugged latch]

yo this is Ai is so easy yo, oml, its so much fun

i be makin images all day longgggggggggg

[rugged latch]

munchu, i sry u having issues wit it, really :<
im not gonna say its, "easy" to get workingg, but it is envloved, u kinda gotta know a lil bit about computers and software to really know what ur doing

there are guides u can follow

i can send a video that ehlped me first time

[primal pendant]

i followed guides many guides none have my problem it's only suppose to be very long first run but mine is everytime i open it

my thing just loves to compile all day

[rugged latch]

when u first launch stable by double clicking the "webui-user.bat", it will need to downlaod dependencies

things like pytorch

u alos need to have Python installed

[neon orchid]

Is there any difference between using --lowvram and --medvram?

[rugged latch]

medram uses half and low ram uses like barley

i think

[primal pendant]

yeah i know all that i tried and had working it just after the very first long run before it creates image i can create many images fine but if i close then reopen i have to wait about an hour for it to compile every time i want to use it

[rugged latch]

ur telling it that u dont have enough ram when u use those args '

i believe

[rugged latch]

ahhh

[primal pendant]

i don't know how to fix

[rugged latch]

that might be a PC hardware related thing

[primal pendant]

such as?

[rugged latch]

what are u running stable on

and is it GPU or cpu

[primal pendant]

5800x 6700xt 32 gb ram 10 64

i run it on gpu

[neon orchid]

but between lowvram and medvram is there any difference?
o.o

[vocal burrow]

what s your gpu ?

[neon orchid]

oh 1sec

8135mb VRAM

[primal pendant]

i think i will take break from it for few days then maybe try yet again this is like 3rd day of tryign i think

[rugged latch]

idk munchu

shouldnt take so long

are u running python 3.10.6

what version of stable diff are u using

[primal pendant]

no i was told to do 3.10.11 by cs1o

[vocal burrow]

Amd, I don t deal much with red team gpus but with """"only"""" 8gb, --medvram might be enough. If you notice that it fills up your whole vram during generation, switch to --lowvram.
Also here s some guide for you : https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides#amd-automatic1111-with-zluda
(difference between lowvram and medvram is basically how much it s gonna split the model into small chunks, therefore how many times it will have to load and unload stuff from your gpu. When it comes to speed. --lowvram is the "slowest" ,--medvram is "slow", running without any of them is the fastest option)

@neon orchid

[neon orchid]

ohhh okay

[rugged latch]

u need stable diff 3.10.6

[vocal burrow]

did you update your gpu drivers ? did it update automatically ?

[vocal burrow]

no 3.10.11 is fine

[rugged latch]

o

[primal pendant]

drivers are fully updated yeah

[neon orchid]

I'm going to try to use --lowvram

[vocal burrow]

yes but did they got updated since last time you launched ?

[vocal burrow]

IIRC, each time you update your gpu drivers it s gonna have to "recompile" zluda stuff

[rugged latch]

try driverbooster

[vocal burrow]

(Cs1o is the amd guy here, not me)

[primal pendant]

i up[dated my gpu drivers quitea few days ago not recently

yeah cs1o was helping

[vocal burrow]

can you share some startup log ?

[primal pendant]

i really don't wnanna do clean install of windows 10 considering i did only few months ago

[neon orchid]

and the difference between time is a lot or a little? That's my last question 🙂

[primal pendant]

i can't i got pissed i removed everything nothing left i deleted all

but i did post earlier texts

[vocal burrow]

yes it s significant.

[neon orchid]

okay, thanks Aryetis and CS1o

for help me ^^

Oh, truth, another thing.
But the result will always be the same, right?
So the resolution of the drawings doesn't have a negative influence, does it?

[rugged latch]

ya it shouldnt have to recompile evrytime u laumch it <v>

[vocal burrow]

looks like it can t write there, check that you have permissions for it

[rugged latch]

to get to ur environments, just type this and click that

[vocal burrow]

or put
set TRANSFORMERS_CACHE=SomeFolderYouKnowForSureYouCanWriteTo
in your webui-user.bat

[vocal burrow]

the higher the resolution the more vram will be used

[neon orchid]

Ohhh okay, thanks a lot!

[rugged latch]

pandu that shoulda been obvious xD

qvq

think of it like when u play a game, with high res textures, and ur on a low end pc

pc wants to die

bc lacks suffiicent ram and power

[neon orchid]

And... out of curiosity, how do you Loras train?

[rugged latch]

that is out of my scope

i wan know too

xD

i heard theres an extension for it

[neon orchid]

I was always curious hehe

[primal pendant]

@vocal burrowignore that one i really broke things trying to fix earlier the last working one that loved to complile every start up run for an hour was this

this was after the massive long 1hour first run

this was me simply closing and tryin again

[merry wing]

Hey guys can you help me creating WeChat account for valorant mobile WeChat is ban in my country 🥺🥺

[ornate elk]

Maybe its really a permission thing that the compiled file can't get created in appdata too and only in temp

When I'm at home I send you the zluda compiled file location (path)
To check

[primal pendant]

i will need to start over i got pissed at it and removed everything

[ornate elk]

Oh okay

[primal pendant]

but sure i will add all to folders to full permissions if i kniw where

i did for all the folders i know of

also i really messed up on my last try thts when i got alot of permission error i was trying to fix and made far worse afterwards

matplotlib ect

[ornate elk]

Yea you Normaly should never need to add folder permissions to anything

Because on normal systems there shouldn't be permission errors in the first place on system folders

[vocal burrow]

seems good so far

[primal pendant]

that was earlier one yeah for first run after 1 hour or close to that created images all seemed fine i closed the program after and reopened and that's when the long compiling happens

like i wouldn't mind if it was 5 mins but it an hour before it stops it's like it's that very first run all over again each time i close it and open it

[valid iron]

heyo my stable-diffusion-webui folder is starting to take up lots of room, is there a way to clear it out without needing to change much? im not sure how oython etc deals with temp and cache files so i dont wanna guess lol

[crystal nacelle]

My AMD been dropping frames like a mofo, and its been getting worse. NOw that I think about it, it all started when I was running Stable DIffusion earlier this year, before the ZLUDA compatibility patch came out. Sometimes my webui would crash while rendering and it wouldnt work again till i restarted it. could this have done some kind of damage to my gpu over time?

[ornate elk]

Open up a CMD and run
Pip cache purge
That cleans the temp stuff.
And if you want to store the models on a different drive and link it to the webui, thats possible too

[ornate elk]

I doubt that.
But what's your GPU ? And check the temps of it while gaming

[valid iron]

ill try that thank you

[vast magnet]

hey can anyone guide me with image inpainting using stable diffusion?

[proper swallow]

Im having issues with ForgeUI. First off, dont see the actual image, I see this image (attached). Second, when I try to upscale it says after it is completed: FileNotFoundError: [WinError 3] 😦

[ornate elk]

set the inpaint area to "mask only", then set the resolution to a sqare resoltuin depending on the model you use.
for 1.5 its 512x512
for sdxl/pony its 1024x1024
768x768 can work for both too

then adjust the denois, start with 0.5

[ornate elk]

make sure the webui is whitelisted in any browser adblocker

[vocal comet]

I just updated my comfy-ui to ComfyUI: 2736d1cdf5 and when I open it this is what I get on any browser and I am using Manager: V2.51.2 I updated my nvidia drives and rebooted with no difference

[primal pendant]

@ornate elkhey so i'm back at pc and alot calmer than earlier

so where does compiling file hide ?

first thing i'm gonna do is reapir all permissions there are tools to do this

[vocal burrow]

if there's no permissions error, most likely there is no permissions to repair. And if there are permissions errors then it's telling you what folder to look at in the logs

[primal pendant]

did these

04 - Register System Files
Start (01/10/2024 21:38:04)
Running Repair Under Current User Account
Running Repair Under System Account
Done (01/10/2024 21:38:24)

22.01 - Repair bat Association
Start (01/10/2024 21:38:24)
Running Repair Under Current User Account
Running Repair Under System Account
Done (01/10/2024 21:38:26)

22.02 - Repair cmd Association
Start (01/10/2024 21:38:26)
Running Repair Under Current User Account
Running Repair Under System Account
Done (01/10/2024 21:38:28)

22.03 - Repair com Association
Start (01/10/2024 21:38:28)
Running Repair Under Current User Account
Running Repair Under System Account
Done (01/10/2024 21:38:30)

22.04 - Repair Directory Association
Start (01/10/2024 21:38:30)
Running Repair Under Current User Account
Running Repair Under System Account
Done (01/10/2024 21:38:32)

22.05 - Repair Drive Association
Start (01/10/2024 21:38:32)
Running Repair Under Current User Account
Running Repair Under System Account
Done (01/10/2024 21:38:34)

22.06 - Repair exe Association
Start (01/10/2024 21:38:34)
Running Repair Under Current User Account
Running Repair Under System Account
Done (01/10/2024 21:38:37)

22.11 - Repair reg Association
Start (01/10/2024 21:38:37)
Running Repair Under Current User Account
Running Repair Under System Account
Done (01/10/2024 21:38:39)

31 - Restore UAC (User Account Control) Settings
Start (01/10/2024 21:38:39)
Running Repair Under Current User Account
Running Repair Under System Account
Done (01/10/2024 21:38:41)

it don't tell me if they was needed or not but no harm right

[vocal burrow]

probably useless

(for this situation)

[primal pendant]

hmmmm maybe but i'm out of ideas as why it like to complile afetr the very long first run i close and does it all over again when i start up

[vocal burrow]

what does the log says for the second long start up ?

[ornate elk]

its under:
C:\Users\YOURUSERNAME\AppData\Local\ZLUDA\ComputeCache

the zluda.db file should be 1.8gb, or maybe smaller idk if it depends on the gpu

[primal pendant]

hmm ok thanks

all permissions are fine i checked

but that might be due to fix i just ran

i will dlete it anyway

right will try one last time

[lavish void]

why is the image blurry?

[primal pendant]

so afr so good

so as this complies that one in hidden program data should be filling up right ?

40.1mb 48.5 slowly going up

[ornate ruin]

Is there a way I can change the save location to my one drive so they are immediately accessible to me from other devices?

Or sync the txt-img folder with OneDrive?

Probably not because you can only select a save location within the stable-diffusion folder I believe

Never mind it works perfectly fine sorry that's pretty neat. Idk why I thought it wouldn't work..

[primal pendant]

so all is working fine done a fair few images testing 83.2 mb is that hidden one

now i guess ultimate test close and reopen

success i'm done so it really was messed up permissions things causing that grief for 3 days must of been that hidden one all along not having the permissions

[ornate elk]

Oh it works?

Which folder permission was the issue?

[primal pendant]

yeah i can close and open up no more compiling for an hour each time less than 1 minute to launch ui

i have no idea as i fixed the lot

[ornate elk]

Okay ^^ glad it worked now

[primal pendant]

tweaking.com

i used

[ornate elk]

Okay

[primal pendant]

so couldn't tell which as i just reapired all to be on the safe side lol

[ornate elk]

xD alright

[primal pendant]

nice tool though and free#

[lavish void]

tweaking.com

🤣

[primal pendant]

yeah was a whole bunch

i did bat and cmd and registry

oh directory to

[ornate elk]

And file permission I guess

[primal pendant]

think even did uac

[ornate elk]

Okay

[primal pendant]

so clearly something was broken asf

[ornate elk]

Yep

[primal pendant]

btw i have webforge one is the automatic1111 better ?

[ornate elk]

Automatic1111 has a better extension support but forge has flux support.
You can also use both and share the models folder. Thats what I do with all webuis

[primal pendant]

hmmm how much rooom is it to run more than one ?

[ornate elk]

Maybe around 7gb per webui

[primal pendant]

oh

[ornate elk]

Oh seems a bit full there

[primal pendant]

yeah always in the red

😭

how much space do you have as you been doing for so long right ??

[ornate elk]

Bought a additional 2tb SSD last month so now at 5TB

[primal pendant]

some models are 20 gb oh nice whats your go to brand ?

[ornate elk]

But I moved a lot of stuff on my old PC and an external drive

[primal pendant]

minesa real mix

[ornate elk]

1.5 models are 2-4gb
Sdxl and pony are 6gb
Flux can be 7-23gb

Oh I mean Stable diffusion models xD

[ornate elk]

My drives are all Samsung right now.
But in my old PC I had WD too

[primal pendant]

would flux work on 6700 xt ?

[ornate elk]

Could work yes

But will be slow and needs a lot of ram too

[primal pendant]

vram or system ram ?

[ornate elk]

System ram

[primal pendant]

i have 32

[ornate elk]

Thats good then

[primal pendant]

yeah 32

i do have 2nd pc nvidia gtx 1080 not even attempted that one

[ornate elk]

I have the same card in my old PC xD

Great GPU

[primal pendant]

msi gaming x ?

[ornate elk]

Gainward Phoenix GLH

[primal pendant]

oh

never tried them once

[ornate elk]

Its overclocked to 2ghz xD

[primal pendant]

my go to are asus/msi/xfx/gigabyte/sapphire i think my 6700 xt is which a first for me to

[ornate elk]

What's your 6700 brand?

[primal pendant]

sapphire

[ornate elk]

My 7900xtx is Saphire

[primal pendant]

pulse i think

[ornate elk]

Saphire makes the best AMD Custom gpus

[primal pendant]

yeah so i heard

[ornate elk]

After that comes Asus and msi

[primal pendant]

well asus are greedy

[ornate elk]

True

[primal pendant]

saphire best value

so whats you go to creating ai images? anime type or more relistic

[ornate elk]

Anime mostly but did realistic stuff too

[primal pendant]

any recommendation ?

[ornate elk]

Aam anime mix is a great anime model

[primal pendant]

ok will check that out

[ornate elk]

The sdxl variant is really good

[primal pendant]

it;;s xl not 1.5 still can use ?

[ornate elk]

You can use both

Sdxl should work without problems

[primal pendant]

6.5 gb ouch lol

[ornate elk]

But in forge enable the "never oom" button at the bottom to be safe, to not run out of vram

[primal pendant]

the what now ?

[ornate elk]

Yep sdxl models are heavy xD and slower than 1.5, but have a better quality due their native res of 1024x1024

[primal pendant]

after the amount of compiling i seen in last few days all is quicker lol

each start up run was 1hour plus

[ornate elk]

Yea that was crazy

[primal pendant]

where is this vram one ?

this
?

[ornate elk]

Yep

Just enable it if you go for higher resolution or if you use upscaling

It will prevent an out of memory error if it can

[primal pendant]

oh only then ?

[ornate elk]

It won't hurt to leave it enabled all the time as it detects when the resolution is critical, and only then works

[primal pendant]

ok#

any other settings i should know?

as complete newb

[ornate elk]

If you go for anime images I like the
Boorutag autocomplete extension

Its a nice help for prompting

[primal pendant]

where is that ?

[ornate elk]

Click on extensions, then on Available, then on Load from.
Then search for Boorutag autocompletion and install it

[primal pendant]

lol driver update jsut noticed

[ornate elk]

Oh its from today 😮

[primal pendant]

yeah like hell am i doing that

[ornate elk]

Better wait a week

[primal pendant]

well i got doubts and right now it's working properly lol

kinda scared to do anything

3 god damn days

[ornate elk]

Its fine to do the updates, but it could trigger a recompiling

[primal pendant]

yeah that to

i seen enough of that in the last 3 days

😭

[ornate elk]

True

I'm off for tonight. Already late again.
Feel free to ask if you have any questions

[primal pendant]

ok thanks alot will do thanks for the help

[ornate elk]

No problem 🙂

[primal pendant]

was you lucky enough to get any of these leaked check points

[neon orchid]

@ornate elkWhen you have time, would you give me a hand to try with Zluda I think it was

please :c

[proper swallow]

@ornate elk ok i will try again today

[proper swallow]

@ornate elk it solved the image problem, I can now see the image but I still have FileNotFoundError: [WinError 3] Cant find path, issue :/

[craggy epoch]

any idea why I get the following when I go into the Lora tab in A1111:

It used to have icons there, but this is a clean install on Pop_OS, but I have seen it with installs on Windows, Debian, Mint, etc.

the last time it happened was after I installed the civit browser, so I thought it was related to that. However, this is a clean first run

even more strange, it appears to only be in Chrome... Edge is just fine:

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker

[wooden marten]

Hi everyone, does stable diffusion 3 medium support IP adapters?

[charred rover]

Hi! Does Comfy work faster on linux compared to win11? Does TensorRT work faster? I saw someone say it s night and day but chatgpt said it s the same

[wooden marten]

@charred rover I have the same question. I kind of imagined that if comfyui integrated internal optimizations to deal with the gpu, these plugins won't be needed.

[neon orchid]

Guys, I have a question.
In which part of Stable Diffusion should I increase to improve the quality of a drawing?

Sampling Steps?
Widht - Height?
CFG Scale?

[charred rover]

everything + experimenting with prompts like 'dramatic light'

[neon orchid]

Dramatic Light?

Right now I'm trying with

[charred rover]

the main one is steps, yes, but I get results that reflect the prompt better with 30 than with 100 which is crispier

[charred rover]

also get a lora for it (XL)

[neon orchid]

768x768
Sampling Steps = 25
CFG Scale = 10

oh

30 steps?

[charred rover]

I m also not a pro but usually do 900*900 30 and 4-5.5

[neon orchid]

It's hard for me to find a Lora of X character that I want to use.
I've already searched and I haven't found any XD

[charred rover]

I dont understand

oh i see

[neon orchid]

So according to what you're telling me... if I add more steps will the drawings come out weird?

[charred rover]

no, just more boring and more quality)

actually sometimes it s great

[ornate elk]

More steps doesn't mean better quality

At each step the ai tries to change something in the image.
More steps = more changes.
At some point there isnt much to change anymore so no need to use more

25-35 is usually a good value

[ornate elk]

Hey, at which step do you need help?

[charred rover]

could you please answer about linux? we are curious

[mortal compass]

I'm getting this error message when trying to swap faces with ReActor in Forge
2024-10-02 14:34:56.2618144 [E:onnxruntime:Default, provider_bridge_ort.cc:1992 onnxruntime::TryGetProviderInfo_CUDA] D:\a\_work\1\s\onnxruntime\core\session\provider_bridge_ort.cc:1637 onnxruntime::ProviderLibrary::Get [ONNXRuntimeError] : 1 : FAIL : LoadLibrary failed with error 126 "" when trying to load "C:\Ai_Root\Stable-Diffusion_Forge\webui\venv\lib\site-packages\onnxruntime\capi\onnxruntime_providers_cuda.dll"

I asked chatGPT about it and it said "This error is related to ONNX Runtime attempting to use CUDA, but it's failing to load the necessary onnxruntime_providers_cuda.dll file." Which I think could be correct. ChatGPT also gave me potential causes; CUDA Toolkit not installed, Incompatible CUDA version, Path issue or Mismatch between PyTorch/CUDA and ONNX Runtime versions. I could only guess which one it could be. Any ideas?

[ornate elk]

There isnt a performance advantage for Nvidia cards on linux.
Only for AMD cards its better with stable diffusion

[remote cosmos]

I'm having a really hard time getting OpenPose to work.

[ruby yoke]

explain please

[remote cosmos]

Here are my settings

I have installed ControlNet 1.1.455 and have the openpose_dw preprocessor, along with an SDXL OpenPose Model (I am using Pony XL as my checkpoint model).

I have set up Openpose on the left to have a pose of a person standing, holding their right hand up, and set the Control setting to "ControlNet is more important", but my images generate without having that hand up at all.

Is there something I'm missing?

[ruby yoke]

start by setting control weight to max

[remote cosmos]

My workflow is to import an image I like the pose of in ControlNet, then use a preprocessor preview, then use the WebUI OpenPose Editor to tweak the pose to my liking, then drag that preview image to the input image.

[ruby yoke]

generate and see how that looks

[remote cosmos]

Got it, will set it to 2 and report back in a few.

[remote cosmos]

So uh

The colors got all messed up and its now giving me a from the back view

Which is... interesting

I noticed that the colors typically get messed up with weights above 1 but that from the back view is new

None of my other generations have ever done that, so it is using ControlNet in some way

[ruby yoke]

and it still didn't raise the arm. yeah, max on most of the sliders is going to over cook stuff