🤝｜tech-support

[ornate elk]

scroll down on the site

there you see the zip files

[sick elk]

oh nooo excuse me.. 😭

[ornate elk]

np 😄

[obsidian gull]

hey now its saying this

[ornate elk]

run the python 3.10.11 installer again
click "modify" click next, check "add python to environment variables"
then finish and relaunch the webui-user.bat

[obsidian gull]

well now it says this 😔

[ornate elk]

delete the venv folder

then relaunch

[obsidian gull]

thank you! its downloading smth now but i think it worked

you helped me out like 6 months ago aswell lmao

[ornate elk]

oh 😄 np ^^ checkout my install guides for the best performance launch arguments for the webui

[obsidian gull]

where can i check it out?

[ornate elk]

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

[sick elk]

3.2 Go into C:\Program Files\AMD\ROCm\6.1\bin\rocblas folder.
There copy and rename the library folder to old_library
3.3 Open the .zip file and drag and drop all files of the library folder into the library folder. Not into the old_library folder.

it's like, duplicate "library" and rename "old library" the second files. so i extract all datas in "library" (first files) ?

[ornate elk]

correct 👍

[sick elk]

okay thanks

[obsidian gull]

@ornate elk whats the best sampling method that i can use?

its defaulted to DPM++ something

[ornate elk]

DPM++ 2M is good or Euler a
Or DPM++ SDE (but that takes longer)

[drowsy creek]

do the A1111 flags to not require cuda/rocm and just use CPU not work in the "better" forge?

I copied over my command args from A1111

[brittle lantern]

What’s the best way to speed up flux and have good quality (what Lora combos or whatever checkpoint is needed for quick but good gen quality)

[ruby yoke]

have you tried using the ggufs for it?

[brittle lantern]

I would but I rn I only have my 16gb ram MacBook Pro m2 pro and idk what GGUF to go for that will actually not take 20 years to gen

[ruby yoke]

GGUF is a different format, it's smaller, should work better.

[brittle lantern]

What size would be the best for my 16gb of ram?

[ruby yoke]

well - i run flux dev in 16 gig vram, but i'm on windows 11, and on a desktop. you're on mac laptop. i might suggest that you should look into using comfy on one of the run comfy online free websites OR consider renting some GPU time on places like vast.ai

laptops are not really designed for this and comfy doesn't really like macs

[brittle lantern]

I think I have $10 on vast AI I have not used yet but the problem is it takes a while to transfer a model into its storage and idk how much credits that takes Ik storage and the hardware charges different I’m pretty sure

[ruby yoke]

i don't know that either. you could ask them perhaps?

[brittle lantern]

Possibly

[ruby yoke]

if you search google for run comfy online free there are some other options

[brittle lantern]

I’ll try u think with Zluda my 6600 xt with 8Gb vram could use GGUF

Would I have to use Forge for that low vram?

[ruby yoke]

i can't imagine why

just make sure you use the GGUF loader for comfy, not the checkpoint loader

[brittle lantern]

I see I also don’t know how the different GGUF sizes work

[ruby yoke]

start here https://huggingface.co/city96/FLUX.1-dev-gguf

[brittle lantern]

I’ve seen some people run Q8 on 12gb VRAM and some run lower on lower vram

[ruby yoke]

he's got a nice chart on the right side of that page that explains some stuff

keep in mind that the number of bits is how far out in decimal places your numbers will go - and the more decimal places you have, the more precise your math is

[brittle lantern]

Gotcha

I’ll try Q4 on the 6600 xt

Problem of problem of having a Mac and AMD GPU that isn’t on the higher end

😢

[ruby yoke]

i know. i sympathies. it can be very frustrating. but there are silver linings. remember, your mac is an incredible graphics machine. maybe generate online and then use your mac for post production

[brittle lantern]

That is true I don’t have a issue with “heavy” video editing or photo editing

Running blender without any form of “RT” that blender has works quite well

I may not have a M4 chip and a older M2 pro chip but it still works great for what it is

[ruby yoke]

i know. you have an incredible machine. don't try to make it be the only tool, though. remember - if you're a 5 star chef, your kitchen is full of tools for different things

[brittle lantern]

Very true actually

[ruby yoke]

try out the online options to create the content with, and then pull that into your mac, and make some incredible stuff out of it

[brittle lantern]

That sounds like a good idea so is Vast my only good value option

Or is there any other services?

That’s worth checking out

[ruby yoke]

no, there are lots of places to rent gpus. replicate is also an excellent option

just search google for rent gpu. there's also amazon web services - AWS

and other places.

there are even a couple places that'll let you run comfyUI online for free or almost nothing

just have to do a few google searches

[brittle lantern]

Thank you

[ruby yoke]

you're welcome

[ornate elk]

You would need Forge Webui with ZLUDA for the 6600xt and then maybe a Q4 gguf can work

Also the t5 q8 gguf helps in keeping ram usage low

[obsidian gull]

im getting burnt images using a flux. 1 D model

how can i fix it?

what vae should i use for it?

[crystal nacelle]

Hello. Im using ZLUDA on an AMD pc, seeing alot of FLUX Base Models in civitai. 1) I dont want to spend hours upgrading to my stable D to flux, is that okay? Do you recmmend i go thru the upgrade coz its quick and better than my SD1.5?

[rich plover]

Hey all, been a while since I touched SD. Has anyone figured out how to get Automatic1111 (or possibly something better if there is something better now) working with an AMD card on windows?

[brittle lantern]

my gpu driver timed out when doing the first time gen 45 min gen with forge flux 💀

[ornate elk]

Dont run the first time gen with a flux

[ornate elk]

What's your GPU?

[ornate elk]

Hey, yes I made Guides for all common webuis to use with AMD and Zluda as Backend.
Here:
The latest Stable Diffusion install Guides for Nvidia and AMD:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[ornate elk]

AE.safetensor is the official one

[obsidian gull]

Can you send me the link please?

[brittle lantern]

Yh I noticed lmao

[obsidian gull]

i tried finding it but couldnt

[ornate elk]

Here:
https://huggingface.co/black-forest-labs/FLUX.1-schnell/tree/main

[brittle lantern]

@ornate elk doing with illustrations (sdxl) and it just pumps out this

and the gen speed hella slow

[ornate elk]

What's your GPU?

[brittle lantern]

6600 xt running forge

[ornate elk]

Was that the Q4 gguf?

Or which model?

[brittle lantern]

this Illustrations (SDXL) model

[ornate elk]

Oh sry right

What's the resolution?

[brittle lantern]

896x1152

[obsidian gull]

@ornate elkit keeps showing me this error when i try to generate

[rich plover]

tysm

[ornate elk]

Can you try 1024x1024
Sampler Euler a
Disable hires fix
Enable Never OOM for Tiled VAE at the bottom

[brittle lantern]

same cmd output low vram

[obsidian gull]

@ornate elk can you help me out please?

[brittle lantern]

should i go back to normal Auto1111 and leave out flux or try comfy?

[snow breach]

Every time I change checkpoints and generate in forge using any sdxl checkpoint I crash. No error printed, just Press any key to continue. Doesn't happen with SD15 just SDXL, any hints at how I can stop this?

[ornate elk]

Sdxl should work fast normaly.
Make sure you have other programms like Wallpaper engine and games closed

You can try Auto1111 with ZLUDA too

[ornate elk]

How much Ram do you have?

[brittle lantern]

Yh I’ve tried those also idk why it shows the same

[ornate elk]

How much Ram do you have?

[brittle lantern]

16

[ornate elk]

Ah okay then you need to increase the Windows pagefile

That speeds up sdxl performance

[brittle lantern]

I see I remember using Med vram SDXL or whatever that command was on normal Auto1111

And that worked perfectly fine

I’ll assume forge is different and doesn’t support that command

[ornate elk]

Thats true. But it should have set the best options by default

Can you screenshot your forge txt2img settings?

[brittle lantern]

Can do

[brittle lantern]

In the settings menu or the homepage where u gen the images?

heres this @ornate elk

[ornate elk]

Set the schedule type to Automatic also try an other sdxl model for the test

[ornate elk]

Which settings and models did you used?

[rich plover]

Btw, What web ui's do you suggest? I have not touched SD in probably 2 years.

[gentle cipher]

This afternoon the server experienced an issue. Could you please tell me the reason for the service disruption and when it will be restored? @ornate elk

[ornate elk]

What's your GPU?

[gentle cipher]

We have been using your company's text to video API, but the service has been unavailable since this afternoon. Could you please let us know when it will be restored? Please reply as soon as you see this message, and you can send the response to the email [e-mail]. @ornate elk

[rich plover]

6800xt

[ornate elk]

Then I suggest starting with Auto1111 or Forge Webui with ZLUDA

[gentle cipher]

@ornate elk please reply me

[rich plover]

alr thank you!

[gentle cipher]

@ornate elk Are you there?

We are waiting for you

[ornate elk]

I'm a discord community guide and not working at stability ai.
Please use their official support site:
https://stabilityplatform.freshdesk.com/support/home

[ornate elk]

NP, if any step isnt clear let me know

[gentle cipher]

Are your servers functioning normally?

[ornate elk]

Idk I'm not using the API or any Server.

[ornate elk]

Looks like there are issues:
https://stabilityai.instatus.com/

Routing issues but they are working on it

[gentle cipher]

Thanks, when can they fix it?

[ornate elk]

Nobody knows how long it will take

[obsidian gull]

I didn’t tinker with the settings but I used a realistic images model, it base model was flux.1 D something and it said checkpoint trained

Got it from civitai

[ornate elk]

@brittle lantern You can also send a full cmd log to check if its actually using Zluda.

[ornate elk]

Are you using Auto1111 or Forge Webui ?

[obsidian gull]

Auto1111

[ornate elk]

Auto1111 doesnt support Flux rn

[obsidian gull]

Oh, so what should I use if I wanna use flux?

Forge?

[ornate elk]

Forge Webui

[obsidian gull]

Alr I’ll try it later

[wicked widget]

Community guides ARENT employees, be nice to them

doing this out of their free time

[brittle lantern]

I’m currently not at my pc it’s 10:20 am and have not slept all night is it ok I take a nap then I can show u?

[ornate elk]

Sure np

[brittle lantern]

Thank you also so I don’t waste time when I wake up can u tell me quickly where the CMD logs are stored

[ornate elk]

You need to launch forge and then do one image generation.
Then copy the whole cmd log out of the cmd. There is no log file

[brittle lantern]

Ah ok thanks

[snow breach]

4080 Super, 16GB.

[unborn mauve]

is it normal for 3080 to struggle with flux and Pony XL?

[ornate elk]

I mean RAM

[ornate elk]

With flux yes with pony No

[unborn mauve]

hmmm oka

[ornate elk]

For pony make sure you have --medvram-sdxl --xformers in the webui-user.bat

Thats for Auto1111

Forge doesn't have these argument's

[unborn mauve]

oh

then what does forge have?

I read on the forge page that the cmd line prompts are the same as A1111 and I expect them to since it's just like A1111 in UI and file structure

[ornate elk]

Forge deactivated the cmd line args

And uses preset performance settings

[unborn mauve]

waaat

then what do i do

or is it all good by default?

[ornate elk]

Its good by default

[unborn mauve]

neat

[ornate elk]

But you may be better of using Auto1111 if sdxl is still slow

[unborn mauve]

nah it was alright at first, worsened after sometime

maybe there is garbage collection happening in vram or something

[ornate elk]

If you have only 16gb of ram, increasing the windoes pagefile helps

[unborn mauve]

I'll try, thanks

[snow breach]

64gb

[unborn mauve]

where do pony models go

[ornate elk]

Xl

Models/stable-diffusion

[ornate elk]

Okay then run the update.bat in the forge folder and then delete the venv folder.
Then also make sure your Windows Pagefile is set to C drive and enabled. And not for any other drive.

[unborn mauve]

dude something is wrong

my pc's lagging way too much

also is this normal? freeing memory after every generation?

and after I enabled hires fix, that was a sin in the eyes of stable diffusion gods

They have thoroughly smited my pc, they made it potato for a solid 10-12 minutes

[ornate elk]

You need to increase the windows pagefile

[unborn mauve]

to?

[ornate elk]

16000min and 24000max for C

And disabled for all other drives

Also when using hires fix set the hires steps to 10 or your PC will lag and freeze xD

[unborn mauve]

apparently 20 is working smoothly now

but

[unborn mauve]

this for viirtual memory?

isn't normal pagefile size like 2MB?

[ornate elk]

Nope

[unborn mauve]

I see

[ornate elk]

Normaly windows allocates how much it needs (around 16gb)

[unborn mauve]

but again, it's not lagging now

it seems to throw a fit only when switching models

[ornate elk]

Yea because switching models requires RAM

[unborn mauve]

hmmmm

[ornate elk]

Generating requires GPU vram

[unborn mauve]

alright

[iron lantern]

whats a good prompt for tattoos on the breast or arm for SD ???

[deft hemlock]

bruh

this aint even the right server for this stuff

[iron lantern]

Ohh where do I go

[deft hemlock]

unstable diffusion

the opposite of here, they have a lot of evil things

like evil-tech-support

evil-rules-and-tos

evil-announcemnts

https://tenor.com/view/eggman-announcement-speech-sonic-the-hedgehog-report-gif-16600754

anyways

ive tried installing forge

since a11111 was extremely slow

now i messed up

[deft hemlock]

this is my bat file

[ornate elk]

What's your GPU?
Please don't use --skip-torch-cuda-test

[modest heron]

Hello, I used to be able to run forge, now, whenever I try to run it, it shows me this error message, and I’ve tried all the versions available even “the old forge” and it still doesn’t work, I was running 2.3.1 just fine in October

[ornate elk]

Hey, What does your 2.3.1 version shows?

[modest heron]

It shows the same error message, “runtime error your device does not support this version of torch/CUDA

[amber rock]

hey people 😛
I'm looking for some advices in how to use Comfy UI... I managed to create a workflow, using my way for ComfyUI, it's working, generated some cool results, great platform and modularity. However, when trying to use controlnet, just like I used with A1111, the result is not coming out the way I expected.
Here is a print of the workflow:

However, even with a simple prompt, in this case "girl", apparently the control is not acting and controlling the diffusion using the Canny method, what am I doing wrong? Thanks in adv...

[ornate elk]

then update your GPU driver and then delete the venv folder and relaunch

[amber rock]

If this is not the right place to ask this type of question, can you point me to a place where I can find answers to this and other questions?

[ornate elk]

hey, you can try ask in #🧣｜comfy-ui too

[meager quiver]

Ok not sure if this is the right place to ask this but I get this when trying to run forge, did the config file got corrupted or something?

[ornate elk]

Could be so yes

Maybe try a reinstall

The version with torch 2.3.1 is recommended

[deft hemlock]

Mobile 2050 with 4gb vram

If I don't use skip torch test I get an error that I have no idea how to fix

[uncut jewel]

cannot import name 'AutoencoderKL' from 'diffusers' (C:\ComfyUI_windows_portable\ComfyUI\custom_nodes\diffusers\__init__.py) is:issue

Hey!
Does anyone knows this issue ?

It was working pretty fine before, maye because i updated Comfy and stuff, it changed some dependencies with python

Just trying to run Kohya on Comfyui

[ornate elk]

please remove all args you used, then delete the venv folder
also make sure you downloaded the 2.3.1 torch forge version

then it will work. Also make sure your gpu driver is updated

[ornate elk]

dont.
Kohya is a heavy tool only made for training.
Its not recommended to integrate it in comfyui
Its a seperate tool and should be used as such

[uncut jewel]

Yeah i know, but Kijai made a nice integration on Comfy, and it was working fine until now 🥲

[ornate elk]

okay, make sure you have Python 3.10.11 installed and not 3.10.6

[uncut jewel]

I'm even using Python 3.12.7 lol

[ornate elk]

yea thats not recommended for kohya or any webui

3.10.11 is the most compatible version you can use (kohya lists it as requirement)

comfyui works with 3.11 too

[uncut jewel]

I see, i see, Thanks for the advices man! 🙂

[ornate elk]

no problem 🙂

[obsidian gull]

@ornate elk

https://github.com/lllyasviel/stable-diffusion-webui-forge

is this the one

alr

[obsidian gull]

@ornate elk i got forge and i tried to generate a image but now its showing this

[ornate elk]

which model did you tried?

flux needs clip l and t5 text encoders + vae

try an sdxl model and it will work

[obsidian gull]

im trying to use a flux model

[ornate elk]

then you need the clip text encoders and the vae

[obsidian gull]

i have the ae.safetensors vae but what are clip text encoders?

[ornate elk]

files needed for flux to work

here get the ae.safetensor vae file:
https://huggingface.co/black-forest-labs/FLUX.1-schnell/tree/main

here get the t5 q8 text encoder:
https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main

and clip l
https://huggingface.co/comfyanonymous/flux_text_encoders/tree/main

[obsidian gull]

there are a bunch of files, which one do i need to get?

[ornate elk]

the ones i mentioned

[obsidian gull]

oh okay

[obsidian gull]

where do i put the text encoder and clip?

[desert vector]

I'm using WebUi Forge, just got a new preprocessor, where do I put it? I know where to put the model already, but just not sure about the preprocessor.

GOt the Anyline preprocessor to be exact

[ornate elk]

Into models/text-encoders

Or models/clip

[obsidian gull]

well i put everything where its suppose to be and it just gets stuck at the last line then crashes

is it possibly an issue with my gpu?

not enough vram?

[ornate elk]

Did you used controlnet?

Try without first

[obsidian gull]

i didnt really tinker with anything, how can i stop using controlnet?

[ornate elk]

Just make sure its closed

What was your GPU again?
You also may need to use a smaller flux model

[obsidian gull]

rtx 4060

[ornate elk]

With 8gb vram?

[obsidian gull]

yep

[ornate elk]

Then you need the nf4 or Q4 flux model

Fp8 and q8 are to large

[obsidian gull]

oh i see

alright thanks

[ornate elk]

Np

[rich plover]

Hey @ornate elk , sorry for the ping. I'm getting an error after following the steps.

'"D:\SD-Zluda\stable-diffusion-webui-amdgpu\venv\Scripts\activate.bat"' is not recognized as an internal or external command,
operable program or batch file.

and then im also getting

Installing torch and torchvision
D:\SD-Zluda\stable-diffusion-webui-amdgpu\venv\Scripts\python.exe: No module named pip
Traceback (most recent call last):
File "D:\SD-Zluda\stable-diffusion-webui-amdgpu\launch.py", line 48, in <module>
main()
File "D:\SD-Zluda\stable-diffusion-webui-amdgpu\launch.py", line 39, in main
prepare_environment()
File "D:\SD-Zluda\stable-diffusion-webui-amdgpu\modules\launch_utils.py", line 603, in prepare_environment
run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True)
File "D:\SD-Zluda\stable-diffusion-webui-amdgpu\modules\launch_utils.py", line 120, in run
raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't install torch.
Command: "D:\SD-Zluda\stable-diffusion-webui-amdgpu\venv\Scripts\python.exe" -m pip install torch==2.3.1 torchvision --index-url https://download.pytorch.org/whl/cu118
Error code: 1

[rich plover]

Nevermind apparantly it didnt install right

[marble folio]

Anyone know of a reliable way of modelling yourself in Forge? - I've tried a couple of fluxgym methods but they hung and didn't complete. A similar method worked for me ages ago in SD 1.5 and is still working, but 1.5 isn't as good as Forge from prompts imo, so I'd rather apply it to Forge too.

[rich plover]

Im getting this error and have absolutely no idea why. The image generates, then it disappears, and the error is there

it thinks im using xformers it seems

and i cant figure out why

[ornate elk]

Can you share your full cmd log?

[rich plover]

I’ve just moved to forge web ui and it works fine

[sick elk]

How can i upgrade quality ?

[wicked widget]

image qual?
1.5 is best using 512x512 res, XL 1024x1024, Flux and 3.5 1440x1440

[vocal burrow]

flux and sd3.5 are also trained for 1024x1024, not 1440x1440

[sick elk]

Thanks I will try that’s

[wicked widget]

ah ty for the correction

i got my info here
https://www.reddit.com/r/StableDiffusion/comments/1enxdga/flux_recommended_resolutions_from_01_to_20/

[uncut jewel]

Heyo, sorry to bother, i was running a workflow, then suddenly it's stopped to work. I literally just refreshed the page the load a new model and now it's not working, while the model that i upload in my comfy as nothing to do with the error. It's an error on the Ksampler ?!

[modest heron]

Hello, I would need some help. I used to be able to generate plenty of images, but now, using the same model and the same Lora, all I get are these kind of pictures:

[wicked widget]

check ur vae

[modest heron]

Sorry I am a beginner with all that stuff, I checked the VAE folder, what am I supposed to look for though?

[ornate elk]

Go into the Settings, search for tiling
And disable it, then hit apply

[modest heron]

What if tiling wasn’t activated?

[ornate elk]

Thats good.

Then show your txt2img settings

[modest heron]

here

[ornate elk]

Ah okay you need to use an other pony model

This version is bad.
And you need to use the pony specific prompt words

[modest heron]

What pony model would you recommend?

[topaz ocean]

Forge UI omits ControlNet Scripts. Can they be installed? ...ControlNet in Forge UI preloads the extension. By comparison, in A1111 ControlNet needs to be installed. And, when creating a Script in A1111, say X/Y/Z Plot, you can choose an "X type" from this ControlNet list (attached). However, in the Forge UI interface, these Script items are not to be found! I assume that there is an extension that puts these items in Forge UI. Do you know which of the ControlNet extensions would do that, or some other way?

[deft hemlock]

same error

File "F:\ai\venv\lib\site-packages\gradio_simple_templates\simpledropdown.py", line 6, in <module>
from gradio.components.base import Component, FormComponent
File "F:\ai\venv\lib\site-packages\gradio\components_init_.py", line 1, in <module>
from gradio.components.annotated_image import AnnotatedImage
File "F:\ai\venv\lib\site-packages\gradio\components\annotated_image.py", line 13, in <module>
from gradio import processing_utils, utils
File "F:\ai\venv\lib\site-packages\gradio\processing_utils.py", line 22, in <module>
from gradio import utils, wasm_utils
File "F:\ai\venv\lib\site-packages\gradio\utils.py", line 56, in <module>
from gradio.data_classes import BlocksConfigDict, FileData
File "F:\ai\venv\lib\site-packages\gradio\data_classes.py", line 16, in <module>
from pydantic import BaseModel, RootModel, ValidationError
ImportError: cannot import name 'RootModel' from 'pydantic' (F:\ai\venv\lib\site-packages\pydantic_init_.cp310-win_amd64.pyd)

[civic bramble]

Can somebody tell me on civitai, what are these used for and I just noticed I got more added. THanks for the help

[ornate elk]

Try install the webui like stated in my install guides from the pinned messages.

[deft hemlock]

but the thing is i dont a1111 due to how much it slows my entire pc when it reaches 95% and 99%

i installed forge because people told it me it wont slow my pc

will it still work using the same steps as a111?

[spark pond]

Came back to SD just now after a 6 months pause and somethings off.

New drives so completely new install, but it just isn't the same anymore. It is disgustingly slow and I don't know why.

[wicked widget]

buzz, 500 u can train a lora. try it on their site

or save up and get profile cosmetics

its all just civit ai currency stuff

[civic bramble]

ok thanks @wicked widget I want to try in oneday...

😄

[wicked widget]

np

doing a lora on civit is brain dead easy

its made for beginners /anyone make an decent lora

[civic bramble]

I never done it before...so thats how I need to learn...is there a tutorial u know I can watch to help me learn?

[spark pond]

Like it's a normal 960x540 pic, 35 sampling steps, just some normal prompts, running on noobauXLNAIXL checkpoint and Anything VAE and it takes more that 10 min for one image?

Something seems oddly off and I don't think I installed anything differently this time.

[wicked widget]

https://youtu.be/70H03cv57-o?si=hhvkgEMXYlvyPHDA

[wicked widget]

hdd or sdd

[spark pond]

main m.2

[wicked widget]

wild

[spark pond]

had it on my secondary m.2 last time but running it on my main one right now.

The picture doesnt even finish

[wicked widget]

hmmmm.

[spark pond]

need a screen?

[wicked widget]

try

[spark pond]

maybe I'm missing something?

[wicked widget]

try it on a different drive, pref ur main one

see if it does the same

if not

[spark pond]

I am on my main drive right now

[wicked widget]

ah well

there goes my idsa

[spark pond]

git, python and sd all installed on my main m.2

[wicked widget]

did u add python to path

when installing

[spark pond]

i did not

[wicked widget]

it has a lil checky

reinstall

do that

reboot

retry

[spark pond]

reinstall which of them?

[wicked widget]

python needs path

prolly why its not working tbh

[spark pond]

so reinstall python

[wicked widget]

ye and make sure you add to path

[spark pond]

i remembered seeing something of a missing path

[wicked widget]

if you still get errors paste the console in here

[spark pond]

I can't remember how to add it to path, sorry, too late for thinking on my own

[wicked widget]

https://phoenixnap.com/kb/add-python-to-path

[spark pond]

i saw the little checky on the first page ahahaha

should i disable the path length limit?

[wicked widget]

yes

its good practice to reboot pc when doing system changes like adding stuff to path afterwards

[spark pond]

yea yea, will do, thanks for reminding

[wicked widget]

np

[spark pond]

faster than before, but still way too slow

and I don't think this is "a tall girl"

[wicked widget]

do you have a vae

which mdoel

are you also using

[spark pond]

yes, have been using anythingModelVAEV40_v10

[wicked widget]

idek what that is lol

which model

[spark pond]

used noobaiXLNAIXL checkpoint xD

[wicked widget]

ahh ya, is that an xl vae? and loaded

[spark pond]

the checkpoint is an XL

and that is causing the problem

[wicked widget]

is the vae for an xl model

cos if not

poorf

poof

try it with just standl sdxl vae

[spark pond]

even without the VAE it runs horrible

[wicked widget]

and if thats a vpred model youll need to use advance sampling or w,e, i dont rly use it so i dont know much

[spark pond]

going back to the standard v1.5 sd checkpoint, it takes 2 seconds

the checkpoint seems to be the main problem here

[wicked widget]

that can happen yeah

[spark pond]

also with None VAE I got my green goo xD

[wicked widget]

well.. yea

you need a vae

[spark pond]

let me just grab some stuff I've used back then

connections not the best so it takes a minute, please hold the line

can a VAE be a safetensor file? it can, right?

[wicked widget]

yup

[spark pond]

doesn't have to be .pt. safetensor should be better, right

back in business, should have been the XL checkpoint causing me trouble. Looking further into it when I have the time.

Still thanks for all the support ^-^

[spark pond]

u are lovely

maybe the VAE isn't the one xD

or maybe I'm wrong, should this happen at the end of every image? XD

only happens on that VAE xD

restarting didn't fix that xDDD

i"ll find my way through it again, have a good one ^-^

[ruby yoke]

that VAE isn't written for the model you're using it with

[ornate elk]

checkout my install link for forge

[ornate elk]

whats your gpu?

remove --opt-sdp-attention --precision full --no-half

[deft hemlock]

the pins say a1111 for nvedia and amd

[ornate elk]

the first link

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

[deft hemlock]

can confirm

i am a stupid idividual

[marble siren]

Hello, i need help with face sawp in A1111

[sonic iron]

I was using the node manager in comfy to install a custom node and suddenly I'm getting this when I go to start it back up

Traceback (most recent call last):
  File "E:\AI Art\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\main.py", line 132, in <module>
    import execution
  File "E:\AI Art\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\execution.py", line 13, in <module>
    import nodes
  File "E:\AI Art\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\nodes.py", line 22, in <module>
    import comfy.diffusers_load
  File "E:\AI Art\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\diffusers_load.py", line 3, in <module>
    import comfy.sd
  File "E:\AI Art\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 6, in <module>
    from comfy import model_management
  File "E:\AI Art\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 145, in <module>
    total_vram = get_total_memory(get_torch_device()) / (1024 * 1024)
                                  ^^^^^^^^^^^^^^^^^^
  File "E:\AI Art\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 114, in get_torch_device
    return torch.device(torch.cuda.current_device())
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "E:\AI Art\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\cuda\__init__.py", line 940, in current_device
    _lazy_init()
  File "E:\AI Art\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\cuda\__init__.py", line 310, in _lazy_init
    raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

E:\AI Art\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable>pause
Press any key to continue . . .```

[sonic iron]

reinstalling comfyui fixed it

[marble siren]

how to get a better deep fake? i have some little pixels under face just a little it's pretty close to the perfection

[deft hemlock]

F:\ai\reforge>git clone https://github.com/Panchovix/stable-diffusion-webui-reForge.git
Cloning into 'stable-diffusion-webui-reForge'...
fatal: unable to access 'https://github.com/Panchovix/stable-diffusion-webui-reForge.git/': error setting certificate file: E:/ai/Git/mingw64/etc/ssl/certs/ca-bundle.crt

i get this in the cmd part

i have an E drive but its the system backup which is 13mb, i checked with hidden hiddons enabled and there isnt a single file in there

i double clicked and gave it permission

crt file asking for human rights at this point

i told it to git config --global http.sslverify "false" itself, i think it listened

[fair oxide]

I want the link invite to add the bot to my server

[spark pond]

using a RTX 4070 ti

[spark pond]

did that for now, but currently no time for testing, work is calling me in xD

[deft hemlock]

I FIXED IT

now i have a new issue

my loras dont appear

[ornate elk]

would suggest trying to get normal Forge work first

Loras only show up when a compatible model is selected

[ornate elk]

hey, what do you use for faceswap?

[deft hemlock]

i have noobxl selected but i cant see any of my loras

[ornate elk]

are they illustrous loras?

[deft hemlock]

mix of sdxl pony and illustrous

i seperated them by folder

like how a1111 has buttons u can seperate ur loras based upon

[ornate elk]

yep

make sure the webui is whitelisted in any browser adblocker

they hide loras sometimes

also try change the model so sdxl and refresh the lora page

[deft hemlock]

still nothing

[deft hemlock]

base sdxl?

[ornate elk]

any sdxl

[deft hemlock]

do u think its this?

[deft hemlock]

noobxl is selected

[ornate elk]

nope

i would just use forge

and not reforge, it has some bugs

[deft hemlock]

reforge had the same thing

i installed both reforge and forge

both had the same issues

[ornate elk]

ah okay

[deft hemlock]

the checkpoints appeared normally but the loras didnt

[ornate elk]

then we focus on forge first

restart it

and then select and other sdxl model

or pony one

[deft hemlock]

i did for forge

it had 4 buttons ontop

i chose sdxl and the loras didnt appear

[ornate elk]

and its whitelisted in the adblocker?

and you restarted the browser too?

[deft hemlock]

whitelisted

[deft hemlock]

yes

[ornate elk]

do they show up in any other webui?

can you share me the full cmd log of forge?

[deft hemlock]

they showed up in a1111

i can just see the checkpoints

nothing in loras

idk if this helps

i moved sdxl realistic vision and it cant see that in checkpoints

it can only see noob

[ornate elk]

your folder paths are wrong in the webui-ui user.bat

[deft hemlock]

wait no forgest this part

i just moved the wrong file

but the loras is correct

[deft hemlock]

what should it be?

[ornate elk]

show me the wbeui-user.bat please

there are also a lot of unneeded commandline args that forge cant use

[ornate elk]

your path to A1111 needs to use / instead of \

[deft hemlock]

i see

ok lemme change it

[ornate elk]

and remove --xformers --lowvram --no-half-vae

forge already uses the best settings

[deft hemlock]

should i change all \ to / ?

[deft hemlock]

hell yeah

[ornate elk]

only here:

[deft hemlock]

done

ill restart

while its restarting

thanks for putting up with my idioticy

damn i got an error now

[ornate elk]

delete the venv folder and relaunch

[deft hemlock]

oh yeah forgor about it

where was it again

i forgot

[ornate elk]

in the stable-diffusion-webui-reforge foulder hopefully

[deft hemlock]

no venv folder there

[ornate elk]

okay strange

[deft hemlock]

very

[ornate elk]

then add back --xformers to the webui-user.bat and relaunch

[deft hemlock]

okay

[deft hemlock]

still no lora

:( i get no loras

[ornate elk]

Then add a "" around the webui path

"Path to Auto1111"

[deft hemlock]

like this?

[ornate elk]

Your lora dir folder is incorrect too

It has to be %A1111_HOME%\models\lora

[deft hemlock]

i see

okay

but the "" are correct right?

should i include that arrow thing in the end?

[ornate elk]

Nope

You can also just copy the perfect example from my forge guide and only change the path to Auto1111

[deft hemlock]

okay

but the "" are correct right?

[ornate elk]

Yes

The path at the top is correct and the only thing you have to edit when copy the example

[deft hemlock]

i see

found the venv folder

should i delete it?

[ornate elk]

Not needed for now

[deft hemlock]

this is all correct

should be

https://tenor.com/view/hot-ones-venom-venom-hot-ones-venom-milk-hot-ones-venom-milk-gif-12373438466562356314

[ornate elk]

Looks good

[deft hemlock]

how did i mess up so badly that file explorer crashed

[ornate elk]

Wtf xD

[ornate elk]

You didnt copyied my example

Now I see it. All the \ are wrong

[deft hemlock]

damn

[ornate elk]

Please just copy my example from the guide

And replace the whole content of the webui-user.bat with it

Then only change the one path at the top

And everything should work

[deft hemlock]

alright

shoyld be good now

it gave me some errors regarding venov, so i deleted it

[ornate elk]

You may need the " " again for the path

Because the path contains a folder with a . in the name

[deft hemlock]

in the home

i added it and it gave me an error

i tried "F:/ai/sd.webui/webui"

it said " "F couldnt be found

so i added a space and the cmd opened and closed immediately

[ornate elk]

Strange

And without the " " ?

[deft hemlock]

i think it worked

im testing it out

sorry csio for wasting ur time

[ornate elk]

No worries

[vast crest]

hello guys

In your opinion, why does webui1111 crash? Even if I don't use it, after a while, this is what happens

[ornate elk]

It says out of memory

What does the CMD shows?

[ornate elk]

Okay so the webui is still running just the browser tab crashed

Try an other browser as chrome needs a lot of RAM

[vast crest]

I first reset the chrome settings, then since it continued to crash, I decided to uninstall it. Once reinstalled, however, it continued to crash 😦

Chrome has never given me any problems 😦

[ornate elk]

which text2img setting do you use and whats your gpu then?

when it crashes

[vast crest]

640 x 960 - dpm 2m sed karras gpu 4070 oc

[fierce moat]

Izzit weird that I ran LTX 0.91 faster than LTX 0.90

[ornate elk]

with controlnet and hires fix enabled or just default ?

[vast crest]

I only use default

[ornate elk]

Hmm okay, what's in your webui-user.bat?

[vast crest]

this

git pull

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --xformers --no-half-vae --autolaunch

call webui.bat

[ornate elk]

Looks okay

Then you should try if it happens with other browsers like Firefox too

[vast crest]

ok

[marble siren]

im using ReActor but as well i have roop

[ornate elk]

Roop is outdated and breaks the webui.

[dapper timber]

hello

having trouble installing stable diffusion

[kindred crater]

I've installed ComfyUI on WSL ubuntu 22.04, no errors on installation.
rocminfo returns my gpu
however, when doing python main.py it's erroring to no hip devices. Any ideas? :|

acrivec@MainPC-Cooled:~/ComfyUI$ python main.py
Traceback (most recent call last):
  File "/home/acrivec/ComfyUI/main.py", line 132, in <module>
    import execution
  File "/home/acrivec/ComfyUI/execution.py", line 13, in <module>
    import nodes
  File "/home/acrivec/ComfyUI/nodes.py", line 22, in <module>
    import comfy.diffusers_load
  File "/home/acrivec/ComfyUI/comfy/diffusers_load.py", line 3, in <module>
    import comfy.sd
  File "/home/acrivec/ComfyUI/comfy/sd.py", line 6, in <module>
    from comfy import model_management
  File "/home/acrivec/ComfyUI/comfy/model_management.py", line 145, in <module>
    total_vram = get_total_memory(get_torch_device()) / (1024 * 1024)
  File "/home/acrivec/ComfyUI/comfy/model_management.py", line 114, in get_torch_device
    return torch.device(torch.cuda.current_device())
  File "/home/acrivec/.local/lib/python3.10/site-packages/torch/cuda/__init__.py", line 940, in current_device
    _lazy_init()
  File "/home/acrivec/.local/lib/python3.10/site-packages/torch/cuda/__init__.py", line 319, in _lazy_init
    torch._C._cuda_init()
RuntimeError: No HIP GPUs are available

[ornate elk]

Hey, checkout the first link of the pinned messages.
There are all install guides.

[dapper timber]

yo thanks i need some help with troubleshooting atm

cant clone the git file

[small shell]

guys, im using stable diffusion, more precisely forge through stability matrix and my adetailer is not working, it pastes a square on top of the image face. how can i fix this? it also started happening just yesterday, think after last update.

[undone locust]

hi guys , i need some help to convert photo to vector (coloring book style) for engraving with my laser
First photo, style i want
Second , it's my real photo
third : I try controlNet Canny with prompt, but it's to many details
Last photo same prompt without controlNet looks ready for vector studio

Thank 🙂

[fierce moat]

https://civitai.com/models/880102/line-drawing-ce-flux

[undone locust]

It's flux ? i can use ControlNet to copy my real photo with this ?

[fierce moat]

I think image to image is hard for this one, flux have control net tho, but I am not sure it whether helps

[undone locust]

Hmm ok thank 🙂
I think the result with my prompt and controlnet is good but just to litle detail , maybe with photoshop i can do the job but i asking here before if some one have magic trick haha

[fierce moat]

i found lora for sdxl too: https://civitai.com/models/136348/coloringbookredmond-coloring-book-lora-for-sd-xl

[ruby knot]

Hi does anyone know how to disable saving the prompt to image file name on automatic1111? I can not seem to workout how to do it or find any info on how to do it.

[marble siren]

so you recommend to delate that extension?

[undone locust]

i will try ty

I have found something with controlnet and prompt , looks good 🙂 (vector svg)

From this

[ornate elk]

Yep and then the venv folder.
That fixes the webui

[ornate elk]

Do you get an error

[ornate elk]

In the Settings of Auto1111 at the first few settings should be image name pattern

[marble siren]

should i have also choose one of option in box where it is written "None"? for deep faking?

[ornate elk]

Is that in Reactor?

[marble siren]

yes

[ornate elk]

Then nope. Not needed, try whithout first

[ruby knot]

I’ll check that thanks

[ruby knot]

I dont see it? Can you send a screenshot with it highlighted please?

[ornate elk]

here:

datetime is the best as you cant get duplicate file names its just date and time. and they are sorted automaticaly

[ruby knot]

It’s blank atm, also I forgot to mention for some reason it also isn’t saving preview grids when I tell it to do a batch of images.

[ornate elk]

grid options is beneath the options of my screenshot

[ornate elk]

blank means its using default (prompt)

[ruby knot]

I did change those settings and their still not showing when it shows the images

[ornate elk]

make sure the webui is whitelisted in any browser adblocker

these are my grid settings:

dont forget to hit apply

[ruby knot]

I haven’t had that issue before on my laptop and I use the same browser and ad blocker

[ornate elk]

browsers and adblockers update

so the issue can occur randomly

[ruby knot]

Hmm maybe didn’t think of that the issue being due to a random update I mean

[ornate elk]

better whitelist the wbeui and your fine

we had missing loras, missing img2img previews and stuff like that because of adblockers

[ruby knot]

Ah is that why I have to edit lora names sometimes or they don’t show up

[ornate elk]

could be related yes

[ruby knot]

Tried add the ip adress for a1111 and it won’t let me

[ornate elk]

which adblocker do you use? 😮

[ruby knot]

Ad block plus like I always do

[ruby knot]

I’m not sure which other ones are even safe to use so I’ve always used that one

[ornate elk]

ah okay, wouldnt recommand that adblocker because they get paid by advertisement companys to sneak trough their block list.
Better use ublock origin, its open source and the best adblocker

[ruby knot]

Does it value privacy tho ? I don’t want my info and payment details etc stealing

[marble siren]

what to do if on my face swap ears are not looking real? i already tick options" "face mask correction, soft inpainting, denoising strenght on 0/0.11" i as well add some prompt to img2img. what i could do else to correct face swaps?

[ornate elk]

yea its more private than adblock plus xD

[marble siren]

hyper-realistic, photo-realistic, ultra detailed skin, perfect anatomy, symmetrical features, soft natural lighting, high-resolution 8K, HDR, depth of field (DOF), texture detail (e.g., pores, imperfections, freckles) - postivie prompts

[marble siren]

deformed features, bad anatomy, blurred skin, low texture resolution, asymmetrical face, cartoonish effects - negative prompts

[ornate elk]

@ruby knot it also got checked by firefox and trusted by milions, + its opensource

[ruby knot]

Sounds decent, I use Firefox browser anyway

[ornate elk]

then its the best adblocker you can use with it, librewolf and other firefox forks native ship with ublock

im on firefox + ublock too

in the whitelist i have added: http://127.0.0.1:7860/ for auto1111

[ruby knot]

I wonder if Firefox vpn is decent?

[ornate elk]

they are using Mullvads VPN in the backend and just put their logo on it. so yes its on of the most private ones you can use

[ruby knot]

I think I’ve heard of that

[ornate elk]

swedish company with "no log" policy and you can even send them money in an envelope to pay completely anonym xD

[marble siren]

?

[ruby knot]

Damn

[ornate elk]

thats a question for #📝｜prompting-help
there you also can share an example maybe

[ornate elk]

let me know if the whitelisting in ublock fixed the issues

if the grids still dont save you can also reset all auto1111 settings by deleting the config.json and ui-config.json

[ruby knot]

I will do I just got ublock now

[ruby knot]

The config file is useful it will save me ages putting the same default settings in every time I load a1111

[ruby knot]

Still not showing grids

[ornate elk]

strange maybe then try the full reset by deleting the two configs

[ruby knot]

Ok it is saving some grids but not showing them in the previews like the other images

[ornate elk]

so you cant see them in the browser but they get saved?

[ruby knot]

Yeah some of them did not many though

[ornate elk]

strange

is your webui updated?

andy errors in the cmd log?

also do you use the x/y/z script or does it happen when using batch count or size?

[ruby knot]

I don’t know I use a portable mod someone made

[ruby knot]

I use whatever the default is

[ornate elk]

hmm okay

a version "someone" made doesnt sound that good xD

because auto111 does provide a portable mod aswell

[ruby knot]

I forget the name

[ruby knot]

where?

[ornate elk]

Here:
https://github.com/AUTOMATIC1111/stable-diffusion-webui/releases/tag/v1.0.0-pre
download the sd.webui.zip
Extract it and run the update.bat
then the run.bat to launch

[ruby knot]

The one I originally used was an easy install one and I can keep a backup of it all on a different HD for if an extension breaks the install but if I try use that back up on my pc too it doesn’t work.

[ornate elk]

if you use the version from above you can easily fix any issue by deleting the problematic extension + the venv folder inside the webui folder and then its resetet

[ruby knot]

But with that one after installing it could I copy the whole install folder etc to another computer and it just work because it doesn’t normally due to gpu being different.

[ruby knot]

I tried that one the ones I use but it doesn’t always fix it

[ornate elk]

you can do that but you need to delete the venv folder before the first launch again

[ruby knot]

Oh wait what’s venv folder?

[ornate elk]

thats the folder most webuis have that stores core data of the programm. so if a extension messes something up and you delete the extensions the venv is still broken. Until you delete the venv and it gets recreated clean

[ruby knot]

I also split mine into 3 main folders with a different portable copy of Firefox for each one. you know SD1.5 SDXL & PDXL

Due to loras being a mess in the ui

[ornate elk]

why that?

you can just create subfolders in the lora folder to sort them 😮

[ruby knot]

Because of all the Loras and models being confuseing to keep track of what works with what, the gui is bad for that imo

[ornate elk]

the folders really help there imo

[ruby knot]

I have six million loras and checkpoints

I carnt even see them when it’s showing me the folders on the gui due to having so many

Nearly 2tb worth

[ornate elk]

wow

then true maybe splitting it up is a good idea

[ruby knot]

I like playing around with them and seeing what’s decent and what’s not so I have a lot.

[ruby knot]

You wouldn’t happen to know where I could find a decent tutorial on how to merge checkpoints and loras into checkpoints would you? I’m interested to have a go at messing around with that too,

[ruby knot]

Also thanks for the help

[dapper timber]

yeah

i tried so many ways

[ornate elk]

checkpoints can be merged in the normal auto1111 webui in the checkpoint merger tab. if needed i can send you my note "how to"
For merging loras in to checkpoints you would need to use the supermerger extension

[dapper timber]

used a different path, reinstalled python, used vpn...

nothing is working

[ornate elk]

can you show me the full cmd log ?

[dapper timber]

sure lemme just run it again

Microsoft Windows [Version 10.0.26100.2605]
(c) Microsoft Corporation. All rights reserved.

C:\AI>git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
Cloning into 'stable-diffusion-webui'...
remote: Enumerating objects: 34819, done.
remote: Counting objects: 100% (28/28), done.
remote: Compressing objects: 100% (20/20), done.
error: RPC failed; curl 92 HTTP/2 stream 5 was not closed cleanly: CANCEL (err 8)
error: 6877 bytes of body are still expected
fetch-pack: unexpected disconnect while reading sideband packet
fatal: early EOF
fatal: fetch-pack: invalid index-pack output

C:\AI>

[ruby knot]

I downloaded some more advanced ones for doing it

[ruby knot]

Where is the venv folder?

[ornate elk]

normaly in the webui folder

[ruby knot]

I don’t see it?

[ornate elk]

can you screenshot what you have?

[ornate elk]

are you located in asia, russia or iran?

if so you need a VPN

[dapper timber]

asia

yeah

let me use it and run it again

same error

[ornate elk]

which country did you connect to?

[dapper timber]

netherlands

[ornate elk]

should be good

[ruby knot]

This is from the other install I use in my laptop I mean I was going to see if what you said would make it work on my pc too but I don’t see the folder.

[ornate elk]

try open up a cmd and run
ipconfig /flushdns
that clears the DNS entries

[ornate elk]

is it already installed or just downloaded?

because the venv gets created on the first launch

[dapper timber]

on the same file path?

[ornate elk]

doesnt matter

[ruby knot]

It installed when I dl it it was an easy install

[dapper timber]

okay

damn

[ornate elk]

sry i dont get it, is that the version linked from me before or the "other" one from you?

[dapper timber]

it worked

omg

i spent 2 hrs

on this

[ruby knot]

The one I use on my laptop from backup

[dapper timber]

tysm

[ornate elk]

no problem 🙂 the issue was that your PC still had no route to the adress because it was cached in the DNS

[ruby knot]

Weird it says dubious ownership detected if I try run it on my pc instead

[ornate elk]

ah okay thats strange than, is that the version you cant copy over? because that would make sense. If a webu doesnt have a venv it stores everything in %appdata% of the local users on the PC

[ornate elk]

thats normal as it got created with an other user the other PC didnt knew

[ruby knot]

How do you fix that then?

[ornate elk]

it should show a command you have to paste inside a cmd thats running in the same folder

[ruby knot]

Oh and if I load it offline it says python not found instead

[dapper timber]

oh i see, still thanks i really appreciate it

its downloading now i hope everything goes right now

[ruby knot]

Wait I’m still online wth

[ornate elk]

np, yep it should work. after the install you can turn of the VPN.
But if you want to use community extensions you need to change your DNS

[ruby knot]

I swear it always seems to have six million errors no matter what.

[dapper timber]

its currently downloading some pytorch, is it okay if i turn vpn off midway

[ruby knot]

I wouldn’t think so

[dapper timber]

yeah ig ill just wait

[ornate elk]

i would leave it on until its done with everything

[dapper timber]

and then turn off

[ruby knot]

Ig?

[dapper timber]

yeah that seems right

i guess

[ruby knot]

Oh ok

[dapper timber]

yo btw

if i may ask

how did u gain knowledge about all this

dns and all

[ornate elk]

im a sysadmin and helped a lot of people here with SD and technical problems

[dapper timber]

o-o

seems interesting man keep it upp

[ornate elk]

thx for sure ^^

[ruby knot]

Is there a newer version of the portable one you linked me to? It says that’s a pre release and from 2023

[ornate elk]

its the latest version, as i said you have to run the update.bat file inside it, it then downloads the latest updates

[ruby knot]

Oh ok my bad

[ruby knot]

What about those pt files do I need those too?

[ornate elk]

which .pt files?

[ruby knot]

Model pt and 2 vae pt files

[ornate elk]

where are those?

ahh nope

not needed

but before using run.bat
edit the environment.bat and change that to 0
SKIP_VENV=0

if not it will skip the venv creation and then you have the same problem as before

[ruby knot]

What is venv for ?

[ornate elk]

or wait ignore that

you want it portable

[ruby knot]

Yes

[ornate elk]

then it should work with it set to 1

[ruby knot]

Change nothing then ?

[ruby knot]

Do I run update first or run first ?

[ornate elk]

update first

[ruby knot]

Also so far it looks the same as the easy install version I got ages ago but with a different name

[ornate elk]

okay good

but with this version we know its the clean original one

[ruby knot]

Hmm I tryed load update first and it’s just a blank screen

[ornate elk]

complete blank ? 😮

[ruby knot]

Hopefully their wasn’t anything dodgy in the one I used before

It did something randomly now

No idea what tho

Then told me to close it so I’ve loaded run

It says it’s downloading PyTorch atm

[ornate elk]

i guess your fine, but always recheck the official sources, running some bat files from anywhere can do anything

[ruby knot]

That’s a bit worrying and I’m normally the one that’s really careful ful and warning other people too

[ruby knot]

Oh also why does it always say AMD64 when it’s for Nvidia ? Plus my laptop isn’t AMD but it still worked.

[ruby knot]

My pc is AMD with Nvidia gpu

[ornate elk]

that AMD is refering to the AMD CPU architecture not based on AMD Chips but on AMD as the founder of this architecture

its used with any modern CPU

so even with a Intel CPU it says AMDx64 in python 64bit

[ruby knot]

Who else makes chips like AMD?

[ornate elk]

i mean its not hardware based at all

[snow breach]

Quick one; do I need to use the base image to train LoRA against? If not, can any one recommend a few SDXL models I might get better results from other than the base SDXL model?

[ruby knot]

I know of amd intel Samsung snapdragon mediatek and one I forget

And Apple not including one I forget not sure how I forgot to put Apple there too.

[ornate elk]

Samsung, snapdragon and mediatek and Apple using the ARM Architecure.
INTEL and AMD using the x86 Architecture, but intel wanted a new one for x64
AMD got an idea to made software that is x64bit based work on x86 instead so thats why it say AMDx64
if you install Python 32bit it will show "Intel x86"
More info can be found here:
https://en.wikipedia.org/wiki/X86-64

[ruby knot]

Apple uses arm too

Arm is better in some ways

[ornate elk]

ARM is great for Mobile devices

[ruby knot]

Even Apple main computers are arm now I’m pretty sure?

[ornate elk]

yep they are on their 4th gen of ARM now

[ruby knot]

They are good

[ornate elk]

powerfull machines with less energy consumption

was their best move to get on ARM

[ruby knot]

Yeah exactly wish I could get a MacBook Pro

Apple is the only arm computer stuff I’ve heard of that isn’t trash so far for desktop os devices I mean

[ornate elk]

nope you can train on any sdxl model. for realism try juggernaut xl or realivis xl

juts make sure you dont use hyper models or any special variants

[snow breach]

Great, thanks. Ok, so nothing that uses lightning for example.

[ruby knot]

I hope I can get training and control net etc working this time, it’s been a failure for me so far

[ornate elk]

snapdragon made some nice arm laptops, snapdragon x elite

[ruby knot]

Oh? Last I heard windows is bad on arm.

[ornate elk]

for training get Khoya_ss or OneTrainer, dont try to do it inside the webui
controlnet should work fine

[ruby knot]

I have an 8core cpu and Nvidia 4060ti 16gb and 64gb ram

[ornate elk]

16gb vram is good for training

[ruby knot]

That’s a bit more reassuring but I’d guess it will be slow?

[ruby knot]

Oh also is Ai video possible on my specs ? I mean I know you normally need 24gb gpu but still

[ornate elk]

hmm depends but i guess maybe between 1 hour and 2-3 max

maybe even less

[ruby knot]

With how many images?

[snow breach]

Maybe I'm doing something wrong, I've got a 4080 Super and it takes me 3.5hrs to do SDXL training on a 20 image dataset with 25 reps.

[ornate elk]

cant say, i didnt trained a lot of loras

[ornate elk]

do you use khoya_ss ?

[snow breach]

Only uses 22% of my GPU though. Yes, with cuda 12.4

[ornate elk]

hmm is the vram maxed out ?

[ruby knot]

So it should be doable for me even tho my gpu isn’t as good?

[snow breach]

Depends, some times its close, sometimes its only 50% vram used.

You can totally train SDXL with 16GB ram. I was training SD1.5 easily even on my old 1080.

[ruby knot]

Where do you get that?

[snow breach]

I'm having issues getting flux to run though. Waiting for better support for SD 3.5.

[ornate elk]

with these new tools i dont think so, we have to wait until they get optimised for lower specs
but that seems good so far as i heard the work on making q8 models work

[ornate elk]

https://github.com/bmaltais/kohya_ss

[ruby knot]

I don’t use 1.5 much now as it’s trashy compared to xl most the time

[ornate elk]

flux should easily run on your card

make sure you dont use the 23 gb model

[ruby knot]

I tried flux on mine and it’s slow af

[snow breach]

I'm not its just that my front-ends are always crashing out for some reason. I've tried comfy and forge and they just usually give me "Press any key to continue" on the console after I click Generate.

[ornate elk]

with 16gb vram i would try useing the fp8 or Q8 flux models

[snow breach]

I tried both, dev-8 and Q8

[ornate elk]

how much RAM do you have?

[ruby knot]

Damn I hope it’s soon I want to make some videos for my music I made.

[snow breach]

64GB ram, 24gb swap file.

on C drive 🙂

[ornate elk]

sounds perfect

which flux mdoel did you tried?

oh both, i see

but it shouldnt crash, make sure you dont use the t5 fp16 encoder

that is to heavy

[ruby knot]

I don’t like swap files surely they will wear out your hard drive a lot faster, some phones force that on you now too

[snow breach]

flux1-dev-fp8 16.8gb and flux1-dev-q8_0 12.4gb

[ornate elk]

true but not that you will notice the degredation that fast

[ruby knot]

Oh also is it correct that flux won’t work in a1111? I couldn’t get it to work anyway had to use Comfy ui which is confusingly complex

[ruby knot]

Still would rather not wear out my ssd faster

[ornate elk]

yep its not supported in a1111 currently.
You can use Forge Webui for that. it looks like auto1111 so its not as complicated as comfyui

[snow breach]

I think I had put only the two normals in there:
clip_l 240kb
t5xxl_fp8_e4m3fn 4.7GB

and then I tried adding in:
ae 327kb

[ornate elk]

the ae file is 327 kb?

[snow breach]

^yup

[ornate elk]

then its corrupt

[snow breach]

oh

let me to grab that again then

[ruby knot]

How do I check if it’s using swap too ?

[snow breach]

oh I'm sorry, i've been using KB instead of MB

[ruby knot]

That’s so tiny on my phone

[snow breach]

total boomer

[ruby knot]

Who?

[ornate elk]

you have to go into your windows pagefile settings to check

[snow breach]

me

[ruby knot]

Oh

[snow breach]

ok. maybe I'll re-install forge and try flux again.

[ornate elk]

yep, let me know when you test it again

[ruby knot]

I really want to retry flux but I think it’s way too slow for it to be worth it for me atm so I hope they make more optimised mods so it’s faster on lower hardware

[snow breach]

will do, also do i need AE? or just the clip and t5

[ornate elk]

ae is needed

for flux

[ruby knot]

Damn that a1111 is still installing

[ornate elk]

a 3060 12gb gets 2-3s/it on flux fp8 / q8

[ruby knot]

I don’t know what that means

[ornate elk]

its like 2 mins for one image at a 1024x1024 resolution

[snow breach]

its not too bad. You can live with it!

[ornate elk]

My 7900XTX using ZLUDA gets 1-2 s/it = 1:30

[ruby knot]

I think if I remember correctly when I tried it it was 1 to a few mins for 1 image on 4060ti 16gb 1024 by 1024 which feels like forever compared to 16-30 secs on SDXL

[ornate elk]

yep sounds normal then

depending on the resolution it can take really long

[snow breach]

what resolutions is the flux model trained on?

[ruby knot]

It wouldn’t be so bad if you didn’t get loads of bad outputs until you tweak the prompt enough

[ornate elk]

true

reminds me of late 2022 when i used my GTX1080 on a first 1.5 7gb model and got an 512x512 image in 1-2 minutes

[snow breach]

that is why i just use the scripts to setup different checkpoints and batch search replace prompt keywords. pump out a 10x20 grid and walk away for an hour do something useful and come back 🙂

[ruby knot]

It’s also why I rarely try render 1080p images

[snow breach]

exactly me like 1 year ago lol

[ruby knot]

That sounds better than I would have expected?

[ornate elk]

with the 1.5 default model quality the outputs where bad but it was crazy that something like that could work localy, and now where here with flux and SD 3.5

[ruby knot]

Ah damn I forgot to put the checkpoint in the checkpoint file folder so now it’s wasting time downloading that too

[ornate elk]

you can close it and put them in then relaunch

[snow breach]

i tried the forge branch that supports 3.5, had about as much luck as flux, is that what you're working with?

[ruby knot]

I’ve seen some stuff on 1.5 recently that I didn’t know was possible with it

[ornate elk]

i use the normal forge as dev branch is experimental

[ruby knot]

I don’t want to interrupt it now incase it screws the install up

[snow breach]

normal forge supports 3.5??

[ornate elk]

nope

[snow breach]

oh haha what are you using for sd3.5?

[ornate elk]

comfyui

[snow breach]

damn ok.

[ruby knot]

Would 3.5 work on my setup ?

[ornate elk]

3.5 medium should work

[ruby knot]

Would it be as slow as flux ?

[ornate elk]

or even large if a smaller model gets used

[snow breach]

i'm guessing onetrainer is the only thing that supports 3.5 also

[ruby knot]

That comment is confusing ?

[ornate elk]

sry, there is also SD 3.5 large

and then you could use a fp8 /q8 model of it

[ruby knot]

Yeah that’s the same for flux too

[ruby knot]

What’s xformers ? On all the installs I tried it’s always errored and said it doesn’t have it

[ornate elk]

oh thats something you should use in auto1111. its a free performance boost

for nvidia cards only

[ruby knot]

Oh how do I get it?

[ornate elk]

you edit the webui-user.bat file thats in the webui folder
there you add --xformers to the commandline_arg= line

then relaunch

[ruby knot]

Oh ok

[ruby knot]

Then remove it after?

[ornate elk]

nope

then it would be deactivated

[ruby knot]

Oh ok

[ruby knot]

Pruned mean means smaller right ? If so that’s weird as I have a copy of it’s that’s only about 2gb and the one it’s auto downloading is about 4gb

[ornate elk]

yep. but the default model that auto1111 downloads is the first 1.5 model its a bit different (pruned and not pruned are the same size)

you can just delete it later

[ruby knot]

Oh ok

[ruby knot]

What is the difference between those 2 then if they are the same size ?

[ornate elk]

the non pruned is made for training

the pruned is for generating

[ruby knot]

Oh

[ruby knot]

I think it finally finished installing

[ruby knot]

Is it weird that I’m running the ai off an external ssd in a 10gbps enclosure, maybe running at 5gbps atm?

[dapper timber]

Yo

I ran stable diffusion in my pc

but the output was absolute garbage

do i need to change some settings or something to get the best output

[ornate elk]

if you use the base model 1.5 you should get an better model (checkpoint) on civitai.com

[ornate elk]

hmm it shouldnt affeect anything as the ssd just moves the mocel into the RAM

then the image gets generated on the PC with the GPU + RAM

[ruby knot]

I’m doing that as I didn’t put the ssd in my pc yet and I want to sort out my ai setup and make a back up of it too, won’t still be faster then a normal hard drive anyway ? Tho it is mental using a 7500 speed ssd at 5-10gbps

[ornate elk]

its for sure faster than an hdd

[dapper timber]

there are many checkpoint models, which one to get?

[ornate elk]

what was your gpu again?

[dapper timber]

2050

lol

[ornate elk]

and your using forge or a1111?

[dapper timber]

a1111

[ornate elk]

make sure you have --xformers --medvram in the webui-user.bat

[dapper timber]

yeah i do

[ornate elk]

a good 1.5 based model would be Dreamshaper v8 for semi realism and Juggernaut reborn for realism

[dapper timber]

i downloaded noobai

xl

what does that mean

[ornate elk]

its a larger model and for that type sdxl/pony/illustrious you should use forge webui

[dapper timber]

whats the main diff between forge and a1111

[fair oxide]

anyone know what a good price is to pay to have someone get stable diffusion running on my pc? i keep running into a bunch of errors. thanks.

[ornate elk]

forge uses less resources, is better suited for lower end gpus

they have the same ui

[ornate elk]

you can ask here and i help for free xD

just let me know whats your gpu and what do you try to install (auto1111, forge etc)

[fair oxide]

once i get back to my computer i’ll show you. i was able to get past the first 3 errors by googling it but i can’t figure this one out.

[ornate elk]

if you want to do everything correct, checkout my install Guides:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides
if then something doesnt work feel free to let me know

[fair oxide]

do i need to uninstall anything since i tried it once

[ornate elk]

depends on what you installed, what your gpu is and which guide you followed

better to start clean of course

and python should be 3.10.11 64bit and not higher (not 3.11 or 3.12)

[dapper timber]

are these models able to browse internet

[ornate elk]

no, models for image generation are trained fix

[fair oxide]

that’s the python i have.

[fair oxide]

i have AMD should i pick anyone of these on your guide

oh wait i see

nvm

[ornate elk]

whats your gpu model?

[fair oxide]

amd radeon

[dapper timber]

*the

[fair oxide]

rx6800

[ornate elk]

ah okay good

then i recommend trying the Auto1111 with ZLUDA Guide or the Forge-Webui with ZLUDA

[fair oxide]

bro if i send you $20 can you just do this shit for me ? 😂

[ruby knot]

Do you know anything about Ai music generation ? I can actually make music my self but I’m interested in messing with Ai too

[ornate elk]

sure we can do that xD

[fair oxide]

i sent you a dm

[ornate elk]

localy there isnt much, but there are some websites who offer music generation

[ruby knot]

Yeah I have Suno