🤝｜tech-support

[loud pond]

I've tried lower resolutions but they always give me horrible faces

The model is the ponydiffusionV6XL

[plain pond]

aw.. i got new problem.. again and again.. sorry

[loud pond]

I'll give it a try again tho

[ornate elk]

roop is deprecated and wont work. you have to delete the roop extension and the venv folder again.
Then you can install the Reactor Extension for Faceswap

also check the folder permissions of your stable-diffusion-webui folder and maybe give yourself "all" permisions in the properties/security tab

[loud pond]

Thanks! It works, any reason as to why it has to be 768x768 and not something like 800x800 or 512x512

[ornate elk]

Pony is sdxl based and bot hare trained on 1024x1024
but 768x768 can work too mostly. 800x800 can work too, you have to try how far you can go

[loud pond]

Oh alright thanks a ton!

[ornate elk]

also you can try enable FP8 mode for SDXL in the settings

that should reduce vram usage

[plain pond]

okk.. like. all of this.. must chek the full control?

[ruby yoke]

just check allow for all of those

[ornate elk]

click on Users

then checkmark all "Allow"

then click on your username and do the same

[plain pond]

ow ok.. actually the user is already checkmark all.. but my username is not yet.. let me chekit wait...

[loud pond]

How do I enable FP8 mode? Where do I go in settings?

[ornate elk]

in the settings search for FP8 on the top left

then enable it for sdxl and click apply

[loud pond]

Oh alright what will it do?

Does it reduce any other features or something?

[ornate elk]

it loads sdxl and pony in fp8 mode which reduced vram usage and can improve the speed

[loud pond]

Oh alright thanks!

[loud pond]

gave me an error

[ornate elk]

hmm okay

then it wont work for directml it seems

[loud pond]

oh alright, still thanks for the help!!

[ornate elk]

np, what was your gpu again?

[loud pond]

Radeon 480

[loud pond]

I need to upgrade it soon. Any advice on gpus?

[ornate elk]

depends on your budget and what you want to do with it (gaming, ai, rendering) etc

but the gpu should have much VRAM

[loud pond]

Like something which will let me render a huge image

And probably do some gaming on the side

[ornate elk]

the more vram the better. more vram gets you done larger images or faster upscaling

[loud pond]

Alright, thanks I'll look into a gpu with a ton of vram then

[ornate elk]

yea i would go for 12gb vram as minimum

8gb will work too but 8gb is not future proof

Nvidia currently is more compatible with AI stuff but also more expensive than AMD.
low budget would be a 3060 12gb
mid range: 4060ti 16gb
high end: 4070ti 16gb, 4080 16gb or 4090 24gb

if you like linux or your fine with zluda then for AMD:
low budget: 6750xt
mid range: 7800xt 16gb, 7900gre 16gb
high end: 7900xt 20gb, 7900xtx 24gb

[loud pond]

So anything 24gb is overkill right?

[ornate elk]

24gb is overkill for the normal users that just want to generate some nice images

24gb is needed if you want to run crazy workflows or video generation etc but also very large models like FLUX

but FLUX got more optimised by the community so it runs also on 12gb vram

[loud pond]

Oh alright got it, so for me it would prob be something with 16gb. Thanks I'll prob look into the 4070ti then

Thank you so much!

[plain pond]

got this error

[ornate elk]

yea the 4070ti is a beast for ai and gaming for its nvidia "lower" price tag

[ornate elk]

seems like a connection error

make sure you have python 3.10.11 64bit installed

[plain pond]

so i need use vpn or smt?.. wait let me check the python

[ornate elk]

first try update your python version

[wet flame]

using forge ui but right after start the cmd closes no matter what

i ran it only one time

[ornate elk]

delete the venv folder and relaunch

[plain pond]

okk.. am sorry can i ask sir.. actually what happend with this situasion?.. because its never been happen before i reset the windows

[ornate elk]

python ssl verify error, could be caused by an antivirus programm (not windows defender), or by a regionblock of a government or ISP (like in asia or russia) or by the wrong computer time and date

if your from asia you can try change your DNS setting in the PC or try using a VPN

[plain pond]

okay now i think this makes sense.. after i reset the windows.. time and date is broken.. so i can't use the automaticly time and date.. because its not work & never change.. even i select region right way its not working.. so i need to set it manually... and of i change the dns to warp.. and then after that.. i install automatic 1111 and of course lot of error come out.. its about 1-2 days i can't fix that lot of error.. so i reset the windows twice.. and still same the first problem is date and microsoft things.. and still give me the erors things rn..

[ornate elk]

You should change the NTP server

That can fix time and date errors
And change your DNS to 8.8.8.8 for Google or 1.1.1.1 cloudflare

[chilly knot]

RuntimeError: CUDA out of memory. Tried to allocate 20 MiB (GPU X; X GiB total capacity;...

i use low ram mode .

[ornate elk]

What's your GPU ?

[chilly knot]

1050 gtx

4 gb

[plain pond]

yea i use cloudflare dns..

[chilly knot]

i try run ipadapter

it seems that now more memory is occupied than before, as if there were some bindings left

[ornate elk]

ok then try change the ntp server and reboot and then enable set date time automatically

[ornate elk]

after getting an out of memroy error you should relaunc the webui-user.bat
make sure to enable the lowvram checkbox at controlnet

[chilly knot]

i try this and res 256 256

it help me if i buy more memory?

[plain pond]

i dont know what is ntp server.. it safe to change? https://www.youtube.com/watch?v=NUJaMvbtmsM is this tutorial is work?

[ornate elk]

not its vram thats to less

[chilly knot]

now it use 95% if i start it

gpu can use RAM

[ornate elk]

try this: its much easier
https://www.windowscentral.com/how-manage-time-servers-windows-10

and set the time servet to your local country ntp

you can search it here: https://www.ntppool.org/en/

[plain pond]

ok let me try..

[plain pond]

this is not wrong right?.. idk why am scared sir haha

[ornate elk]

nope, not the example server, click on the link again i updated it

to a better site, that explains it more

https://www.windowscentral.com/how-manage-time-servers-windows-10

and then you need to find your country here and look for the ntp server.

https://www.ntppool.org/en/

or tell me xD

then i can search you the right ntp server

for example here you see different ntp servers:

you just need to find the one for your country and add it here:

[plain pond]

like this?

[ornate elk]

exactly 👍

then click on update now

and it should snychronize the time

you can also try restart the PC

[plain pond]

thanks.. so i need do this too?

[ornate elk]

nope

[plain pond]

ok thanks let me reset then..

[unkempt berry]

Quick question for everybody, I remember on automatic111 you could install an extension that would show an image wall of your checkpoints and loras, and then you could auto assign preview images to each checkpoint and lora. I'm using Forge now, and it has the "wall" for checkpoints, but I can't figure out how to give them preview images. Does anyone know how? Thanks

[plain pond]

im back.. is this ok if i not sync it?? is actually same with the time but nit exactly same..

[ornate elk]

idk what time it is at your location

verify with a smartphone clock

if not disable and re-enable the two buttons

[ornate elk]

hover over, and click on the icon on the top right.
if you then click "Replace preview" it will use the last generated image

[plain pond]

ah.. now it same

let me check the au1111

[torn lotus]

excuse me... I'm trying to use move_blk but the source is not an array, any workarounds?

[plain pond]

still same.. maybe i need to chane to google dns

still same.. maybe i uninstall malwarebytes now

[tranquil laurel]

Hi. I purchased 1000 credits for the API 10 minutes ago but I still get the error 'You lack sufficient credits to make this request. Please purchase more credits at https://platform.stability.ai/account/credits and try again.'

Is there something to do ? Should I wait longer?

[torn lotus]

lol @ me wrong place

[plain pond]

still not working.. ow maybe i need add --lowvram wait

[plain pond]

now its stuck...

[ornate elk]

whats your gpu? you dont need --lowvram mostly

[plain pond]

yea actually dont need it but just idk the problem still not solve it

intel core i5 10300h, rtx 2060 max q 6gb vram, ram 16gb

[ornate elk]

6gb vram needs --medvram not --lowvram

and --xformers

[plain pond]

dayum thats work man.. thank you very much master.. can i ask again What do you think that could happen because?.. i mean idont use medvram command and i dont have any issue with automatic1111 before reset.. but after i reset.. just boom.. i got tnt of error..even i update the gpudriver etc.. still like that.. dayum

[ornate elk]

You could try reinstall Auto1111

[plain pond]

i do about 4-5 times maybe and i reset windows twice and reinstall again still not working.. that why iam here.. i need fixed from u master lol

[ornate elk]

Do you still get the SSL error?

[plain pond]

nahh

u fixed

but just for now

[ornate elk]

What's the new error then

[plain pond]

nah iam just ask.. like what happend..

sorry if my english bad

but thanks for now it works

[ornate elk]

Oh you mean why it happened

[plain pond]

yaa

like why

like before i reset the windows that never happend

[ornate elk]

Are you on windows 10 or 11 ?

[plain pond]

windows 10 home

[ornate elk]

Hmm okay, should work normal

[ornate elk]

Good that it works now 🙂

[chilly knot]

ssl error when download something?

[fair oxide]

Hi, I am trying to use the WebUI with the Modern UI extension, but when I enable this extension, or any, then apply changes the extension is not used, and is still disabled

[tall geyser]

I'm a newbie and I can't figure out how to get Loras to work. I am running Flux locally with the WebUI Forge. I downloaded a lora.safetensors file from huggingface for the Lora I want to use, put it in Models/Lora and restarted WebUI Forge. With or without the trigger words, I get the same style. Removing the Lora file and restarting and re-prompting, I get the same image style. So, the Lora doesn't seem to have any effect when it's installed and I use the trigger words. What am I doing wrong?

[charred rover]

I could only install an older xfromers version with an older torch 2.0.1 and it gave me -10% inference speed so I gave up on the xformers

[lofty kiln]

My Card works everywhere any I know 100% postal code is correct , why is it complaining on for stability AI
I tried 2 different addresses and still same issue

[vocal burrow]

This channel is community maintained and about technical problems with local installs of SD and derivatives.
For problems with official SD websites you want to ask them directly through their support portal https://stabilityplatform.freshdesk.com/support/home.

[steady thistle]

Ah, I see Flux doesn't support openpose yet.

[glossy pumice]

Help it keeps crashing when it's just about to finish an image

I was only able to make one image

@ornate elk

Is forge good

[ruby yoke]

People that use it seem to like it

Personally, i prefer comfy

[median geode]

what's going on

[ornate elk]

whats the cmd error log?

[median geode]

@ornate elk @vocal burrow bro why does the extension connect fail over and over again? i can open github sucessfully, but sd can't access the github server. can you help me T_T

[glossy pumice]

No my lap goes to restart

Is that bad

I'm using lowvram and everything

[glossy pumice]

Is comfy or forge good

I don't understand what's the difference and stuff

[ruby yoke]

they are interfaces that you use in order to generate with the model. I prefer the way comfy works. what you should probably do is watch some tutorials for each of them on youtube to get a feel for how they look and what you do with them

[flat barn]

Hey, can anyone help me with motion lora? Its not working when trying to execute the prompt

[ruby yoke]

any errors in the console log?

[flat barn]

yup. where can i post them?

[ruby yoke]

right here

[flat barn]

So whenever i want to pan zoom or else it says this:
File "F:\Main\AI\Automatik1111\stable-diffusion-webui\extensions-builtin\Lora\networks.py", line 321, in load_networks net = load_network(name, network_on_disk)
File "F:\Main\AI\Automatik1111\stable-diffusion-webui\extensions-builtin\Lora\networks.py", line 254, in load_network
raise AssertionError(f"Could not find a module type (out of {', '.join([x.class.name for x in module_types])}) that would accept those keys: {', '.join(weights.w)}")

So it just skips it and everything else works fine

[ruby yoke]

you should probably take that error back to the developer and see if they've fixed that

[flat barn]

Well i just followed instructions on stable-diffusion-art so normaly it should work i guess? Maybe reinstalling sd would help? i dont know

[karmic hollow]

anyone here with experience making comfyui nodes? I have a question regarding how it works with python

[ornate elk]

Try change your DNS

[ornate elk]

What's your GPU?

[ornate elk]

Motion lora only works with AnimateDiff model 1.2

[flat barn]

Where do i see the current version or how do i get exactly this one?

So it is animating the video just not the zoom or else

[glossy pumice]

I did see comfy is more resource taking right

[glossy pumice]

Nvidia mx550

[ruby yoke]

Maybe

[glossy pumice]

Also have a built in 2gb GPU so it says combined 4gb

[glossy pumice]

Is forge better for lower end

[ruby yoke]

You really dont want to use an integrated gpu unless you have no issues waiting for hours for imafes to finiah

[ruby yoke]

Online is better, just rent some gpus

[glossy pumice]

At this point I wouldn't mind waiting hours but it just crashes

[glossy pumice]

How do I do that

Also can you teach me what's 1.5, sd3 sdxl and stuff

What's it mean

[ornate elk]

That is not supported by stable diffusion sry

[glossy pumice]

Not even a little?

Anything I can do at all?

[ornate elk]

Have you used --lowvram ?

[glossy pumice]

Yeah

[ornate elk]

And a 1.5 2gb model?

[glossy pumice]

I think it is

That's what I was gonna try next search alot of 1.5

Is there anything I can do on the webui maybe

[pulsar harness]

yo i have linux mint 22 and 1.10 webui when i try to launch it i get

"Traceback (most recent call last):
File "/home/me/stable-diffusion-webui-1.10.0/stable-diffusion-webui/launch.py", line 48, in <module>
main()
File "/home/me/stable-diffusion-webui-1.10.0/stable-diffusion-webui/launch.py", line 39, in main
prepare_environment()
File "/home/me/stable-diffusion-webui-1.10.0/stable-diffusion-webui/modules/launch_utils.py", line 381, in prepare_environment
run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True)
File "/home/me/stable-diffusion-webui-1.10.0/stable-diffusion-webui/modules/launch_utils.py", line 116, in run
raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't install torch.
Command: "/home/me/stable-diffusion-webui-1.10.0/stable-diffusion-webui/venv/bin/python" -m pip install torch==2.1.2 torchvision==0.16.2 --extra-index-url https://download.pytorch.org/whl/cu121
Error code: 1"

any fix?

[ornate elk]

Nope

The GPU is just to old

[glossy pumice]

So what else can I try

[ornate elk]

You could use the very slow CPU mode

[indigo dirge]

I added RAM..but it shows mt3600 at the bottom instead of 3600 hz?

What's the problem?

[glossy pumice]

Ok please tell me how

[ornate elk]

In the webui-user.bat you have to add --use-cpu all --skip-torch-cuda-test --no-half --opt-sub-quad-attention --precision full

Then save and relaunch

[glossy pumice]

Alright thank you

I'll tell how it goes

[ornate elk]

Yes its animating the video normaly

What motion module do you use?

[flat barn]

mm_sd_v15_v2.ckpt and the v2_lora's

[ornate elk]

That should work

[ruby yoke]

Those are different models. You can think of them as different books on a bookshelf

[flat barn]

I will just reinstall it later. But thanks for the help :)

[vocal burrow]

unstable internet connection

[indigo dirge]

What is the difference between mt/s vs hz/s
I have two hz pieces
I bought two pieces and found them to be mt
Should I return them or is it not a problem?

I hope someone can answer me?

[vocal burrow]

MTransfer/Second

just another way to measure speed

[gilded cradle]

hey, I have a problem, when I enter the prompt, a gray square is constantly generated regardless of the model I use

[ornate elk]

Sketch style isnt a model. Move Ir to models/Controlnet

[gilded cradle]

I don't see this folder

[ornate elk]

You then need the Controlnet extension first

Or wait

It could be a Lora

Check the description of the file where you got it from

If its a lora it goes into models/lora

[gilded cradle]

and what next? I moved it to Lora, ran it and I don't see a style to choose

[ornate elk]

You have to use the lora

Select it at the lora tab in txt2img

[empty cipher]

I need to talk with someone who really understands LoRA training

I've tried everything. Low learning rate, high learning rate, big model, small model, large batch size, small batch size, small dataset, big dataset, automated captions, hand-written captions, no captions,

Everything I can think of. I've attempted at least 100 times to train this LoRA for style, and nothing works

I need help -_-

[deft crystal]

Why, after updating to the latest version of Forge, does it take so long to generate an image.
This took 4 min. 47.6 sec.

[valid iron]

getting this error now? any ideas?

i think ccleaner issue after deleting duplicate files is the issue but not sure]

[indigo dirge]

I have a 3600 gb ram which one should i choose from this list?

[tall geyser]

I'm a newbie and I can't figure out how to get Loras to work. I am running Flux locally with the WebUI Forge. I downloaded a lora.safetensors file from huggingface for the Lora I want to use, put it in Models/Lora and restarted WebUI Forge. With or without the trigger words, I get the same style (which is not what I expect from the Lora). Removing the Lora file and restarting and re-prompting, I get the same image style. So, the Lora doesn't seem to have any effect when it's installed and I use the trigger words. What am I doing wrong?

[chilly knot]

maybe due to the wrong version torch consumes more vram?
before i can generate with ip adapter without problem now i have out of memory

[meager carbon]

hey, fresh install of windows here. im trying to get the zluda stuff to work, but comfyUI crashes after like 20 sec of trying to generate an image, and the regular 111ui wont even launch.

[ornate elk]

Hey, let's first try to get Auto1111 working then the others will work too

[meager carbon]

well, auto111 works less than comfy haha

comfy atleast launches and lets me start image gen

[ornate elk]

@meager carbon Yep but still Auto1111 needs to work for easier troubleshooting

What's your GPU again, the 7900xtx ?

Rocm HIP SDK 6.1 installed ?
How does your environment path settings look like?

Python 3.10.11 64bit installed too?

Radeon TM Graphics disabled in the device manager?

For auto1111 try delete the venv and .zluda folder and relaunch and show me your webui-user.bat

Adrenalin 24.9.1 installed and PC rebooted?

[fickle panther]

Sometimes when i generate on forge with sdxl, it just stops at cleanup minimal inference memory and then just slows my entire pc to a crawl with 100% constant gpu usage.

I notice it specifically only happens when using nkmd and esrgan upscalers

[ruby yoke]

Did anyone get hold of you for this?

[empty cipher]

No 😦

Sleep now though

[pale magnet]

I also am in dire need of help about LoRA, if there is two kind souls (or one really kind soul) out there

[tacit trench]

Its not a problem.

[ruby yoke]

Ping me when youre back online

[rough locust]

is it possible to run forge webui server in linux

[vocal burrow]

yes

[rough locust]

how can i connect because its not available on port

i mean how can i share

it says share=true in load()

but where is load

i figured out i guess

its --share

[vocal burrow]

you have to set the --share --gradio-auth username:password in the webui-user.sh command line arguments

[rough locust]

thx

[vocal burrow]

there is a line in there for command line arguments, make sure to uncomment it

[rough locust]

is there a list for command line args

in any file

found it

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings

[chilly knot]

maybe due to the wrong version torch consumes more vram?
before i can generate with ip adapter without problem now i have out of memory

[empty cipher]

Ping

[ruby yoke]

hi. can you tell me what it is you were needing help with, please

[signal sigil]

Ah thx for replying m8, i appreciate it.
I guess i will give that a try too then.
thx again for the reply!

[empty cipher]

Well, truth be told, I don't know. All I can say is that I've done the work of trying every combination of hyperparameters I can think of, as well as a number of different configurations of dataset, and I can't get the LoRA to look anything like the style I want to replicate or even halfway decent at all

I've got some kind of misunderstanding of the basics, would be my guess

[ruby yoke]

okay, let's start with, what is it you're trying to train a lora to do

[empty cipher]

🙂

I want to replicate a specific artist's style. Not for profit or anything like that mind you, just for personal use because I love their style

[ruby yoke]

so you are training what is called a style lora. to start with, how many images of their specific style do you have in your dataset?

[empty cipher]

I've been able to gather over 700 images

[ruby yoke]

okay. i'd like you to carefully read through this https://huggingface.co/blog/alvdansen/training-lora-m3lt

[empty cipher]

Though I know I probably don't need that many. I've tried with as few as 100 and as many as all 700. I also once tried taking my 100 and artificially expanding the dataset using crops and rotations

[ruby yoke]

i'd probably go up to around 1000 if i was doing it. but go read through that article

[empty cipher]

So, I've got a few general questions

First - what is the ideal dataset for training a style? How large, and what would it contain?

Second - captions. I've tried hand-written captions, automated tagging, and no captions. None seem to make any difference.

[ruby yoke]

there is no such thing as an ideal dataset - for anything. except - the more information you can feed the AI, the better. however, you have to modify that by what weights are you trying to update

[empty cipher]

Stable Diffusion 1.5, btw

[ruby yoke]

you are writing an errata sheet, and updating the information the base model has with new information when you use a LoRA - rather than retraining the entire model

[ruby yoke]

doesn't matter which one, the reason is the same

retraining a base model just to add some new information is extremely expensive. you train a lora, and use that with the model to update the information on the fly, while generating

[empty cipher]

That I do understand

[ruby yoke]

you have to tell the AI what things specifically are. your captions should be 1: what is in the image and 2: terms you intend to use in the prompt. if you give it a hundred pictures of red shoes, you lable them all 'cow' and then you use the lora and prompt for "a cow eating grass" - it's going to try to draw red shoes in a field eating grass. but because it has seen so many real cows, it'll mix those in too

[empty cipher]

That I also understand

[ruby yoke]

for a style, you want it to be very specific. if you scan all the pages in a coloring book that hasn't been colored, put them in a data training set, use labels like "black and white line drawing of a cow" "black and white line drawing of a fish" and so on - and give it a UNIQUE keyword - then when you use the lora, and you use the keyword in your prompt, and you say something like "a landscape with a barn in winter" - that keyword will trigger the AI to use the lora, and it'll then draw you a black and white line drawing of a barn in winter-ish. but if you also prompt for a black and white line drawing - it'll use the information in your lora as well as any information it learned previously about black and white line drawings to create your image.

you want to consider what it already knows, and what weights you need to update

[proper swallow]

Why is this happening. keeps crashing at 97%

[proper swallow]

I have 3060 12 gib vram

[jolly whale]

does anyone here have experience with the SD API? i'm making a discord bot for my server that'll let members generate images via my local install of forge. Everything is working great and the bot properly outputs images in the discord chat.

I am having an issue with vram managment though. i want the model to immediately unload from my vram after generating to save my system resources when not being actively used.

this is my current code for handling the unload: unload_response = requests.post("http://127.0.0.1:7861/sdapi/v1/unload-checkpoint") which appears to be the correct endpoint for SD's API but it doesn't work. any help is greatly appreciated 🙏

[wheat current]

hello, im wondering how i can install a model on the stable diffusion web ui? I cannot find any guides online. I read something about environments but idk how to access it, and im meant to type a command into it to download something

[knotty turret]

urllib.error.URLError: <urlopen error [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:997)>

anyone know how to fix trying to use openpose but when try to preview

[ornate elk]

Hey, if you have a webui installed you just need to download some models (files that are 2-6gb)
And then put them into the models/stable-diffusion folder.
You can find models on Civitai.com

[wheat current]

o thank you! how about this step then? i got this on a site called huggingface

[ornate elk]

Network error. Could be caused by antivirus Programm (not defender), ISP or gov blocking, or wrong PC time and date

[ornate elk]

Not needed at all, thats stuff for devs

[wheat current]

oh ok thanks so muchh

[knotty turret]

dont have antivirus

gov blocking??

so i can do nothing?

[ornate elk]

You can change your DNS or use a VPN

[knotty turret]

i using the dns from google

what should i use

any suggestion?

[ornate elk]

That should work or 1.1.1.1 cloudflare

[knotty turret]

ok i will try

[knotty turret]

need change ipv6 too?

or only ipv4

[ornate elk]

Only 4

[knotty turret]

restart pc?

f same error after change my dns
my time on pc its fine

[ornate elk]

You can open up a CMD and run
Ipconfig /flushdns

[knotty turret]

Ok

[ornate elk]

If you still get the error share the whole cmd log please

[knotty turret]

Ok

[ornate elk]

Hmm strange

[knotty turret]

maybe the model?

[knotty turret]

trying to use for sdxl model

[ornate elk]

Oh yes, try an other model

[knotty turret]

https://huggingface.co/xinsir/controlnet-union-sdxl-1.0

do you know another that works

for pony v6?

[ornate elk]

The bottom one from here:
https://huggingface.co/lllyasviel/sd_control_collection/tree/main

[knotty turret]

that one?

on bottom

sorry

Ok

[ornate elk]

That should work for sdxl and pony

[knotty turret]

alright

[knotty turret]

same error😭

[ornate elk]

What's your forge version?

[knotty turret]

i mean

did git pull

so i believe lastest one

[wheat current]

what am i meant to do with all of these files? just drop them in the models folder?

[knotty turret]

wake up luffy need you🫠

[ornate elk]

You only need the .safetensor

[ornate elk]

Can't help much with forge

[knotty turret]

😭

[ornate elk]

It uses different controlnet Settings

[empty cipher]

So, TL;DR, I'm using SD to create a big dataset of images in a style I've engineered from multiple LoRA and a specific set of keywords

What is "the ideal" dataset, in theory, such that I can train a LoRA on this style for another (more flexible) model?

[marble frigate]

anyone had this bug when trying to use animateDiff

[main rune]

@ornate elk I currently can run Pony, but the flux models always hit this error:
Error named symbol not found at line 90 in file D:\a\bitsandbytes\bitsandbytes\csrc\ops.cu

[buoyant heron]

im losing my mind i dont understand why this shit so hard

now i got it to work
Total VRAM 1024 MB is says why is that?

also its not using my gpu for somereason only my cpu

crashes at 75 percent always

Reconnecting then crashes

[buoyant heron]

goes to here

[ruby yoke]

if you run swarm, rather than what you're trying to do, it might be a lot eaiser

[ruby yoke]

cause you have an AMD cpu at a guess, and AMD loves to take over and do the GPU's job

[buoyant heron]

i got an intel cpu
and how do you run swarm?

[ruby yoke]

that's a question better asked in the #🐝｜swarm-ui channel

[vapid quiver]

How do I convert text into an image?

[ornate elk]

AMD doesn't work with the flux nf4 models.
You have to use the gguf models

[main rune]

Ah well that explains a lot

[drifting tundra]

hello, does anyone know why face restore delivers greyed out faces like these?

Using ComfyUI

[tacit trench]

Can you share me the workflow?

[drifting tundra]

Is it fine to upload a full image here or in private?

[buoyant heron]

anyone know All available backends failed to load the model.

idk why this shows

[split otter]

Why do I get this every time I try to apply a preprocessor? except Canny

[main rune]

@ornate elk it seems like if I load LORAs I get this error
Exception Code: 0xC0000005
not sure if it means anything but it somewhat annoyingly lack examples and precedents out there for me to try and diagnose myself

[umbral relic]

How to use ssl?

reverse proxy on https server is working but sometimes buggy and i have troubles redirecting connections from outside my local network directly to the api (a webui i made in php/java is able to connect using the proxy from the web only if both servers are on the same machine).

Accessing the api using my ui on other servers is giving issues (ssl, cors and proxy revelant)

Enabling ssl on the SD server will remove all these issues.

[azure magnet]

Can i set comfyui so it check and do the update every time i run it?

[ornate elk]

Which lora?

[fair oxide]

Is there no way to run scripts from the extras tab? I use ABG (background remover), but when I use the extras tab, it re-adds background color to the scaled image

[plain ocean]

I am getting this error while installing stable diffusion on amd OSError: [WinError 126] The specified module could not be found. Error loading "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\lib\fbgemm.dll" or one of its dependencies

[solemn moon]

when ever i see venv i just delete and rerun

delete the folder i mean

[plain ocean]

You mean, i should reinstall it?

like completely

[solemn moon]

no delete venv folder

[plain ocean]

oh

then run webui again?

[solemn moon]

yeeeaaaa

might work

just a bit of a dwnld

[plain ocean]

It also is not using zluda which i have downloaded

it uses the zluda which it downloaded on its own

how do i make it to use the zluda that i downloaded

[solemn moon]

no idea bro

[plain ocean]

dang alr thx brother

[solemn moon]

have you tried gpt

helped me solve a lot

[plain ocean]

gpt??

[solemn moon]

The error you're encountering is related to missing or incompatible dependencies for AMD GPUs when installing Stable Diffusion. Specifically, the error points to the failure in loading fbgemm.dll, which is a library used in PyTorch for optimized matrix operations on CPUs, and it might not be compatible with AMD setups.

Here are some steps to troubleshoot and resolve this issue:

1. Ensure Proper PyTorch Version for AMD
   Make sure you are using the correct version of PyTorch for AMD GPUs, as not all PyTorch builds support AMD. AMD GPUs typically require ROCm (Radeon Open Compute) for PyTorch.
   Follow these steps:

Uninstall the current PyTorch version:
bash
Copy code
pip uninstall torch torchvision torchaudio
Install the ROCm-supported version of PyTorch (for AMD):
bash
Copy code
pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/rocm5.2
2. Check Missing DLLs
The error could be due to missing dependencies (DLLs). Use a tool like Dependency Walker to check if fbgemm.dll or its dependencies are missing. Here’s how to do it:
Download Dependency Walker and open fbgemm.dll with it.
See if there are any missing DLLs, particularly related to Windows system files or AMD drivers.
If there are missing system files, you might need to update your system or install missing Visual C++ Redistributables.
3. Update Visual C++ Redistributables
Install or update the Visual C++ Redistributable packages, which are required by many DLLs.
Download the latest version here and install both x86 and x64 versions.
4. Check GPU Drivers
Make sure your AMD GPU drivers are up to date.
You can download the latest drivers from the AMD official website.
5. Use DirectML for AMD GPUs
DirectML is a machine learning backend for DirectX, supported by Windows, which can help run PyTorch models on AMD GPUs.
To use PyTorch with DirectML, you can install it as follows:
bash
Copy code
pip install torch-directml
You may also need to modify your Stable Diffusion setup to use DirectML instead of standard CUDA or ROCm.
6. Reinstall PyTorch with Dependencies
If nothing works, you can try reinstalling the entire PyTorch setup with the correct dependencies:

Remove the existing venv (virtual environment) and recreate it:
bash
Copy code
rm -rf venv
python -m venv venv
Then activate the virtual environment and install PyTorch:
bash
Copy code
venv\Scripts\activate
pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/rocm5.2
7. Modify Environment Variables (if needed)
Sometimes, missing dependencies are due to incorrect environment variable paths.

Make sure that the PATH environment variable includes paths to the torch and venv directories.

[solemn moon]

chat gpt, AI bruh

[plain ocean]

oh

[solemn moon]

pro tip, post your command prompt log thingy and your syetem specs, helps people fully see whats heppening

[chilly knot]

when i start install.py-ModuleNotFoundError: No module named 'launch'

when i try install launch-launch 0.1.3 depends on flowdas-meta<1.1 and >=1.0.1

[plain ocean]

oh

ok

my system specs are:

Intel i5 12400f

Amd rx 6600

32gb ram

[solemn moon]

just for the record, those people to see the specs are not me 😅

[plain ocean]

I got stable diffusion to work just a month ago using the same process

but now it shows this error bruh

Same error again: after deleting the venv folder and reinstalling it

[solemn moon]

the problem with going with this is that the problem is way less of a fix

[ruby yoke]

have you gone through all of the AMD guides pinned at the top of this channel?

[solemn moon]

i would wait for @ornate elk or @vocal burrow

[plain ocean]

No what im trying to tell is

i downloaded stable diffusion using the same process a month ago

and it worked

but now it doesnt??

[solemn moon]

ahh yea

what are your command args?

in yr bat file?

[plain ocean]

is visual c++ 2015 needed to run stable diffusion??

[ruby yoke]

when you deleted it, are you sure you removed everything correctly?

[plain ocean]

yes

[solemn moon]

my command args changed

[plain ocean]

my command args are --use-zluda thats it

[ruby yoke]

not stable itself. which interface are you running it in?

[plain ocean]

a111

amd for forge by ishqqytiger

*forge for amd

[ruby yoke]

suggest you get rid of that, install swarm instead, and run stable in comfyUI through swarm

[plain ocean]

no the thing is it works the smoothest in forge webui for amd

and runs without any errros

[ruby yoke]

it really doesn't. also, both forge and a1111 have had a number of changes in the last month. so that might be why stuff isn't working now

[ornate elk]

You need to follow my AMD guides from the first link of the pinned messages

Automatic1111 with ZLUDA

[plain ocean]

Ok ill give it a try

Thx

[solemn moon]

Just want to take a moment to appreciate A1111 and its ease

[plain ocean]

Like this?

And yeah im using python 3.10.6

is it good?

[plain ocean]

Oh that could be why im having errors

I did everything as mentioned in the guide ran into the same error again

See this is the full cmd thingy

I forgot to tell. I had uninstalled DotNet 8.0 Should i reinstall it?

and shall i try after installing both .net and visual c++??

[ruby yoke]

do yourself a huge favor. do what @ornate elk just told you. pull up his AMD guides that are pinned in this channel and follow them exactly. but first make sure your sytem is up to date

and if you are still stuck, talk to him - he's the AMD expert

[plain ocean]

I literally did the exact thing said in the guide brother and still ran into the error

It says i cant message unless im a friend soooo

[ruby yoke]

so you post here and you ping @ornate elk - he was just here, and i'm pretty sure it won't take him long to answer you. And i would suggest you reinstall dotnet if you haven't just becase other things on your system probably need it

[ornate elk]

Post the full CMD log

[balmy viper]

I have a base model of 7GB that I've created with merging, and I want to continue merging it with other models to make it better, but I dont have enough RAM. If I upgrade from 32GB to 64GB, whould it be enough, or is there some problem with merging a model that has been created by merging other models?

[azure magnet]

Can i set comfyui so it check and do the update every time i run it?

[crystal meteor]

hi hi, im a bit new to stable diffusion and I can't get it to work. I followed the steps on https://stable-diffusion-art.com/install-windows/#Installing_on_AMD_GPU
as I am running a 6900 XT but I keep getting "RuntimeError: Input type (float) and bias type (struct c10::Half) should be the same" anyone got any advice?

in my user.bat I ended up with this line:

set COMMANDLINE_ARGS= -skip-torch-cuda-test -disable-safe-unpickle --precision-full --no-half --precision full --upcast-sampling

[solemn moon]

@ornate elk here’s the log

[crystal meteor]

I got it working but now its using CPU instead of GPU 😭

[ornate elk]

You need to follow my AMD Automatic1111 with ZLUDA Guide

The link is in the pinned messages of this channel

[plain ocean]

Hello

[crystal meteor]

I see, ill give it a try, thanks :)

[plain ocean]

Stable diffusion worked but this is the error im getting now

[ornate elk]

NP, let me know if you have any questions

[plain ocean]

CUDA error: operation not supported
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

What does this mean??

@ornate elk What to do brother?

I followed your exact guide and did everything again from scratch

[main rune]

turns out it was some weirdness with something else, it's fine now

[plain ocean]

Sorry for the late reply

[plain ocean]

Sorry for the late reply again

[azure magnet]

Can i set comfyui so it check and do the update every time i run it?

[crystal meteor]

it took a bit to un-bork the instalation because of leftover stuff but it works great now :D thanks

[royal sapphire]

I have been trying for days to achieve the same results as XY on my local installation. However, I am failing.

I know that they use Stable Diffusion XL but nothing else. I have a sample prompt here.

"Create a cartoonish illustration of a muscular, anthropomorphic rat. The rat should have exaggerated, bulging muscles and a confident expression, resembling a bodybuilder. It should be holding a red dumbbell, showcasing strength. The rat should have large ears and a cartoonish face with a smirk. Incorporate a bright and bold color scheme with black as the background to make the character stand out. At the bottom of the image, add the text "GYM RAT" in a large, bold, red font that looks hand-draw"

Attached is the comparison.

Can anyone help me how to get a similar result locally? I have tried it several times and the results were always better with stablediffusionweb.

[ornate elk]

Nice np!

[ornate elk]

Can you show me the settings you tried when getting the error?

[steel geyser]

have an interesting problem. im using a custom node that batches images from a directory, in order to feed the different faces into a faceid ipadapter. unfortunately, when i do it this way rather than use the tedious default 2 way merge batch images node, it generates a new image for each individual face, rather than averaging all the faces together for one single generation. any ideas??

[knotty turret]

More easy to see:
idk what wrong

try to start prompt but then insta pause

[lucid summit]

Anyone know if there's an agent scheduler for forge?

https://github.com/ArtVentureX/sd-webui-agent-scheduler

[ornate elk]

That could work in forge

[lucid summit]

It doesn't show up in the UI properly

https://github.com/ArtVentureX/sd-webui-agent-scheduler/issues?q=forge

There's this but it's kinda ass

[knotty turret]

@ornate elk

[ornate elk]

Are you on AMD?

[neon orchid]

Guys

One question, can Pony diffusion models be in stable diffusion?

or would it cause any error?

[knotty turret]

When you ask it what piece of my pc do you need to know

but i 90% sure that i not using AMD

[chilly knot]

When try CUDA provider
FAIL : LoadLibrary failed with error 126 "" when trying to load Stable diffusion\stable-diffusion-webui\venv\lib\site-packages\onnxruntime\capi\onnxruntime_providers_cuda.dll"

need change onnx version?

[ruby yoke]

i'm not sure what you mean by

'be in stable diffusion'

can you explain what you're thinking, please

[neon orchid]

The thing is that I have Stable Diffusion

And there are base models that I want to use but they say it's for Pony Diffusion

And my question was...
If I use those Pony Diffusion models in Stable Diffusion would it cause any errors?

Por example

[ruby yoke]

okay - so you can think of each base model as a book. and things like loras as revisions to those books. you can't use a lora that was written for pony diffusion with anything other than pony. the words that needed to be updated wouldn't be on the same lines and pages in any other book

[neon orchid]

Oh

Damn... there are many models that I like but they are from Pony Diffusion

Well, thanks for clarifying my doubt.

😦

[ruby yoke]

you're welcome. i'm sorry but if you use the wrong revision with the wrong book, instead of changing a sentence from "the burning fire" to "the blazing fire" you might get "the chirping house door"

it doesn't work at all well

[neon orchid]

Now it makes me doubt whether to continue using Stable Diffusion

[ruby yoke]

stable diffusion which? Sd1.5, SDXL, SD2.1, SD3-2b-medium... also, pony diffusion has a LOT of internal issues. so if you want a more flexable model, you'd be better off using stable not pony

[neon orchid]

where can i see that?

The Stable Diffusion Model

I don't know which one I have

[ruby yoke]

let's start with what interface are you using to generate with? auto1111? forge? comfyUI? WebUI? something else?

[neon orchid]

uhmm....

Sorry I'm bad with that, where the heck do I know what I use?

But i think...

Automatic1111

I use this

[ruby yoke]

alrigh, after you run that, there should be a place to choose the model you want to use. can you find it and tell me what models it has availble?

[neon orchid]

?

is this?

[ruby yoke]

no. have you run auto1111 before?

[neon orchid]

Yeah

I have stable diffusion for 1-2 weeks I think

[soft fiber]

that should be the correct directory iirc, check the stable-diffusion one

[ruby yoke]

okay, please run auto1111 and after it's running, look around on the interface and see if you can find a place to choose which model to use

[knotty turret]

not using AMD

[main rune]

@ornate elk seeing how I have both HIP SDK installed, does that affect things like my constant out of memory errors?

[fair oxide]

Is it possible to make a good functional Lora with 3 different concepts? Now these concepts are somewhat similar. Like for an example 1 is about thighs, and other about chest ect. Or do I need to have 3 separate Lora’s for this

[ornate elk]

It shouldn't, but you can uninstall SDK 5.7

Then show me the txt2img settings when you get our of memory

[visual bane]

I am not able to use controlnet and scripts both together and when generating image the controlnet images are not coming. can anyone tell me further ?

[uncut heron]

can someone explain what i need to do to solve this problem - https://github.com/comfyanonymous/ComfyUI/issues/5004

[sleek meadow]

Please tell me, has the price for generation gone up? It used to be 0.96 credits for 4 pictures, now it's 4.8
On https://dreamstudio.ai

[vocal burrow]

depends of the model, resolution, etc

[sleek meadow]

But I did not change these settings.
Earlier in the billing panel it was written that 1000 credits is about 5000 images. There is no such thing now.

[vocal burrow]

It depends, maybe you remember the pricing before sd3 launch and they changed the settings in order for sd3 to be the new default model.

If not, that s probably something to discuss with their support https://stabilityplatform.freshdesk.com

[lone kettle]

Hi Guys i am looking to achieve an use case where i have a template image as attached.

I want to in paint provided child face in to the original image.
With face swap models it works pretty well but the hair is as it is,

It doesn't modify the hair which is a prime requirement.

Can i have some help here ?

[steel garden]

Does anybody know of an alternative for agent-scheduler that works for ForgeUI?

[lucid summit]

for those on mac what's your workflow r yall just using serverless like runpod or vast?

[twilit gale]

Hey everyone! I'm working on a large-scale 360 equirectangular video project with over 32,000 frames and need some advice. Specifically, I'm looking for help with:

Controlling latent images and noise for smoother transitions
Mapping latent control to a camera track
Tips or best practices when working with equirectangular projection in animation

I have prior experience with A1111, Stable Diffusion, and ControlNet on a similar project from two years ago, but this one’s a bit more complex. Any insights or techniques would be super helpful! I'm happy working with python, json.

I'd really appreciate some explaining on how to use your own noise layer. and how i'd go about building my own transformation plugin for equirectangular projections in A1111 or other ui or alternatively how I can use software like blender/TouchDesigner to take my rendered image transform it based on camera tracking data (along with the noise layer) and layer them all with the next frame from my reference video, to use as my init for my next frame.

[flint night]

nvm

[main rune]

uninstalled 5.7, this happens
Failed to load ZLUDA: Could not find module 'V:\SD-Zluda\stable-diffusion-webui-amdgpu-forge\.zluda\nvrtc64_112_0.dll' (or one of its dependencies). Try using the full path with constructor syntax.

[ornate elk]

Delete the .zluda folder and relaunch

[ruby yoke]

With a LoRA, you are updating the weights of the model so you can add information without retraining the entire model. so yes, you could - but are you always going to want all of those weights changed every time you use the LoRA?

[humble flare]

Could anyone help me troubleshoot why DoRAs dont seem to work anymore on my systems?

It used to work until I updated FORGE. Then, I tried some code fixes - those didn't work. Someone advised me to switch back to A1111 where DoRA support is automatic.

Just got A1111 working again and i'm getting outputs like this:

[main rune]

Way ahead of you, now it just crashes and won't launch

already tried deleting .zluda and the pycache as well assuming the previous files was the issue

[woven zealot]

Are there any resources to help me use more of Automatic1111 on a bad GPU?
I don't mind the rendering time (that much), I just wanna be able to generate images above 512p and upscaling above 765p without running out of VRAM (at least running out of it that often) even if it means images will take a long time

[solemn moon]

What’s your setup

[woven zealot]

i5, GTX 1050 3GB, 12GB RAM

[solemn moon]

Did you add to your command args?
Add xformers?

[woven zealot]

Nope. I have not started optimizing it at all yet, other than using firefox and closing everything I don't need. So literally any ideas will be useful

[solemn moon]

Okay you know your .bat file you use to run Auto

[woven zealot]

Yuup

[solemn moon]

Right click and edit with notepad

Ahh shoot I’m not at my laptop right now
But I’ll forward what to change it to in a couple 10 mins 😅🤦‍♂️

[solemn moon]

Or maybe one of the code bros here knows off hand what to add

[woven zealot]

Oh that's actually perfect because I broke something a bit earlier so I'm trying to fix that first hehe

[woven zealot]

Alright I got it sorted back out. Now I can do the optimizing if anyone can help me

In the meantime I'm gonna check on xformers 34N3 mentioned

NotImplementedError: No operator found for `memory_efficient_attention_forward` with inputs: query : shape=(2, 4096, 8, 40) (torch.float16) key : shape=(2, 4096, 8, 40) (torch.float16) value : shape=(2, 4096, 8, 40) (torch.float16) attn_bias : <class 'NoneType'> p : 0.0 `cutlassF` is not supported because: xFormers wasn't build with CUDA support `flshattF` is not supported because: xFormers wasn't build with CUDA support requires a GPU with compute capability > 7.5 `tritonflashattF` is not supported because: xFormers wasn't build with CUDA support triton is not available requires A100 GPU `smallkF` is not supported because: xFormers wasn't build with CUDA support dtype=torch.float16 (supported: {torch.float32}) max(query.shape[-1] != value.shape[-1]) > 32 unsupported embed per head: 40

Yeah I think my GPU is too cooked for that. Unless I'm doing something wrong, my args are

set COMMANDLINE_ARGS=--xformers
set XFORMERS_PACKAGE=xformers==0.0.17

I don't really know what I'm doing though so be patient with me lol

[main rune]

@ornate elk Update: After some restarting and reinstalling of stuff this is what happens now

[solemn moon]

these are my command args
--xformers --no-half --medvram --always-batch-cond-uncond --api --ckpt-dir

[woven zealot]

And what about set XFORMERS_PACKAGE=xformers==0.0.17
Do I get rid of that line or leave it?

[solemn moon]

i would try clear all of yours and try mine

ive got a 1660 and those are mine
maybe you want --lowvram instead of --medvram

[woven zealot]

Yeah I'm getting a launch.py: error: argument --ckpt-dir: expected one argument error when launching

[solemn moon]

oh wait my bad take that out

--xformers --no-half --medvram --always-batch-cond-uncond --api

thats me changing my directory of folders to another harddrive

[woven zealot]

Yeah I still get the same error as when the first time I added xformers

NotImplementedError: No operator found for `memory_efficient_attention_forward` with inputs: query : shape=(2, 4096, 8, 40) (torch.float32) key : shape=(2, 4096, 8, 40) (torch.float32) value : shape=(2, 4096, 8, 40) (torch.float32) attn_bias : <class 'NoneType'> p : 0.0 `cutlassF` is not supported because: xFormers wasn't build with CUDA support `flshattF` is not supported because: xFormers wasn't build with CUDA support dtype=torch.float32 (supported: {torch.bfloat16, torch.float16}) requires a GPU with compute capability > 7.5 `tritonflashattF` is not supported because: xFormers wasn't build with CUDA support dtype=torch.float32 (supported: {torch.bfloat16, torch.float16}) triton is not available requires A100 GPU `smallkF` is not supported because: xFormers wasn't build with CUDA support max(query.shape[-1] != value.shape[-1]) > 32 unsupported embed per head: 40

[solemn moon]

Okay 1 sec

[woven zealot]

Thank you so much for helping me with this though

[solemn moon]

lol im just trying

try build xformers
ive forgotten how to do it

https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/2103

[woven zealot]

Oh hold on, I don't even have CUDA installed. I swear I thought I had. That could be it

[main rune]

I have come to the conclusion that for my particular setup (5600X with 6600XT on Windows) ROCm 6.1 is practically useless, nobody talks about it except to mention it probably doesn't work and any troubleshooting in the reddit or discord just defaults to 'just use 5.7'

[little gate]

Hello everyone. I am getting CUDA out of memory errors when I try to make larger images. I have tried and tried different steps. The computer is a high performance gaming laptop that I use for photoshop and gaming, but for some reason it's acting like my VRAM is super low. I have tried looking things up online and just can't get it working. Any help would be great.

[main rune]

find out what your mGPU is first

[little gate]

I have 2 GPUs

[main rune]

uh oh

iGPU

[little gate]

?

also apparently when I go to check my settings, my screen displays are being used by different gpus...

this is the error

OutOfMemoryError: CUDA out of memory. Tried to allocate 26.78 GiB. GPU 0 has a total capacty of 6.00 GiB of which 4.33 GiB is free. Of the allocated memory 620.47 MiB is allocated by PyTorch, and 45.53 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

[woven zealot]

@solemn moon followed the guide and it's still the exact same error... Is there a test I can do to see if I even installed xformers correctly?

Oh hold on, when I launch A1111 I see this in console

No module 'xformers'. Proceeding without it.
*** Cannot import xformers
    Traceback (most recent call last):
      File "C:\Users\Fisi\stable-diffusion-webui\modules\sd_hijack_optimizations.py", line 160, in <module>
        import xformers.ops
    ModuleNotFoundError: import of xformers halted; None in sys.modules

So it's not installed... How

[solemn moon]

Hmmm

Trying to remember back

[woven zealot]

Just to make sure, I'm redoing all the steps again

[solemn moon]

Try —reinstall xformers in command args

Also have to check the versions you’ve stalled
Something like cuda 3.10 or something

Maybe post the full log so other can also see exactly what might be the issue

[woven zealot]

3.10....
What if mine says 11.3 LOL

[solemn moon]

Oh no wait that must be python 3.10

OMG im probably messing up your pc 😅🤷‍♂️

[woven zealot]

Yee I got both 3.10.6 and 3.10.11

[solemn moon]

But we trying 😂🤪

[woven zealot]

It's fine, I'm desperate hahahah

Aight lemme get full logs

After I do --reinstall xformers

@solemn moon 'No module xformers' ..... 'please reinstall xformers'
That's... what I'm doing though lmao

Full boot up console

venv "C:\Users\Fisi\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr  5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: v1.10.1
Commit hash: 82a973c04367123ae98bd9abdf80d9eda9b910e2
Launching Web UI with arguments: --reinstall-xformers --no-half --lowvram --always-batch-cond-uncond --api
C:\Users\Fisi\stable-diffusion-webui\venv\lib\site-packages\timm\models\layers\__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers
  warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning)
WARNING:xformers:WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for:
    PyTorch 2.0.0+cu118 with CUDA 1108 (you have 2.1.2+cu121)
    Python  3.10.10 (you have 3.10.11)
  Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers)
  Memory-efficient attention, SwiGLU, sparse and more won't be available.
  Set XFORMERS_MORE_DETAILS=1 for more details
No module 'xformers'. Proceeding without it.

I'm gonna try and downgrade to 3.10.10 while waiting for someone

[solemn moon]

yo

this is gpts response

How to fix:
Reinstall xformers: To make sure xformers is compatible with your current setup, you can reinstall it following the steps on the official xformers GitHub page. Here's the general process:

Activate your virtual environment.
Uninstall the current version of xformers:
bash
Copy code
pip uninstall xformers
Reinstall the correct version of xformers that is compatible with your setup (PyTorch 2.1.2 and CUDA 12.1). You may need to install it using a custom wheel based on your CUDA version:
bash
Copy code
pip install xformers --force-reinstall
Verify CUDA version: Ensure that the CUDA version installed in your environment (CUDA 12.1) is compatible with your PyTorch version. If not, you may need to adjust the installation by selecting the appropriate version from the PyTorch installation guide.

Check if you need xformers: If xformers features (such as memory-efficient attention) are not essential for your use case, you can proceed without it, as the warning suggests. However, reinstalling xformers with the correct dependencies should resolve the warning if you need its functionality.

[woven zealot]

Aight let's do it

[woven zealot]

pip uninstall xformers - done
Install CUDA 12.1 - done
Install PyTorch 2.1.2 - According to the logs I last sent, already been done so skipping that
pip install xformers --force-reinstall - done
Let's see if that did anything

[solemn moon]

LG!!!!

[woven zealot]

You are running xformers 0.0.17. :0

[solemn moon]

WWWOOOOOHHHOOOOO

[woven zealot]

Well shit

So

I have xformers now, finally

But when I try to generate an image I get the exact same error as before

[solemn moon]

omg wth

and i bet its an easy fix us non code bro's just dont know

[woven zealot]

Of course it works perfectly fine if I remove --xformers from args

[solemn moon]

hmmm

Maybe we need more technical help

Maybe start fresh on the boot

Then press generate with a prompt and xformers in the bat, and post the full log for someone else to help

[woven zealot]

Ima try chatgpt's solution first. What do you mean by start fresh on the boot?

[solemn moon]

like start Automatic again, run the .bat file so the log is clean from start

just so whoever is reading can see a clean log, easier to read i dunno 🤷‍♂️

[little gate]

I effectively can't generate images over 512/768 or I get an Out of Memory error. anybody have suggestions?

[woven zealot]

We're working on the same issue then hehe

Apparently first step is to get xformers which I'm struggling with right now

[little gate]

I have xformers

basically tried everything

[solemn moon]

Update your drivers

[little gate]

did

[solemn moon]

Delete venv folder and run .bat again

It will reinstall the folder

[little gate]

I... dont know what that stuff is

[solemn moon]

It’s a folder in your stable diffusion folder

It’s called venv

Delete it then run .bat file again

No idea what that does but sometimes it fixes whatever issue I have 😅

[woven zealot]

Okay I found the possible reason for my issue
Your GPU needs compute capability of 7.5 for xformers, and my GPU only has 6.1
So I gotta find a way to make xformers run on 6.1

ChatGPT says it is possible

[solemn moon]

We need @ornate elk to save us 😭

[woven zealot]

Bro pinged the Avengers

[solemn moon]

Or @vocal burrow

[woven zealot]

Just to TL;DR if they come

I get this error if I try to generate an image with xformers

NotImplementedError: No operator found for `memory_efficient_attention_forward` with inputs: query : shape=(2, 4096, 8, 40) (torch.float32) key : shape=(2, 4096, 8, 40) (torch.float32) value : shape=(2, 4096, 8, 40) (torch.float32) attn_bias : <class 'NoneType'> p : 0.0 `cutlassF` is not supported because: xFormers wasn't build with CUDA support `flshattF` is not supported because: xFormers wasn't build with CUDA support dtype=torch.float32 (supported: {torch.bfloat16, torch.float16}) requires a GPU with compute capability > 7.5 `tritonflashattF` is not supported because: xFormers wasn't build with CUDA support dtype=torch.float32 (supported: {torch.bfloat16, torch.float16}) triton is not available requires A100 GPU `smallkF` is not supported because: xFormers wasn't build with CUDA support max(query.shape[-1] != value.shape[-1]) > 32 unsupported embed per head: 40

I'm suspecting that this is because my GPU only has compute capability of 6.1 (out of 7.5 required). ChatGPT says it is still possible to install xformers for 6.1, but already warns me that cutlassF and tritonflashattF still wont work, which is part of that error

[woven zealot]

Man, I'm on such freaking side quests to make it work. I'm literally installing extensions for visual studio

Hehehe you know it's fun when even shutting down your PC gives you an error

[ornate elk]

The error says it can't find the correct rocm lib files. Because you need these:
https://github.com/likelovewant/ROCmLibs-for-gfx1103-AMD780M-APU/releases/download/v0.6.1.2/rocm.gfx1032.for.hip.sdk.6.1.2.7z

Replace the files in the library folder of rocm/6.1/bin/rocblas/library

And put the rocblas.dll into the rocblas folder

[ornate elk]

Hey, what's your issue?
And GPU?
And what's inside your webui-user.bat?

[main rune]

great timing, I was just about to try restarting everything with 5.7

[ornate elk]

It should work without a problem with hip SDK 6.1

It just need the correct library files

[main rune]

fingers crossed

[woven zealot]

1050
Yes, I know I'm cooked
But I just wanna be able to generate images above 512x512, even if it hurts the rendering time
I wanna get my A1111 as optimized as possible
Wanted to start with installing xformers, but if I install it and use it in my args of webui-user.bat, when I try to generate an image I get

NotImplementedError: No operator found for `memory_efficient_attention_forward` with inputs: query : shape=(2, 4096, 8, 40) (torch.float32) key : shape=(2, 4096, 8, 40) (torch.float32) value : shape=(2, 4096, 8, 40) (torch.float32) attn_bias : <class 'NoneType'> p : 0.0 `cutlassF` is not supported because: xFormers wasn't build with CUDA support `flshattF` is not supported because: xFormers wasn't build with CUDA support dtype=torch.float32 (supported: {torch.bfloat16, torch.float16}) requires a GPU with compute capability > 7.5 `tritonflashattF` is not supported because: xFormers wasn't build with CUDA support dtype=torch.float32 (supported: {torch.bfloat16, torch.float16}) triton is not available requires A100 GPU `smallkF` is not supported because: xFormers wasn't build with CUDA support max(query.shape[-1] != value.shape[-1]) > 32 unsupported embed per head: 40

Of course removing --xformers from the args fixes it

[ornate elk]

What else is in your webui-user.bat ?
And can you post a full cmd log?

[woven zealot]

Yup, hold on

venv "C:\Users\Fisi\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr  5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: v1.10.1
Commit hash: 82a973c04367123ae98bd9abdf80d9eda9b910e2
Launching Web UI with arguments: --precision autocast --xformers --no-half --lowvram --always-batch-cond-uncond --api
C:\Users\Fisi\stable-diffusion-webui\venv\lib\site-packages\timm\models\layers\__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers
  warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning)
WARNING:xformers:WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for:
    PyTorch 2.0.0+cu118 with CUDA 1108 (you have 2.1.2+cu121)
    Python  3.10.10 (you have 3.10.11)
  Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers)
  Memory-efficient attention, SwiGLU, sparse and more won't be available.
  Set XFORMERS_MORE_DETAILS=1 for more details
=================================================================================
You are running xformers 0.0.17.
The program is tested to work with xformers 0.0.23.post1.
To reinstall the desired version, run with commandline flag --reinstall-xformers.

Use --skip-version-check commandline argument to disable this check.
=================================================================================
Loading weights [fac1978d0b] from C:\Users\Fisi\stable-diffusion-webui\models\Stable-diffusion\v1-5-pruned-emaonly.safetensors
Creating model from config: C:\Users\Fisi\stable-diffusion-webui\configs\v1-inference.yaml
Running on local URL:  http://127.0.0.1:7860

To create a public link, set `share=True` in `launch()`.
C:\Users\Fisi\stable-diffusion-webui\venv\lib\site-packages\huggingface_hub\file_download.py:1142: FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.
  warnings.warn(
Startup time: 62.5s (prepare environment: 25.2s, import torch: 15.8s, import gradio: 7.7s, setup paths: 5.5s, initialize shared: 0.7s, other imports: 2.2s, list SD models: 0.8s, load scripts: 2.4s, load upscalers: 0.2s, initialize extra networks: 0.1s, create ui: 0.9s, gradio launch: 1.0s, add APIs: 0.4s).
Loading VAE weights specified in settings: C:\Users\Fisi\stable-diffusion-webui\models\VAE\kl-f8-anime2.safetensors
Applying attention optimization: xformers... done.
Model loaded in 40.4s (load weights from disk: 0.9s, create model: 2.2s, apply weights to model: 30.2s, apply float(): 0.1s, apply dtype to VAE: 0.3s, load VAE: 3.8s, load weights from state dict: 0.2s, hijack: 0.3s, load textual inversion embeddings: 0.1s, calculate empty prompt: 2.1s).

[ornate elk]

This was the error. Dont set variables yourself. You installed and old incompatible xformers version (0.0.17)

Remove that xformers package line from the webui-user.bat

[woven zealot]

To be fair I probably went through a lot of versions. Overall I reinstalled xformers like 10 times, each time with a different way

[woven zealot]

Aight

[ornate elk]

Then delete the venv folder and relaunch the webui-user.bat

Then it should work

[woven zealot]

I mean, yeah, it will

Like I said, if I remove --xformers from the launch args then everything works fine

But idk how else am I supposed to use it

[ornate elk]

Keep --xformers

And delete the venv

Then it will automatically install the correct version

[woven zealot]

Hold on, so do I delete --xformers or keep it. Which line do you want me to delete

[ornate elk]

If that won't work, open up a cmd and type
Pip cache purge
Then delete the venv folder

[woven zealot]

Be patient with me, I have no clue what I'm doing hahaha

[ornate elk]

Keep

And delete the venv folder

AFK now, driving to work

[woven zealot]

Okay, deleted venv, relaunching a1111

Now a1111 doesn't even launch

https://pastebin.com/DfedzLEZ

The red part being

ERROR: THESE PACKAGES DO NOT MATCH THE HASHES FROM THE REQUIREMENTS FILE. If you have updated the package versions, please update the hashes. Otherwise, examine the package contents carefully; someone may have tampered with them.
    torch==2.1.2 from https://download.pytorch.org/whl/cu121/torch-2.1.2%2Bcu121-cp310-cp310-win_amd64.whl#sha256=9925143dece0e63c5404a72d59eb668ef78795418e96b576f94d75dcea6030b9:
        Expected sha256 9925143dece0e63c5404a72d59eb668ef78795418e96b576f94d75dcea6030b9
             Got        944fd7e44d88caaa600516b8f3ee45d44ab5e3b23afc149836182ffacc0cd264

[ornate elk]

#🤝｜tech-support message

[woven zealot]

Oh, my bad

[main rune]

SD is running now, need to gen an image first to be sure but just making it this far is already good.
I expected there must be something missing for the 6.1 files but there's literally no mention of it existing anywhere as far as I can tell so I can only assume this is a recent thing

[woven zealot]

Nah I still get the same error

Purged the cache, deleted venv, relaunched a1111 and I get the same error

Even if I remove all args

So now I can't launch a1111 at all hehe

[main rune]

hehe

[ornate elk]

Yea I will update my Guides for 6.1 maybe this weekend

[ornate elk]

What error do you get now?

And please show me the content of your webui-user.bat

[woven zealot]

@echo off

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

call webui.bat

As vanilla as it gets

You sure it was purge cache and then delete venv and not the other way around?
Because that's what I did and now it's actually redownloading something

[ornate elk]

Yea the venv will be recreated

With the correct files

Pip cache purge just removes old install stuff

[woven zealot]

Yeah I first purged the cache and then deleted venv
What that did is that I can no longer open A1111 at all, even with empty args

But I did it the other way around
Deleted venv and THEN purged cache and now it works again

[ornate elk]

Shouldn't matter normaly xD

[woven zealot]

IT ✨

I'm gonna now readd --xformers back into args and see what happens, if it works or not

[ornate elk]

But good that it works now. Then you need to add --xformers --lowvram --no-half-vae
To the webui-user.bat

[woven zealot]

installing xformers

It's cooking

Okay, A1111 launched. Let's see if I can generate an image

Oh god, it works. It was so easy. I was literally going on side quests of installing a super specific version of CUDA and building my own build of pytorch out there, while all I had to do was delete venv XD

Thank you, you're a wizard ❤️

[ornate elk]

No problem ^^

[woven zealot]

Now that this is done... Anything else I can do to be able to squeeze more out of A1111 on a lowend PC?

[ornate elk]

Yea you can install the tiled VAE extension. That can help to not run out of vram that fast when upscaling

[static gust]

can someone help find an art style model

[woven zealot]

How do I search for them properly/know that they're the right VAE?

[ornate elk]

Click on the extensions tab, then on Available, then on "Load from"
Then you'll get a list.
There search for Tiled

[woven zealot]

Holy
I can actually generate 1280x720 images now... Something I couldn't do on my old RTX 2060 (that also didn't have xformers)

(Don't ask how I downgraded to 1050)

[ornate elk]

Okay was about to ask xD

[ornate elk]

Thats something maybe for #📝｜prompting-help

[woven zealot]

Well, if you insist. I worked from home for 3 years and it was the PC the company provided me with. I used it as my own personal PC even though I wasn't allowed to. I don't work there anymore though so I had to give it back

[ornate elk]

Ahh

[static gust]

oh ok ty

[woven zealot]

I was gonna ask if there's anything else like xformers that would help me not run out of VRAM when generating an image, but honestly, I should be happy I can do 1280x720 now, if not more hahahah

Thank you ❤️

[ornate elk]

Best you can do is use 1.5 2gb models and then you can try hires fix but only with the tiled VAE extension

With that you can try upscale them

[woven zealot]

What the hell??
xformers is actual magic
I'm even doing 1920x1080 on a GTX 1050

With ETA of 20 minutes, but it's doing it!

[ornate elk]

Yea its a must have for Stable diffusion

Better generate at lower resolution

[woven zealot]

Yuuup hahahah
But it's good to know I can do it!

[ornate elk]

Because using to high res causes deformation and duplicate's

[woven zealot]

That I don't have to be paranoid about VRAM

I only did it as a benchmark

[main rune]

woop, I probably shouldn't have tested with a new model, it's taking its sweet time

took 8 minutes, but I can say the files worked

[ornate elk]

If you used an sdxl or pony model you can try add --medvram-sdxl to the webui-user.bat

That can speed it up

[main rune]

I'm loading it for the first time, I think it would be faster on the second go around

3 minutes, faster now

[main rune]

Interesting, I am a bit slower with 6.1 but it's less stressful on my PC

Like, significantly so

[ornate elk]

Yea its much better optimised

And has a longer support

Also make sure you get the tiled VAE extension when you want to upscale

[main rune]

quick question, does different models affect speed

[ornate elk]

Yes

Sdxl and pony need longer as they are bigger in size and trained on 1014x1024

[main rune]

and I'm betting even moreso for any derivatives of pony

[vocal burrow]

no as they're essentialy the same model but tweaked.

[hot trout]

Hi!
I have installed Stable Diffusion multiple times (using Python 3.10.11, and I also tried version 3.10.6) on an AMD system, following the instructions step by step. I downloaded the models from the Civitai website. Every time I encounter the following error messages:

raise EnvironmentError(
OSError: None is not a local folder and is not a valid model identifier listed on 'https://huggingface.co/models'
If this is a private repository, make sure to pass a token having permission to this repo either by logging in with huggingface-cli login or by passing token=<your_token>

SSLError: not enough data: cadata does not contain a certificate (_ssl.c:3989)

How can I resolve these issues so that I can finally use the program?

[ornate elk]

Hey, can you share the full cmd log?

[solemn moon]

@woven zealot have you heard of control net?

[woven zealot]

yUUP

[solemn moon]

Kkkk

The diffusion gods blessed us with CN imo

[muted scarab]

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
how can i update torch and download xformers, it says it can't locate the file

[ornate elk]

Are you on windows or Linux

[muted scarab]

windows

this is in a venv or something, made by typing cmd in folder directory bar in stable diffusion folder

[ornate elk]

Auto1111 will download the latest torch version if you delete the venv folder

Also in the webui-user.bat oyu have to add --xformers
So it will install it too

[muted scarab]

where is the venv folder

[ornate elk]

Inside the stable-diffusion-webui folder

If you didnt installed stable diffusion correctly you can checkout my install guides in the Pinned messages of this channel

[muted scarab]

ok i found it

delete it then what

i did install it correctly but torch and xformers outdated then i deleted xformers as part of a guide to update

[ornate elk]

What's in your webui-user.bat?

And what's your GPU

[ornate elk]

Okay there add --xformers

After the --medvram

Then launch the webui-user.bat

[ornate elk]

You maybe need to update your webui too if you installed it not in the last weeks

For that you can add
Git pull
In the line before Call webui.bat

[muted scarab]

probably should

[ornate elk]

Git pull

[ornate elk]

Oh you already had it in the first line xD

[muted scarab]

gg

[ornate elk]

Then you can run the webui-user.bat

It should download the latest torch stuff + xformers

[muted scarab]

it completed

ty

[ornate elk]

Oh that's old

What does the cmd shows?

Post the full log

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

[muted scarab]

is this going to delete any lora,checkpoints or outputs

well, too late i already did it

[ornate elk]

Nope it wont

The venv is only for the Backend stuff

Torch, xformers and other depencies and librarys

[plain ocean]

Ok ill record the video on how i install stable diffusion and post it here to show what happens

idk why i keep getting some error

One question shall i install sd next?? or a11 forge webui for amd??

*a1111

[ornate elk]

What's your GPU ?

[plain ocean]

rx 6600

[ornate elk]

I would start with A1111 for AMD because if that runs. The rest can be easily installed too

[plain ocean]

ok ill go with a1111

Ill make sure to record the steps which i did

[ornate elk]

Okay

[muted scarab]

https://tenor.com/view/for-you-cat-flower-flower-cat-cute-cat-cat-holding-flower-gif-15053298681215141616

[plain ocean]

Alr got the same error and also recorded the video now ill edit it into one video and send it

[plain ocean]

How shall i share it??

its done

the video is done

ok here it is

https://streamable.com/msdkew

here is the video

@ornate elk

[plain ocean]

You there mate?

[granite bolt]

I'm using Animagine in SDXL and it only generates a slightly previous masterpiece paint job when I put in the ”newest” tag. I'd like to generate a more modern thicker paint job, but how can I do that?

[plain ocean]

@ornate elk ?

[sage scroll]

Hey, i have a Problem with the Replacer Extension. All of the video looks like this (its supposed to change the clothes to a dress)(i added the blue scribble to censor) Anybody know why this happens? (I atteched the settings as well, the rest is kept on Default)

[plain ocean]

IT WORKED LETS GOO!!!!! Thx So Much CS1o :)))

But look at this

Why is the bar already at 95% when i start the gen

[little gate]

Man... I am having such trouble. I switched my webui to include accelerate with openvino, but now my generations are so slow, and they are coming out crappy.

still nobody around to help

?

[plain ocean]

Happens everytime

There is almost no one to help

You just gotta figure it out man

[little gate]

... thats... not a thing... if I could "just figure it out" I wouldnt be asking

lol

[plain ocean]

well my bad

i dont know about openvino but let me see

is it like cuda or something??

Cause i have seen some reddit posts saying its not that fast

i dont anything further than that

[little gate]

No, so I was using xformers and cuda and all that, but I kept getting a memory error because my nvidia card has under 6 gigs of available vram

so I switched to the openvino optimizer thing so I could run it on my intel graphics card, but it's sooooooo slow now

and the outputs are not generating basically anything worth keeping

[main nest]

btw how was I supposted to faceswap

was it face fusion

[ornate elk]

Nice, no problem 🙂

[little gate]

Got time to be a guide for someone else?

[plain ocean]

Did you try forge web ui??

Like you can try that

it uses less vram and you can control the vram too

It might work

[little gate]

I havent

[plain ocean]

Ok here

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides#nvidia-forge-webui

look at this guide from @ornate elk

Try that it is far better than a1111

[plain ocean]

you there?

[little gate]

ok ill give it a go

[plain ocean]

try it 🙂

[little gate]

thank you

[plain ocean]

no problem anytime 🙂

[main nest]

can yall help me installing face fusion

[plain ocean]

What is that?

[main nest]

Face swaping

[plain ocean]

Oh

which gpu?

[main nest]

4070 super

12gb

[plain ocean]

Dang

https://youtu.be/TD57PYBTXqI?si=nTPJQ9f1Tu3m9iem

try this tutorial it might help

[ornate elk]

Sure

[plain ocean]

a bit too late XD

[little gate]

Im gonna try this forge web ui

[plain ocean]

Ok go on, let me know how it worked 🙂

[ornate elk]

That high it/s + the 95% is a visual bug with 1.5 models

[ornate elk]

What's your GPU ?

Openvino needs a different webui version

[little gate]

Intel(R) UHD Graphics

[little gate]

Yep, I downloaded it

[ornate elk]

You said you have a nvidia card too?

[little gate]

or I have NVidia GForce GTC 1660 Ti

[ornate elk]

Yea for sure use that

Dont even try openvino with an Intel UHD xD

Its like running in CPU mode

[little gate]

But the other one, no matter what I do in the editing the webui.bat it always either crashed, or gave me errors

[ornate elk]

We can fix that

[little gate]

also... where do I put LORAS for the forge?

[ornate elk]

Models/lora

[little gate]

it doesnt appear to have a premade folder for that

do I make one?

[ornate elk]

It should get created at the first launch

[little gate]

oh

I just made it

and pasted my loras

[ornate elk]

Shouldnt be a problem then

[little gate]

do I change the .bat to have the --xformers and all that?

[ornate elk]

Not for forge. As it uses the best args already

[little gate]

ok, do I change anything in there?

[ornate elk]

There are some args that could help but I'm not at the PC rn to check

But first try with default

[little gate]

ok

[plain ocean]

Oh thx

[little gate]

stderr: ERROR: Could not install packages due to an OSError: [Errno 2] No such file or directory: 'C:\Users\Valued Customer\Documents\StableDiffusion\stable-diffusion-webui-forge-main\stable-diffusion-webui-forge-main\venv\Lib\site-packages\transformers\models\deprecated\trajectory_transformer\convert_trajectory_transformer_original_pytorch_checkpoint_to_pytorch.py'
HINT: This error might have occurred since this system does not have Windows Long Path support enabled. You can find information on how to enable this at https://pip.pypa.io/warnings/enable-long-paths

Press any key to continue . . .

[little gate]

got the above error

[ornate elk]

You installed forge not the correct way.
You have to download the version from my Guides link and dont drop into the documents folder

[little gate]

ummmm...

[ornate elk]

But the error can also be causes by pythons path length limit

What's your Python Version?

[little gate]

no idea?

[ornate elk]

open up a cmd and run
pyton --version

[little gate]

3.10.6

[ornate elk]

okay then run the python 3.10.6 installer again and at the last window select "disable windows path lenght limit"

[little gate]

how do I do that? I am not a coder/program guy

[ornate elk]

you just have to run the installer again

no need to code

[little gate]

...

this is all code until I get into the web ui

lol

[ornate elk]

okay

then download the python 3.10.11 64bit version from here (at the bottom)
https://www.python.org/downloads/release/python-31011/

then leave everything at default until the end (there select "disable windows path lenght limit"

[little gate]

should I not get the newest python?

[ornate elk]

no its not compatible

[kind sedge]

hello which a1111 version is most compatible with extentions?

[ornate elk]

the latest

1.10.1

if you need really old extension support then 1.7.0

[little gate]

ok done

[kind sedge]

for some reason it gives me an error tryin to install extentions, such as facechain for example

[ornate elk]

which error do you get?

[kind sedge]

i'll screenshot it for you a second

[little gate]

seems to be doing some further installing now

alrighty, we are at least loaded

should I try it now like I had been previously?

[ornate elk]

yes

try a normal default image generation

[kind sedge]

im reinstalling the latest version to reproduce the error

but look at the pip warning idk if that's normal or not

[little gate]

Ok, got one image, it was awful... trying another now

It has bypassed 50%, and the crash hasn't hit me, thats a good sign

well.... it didn't crash...but uh... what the hell is this lol

[kind sedge]

silent hill

[little gate]

lol

The Mist

[ornate elk]

can you show your settings? xD

[little gate]

or do you mean the settings tab?

[ornate elk]

nope

[ornate elk]

your upscale settings are bad

and the cfg

hires upscaler has to be anything (for example RESRGAN4x+), then set hires steps to 10, denois to 0.5
Then set the CFG to 7 and leavte it there

and set the normal sampling steps to 30

[little gate]

I am trying to follow these types of steps

https://civitai.com/models/373778/capcom-cps3-sprite-style

what this guy suggests

[ornate elk]

thats a lora

[little gate]

yes

[kind sedge]

what's ur gpu?

[ornate elk]

he uses these settings:

Steps: 30, CFG scale: 7, Sampler: DPM++ 2M Karras, Seed: 3441237576, VAE: vae-ft-mse-840000-ema-pruned.safetensors, Size: 512x768, Model: monstercoffeebangMix_v11NoVAE, Version: v1.6.0, Model hash: c7a3d16416, Hires upscale: 2, Hires upscaler: None, ADetailer model: face_yolov8n.pt, ADetailer version: 23.9.1, Denoising strength: 0.5, ADetailer mask blur: 4, ADetailer confidence: 0.3, ADetailer dilate/erode: 4, ADetailer inpaint padding: 32, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, Clip skip: 2

[kind sedge]

u ever used facechain on the latest version @ornate elk ?