🤝｜tech-support

[vocal burrow]

We had way too many people operating scams in dm

[smoky schooner]

How should I look at it? ComfyUi is on my desk

[vocal burrow]

screenshot its content

[agile sun]

i did it and it works for me i just had to add the necessary addons to the .bat file. i have a question, does this latest version of ‘a1111’ have any built-in filters, anti-NSFW i'm doing NSFW/SFW and the NSFW stuff isn't coming through, is it some special ‘a111’ you sent me that's blocking it if so how can i launch that

[vocal burrow]

no there shouldn't be.

[vocal burrow]

I'm quite confused. How did you install comfyui ?

cause it seems to be missing lots of files

[smoky schooner]

https://www.comfy.org/download

from here

[vocal burrow]

and then you selected the "download for windows" I guess ?

[agile sun]

model 10 and 20 image apart from the words SFW where exactly did the creator of this model build in the filter or what

i don't get it in this version of ‘a111’ it feels like it's the same as before or maybe ‘a1111’ has some built-in content-blocking feature in the settings by default 'a11111'

[smoky schooner]

yes

i selected the nvidia card and done

[vocal burrow]

ok it's their new-ish setup then I'm not too familiar with it but let's roll with it.

First make sure that you have updated your nodes and everything. Their should be an "update all" option in comfyui

[smoky schooner]

Got it

No updates available that’s means my nodes are updated

[vocal burrow]

you're searching for, delete this

[smoky schooner]

Okay got it now

[vocal burrow]

what's in there?

[smoky schooner]

The ultralitycs

[vocal burrow]

is there more information if you click on it ?

[smoky schooner]

Nope just these

[vocal burrow]

I have to move pretty soon so I'll just let you know how to install ultralytics manually as it seems to be the error from your first post.
1/ go in this .venv folder
2/ click in the top bar of file explorer, write cmd then press enter to open a console at this location
3/ then run Scripts\activate.bat
4/ then pip install ultralytics (edited)
5/ restart comfyui

Hopefuly it should fix it. But it's weird that it didn't install it automatically..

[vocal burrow]

Ultralytics

[prisma moss]

lysics

[vocal burrow]

This

[smoky schooner]

And still don’t working bro i swear i hate this fucking

[dire pebble]

Hey everyone, I’m a freelance AI engineer working on AI agents, automations, and practical AI products.
Happy to connect and chat with anyone building interesting things.

[prisma moss]

@ornate elk Oi, has there been a silent update or something to forge? My it/s dropped by half and idk why, across the board (vanilla, classic,neo)? Tried reinstalling driver didnt work

[ornate elk]

forge doesnt get updates since 10 months so nope.
forge neo gets daily updates
but i didnt noticed a speed change there

[prisma moss]

Illd ask u to try with latest nvidia driver but u aint on nvidia

Rolling back did hel ptho

[prisma moss]

Also this is kinda..untrue

Remember the time they did something with pytorch and segm models stopped working?

Even tho nothing updated on my end

[ornate elk]

10 months ago, last update

[prisma moss]

👆

It wasnt getting updates then either

[ornate elk]

that has nothing to do with the webui itself rather than packages getting updated that the webui doesnt set a fixed version for it

[prisma moss]

See

So there is something

[ornate elk]

so no no change for the webui itself

[prisma moss]

Packages which affect the webui tho

[ornate elk]

yes

[prisma moss]

=/=

[ornate elk]

not the same

as the dev didnt intend the change

[prisma moss]

Still it pulled

[ornate elk]

because of git

[prisma moss]

Fuck

[ornate elk]

and and not a fixed version set in requirements versions txt

[prisma moss]

Gotta remove git pull from .bat lol

[ornate elk]

it will always pull the latest and thats why stuff can break

[ornate elk]

no that wouldnt help

git pull just gets the latest stuff from the webuis repo not from the packages

the packages will only pull their versions at the install

when creating the venv

[prisma moss]

Then nvidia broke smth

[ornate elk]

probably, you said a driver rollback helped?

[prisma moss]

Ye

[ornate elk]

yea then they definitly changed something

[prisma moss]

Morons

[ornate elk]

was it slower in forge neo too then?

[prisma moss]

It was in classic

I havent tried neo

[ornate elk]

ah okay

[plain crypt]

Hello, does Stable Diffusion run correctly on Mac for generating videos?

[ornate elk]

probably not that good for videos

but images sure

[plain crypt]

I see, I thought so though, given that I have a good CPU and GPU.

[blissful crane]

Hi all. I've recently upgraded my old laptop, and am now running a 5080 with 16GB vRAM, and 32GB system RAM. I've installed Forge Neo. I used Gemini to answer various questions about my hardware and its compatibility, and it's insistent that my generation speeds are way below what they should be. My only point of reference is my old 2070 laptop, and I was running SDXL on that. I'm now running Flux.1D models. Anyway, running an FP8 or NF4 model to generate a 25 step 1024x1024 takes between 30-50 seconds, depending on whether I'm using LORAs. My helpful AI assistant tells me they should be sub 10 seconds.

Apparently I should see the string "NVIDIA GeForce RTX 5080 Laptop GPU (sm_120)" in the terminal, but I see (native) 0 or something similar. I've previously tried "cudamallocasync" as well. I've tried all suggested fixes to change that, to no avail.

When I run a generation I see BOTH my vRAM and my RAM spike by the size of the model (e.g. 9-11GB). I also see my GPU utilization spike as expected.

Question 1: Does it sound like my generation speed is slow, or about right for my hardware?
Question 2: Does the fact that my RAM is being used alongside vRAM create a slowdown?

Appreciate any advice.

[ornate elk]

Hey, which launch args do you use?
These can help with performance a lot

[plain crypt]

Is it still possible with a good Mac?

[prisma moss]

@ornate elk ok so the halved it/s persists even with the older driver, and only fixes it self after a fresh driver reinstall

but goes back to shit on next windows start up, so illd have to reinstall the driver everytime

got an idea what might be causing this?

[blissful crane]

These are my current ones. --cuda-stream --sage --fast-fp16 --bnb --pin-shared-memory --disable-smart-memory
I've also added the lines
set TORCH_CUDA_ARCH_LIST=12.0
set NVIDIA_ALLOW_UNSUPPORTED_ARCHS=1

[ornate elk]

Ah okay that looks good

[ornate elk]

You should test how fast an sdxl 1024x1024 image with 30 steps is.
That should be under 10 seconds.
Flux 1. dev is slow af even on the latest nvidia gpus. So it won't get to the 10 seconds. Maybe only with a turbo lora.
Nf4 should be faster.

For the RAM usage question, that's completely normal.
The model gets first loaded into the RAM and then into the vram for the actual generation.
VAE, and text encoders also get loaded this way. So if all files are larger then the available VRAM, swapping will start or it goes into shared vram which boath are slower than running completely on the GPU like sdxl does.
Also while a fp8 flux dev model is only 12GB, doesnt mean it will only use 12GB for the image generation. It will also need to load the other files and room for the actual image generation.

[ornate elk]

Thats strange.
First you should make sure your windows is updated.
Then restart and run in a cmd
SFC /scannow
That fixes windows bugs after updates... They often occur.
Then you could try using DDU for the driver reinstall.

[prisma moss]

did sfc, i mean windows 10 is inherently 'out of date' since 11 exists, ill try ddu tho later on

[blissful crane]

Thanks that's really helpful. Yes 30 step SDXL is around 8 seconds, so that's good. It must just be that the model plus text encoder etc is pushing the vRAM.

[ornate elk]

NP, that sounds all normal then

[wooden mango]

getting this error

venv "C:\Users\Owen\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.10.1
Commit hash: 82a973c04367123ae98bd9abdf80d9eda9b910e2
Traceback (most recent call last):
File "C:\Users\Owen\stable-diffusion-webui\launch.py", line 48, in <module>
main()
File "C:\Users\Owen\stable-diffusion-webui\launch.py", line 39, in main
prepare_environment()
File "C:\Users\Owen\stable-diffusion-webui\modules\launch_utils.py", line 387, in prepare_environment
raise RuntimeError(
RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

if I run that it gives me a blue screen

what do I do

[ornate elk]

What's your gpu?

[wooden mango]

NVIDIA GeForce RTX 3090

[ornate elk]

Then you should not use Auto1111 as webui.
Instead follow the pinned messages of this channel and install Forge Neo

Same UI but more updated

Also uninstall python 3.10

[wooden mango]

is that this one: CS1o — 6/7/2024 4:00 PM
The latest Stable Diffusion install Guides for Nvidia and AMD:

thanks for your help

[ornate elk]

np, also for the bluescreen issue, open up a cmd as admin and run:
SFC /scannow
that searches for windows system errors and tries to fix them

[wooden mango]

got this

[ornate elk]

you didnt made the folder that is needed

called ai

[wooden mango]

I know thats a lot of text

oh

[ornate elk]

the folder is needed so you dont get erros based on permissions

also make sure you have uninstalled python 3.10 and installed python 3.13.12

you can then also open up a cmd and run pip cache purge
that will delete old install stuff before launching the bat

[last gull]

can anyone help me regarding lora ?
I've been trying to train a lora so I can generate icons in the same artstyle as some reference icons
used hugging face's diffusers , with train_dreambooth_lora_sd15_advanced script
but I cant seem to be able to get good results , and i am unsure what exactly could be the problem

I couldnt use kohya's as i get error while building wheels ( which I read is a version missmatch as colab uses 3.12 )

for reference , this is the notebook I based my training on :

https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/SDXL_DreamBooth_LoRA_.ipynb

and this is an example reference icon , which want to replicate it artstyle

[silk estuary]

Does anyone know if there is a way to synchronize forge browser UI to "cmd forge" instead of the oposite?
EG my browser crashed in the middle of a several hour metadata fetch from civitai; but cmd forge seems ok. However, I cannot see how it is going since civitai browser plus only reports some steps in cmd and not a full bar with progress or what it is doing in most steps.
So if I reopen the UI, it will not display the ongoing process and it is very likely that the few steps reflected in cmd as part of opening the UI will disrupt the process.

So in short: 1 opening the UI again may stop other ongoing processes, idk. 2 a newly open UI will not reflect ongoing processes from a past open UI that crashed, which makes it near impossible to know if it is stuck.

[prisma moss]

so ddu didnt work, everystart up i basically have to reinstall the driver to get the card back to its full potential

[celest delta]

how do i prevent forge neo from using lots of memory when not generating?

[prisma moss]

there should be a setting to keep or not to keep models in vram

[celest delta]

i have no idea where that would be, there's so many settings

[prisma moss]

you have a search function

[celest delta]

this? Should i put it to 0?

[prisma moss]

no

[celest delta]

then i dunno :p

[ionic citrus]

please can a kind soul help. Im trying to install ComfyUI-Zluda and im getting this

"'.\zluda\zluda.exe' is not recognized as an internal or external command,
operable program or batch file."

[chrome forge]

hey can you help with me something

[ornate elk]

depends on what xD

[chrome forge]

i have a .pt file its an Adetailer and idk how to use it

[ornate elk]

what its called? and which webui do you use?

[chrome forge]

just Eyes.pt im using comfy ui with zluda

cus all of the eyes have color bleeding in them

[ornate elk]

ah okay, do you have a facedetailer node?

there you can use it with

[chrome forge]

how do i get it? (sorry im new to this)

[ornate elk]

do you have comfyui manager installed?

thats mandatory

[chrome forge]

this is thing?

[ornate elk]

there go to install custom nodes and search for facedetailer

[chrome forge]

which one?

[ornate elk]

ah wait i have to check which one

ah okay nope you need the comfyui-impact-pack

there the facedetailer is included

and you also need the ComfyUI-Impact-Subpack

[chrome forge]

the first one? also what does that 4 conflicts mean?

[ornate elk]

yes the first one and the 5

conflicts just mean it uses nodes other extension also use

can be ignored as the impackt pack has the most stuff

you dont need much more

[chrome forge]

does version matter?

[ornate elk]

choose latest

[chrome forge]

okay i intalled both and restarted now wat?

[chrome forge]

i just need to get rid of the bleeding in the eyes (anime style) do i need all this or is this the wrong node?

[ornate elk]

you need that xD

and the small box for the model

[chrome forge]

i need all that to generate eyes like this?

[chrome forge]

this is how my eyes look like rn

[ornate elk]

yep, if you would use forge neo it would be easier as there you can use adetailer xD

[ornate elk]

you can also try use Anima as model, it can do eyes really good!

[chrome forge]

can you send the link?

[ornate elk]

https://civitai.com/models/2383017?modelVersionId=2843915

[silk estuary]

I need some database advice. So I collect tens of thousands of loras to run with forge. I save the metadata using civitai helper and civitai browser plus, which also saves the html.
However, this time while I am sweeping with civitai browser plus, I always get an error in the middle that GPT says is bc of an unexpected response from civitai. So investigating, I discovered that browser plus is dead and no longer updated. And people are complaining about similar things.

My long game was keeping all that data so eventually I can upgrade my wildcard setup to a moltbot or a more advanced bot running forge continuously and trying to optimize parameters to make the finest art.

Some model triggers or other info are not in the triggers field, so keeping the html was very convenient.

I am basically puzzled about how I can solve this to use a different program while keeping my database format and not getting redundant metadata files. Also, I don't know any other program that saves html.

Please, I need some advice.

[hidden night]

IS IT POSSIBLE to get Illustrious models running on an S25 EDGE

[ornate elk]

no

there is no local webui for android phones

[hidden night]

Is it possible to get them running on a PHONE?

[ornate elk]

only with Termux

[hidden night]

What's that

and which phones support it

[ornate elk]

it supports all phones that can install the termux app

but it wont be fast or easy to setup

here is an old guide:
https://gist.github.com/nviet/a4296bae33cbc8738e274951e0a89d4a

[ornate elk]

do you use this version of civitai browser plus?
its the only one still maintained:
https://github.com/anxety-solo/sd-civitai-browser-plus

[silk estuary]

No, I was using blafking's. And civitai page for the extension also said it was dead.
I hadn't even considered there might be an updated working fork somewhere.

Really thank you

[ornate elk]

no problem!

[hidden night]

What about the "Draw Things" app?

On IOS

I managed to get a tiny model running on my iPhone 11 so It can definitely do some local generation on iPhone 17

[ornate elk]

yes but you asked for android

[hidden night]

Oh no

[ornate elk]

ios can use draw things pretty good

but there is nothing for android

[hidden night]

So can iPhone 17 run Illustrious models?

[ornate elk]

probably

idk

[hidden night]

The models are 8 gb

[ornate elk]

iphones have so less of RAM

oh you mean the models

they are around 6gb

[chrome forge]

bros about to explode his phone?

[hidden night]

Justified by the superior operating system in my opinion

[hidden night]

Yeah

[silk estuary]

Wait, iphones have vram?

[ornate elk]

no

they have normal ram that gets used as vram

[silk estuary]

Ah

[ornate elk]

shared memory

[hidden night]

iPhone 17 has 8gb of ram

Illustrious is around 6gb

hmmm

[silk estuary]

Weird, so I can barely run XL architecture in my laptop with 6 GB VRAM; but it could run better in an iphone?

[ornate elk]

it wont run better

because you can use 6gb vram + normal ram for speedup (pagefile)

[chrome forge]

probably will explode

[hidden night]

Probably around 10 people have tried local image gen on a phone so nobody knows

Uh why the hell I'm worrying about illustrious while I got a 5070 ti sitting there

anyways S25 edge, iPhone 17 or Redmagic 11 air

[silk estuary]

hmm not really. I didn't test in newer forge and don't use Comfy; but I can only use XL with no LoRAs or anything else. It gave OoM when I tried a year ago.

[hidden night]

which is a better phone overall?

[ornate elk]

whats your gpu and which webui do you use?

[silk estuary]

Laptop NVIDIA 2060 with 6 GB, 32 GB RAM DDR4, Forge UI

[ornate elk]

if you like bloated ui and a lot of vendor apps go with samsung.
advantages would be long software update support and best camera and display.
Redmagic is faster in charging and has double the battery size,
idk if its as bloated as samsung or xiaomi
Redmagic looks cooler but is 6.85 inches against 6.7 from samsung

[hidden night]

hmmm

[silk estuary]

I can confirm, mine is a samsung

[ornate elk]

have you tried forge classic

its more updated than forge but based on it

supports more performance stuff

[silk estuary]

I didn't know it existed
Thanks

[ornate elk]

No problem make sure to add --cuda-stream --flash --xformers to the webui-user.bat

For the best performance on rtx20 gpus

[silk estuary]

Cool, I use those currently except for the flash which is new

Thank you

[hidden night]

Which model should I use for pixel art sprite generation? Illustrious fine tunes are bad with the pixel art lora

[ornate elk]

this one is good MiaoMiao Pixel:
https://civitai.com/models/1180112?modelVersionId=1749717

[prisma moss]

what arguments do i need on comfy?

its it/s are slower than forge

[ornate elk]

it can be faster than forge, it also supports sage and flash attention

[ornate elk]

whats your gpu again?

[prisma moss]

im using a basic sdxl workflow

4090

so load>prompts>ksampler>vae>output

and ksampler is slower

by like 20%

[ornate elk]

try use --use-sage-attention --disable-pinned-memory

or --use-flash-attention instead of sage attention

[prisma moss]

where do i even put the tags

oh wait

i already have --fast and sage arg

also sage by itself doesnt do anything unless its patched through it at least not in this case

and anyway, with pinned memory its still at 5it/s

[ornate elk]

then try with flash attention it should work directly

[prisma moss]

i mean i cant do that

cuz i dont use comfyui for just image genning

theres wan which works fine

and uses sage

unless i can have both active

[ornate elk]

you can have two launch bat files, one with sage and one with flash

for video gen sage is better, but sage doesnt do anything on sdxl

[prisma moss]

ait trying w flash

still same

[ornate elk]

hmm

[wooden mango]

got this error

when I tried to launch it

[prisma moss]

idk dfq it wants, does it need one for user too?

(yes it does)

[buoyant lava]

Guys I’ve been questioning the Anima dataset and why it has ye pop and deviantart

People say it is poison. And I think I’ve proved that it is poison.

Id like if someone can replicate the issue on their end.

[kindred flame]

any idea how to fix this error, I always got this error when installing the clip idk why

[buoyant lava]

Fun fact: if you include the word “fat” in your anima positive prompt, it deep fries the image.
Probably because deviantart has a ton of fat art which cooks the image.
Why was deviantart even added

[kindred flame]

any idea how to fix this error, I always got this error when installing the clip idk why

[kindred flame]

I used the correct python version and I did everything as instructed but somehow it still doesnt work.

[ornate elk]

What's your gpu?

It looks like you try to install Auto1111 which is outdated

[kindred flame]

rtx 2050

[kindred flame]

hmm how to get the latest version, I already run the update.dat

[ornate elk]

Ah okay with that GPU you should setup Forge classic

[kindred flame]

hmm wdym

[ornate elk]

You find a setup guide for it in the pinned messages

Its based on Auto1111 but more updated

Its the same ui

[kindred flame]

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

aight thanks for the info

[ornate elk]

Np!

[kindred flame]

I should pick "Automatic1111 (Stable-diffusion-webui)" right?

[ornate elk]

Forge classic would be faster with your gpu

[hallow umbra]

hey does anyone know how to have stable diffusion, primarily reforge to stop checking for installation requirements upon starting? It should work without the interent as well, but i really would like to turn this off

[ornate elk]

hey you can try:
--skip-version-check --skip-python-version-check --skip-torch-cuda-test

[novel owl]

hello i have a problem for use comfy on my rx 7900 xt

it dont want launch i have uninstall rocm ddu my gpu driver uninstall comfyui and ré install and it canot launch

i use the installer verssion

[ornate elk]

hey, thats the issue

dont use the installer xD use my setup guides and it will work

[novel owl]

the comfyui dont want launch with your setup guide too

[ornate elk]

which one? the easy (portable) or the mnaul setup?

[novel owl]

this

[ornate elk]

okay, any error log?

[novel owl]

D:\IA\ComfyUI_windows_portable>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
comfy-aimdo failed to load: Could not find module 'D:\IA\ComfyUI_windows_portable\python_embeded\Lib\site-packages\comfy_aimdo\aimdo.dll' (or one of its dependencies). Try using the full path with constructor syntax.
NOTE: comfy-aimdo is currently only support for Nvidia GPUs
Fatal error in launcher: Unable to create process using '"D:\a\ComfyUI\python_embeded\python.exe" "D:\IA\ComfyUI_windows_portable\python_embeded\Scripts\offload-arch.exe" ': Le fichier spÚcifiÚ est introuvable.

[WARNING] offload-arch failed with return code 1
[stderr]

[ornate elk]

seems like the comfyui portable got corrupted

delete and redownload it

and then unpack/extract with 7zip

[novel owl]

okay

[novel owl]

is it compatible with the latest hip sdk ?

the 7 ?

[novel owl]

same issue

D:\IA\ComfyUI_windows_portable_amd\ComfyUI_windows_portable>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
comfy-aimdo failed to load: Could not find module 'D:\IA\ComfyUI_windows_portable_amd\ComfyUI_windows_portable\python_embeded\Lib\site-packages\comfy_aimdo\aimdo.dll' (or one of its dependencies). Try using the full path with constructor syntax.
NOTE: comfy-aimdo is currently only support for Nvidia GPUs
Fatal error in launcher: Unable to create process using '"D:\a\ComfyUI\python_embeded\python.exe" "D:\IA\ComfyUI_windows_portable_amd\ComfyUI_windows_portable\python_embeded\Scripts\offload-arch.exe" ': Le fichier spÚcifiÚ est introuvable.

[WARNING] offload-arch failed with return code 1
[stderr]
Traceback (most recent call last):
File "D:\IA\ComfyUI_windows_portable_amd\ComfyUI_windows_portable\ComfyUI\main.py", line 196, in <module>
import comfy.utils
File "D:\IA\ComfyUI_windows_portable_amd\ComfyUI_windows_portable\ComfyUI\comfy\utils.py", line 20, in <module>
import torch
File "D:\IA\ComfyUI_windows_portable_amd\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch_init_.py", line 156, in <module>
rocm_init.initialize()
File "D:\IA\ComfyUI_windows_portable_amd\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch_rocm_init.py", line 4, in initialize
rocm_sdk.initialize_process(
File "D:\IA\ComfyUI_windows_portable_amd\ComfyUI_windows_portable\python_embeded\Lib\site-packages\rocm_sdk_init.py", line 133, in initialize_process
preload_libraries(*preload_shortnames, rtld_global=rtld_global)
File "D:\IA\ComfyUI_windows_portable_amd\ComfyUI_windows_portable\python_embeded\Lib\site-packages\rocm_sdk_init_.py", line 102, in preload_libraries
cdll = ctypes.CDLL(str(path), mode=mode)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "ctypes_init_.py", line 379, in init
OSError: [WinError 4551] Une stratégie de contrôle d’application a bloqué ce fichier

D:\IA\ComfyUI_windows_portable_amd\ComfyUI_windows_portable>pause
Appuyez sur une touche pour continuer...

[novel owl]

do you want take controll of my pc ?

[ornate elk]

do you have a third party antivirus programm installed?
it seems like something is blocking stuff from comfyui

[novel owl]

no only windows defender

[ornate elk]

hmm strange

[ornate elk]

OSError: [WinError 4551]
Une stratégie de contrôle d’application a bloqué ce fichier

[ornate elk]

do you maybe have the Smart Screen App Control activated in Windows Defender?

that thing blocks a lot of normal stuff falsely

[novel owl]

how can i know it ?

do you want take control of my pc for see ?

[ornate elk]

i can do that if you want

[novel owl]

what software do you want use ?

[ornate elk]

Anydesk works

[ornate elk]

Permission problems caused the issue. Manual install worked

[hallow umbra]

inn the .bat file?

[prisma moss]

@vocal burrow

[prisma moss]

anyone knows if its possible to see the bbox boundary mask like in forge, in comfyui tho? Or to see the confidence score when detecting?

[ornate elk]

Yep you can see it but I'm not at home to show

[prisma moss]

ait please let me know when u get back 🙂

[prisma moss]

this is the current setup

[silk estuary]

So I have Forge in a external HDD. In that same drive I keep the pony and XL loras. Then I keep illustrious in a different drive and 1.5 loras in a third drive.

My free space in system is pretty low. I currently have 11 GBof paging in system and I only have less than 10 GB free other than those.

Forge classic recommends 40 GB of paging. Would it work if I used paging in any of the other drives?

[silk estuary]

Also, I have disabled screen lock and a number of energy comsumption options in windows bc they interferred with A1111 in the past. Do we know if they also interfere with forge and forge classic?

[vocal burrow]

paging on anything else than c: is pretty useless in SD's case

[prisma moss]

if you got ssd yes yes

[prisma moss]

👆

lmao

so what do u say

[ornate elk]

give me a sec

[prisma moss]

what about confidence score?

[ornate elk]

nothing i dont use that xD never used it in forge too

[prisma moss]

i mean its not something to be used

it just says numerically how sure it is what it sees

[ornate elk]

yea idk about that one, ah you mean that square with the number, yea idk if its possible in comfyui

[prisma moss]

aight ty tho

[ornate elk]

np

[prisma moss]

Do you know if opening multiple comfy workflows amd running them, will que them?

As in if i run workflow 1, will workflow 2 (in another tab) wait for the workflow 1 to finish?

[ornate elk]

i havent tested it but i guess it should

[prisma moss]

And do u know if its possible to assign smth like a class to file saving?

For example, if i were to have an image save node where i input the name of what i want the file name to be saved as, but its dybamic in reference to what i put in the prompt or smth

[ornate elk]

it should be possible but idk

i dont use comfyui that much

better ask in #🧣｜comfy-ui channel

[hidden night]

which one? flux.1 kontext dev, dev, dev quantized, schnell, schnell quantized

Oh you sent that one

a sec

[silk estuary]

The same system SSD has other partitions, I could do paging there. Would it work?

[stark torrent]

How much VRAM do you need to make 720p 5 second videos locally (Forge UI, etc)?

[hidden night]

Which one is faster for Image generation? 3090 or 5070 ti?

[prisma moss]

Depends on the size of the model

720p is a longshot

[prisma moss]

Yes

And set it to system managed

[silk estuary]

Thx

[odd steeple]

So im using, for my checkpoint, meinamix and euler a but for the schedule type I cant find a good one because it usually ends up mutating the body. I am also using hires fix R-esgran 4x+ anime6b. How do I stop it from mutating? my negative prompt already has stuff like mutating, disfigured, etc.

[ornate elk]

choose karras as scheduler or leave it on automatic.
the samplers doesnt make the mutating. that comes from wrong hires fix settings (to high denois)

[hidden night]

why karras?

[prisma moss]

why not

[spiral river]

yes

[prisma moss]

mhm

[spiral river]

you can use prompt variables in the image save node, or use something like save image extended

[prisma moss]

i already resolved that

[prisma moss]

do u know about this tho https://github.com/ltdrdata/ComfyUI-Impact-Pack/issues/1204

ill take that as a no xD

[plush vine]

After updating the drivers I get an error
Don't know how to fix it?

[spiral river]

I don't even know what classes are

[prisma moss]

Class is whats being deteced

[prisma moss]

What does the console say

[prisma moss]

oh god its amd

[ornate elk]

because it works good with euler a

[ornate elk]

wrong hip sdk

[fading nebula]

can u check dms

[plush vine]

reinstal, same error:

[ornate elk]

still wrong hip version, you need 6.4.2 + custom gfx library files

or you can try to use native rocm support,
All AMD guides are in the pinned messages of this channel

[plush vine]

extract same rocm files that worked last time?

[hidden night]

Oh man

I wouldn't even bother with AI if I had an AMD graphics card

Guys is it worth trading my 5070 Ti + 1 Stick of 32GB DDR5 ram for 2X 3090?

Triple the vram

I do Image Generation

[ornate elk]

I say not worth.
For image gen the 5070ti beats the 3090 easily.
Also most 3090 cards are from the cry-pto mining era means they could fail more often and the cooling paste needs to be renewed for sure.

Also 32GB ram is worth one of these cards alone xD haha

You also can only use one GPU for one task.
Its possible to outsource different steps of a workflow to different GPUs but idk how good that works

[hidden night]

Hmm ok

and 64gb ram is way more future proofed

If I did video gen, I'd easily do the trade

but ok then

[cosmic furnace]

any good news on updates ?) the rock or something

[wind jewel]

hello. idk where i can ask for help but i have issues building stable diffusion web ui on my pc.

i followed the guide and installed all dependencies but if i double click on "webui-user.bat" i get this error.

i actually have multiple errors but idk what the cause could be.

[still ridge]

Hey guys i keep getting errors

When I try to open up stable diffiser

[wind jewel]

"diffiser"

[still ridge]

Do you know how to fix my issue

[wind jewel]

looks like you have the exact same error as me

i think i fixed it tho.

[still ridge]

Howd you fix it

[wind jewel]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Troubleshooting

one sec..

whoops

wrong link

^^

@still ridge https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/17212

go down to "Fix: we have updated the dev branch"

and follow the instructions there

just open in the cmd window like before but inside "stable-diffusion-webui" folder

and then enter "git switch dev"

then enter,

then tpe in "git pull".

after that try to run stable diffusion again.

[ornate elk]

@wind jewel @still ridge
You boath shouldn't use the outdated Auto1111

If you have a newer GPU you should use Forge Neo instead

[still ridge]

I have a 3050

[ornate elk]

We have the latest updated setup guides in the pinned messages of this channel

[still ridge]

can we maybe hop into a call

@ornate elk '

im new to this\

[wind jewel]

lul

[still ridge]

?

[wind jewel]

i meant this.

whats forge neo btw?

[ornate elk]

Its a webui based on forge which is based on Auto1111

But it gets updated

Auto1111 (stable-diffusion-webui) shouldn't be used anymore as its outdated since over a year

[wind jewel]

ah i see

should i get forge or forge neo?

i have a rtx 3060

[ornate elk]

Yea then go for it

It also supports more models

The UI is the same as Auto1111

[wind jewel]

oh thats nice.

[cosmic furnace]

it seems therock works without tiled vae now ? i dont get any crashes. also mb doing something wrong but i get 2 times slower generation on ZIT with tiled vae -_-

[wind jewel]

i tried to learn comfyUI once but its completly different from a1111. and im still pretty new to this.

[wind jewel]

so thats REALLY nice.

thanks!

[ornate elk]

No problem feel free to ask if any step isnt clear

[ornate elk]

Your missing a space between these two args

[still ridge]

Like this?

[cosmic furnace]

i get pixelated as if low quality image on zit when txt2img

but good quality when i upscale it in img2img

i even tried same 1280 -1280 res there & there . but img2img good quality and txt not 🙁

[ornate elk]

Maybe wrong settings

[cosmic furnace]

https://civitai.com/models/2242173/dark-beast-or

[wind jewel]

@ornate elk i got Forge Neo installed successfully! can i move the build to a different folder?

[ornate elk]

No, moving it breaks stuff.
Your supposed to install it at the place you want first

[wind jewel]

oh shit.

[wind jewel]

the guide sait it should be in C:/AI

thats why i build it there

can i change the build path?

@ornate elk does regional prompter still work in forge neo?

or is there a different extension for forge?

[ornate elk]

There is forge couple

It does the same but easier

From the same dev as forge neo

[wind jewel]

thanks man. 👍

[wind jewel]

what does this mean?

[ornate elk]

that means you could enable that flag but it doesnt improve anything so just ignore it

[wind jewel]

omg forge is SOOO much faster than stable diffusion!!

💀

[prisma moss]

forge is just user interface, you r still using stable diffusion based model (probably)

[plush vine]

amd worked for me at 5.7 version hip, but now i update driver for playing and need to find right rocm build.
well. i install 6.4 hip and extract Optimised_ROCmLibs_gfx1031 to it and now have this error, im reinstal python with add path and (for all users button), also add environment parameters,
Have this error, do anyone now how solve this?

[ornate elk]

try update python to 3.10.11 64bit

then delete the venv folder and relaunch

[plush vine]

instal 3.10.11 delete last pythin verse now this

[ornate elk]

Open up a new cmd inside the stable-diffusion-webui-amdgpu-forge folder by clicking inside the explorer bar and type cmd, then hit enter.
Then run these three commands one by one:
venv\scripts\activate
python -m pip install setuptools==81.0.0
python -m pip install --no-build-isolation https://github.com/openai/CLIP/archive/d50d76daa670286dd6cacf3bcd80b5e4823fc8e1.zip
Then relaunch the webui-user.bat

[hidden night]

Imagine not using Invoke

[rough pecan]

First time setting up SD. Tried setting up ipadapter with comfyui but I keep getting "IPAdapter model not found" even though I have it installed and I can even see it when using IPAdapter Model Loader. What could be the issue?

[quaint rivet]

This usually means a mismatch or missing dependency even if the model shows in the loader.

Check these quickly:

Model is in the correct path: ComfyUI/models/ipadapter/

You are using the right version (SDXL vs SD1.5)

Required CLIP Vision model is installed (very common missing cause)

Restart ComfyUI after adding files

Make sure the file is .safetensors and not nested in extra folders

In most cases, it’s either missing CLIP Vision or SDXL/SD1.5 mismatch.

[chrome forge]

hey gang its me again now my face detailer is not working i have tried changing the sam loader to CPU but still same error

[ornate elk]

whats your gpu btw?
whats the complete cmd log?

[chrome forge]

Gimme sometime I'm doing a chore rn

[ornate elk]

and do you have the impact-pack subpack installed? thats needed for facedetailer

[chrome forge]

yep

@ornate elk here is the full log

here are the errors

bro why the hell cant i send the full log at once

smh

[chrome forge]

its rx 6700xt

[ornate elk]

thats the issue:
Traceback (most recent call last): File "C:\Ai\ComfyUI-Zluda\nodes.py", line 2227, in load_custom_node module_spec.loader.exec_module(module) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^ File "<frozen importlib._bootstrap_external>", line 1022, in exec_module File "<frozen importlib._bootstrap>", line 488, in _call_with_frames_removed File "C:\Ai\ComfyUI-Zluda\custom_nodes\comfyui-reactor\__init__.py", line 23, in <module> from .nodes import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS File "C:\Ai\ComfyUI-Zluda\custom_nodes\comfyui-reactor\nodes.py", line 17, in <module> from insightface.app.common import Face ModuleNotFoundError: No module named 'insightface'

no insightface package
Best would be to uninstall python 3.13
and then install python 3.12.11
then delete the venv folder inside comfyui-zluda and relaunch

[chrome forge]

okay!

[chrome forge]

Note: Python 3.12.11 has been superseded by Python 3.12.13.

what?

[ornate elk]

ignore that

download 3.12.11

oh sry 3.12.10

https://www.python.org/downloads/release/python-31210/

newer 3.12 versions dont have an installer, 3.12.10 is the last one

[chrome forge]

so now i deleted the venv folder how do i download it again?

[chrome forge]

okay i got comfy ui running

this is the first thing i see

and when i check the costum nodes folder the nodes are there so

[chrome forge]

@ornate elk Im still getting the error with face detailer

[ornate elk]

looks diferent know, but would need the whole cmd log

[chrome forge]

how do i send the whole cmd log discord i always cutting them up

[ornate elk]

you can copy and paste it, discord will ask to convert it into a .txt file

[ornate elk]

do you have installed visual studio build tools ?

[chrome forge]

i dont think so

[ornate elk]

thats needed for insightface to be build

you could also manually install the package if you download it here:
https://github.com/Gourieff/Assets/tree/main/Insightface
the one with cp312
then drop that into comfyui-zluda folder.
Then open up a cmd in there by typing cmd into the file explorer bar and hit enter.
then run following commands one by one:
venv\scripts\activate
pip install insightface-0.7.3-cp312-cp312-win_amd64.whl

[chrome forge]

this one right?

[chrome forge]

i got this error

[ornate elk]

thats not enough of the error

[chrome forge]

umm

[ornate elk]

numpy is to new

and pkg resources is missing

[chrome forge]

so now what

[ornate elk]

you have to downgrade numpy and install pkg_resources

with pip

[ornate elk]

open up the cmd like you did before then run these:
venv\Scripts\activate.bat
pip install pkg_resources
pip uninstall numpy -y
pip install numpy==1.26.4

[ornate elk]

ah okay then ignore that one

[chrome forge]

is this okay or

[chrome forge]

is there really no other way to make the face better without face detailer in comfyui like a automated workflow withouth face detailer

[ornate elk]

nope, but with a 6700xt i would use forge with zluda instead

there you have adetailer which is oneclick facefix

[chrome forge]

can i download it without deleting comfy ui?

[chrome forge]

now that i already have the zluda can skip this part?

[ornate elk]

yep

[chrome forge]

now to do this one i opened the webui.bat but it doesnt have COMMANDLINE_ARGS=

im so stoopid

[ornate elk]

webui-user.bat not webui.bat

[chrome forge]

yeah i got it 😭

[chrome forge]

i got this error

[ornate elk]

ohh damn forgot about that bug, but its easy fixable

[chrome forge]

oh how

[ornate elk]

Open up a new cmd inside the stable-diffusion-webui-amdgpu-forge folder by clicking inside the explorer bar and type cmd, then hit enter.
Then run these three commands one by one:
venv\scripts\activate
python -m pip install setuptools==81.0.0
python -m pip install --no-build-isolation https://github.com/openai/CLIP/archive/d50d76daa670286dd6cacf3bcd80b5e4823fc8e1.zip
Then relaunch the webui-user.bat

[chrome forge]

wat to do now

[ornate elk]

the lastcommand is wrong

the url is part of the 3rd command

[chrome forge]

ohhh

[ornate elk]

okay better install auto1111 with zluda, there shouldnt be any issue

you can simply delete the forge zluda folder

[chrome forge]

..\meson.build:1:0: ERROR: Unable to detect linker for compiler clang -Wl,--version
stdout:
stderr: clang: warning: unable to find a Visual Studio installation; try running Clang from a developer command prompt [-Wmsvc-not-found]
clang: error: unable to execute command: program not executable
clang: error: linker command failed with exit code 1 (use -v to see invocation)

A full log can be found at C:\Users\devaj\AppData\Local\Temp\pip-install-7xzghecu\scikit-image_d8a8a9832a99416fac0cece7ad383f6f.mesonpy-jbdr0dls\meson-logs\meson-log.txt
[end of output]

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

Encountered error while generating package metadata.

scikit-image

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

Press any key to continue . . .

[ornate elk]

open up a cmd and run
pip cache purge
then rerun the bat

if that doesn work i guess that auto1111 and forge doesnt ge ta long with python 3.12 and need 3.11 :/

[ornate elk]

ok yep python 3.12 is the problem

auto1111 and forge with zluda boath need 3.11.9 64bit

[chrome forge]

so um

what about comfy ui

[ornate elk]

comfyui needs 3.12.10

but you can have boath installed

then you would need to open up a cmd and run
where python
then copy the path to the 3.11.9 python exe
and paste that path into the webui-user.bat of auto1111 where ist says PYTHON=

then delete the venv folder and relaunch

[chrome forge]

i need to click add to path right?

[chrome forge]

okay i downloaded it now what

[ornate elk]

all explained above

[ornate elk]

Let me know if it worked

[ornate elk]

Can you share the cmd log?

[ornate elk]

Seems there is maybe a model and lora mismatch

Or vae

Make sure all of them are for the same type of model

Can you try if a normal image without any loras or VAE work?

[chrome forge]

okay so i got the txt to image working

now i need to fix the face esp the eyes

how to do that?

[ornate elk]

You need to go into the extensions tab, there click on
Available, then on Load From
Then in the list search for Adetailer
Install and wait
Relaunch the webui

Then you have adetailer as an option in txt2img

Just enable it then and it will fix the face when generating the next image

[chrome forge]

this one?

[chrome forge]

also do i have to wait for the checkpoint to compile everytime i open this

[ornate elk]

Nope zluda compiling is only one time

It can happen after driver updates again

Also make sure to enable Tiled VAE when going for higher resolution or upscaling

[chrome forge]

is this where you keep the .pt files?

the adetailer*

where do i put the embeddings

[ornate elk]

Adetailer files go into models/adetailer

Embeddings go into the textual inversion folder not in models

[silk estuary]

Is there anything newer than openoutpaint for expanding an image?

[quaint rivet]

Yeah

[graceful oar]

am i blind or is there no linux guide for ROCm in https://discordapp.com/channels/1002292111942635562/1002602742667280404/1248758425840648304

the guides are for windows only nvm

anyone knows where can i find one for linux?

[silk estuary]

Does it have a name?

[ornate elk]

Here is a Linux guide for arch:
https://github.com/CS1o/Stable-Diffusion-Info/issues/42

[graceful oar]

great, ty

[chrome forge]

my comfy ui is making images like this

[chrome forge]

i did move the checkpoint to the A1111 setup and moved it back

can that be the issue?

[ornate elk]

nope

[chrome forge]

my goat youre back

ive been waiting for you

save me from this

😭

[ornate elk]

can you show the workflow?

[chrome forge]

i didnt change anything btw

[ornate elk]

there is no clip layer -2

its either 1 or 2

[chrome forge]

it wont let me go into positive

[ornate elk]

then remove it entierly

its not needed at all

[chrome forge]

i tried same issue i added it cus i thought that was an issue

[ornate elk]

whats the model your using ?

[chrome forge]

https://civitai.com/models/376130/nova-anime-xl

[ornate elk]

thats not the one in the picture

there is shows somethin like copax...

[chrome forge]

same issue doesnt matter the model

[ornate elk]

try bypass the loras

[chrome forge]

i tried copax cus i thought the model had issue

[chrome forge]

its the loras i think BUT this image took 240s but normaly generating an image like this only takes about 60-120 seconds

[ornate elk]

you have an upscaler in your workflow that adds time

[chrome forge]

still with upscaler it did not go beyond 160s

[ornate elk]

check your vram usage in taskmanager and the cmd at which steps it takes so long

also tiled vae decode is better than the amd custom vae decode node

[chrome forge]

got prompt
model weight dtype torch.float16, manual cast: None
model_type EPS
Using split attention in VAE
Using split attention in VAE
VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
Requested to load SDXLClipModel
loaded completely; 1560.80 MB loaded, full load: True
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cuda:0, dtype: torch.float16
loaded diffusion model directly to GPU
Requested to load SDXL
loaded completely; 4897.05 MB loaded, full load: True
Requested to load SDXLClipModel
loaded completely; 4647.67 MB usable, 1560.80 MB loaded, full load: True
Unloaded partially: 35.61 MB freed, 1525.19 MB remains loaded, 3.38 MB buffer reserved, lowvram patches: 0
100%|██████████████████████████████████████████████████████████████████████████████████| 29/29 [02:26<00:00, 5.04s/it]
[OVUM_CUDDN_TOGGLE] torch.backends.cudnn.enabled set to False (was True)
Requested to load AutoencoderKL
Unloaded partially: 631.82 MB freed, 893.37 MB remains loaded, 37.51 MB buffer reserved, lowvram patches: 0
loaded completely; 217.50 MB usable, 159.56 MB loaded, full load: True
[OVUM_CUDDN_TOGGLE] torch.backends.cudnn.enabled set to True (was False)
Prompt executed in 182.63 seconds

without upscaler it took 180s also the colors looks wrong also this time i used Nova anime Xl

[chrome forge]

but now how would i use loras tho

[ornate elk]

lower their strenght and check for their model compatiblity

[chrome forge]

i just noticed somthing the images that comfy ui just generated looks like how A1111 would generate why is that? the images generated in comfy ui is much more detailed and better usually

[ornate elk]

the webuis dont have an effect on the image output, if you use the base workflow in comfyui you get the same quality as in auto1111

quality comes from the model, loras, upscaling

[chrome forge]

this is an image that it generated with the same workflow, also lowering the strength of the lora didnt help

[ornate elk]

then drop that image back into comfyui and check whats different in that working workflow

[chrome forge]

what?

[ornate elk]

every image contains the workflow you used for it

drop the image back into comfyui to get the workflow back

[chrome forge]

is there any specific thing for adetailer outside of the webui s? like and individual thing just for face detailer or somthing

[ornate elk]

yes there is facefusion

a standalone tool for face replacement and upscaling and faceswappign etc

[chrome forge]

can it work on anime faces? esp the eyes also can it be run locally?

[ornate elk]

why not use normal adetailer?

there is an eyes only model for it

on civitai

and yes it can run locally but facefusion is more for realism

[chrome forge]

the adetailer node is not working on comfy ui

[ornate elk]

i meant in auto1111

[chrome forge]

oh yeah, btw can i use comfy ui to generate the images then use adetailer in A1111 to refine the eyes?

[deep shuttle]

Hello, I have a problem. I recently upgraded my PC, here are the specs:
Device Name: DESKTOP-IN7OUMM
Processor: AMD Ryzen 7 9700X 8-Core Processor (3.80 GHz)
Installed RAM: 32.0 GB (31.2 GB usable)
Device ID: D4C1394D-2A08-4067-8E67-034613E4E39B
Product ID: 00331-10000-00001-AA844
System Type: 64-bit operating system, x64-based processor
Pen and Touch: No pen or touch input is available for this display
Graphics Card: AMD RADEON RX 9060 XT

I followed the guide in the pinned messages and followed all the steps, but when I run the A1111 installation, I get this:
Launching Web UI with arguments: --use-zluda --update-check --skip-ort --skip-python-version-check
Compilation is in progress. Please wait...

rocBLAS error: Cannot read C:\Program Files\AMD\ROCm\6.2\bin/rocblas/library/TensileLibrary.dat: No such file or directory for GPU arch : gfx1200

rocBLAS error: Could not initialize Tensile host:
regex_error(error_backref): The expression contained an invalid back reference.
Press any key to continue . . .

Just to give you an idea, I also tried installing SDNext and I get the exact same error when I try to generate images.

[prisma moss]

First mistake is amd gpu

https://tenor.com/view/akakichi-no-eleven-bro-bad-news-i've-got-bad-news-for-you-hand-on-shoulder-gif-4737126791591033318

[deep shuttle]

I've been using them since last year, so at this point, I don't even care anymore. The thing is, I managed to install it with three different AMD graphics cards—all of them older and worse—and this one, which is the newest model, is the one giving me the most trouble. I always had issues after the installation when adding extensions and other things, but usually not during the direct installation process itself.

[prisma moss]

Illd love to help but idk sry

Gotta wait for cs1o

[deep shuttle]

No problem, I was actually expecting it. The last time I had these kinds of issues, cs1o was the one who helped me. I just hope I have my computer with me when he/she decides to show up v:

[ornate elk]

Hey, your gpu doesn't need or shouldn't use Zluda at all.
You have to follow the setup guides with ROCm instead

You should also uninstall hip sdk 6.2 and old python versions

[long gale]

Hello.
I recently joined this group.
I still have a lot to learn, but I think it's important to have fun.
I'm looking for people I can easily talk to and who I can become friends with.
I'm all for Japanese people.
I'm interested in Japanese culture.
Nice to meet you!

[prisma moss]

@ornate elk get the guy above me right bro stinks of a bot/scammer, made discord yeaterday

And talks unrelated

[deep shuttle]

So for the first time, I have a GPU that meets all the requirements from that chart in the guides, and I don't need ZLUDA anymore? lol

I went to check the guide again, so I guess the one I should follow now is this one:

Any Webui with TheRock by AMD:

This is a very early setup of getting AMD's TheRock working with WebUIs like Automatic1111 / ComfyUI / Forge and forks.

This can also work with any other program that uses torch.

It currently supports:

• gfx1100
• gfx1101
• gfx1102
• gfx1103
• gfx1151
• gfx1200
• gfx1201

Requirements:

1. Download and install Git 64bit setup.exe from here: https://git-scm.com
2. Requires Python 3.12.10 to be installed before starting.

The steps are very similar for each WebUI.
Setup the preferred WebUI:

And thanks for replying!

[ornate elk]

Rdna4 is so new it doesnt need zluda. In fact its worth to use Zluda with it.
You should take a look at the Forge Neo with ROCm guide or comfyui with rocm guide

The "Any webui with rocm" is for people who like to get other webuis or stuff working with rocm.
As example I used forge neo in that case but the specific forge neo guide is more clear in steps.

[deep shuttle]

I've never tried installing Forge before, maybe I should give it a chance and save myself the headaches with A1111. I'll try it as soon as I get home, thanks for the help.

[ornate elk]

yes please dont try anything else.
dont mess with zluda, uninstall hip sdk 6.2, and older pythons below 3.12
then go with Forge with Rocm guide.
Its very simple and you can ask me if anything isnt clear

Ui is the same as auto1111

[deep shuttle]

I managed to install everything correctly and it's all working fine, but could you tell me which Open Pose Editor extension works for Forge? I'm looking for the one that lets me move the skeleton points in ControlNet

[ornate elk]

should be already integrated if you load an image, select openpose and then click the preview explosion button, and then edit

[deep shuttle]

The edit option doesn't show up for me; it just generates the skeleton and gives me the option to download it.

It should appear in the bottom right corner, shouldn't it?

[ornate elk]

yea normaly, but haven't used it in a long time

maybe the dev removed it

[torn cargo]

hey guys. I am having issues trying to install forge.
note: This error originates from a subprocess, and is likely not a problem with pip.
ERROR: Failed to build 'https://github.com/openai/CLIP/archive/d50d76daa670286dd6cacf3bcd80b5e4823fc8e1.zip' when getting requirements to build wheel

[prisma moss]

sec

[prisma moss]

show me the full console log

and copy paste it

yea this should be it

`venv\Scripts\activate

python -m pip install --upgrade "setuptools<82" wheel
python -m pip install --no-build-isolation https://github.com/openai/CLIP/archive/d50d76daa670286dd6cacf3bcd80b5e4823fc8e1.zip`

(if its haoming forge) this isnt for illyasviell

[chrome forge]

hey man how you doin

i fixed the face detailer RuntimeError: GET was unable to find an engine to execute this computation issue

you need to use a CFZ CUDNN TOGGLE between ksampler and vae decode and turn enable_cuddn to false that would fix the issue

please help others with this info if you see others having this problem ^^

[lusty granite]

hey guys how do i fix this video

https://cdn.discordapp.com/attachments/1447176971892297769/1496815472169975858/20260423_1309_New_Video_simple_compose_01kpwx0ndzet0td72sszqa33e5.gif

[chrome forge]

https://tenor.com/view/son-im-crine-son-im-crine-gif-5176045023480767429

[thorny kestrel]

Hey, I tried installing Forge Neo with ROCm with the guide but ended up with this message: "SystemError: Please update your GPU driver to support cu130 ; or manually install older PyTorch"
My GPU is GPU AMD Radeon RX 7900 XT and I have the most up to date driver installed. Already tried installing an older PyTorch version. Any tips what I can do?

[prisma moss]

Cu130 is nvidia stuff tho

Also @vocal burrow

[ornate elk]

Hey, at which step do you are of the guide?

[thorny kestrel]

The message first appeared at step 7 where it says to run the .bat and reappeared again after setting up ROCm.

[ornate elk]

That means you probably did not run the commands correctly to switch to rocms torch version.

You can check the torch version if you go into the venv/lib/site-packages folder
There screenshot the torch, torchvision folder

Then we see if its rocm or not

[bright dragon]

trying to install AUTOMATIC1111
these are the errors I get when I try to run the webui_user bat

I've tried following a bunch of troubleshooting steps posted by other people having this issue on github and reddit, as well as suggestions from chatgpt, nothing has worked so far. Just did a fresh reinstall of everything and restarted my computer

[vocal burrow]

auto1111's main branch is broken because clip removed some file from their github. cf screenshot
And overall A1111 is completly outdated, better switch to some version of force, comfyui, etc... If you really want to stick with A1111 for some reasons, you can check out their dev branch, it should be fixed on there.

[bright dragon]

I do definitely want to stick with A1111, any idea how to switch to the dev branch? I'm trying to follow some instructions I found online but they're not working either

[ornate elk]

Checkout the pinned messages for the updated guides

There is also how to get Automatic1111 working

[bright dragon]

works now, thanks

[prisma moss]

Install forge instead (a1111 ui but newer and maintained)

[thorny kestrel]

Re-installed everything again and is working now. Are there recommended settings for picture generation or is that base model specific?

[prisma moss]

model specific

if using big models like qwen, flux wan not so much

but for sdxl its 1mpx

[thorny kestrel]

Thank you

[chilly knot]

slow generation time
forge neo gpu 5070 ti
dont understand why
Two days ago, it was generating 5 seconds per image, but now it's 20 seconds.
Meanwhile, the graphics card is running at 100% in the monitoring, but it seems to be consuming insufficient watts.
140 watts, I seem to remember it was 250 watts before.

[chilly knot]

now its work fine
i only off this

strange

[ornate elk]

Did you had boath enabled?

[chilly knot]

now off all and generation work correct

fast

and use all gpu memory

i ebabled this in heavy models

and save settings as defaults

I used to think that these options help avoid errors but don't lead to such a drop in performance.

[ornate elk]

You should never enable Tiled unet

Only tiled vae

[prisma moss]

Does anyone know why my image sometimes turns black? It can happen on hrfix, or base gen, or even at a detailer phase (where the detailed zone gets black) at random,

@vocal burrow @ornate elk 👆

[vocal burrow]

Float precision error ? Anything in the log about that ?

[prisma moss]

Nope not really

I added some cleaning nodes between major parts of the workflow, have yet to check

From my understanding it simply has to do with dirtying the vram from all the loads/unloads

[vocal burrow]

Hard to tell without more infos.

Try a dichotomic approach

Just to isolate an eventual faulty node

Problem is that it s random so you ll have to run a batch of xxx inferences to be sure if nothing shows up in the log.

[prisma moss]

Ait ill do some testing idk when and ill look for errors if any

[vocal burrow]

Good luck if you have some logs share it there we ll take a look

(and the workflow)

[prisma moss]

Does it matter since im running the fp16 launcher, and one guy suggested using --force-upcast-attention arg

[vocal burrow]

That would help if those are float errors but it will coast you more VRAM.

[prisma moss]

@ornate elk 👆

[prisma moss]

no logs

[prisma moss]

i've since applied the upcast arg, will let know if it reappears

[celest delta]

is it safe to modify the max batch count in the .json of forge neo? like 300 instead of 128?

[celest delta]

thanks, dunno why they don't just let us put any number from the start

[silk estuary]

Several dumb but relevant questions: where are models loaded to? RAM or VRAM?
I think my forge is taking too long to swap checkpoints and a quick search advices me to change the number of loaded models to 2.

I didn't measure the time; but it took like almost 15 minutes to do it for 1.5 checkpoints.

Specs: 32 gb ram, 2060 6GB VRAM, ryzen 7 4800H

[prisma moss]

did u disable the igpu? (i assume that cpu comes with one)

[silk estuary]

I didn't, should I? What if the main gpu ever has problems and I need to rely on the igpu?

[prisma moss]

then you turn it back on

until it doesnt

you turn it off

also can try '--cuda-device 1'

or 0 depending on what id is the actual rtx 2060

[silk estuary]

hmm can that be done without a GPU? I was under the impression you'd get a black screen if that happened

[prisma moss]

no

in bios u have an option, it would default to main gpu anyway then

[celest delta]

How do i change the folder where the LORA's are picked? Is it possible to have more than one folder? (i mean two in different places, like different drives and so on)

[celest delta]

tried to use symoblic link to have two different drives, but it doesn't do what i want... like if i move files to the new linked folder, it still has to transfer the files physically so it doesn't change anything in the end, i don't want to move all my loras unless it's my last choice (since it will take a long time)

[sonic pine]

I'm getting this error message when trying to launch forge neo. It was working fine a day ago

[abstract cloud]

Hey guys i was using fourge ui and now i am shifting to comfyui but the thing is my loras dose not work the same

Now the face is changed

[quaint rivet]

That’s normal when switching to ComfyUI your LoRA isn’t being applied the same way.

The face change usually comes from different base model, LoRA strength or missing trigger words. ComfyUI is more sensitive, so even small setting differences will change the result.

[quaint rivet]

If you want, I can take your current workflow and match it to your Forge results exactly same face, same consistency, no guessing. Just send what you are using now and I will fix it for you. @abstract cloud

[abstract cloud]

I am using tigger word correctly

[quaint rivet]

Got it if your trigger word is correct, then the issue is definitely not that.

It’s most likely your base model, LoRA strength or the way ComfyUI is loading the lora

[quaint rivet]

Even small differences there will change the face a lot.

[silk estuary]

There is a way using commandline arguments. add -lora-dir "" and write the folder location in between "". As for if more than one folder can be used, idk. I've been wondering the same for a while. I have illustrious and pony libraries in different drives so I wondered if they could be made to run together in the same instance.

[celest delta]

i found the solution, with symbolic link, at first it wasn't working cause my other drive wasn't NTFS

https://www.howtogeek.com/16226/complete-guide-to-symbolic-links-symlinks-on-windows-or-linux/

[silk estuary]

Cool info. I know of a few programs that do things like this, mostly mod managers for games. I always thought we, mortal users, could not use it. Thanks for the info.

[celest delta]

is there y way on forge neo to fully restore a closed windows? Like i did like 200 pic generation, with lots of different loras, one per pic, and my windows suddenly closed and now it's all gone... Yes i know i can see the image in my output folder, and i can put them in the PNG thing to see which lora was associated it with, but it's laborious instead of just clicking the next image in the browser and seeing the lora associated with it directly

[ornate elk]

Try delete the venv folder and relaunch

[prisma moss]

maybe infinite image browser extension, check DM

[chrome forge]

hey its me again, is there anyway to speed up the generation process?

[ornate elk]

Depends, what's your gpu again? Ans which Webui do you use

[remote thunder]

Hi can anyone help I've downloaded stable diffusion but i can't generate anything

[chrome forge]

comfyui with zluda, rx6700xt

[remote thunder]

it says TypeError: 'NoneType' object is not iterable

[ornate elk]

Hey, what's your GPU?

[ornate elk]

You could try setup sage but you have to check on the comfyui-zluda github how to

[remote thunder]

rtx 5070

[ornate elk]

then you should check the pinned messages of this channel for the updated guides.
There follow the (Nvidia) Forge Neo setup guide

its recommended for new gpus like rtx50 series and has the best performance

auto1111 and normal forge are outdated

[sleek dove]

Hi everyone, I need some help because I'm trying to learn ComfyUi, but there are still some things that are difficult for me. I have Python 13, with ComfyUi portable, and I can't (or so I read) use Pulid with this version of Python. Do you know if there's an alternative or if there's a way to fix it? Unfortunately, I only have 8GB of VRAM. Thank you

[lucid yoke]

hii, what gpu are you using?

[sleek dove]

I'm using an NVIDIA RTX 3050 with 8GB of VRAM, so I'm a bit limited with heavier workflows 🙁

[lucid yoke]

thats not a problem at all with SDXL/ illustrious based models, any reason why spesifically comfyUI?

i used to rock it just fine with a 3070TI with 8gb

oh well got any logs so i see what its saying?

[sleek dove]

Actually, I don't usually have any problems either. I first used automatic1111, then I tried ComfyUi and immediately liked it. I'm not very knowledgeable about what else I could use. I can usually do everything I need even with 8 GB, but my only problem is finding a functional way to transfer faces since I like to train Lora.

[lucid yoke]

hmmm well theres two ways i personally would tackle it.

1. i would downgrade my python version but without logs i cant say for certain. as it shouldnt be a problem normally.
2. use a UI that has an extension for faceswaps like SwarmUI (its basically comfyui but with an extra layer but you might run into the same issue again)

[sleek dove]

Thanks, I'll try that then! If it doesn't work, I'll revert to the version I already had. 🙂

[lucid yoke]

sounds good

[remote thunder]

Guys i tried downloading forge and i didn't work for me, it might be due to me having an rtx 5070, would forge neo work?

[lucid yoke]

oh yeah neo is optimized for 5000 series cards

do you follow the guide in the pinned messages?

[remote thunder]

yeah i just checked

[prisma moss]

@vocal burrow im just letting u know that the command upcast whatevre solved the black issues

been a few days now with thorough testing

so if someone asks the same question u know 🙂

[vocal burrow]

ok, somewhat good news then. Shame that it's gonna cost you more vram tho.

[prisma moss]

got 24

😈

https://tenor.com/view/sukuna-sinister-malevolent-evil-gif-15342583514003527998

[vocal burrow]

something something die of thirst while the other something something

enjoy it then :p

[abstract cloud]

can anyone help me i want to do image to prompt in comfyui

[lucid yoke]

hey, what kind of image to image? like style transfer or using it as a initial image to edit?

[prisma moss]

I wants prompt image hlp plez how to 😭

[lucid yoke]

your asking on how to make better prompts? or want to use image prompting?

[prisma moss]

No no lol, im just poking fun at the guy above

[lucid yoke]

well im sure hes looking for a node that loads a LLM to analize the image and give a prompt for that

if i understand it right

[prisma moss]

If thats that he can use use wd tagger no? If booru tags, i think ollama or florence something something can output NL prompt

[lucid yoke]

possibly, using a seperate instance of like LMstudio could also archive that

swarmUI also has an extension for prompting like that iirc

[rigid nymph]

Anyones app also not working?

[lucid yoke]

What app?

[pseudo imp]

Hey guys! ^^ I'm not sure if this is the right place to ask
Buuut, with ComfyUI, there's a bunch of extensions and nodes that allow you to fill in the mask, (for example, I can just inpaint a circle over the face, and the mask fill will just fill in the rest automatically when I click Run or Generate. Is there an extension that does the same thing for A1111? 👀

[quaint rivet]

Yeah I get what you mean

In Automatic1111 Stable Diffusion WebUI, there isn’t a built-in auto mask fill like in ComfyUI you still have to define the mask manually. But you can get a very similar result using extensions like Inpaint Anything or OpenOutpaint, which let you quickly select or paint an area and have it filled more automatically.

If you want, I can help you set it up to behave almost the same as what you are used to. @pseudo imp

[pseudo imp]

Thank you for replying 😄 That would be awesome thanks. I'll ask a question first before saying yes to the help haha. I've heard of Inpaint Anything, but I'm not really sure how it works exactly, but for example, with my workflow, I use Only Masked a lot to add detail to things, rather than using Adetailer, cus I personally think it looks better that way, and it generates faster than using Adetailer.

What I'd like to know about Inpaint Anything is, can I edit the mask when it masks something automatically? Like, it might end up covering an area I've already edited and I wouldn't want to change!

[prisma moss]

What it is that you are looking for exactly? Inpainting from forge in comfy?

[pseudo imp]

Ooh, I use Forge Neo right now, so I'm aware some extensions don't work.
I did try and get into ComfyUI, but things ended up taking longer with the way I work for my pics.

[lucid yoke]

Hey, dont get lured into random dm's. But if i look at it right you want a extension that detects the face and "fixes" it?

[pseudo imp]

It's okay, I can tell when people aren't bots, don't worry 😂 Someone is helping me in DMs currently. I didn't want to spam the #🤝｜tech-support channel haha!
But nah, just exactly what I'm asking for basically, where I can inpaint/mask an outline around something without completely filling it in.

[lucid yoke]

Becareful of random links because theres definitely a adetailer extension for a1111/forge

Oops misread your earlier message mb

[pseudo imp]

Thank you! ^^ I will be. Adetailer isn't what I'm looking for, I'm okay for finding popular/well known extensions like that haha

Haha that's okay 😄 Thanks for looking out for me though

[lucid yoke]

🫡 goodluck

[vague creek]

Can I ask about a1111 here?

Having a hard time finding a place to find answers

[prisma moss]

sure

first install forge instead of a1111 if u havent (same ui, newer)

[vague creek]

I just did some poking around and found out about forge neo. If this works with 50xx gpu it will end two whole days of angst and grief

[prisma moss]

Sure

[vague creek]

I am so glad I found this discord, im ecstatic

it freegin WORKED!!!

i feel like I passed a kidney stone

[lucid yoke]

oh yeah in the pinned messages here there was a guide, always welcome to ask questions here

[vague creek]

Thankyou kindly

[vague creek]

what do i do with this info? Is it something to edit into batch file? Hint: your device supports --cuda-malloc for potential speed improvements

[prisma moss]

its an argument you can use

[chrome forge]

hey is there anyway to load images 1 by 1 from a folder and upscale one by one?

in comfy ui

[prisma moss]

If you are just looking for batch upscaling illd look at smth like seedvr, or externally gigapixel, if u actually want to denoise something, then not really, maybe with a custom node. But thats an issue in of it self because you can pull out prompt for each upscale for denoise pass

[chrome forge]

yeah i just have image in a folder and i want each image to be upscaled one by one automaticaly (please ping me when you reply or i wont see it)

[prisma moss]

@lucid yoke

@vocal burrow

This server srsly cant catch a break lol

[lucid yoke]

Well its a big server

[round sail]

I am using forge-neo and have managed to get wan2.2 img2vid to work but, I am not sure how to correctly word the prompt to get the correct actions/movements. Also how do you use wan 2,2 LoRAs in the img2vid tab?

Is that TRUE? OR is there just NO OTHER Wan 2,2 img2vid models to laod. I can only find the 14B models for img2vid, yet the txt2vid modesl like wan2.1_t2v_1.3B_fp16.safetensors DO EXIST. Why Is THAT?

[ornate elk]

for img2video only wan 2.2 14b is supported. 5b doesnt work.
txt2video and txt2img works also only with wan 2.2 14b in forge neo

2.1 shouldnt be used anymore as they are far worse than 2.2

[round sail]

OMG, thank you. I am so so tired of arguing with ai to figure out how forge-neo works. I will look for the wan 2.2 version of this:wan2.1_t2v_14B_fp8_e4m3fn.safetensors, which I managed to get to work, but it is not Ideal. I am using the Nvidia Geforce RTX 4070 Ti Super (16gb) Intel(r) Core(TM) i7-147000kf (32gb and the 14B models just barely fit, MY next attempt is to replace the Wan2.1_T5_Encoder.safetensors with the smaller WanSmall-umt5-xxl-encoder-Q8_0.gguf textencoder. Does that sound like a viable option?

Ooops I actually meant the wan2.1-i2v-14b-480p-Q4_K_M.gguf

[ornate elk]

np, and yep sounds, good, gguf is supported also important if you go with high and low noise wan boath need to fit in your vram + a bit of free space
so dont use models that are 14gb, better go with 10-12gb

[round sail]

I cant find any wan 2,2 img2vid model that are smaller than 14b

[ornate elk]

14b is not the size its the version (trained paramaters for the model)

sizes are FP16, FP8, Q8, Q6, Q4 etc

[round sail]

OK, so the Q4_K_M model (11gb) should work?

where can I find info about how to do movement prompts. There doesnt seem to be a lot of places to get information aboy img2vid and especially when it come to the forge-neo implementation

Is there a discord for ForgeNeo? I tried to go to LibreGaming Discord but it wont let me. I am new to discord and cant seem to find a forge-neo one

[ornate elk]

there is no one for forge neo directly but it has a wiki where it shows some example usages of the models

https://github.com/Haoming02/sd-webui-forge-classic/wiki/Inference-References

[round sail]

thx for that link, does the creator of Forge-Neo fork have a patronage site?

[ornate elk]

only a ko-fi

but only for donations not other stuff

[round sail]

Not sure what a ko-fi is (I will look into it). I am really impressed with the neo fork, after using python scripts, A1111 and the Forge-UI this fork seems to be the best one yet.

[ornate elk]

yep its the most updated and feature packed version

also the only one that is really actively maintained

[round sail]

Thanks again, I will review the info you supplied. I am NOT a researcher or a PROGRAMMER, just a user and I love this version. I will look over the GitHub page and keep using this discord server. Trying to use Ai for technical details of new software like Forge-Neo is a frustraiting Nightmare.

[ornate elk]

yep ai doesnt help much in that case, feel free to ask here, a lot of users with knowledge can help

also if your into anime, checkout forge couple extension and the anima models

[round sail]

Ai is telling me this: There are no other WAN 2.2 model releases anywhere:

No WAN 2.2 I2V single-file safetensors
No WAN 2.2 I2V single-file GGUF
No WAN 2.2 I2V fp8
No WAN 2.2 I2V Q4_K_M
No WAN 2.2 I2V 480p full model
No WAN 2.2 I2V 720p full model
No WAN 2.2 I2V “Forge‑Neo ready” model
If they existed, they would be in one of those two places.
They are not.

So, how would I use a wan 2.2 model in Forge-neo?

[ornate elk]

stop using ai it doesnt have any clue

wan 2.2 is the latest open source Wan version out there

Wan 2.5 exists but is paid cloud only

here are the gguf files of wan 2.2 14b:
https://huggingface.co/QuantStack/Wan2.2-I2V-A14B-GGUF/tree/main

[round sail]

That only shows a Wan 2.2 High or Low versions. Can I use just ONE of those. I thought they were required to both be loaded? How does that work?

[ornate elk]

you have to enable the refiner in settings to load boath

using boath gives a better result

low has to be then selected as refiner

you can see it in the examples from the other link

[round sail]

Ok, I found that setting and enabled it. Should I also enable :Switch based on "steps" instead (by default, Refiner swaps the model based on "sigmas" to match Wan 2.2 's behavior)

[ornate elk]

no leave it default

[round sail]

Ok, thx - I will try the Wan2.2-I2V-A14B-HighNoise-Q4_K_M.ggu and Low version. Why do I only see the refiner in the img2vid tab? It sounds like you are saying load the Hi ver in the model check point and then the Low ver n the Refiner. What is the Switch at Point?

[ornate elk]

0.835

is the default one

[round sail]

Ok, Ill play around with that.
Are there any Usefull LLMs for forge-neo instructions. I have been just using the free ai chatbots and they are pitifull. Is there a LLM that I can download for use in LM Studio that would provide better info?

[ornate elk]

hmm idk about that, but maybe gemma4 or qwen 3.5

[lucid yoke]

Youve been using local ai for instructions? But yeah Gemma is amazing for prompts

But for asking help i think only the cloud ones are decent to a degree when debugging or if you mostly know what your doing anyways

[round sail]

Unfortunately, I don't what I'm doing, but I'm learning. Thank God I found this discord. Thx

I have been really upset with the LM Studio Upgrades, has anyone used GPT4Al. I am thinking about trying it out as a replacement ( I had to down grade LM Studio to an earlier version just to keep using it)

[lucid yoke]

I have to say jumping into video ai without prior ai experience is though, local llm's are great for specialized tasks but no where as accurate as the cloud ones

Wan 2.1 is easier to use aswell with the causvid lora and it's still decent but personally in like the latest LTX one as it's fast and does audio too

[round sail]

I could never get the LTX extension to work in Forge-Neo. Have you tried local-tts-studio for Audio, it will replicate a voice from just 2 or 3 small 30second clips. It's pretty accurate. I am not getting into video gen for any commercial use, Im retired and just wanted to play around with it. I have created a dozen or so 5 second vis in forge-neo using wan 2.1, but the last couple of seconds always seem to get weird. I thought Wan 2.2 might be an improvement. I am GUNG HO, on everything LOCAL. (FYI: https://github.com/cunicopia-dev/local-tts-studio)

[lucid yoke]

Ah the old TTS, i mess with a copy of Vibevoice that MS took down. Though no i usually use a instant TTS model (like chatterbox 0.5b iirc) for my llm speech

[regal stream]

My command prompt is saying "Getting requirements to build wheels did not run successfully": What do i do?

[lucid yoke]

Don't join that server, its a scam

Hmm can you post the cmd log?

[regal stream]

wdym?

[lucid yoke]

O a spambot replied to you with a scam link

If you missed it good

[ornate elk]

you probably try to install auto1111, but its outdated.
whats your gpu? there are forge neo or forge classic as alternatives

all guides are in the pinned messages of this channel

[regal stream]

I have a 1650 super

[ornate elk]

okay then go with the forge classic guide

[regal stream]

Im trying to install this to upscale some videos

But yes. Its auto1111

[lucid yoke]

Hmm a1111 is broken since it's outdated yeah

[regal stream]

What do i do instead?

[regal stream]

?

[ornate elk]

you should checkout the pinned messages of this channel, there you find the updated setup guides
there go with the Nvidia Forge classic guide

[prisma moss]

only thing u can upscale is a thumbnail with that thing

[regal stream]

Nuh uh. I got help and it worked perfectly

[neon token]

Does anyone know of any good upscaling methods or have any good worklfows for forge? my usual methods seem to be giving me poorer results now so I'm looking for new methods.

[lime sandal]

4x-UltraSharp ?

https://huggingface.co/lokCX/4x-Ultrasharp/tree/main

[lucid yoke]

I usually use this one
https://openmodeldb.info/models/2x-AnimeSharpV4

[lime sandal]

good to know

[neon token]

I'll give both of those a try

[gilded tartan]

Hi I.m Daniel I would like access to the pdf

[lucid yoke]

Pdf? What

[prisma moss]

🤨

[low bough]

Someone know how to start torch on 6000 series amd gpu?

On rx 6700xt, sd portable

[ornate elk]

Check the forge neo with rocm guide from the pinned messages

It should work for your gpu. If not let me know

[low bough]

where this pinned message? In what chat?

im blind

thx

[lucid yoke]

Don't join that server@low bough

Its a scambot

They meant this guide @low bough
#🤝｜tech-support message

[ornate elk]

The needle icon on the top right

[lucid yoke]

Gosh those bots snipe in so fast some days

[low bough]

thanks for help!

[low bough]

Ai starts attack human race i guess)))

[lucid yoke]

But yeah, all help is offered here but we would never redirect someone to a suspicious server using a brand new account/bot like reply

[orchid patio]

this creator 893019293

[lucid yoke]

? Whats up

[orchid patio]

llik yu wiza family

yu wize kill

[lucid yoke]

i have no idea what your talking about, could you try using google translate?
https://translate.google.com

[wet seal]

Hi, I was planning to upgrade from my really old RX580 and I the most cheapest option I could find were RTX5050 and ARC B580 both for around $300. Others GPU out of that price are abt $100 more and getting out of my budget. My question is: Which one is better for Stable Diffusion SDXL, and also some quantized WAN models in Comfy UI? The B580 have 12gb VRAM which is better from what I know. However, the RTX5050 has CUDA and Tensor cores, and can use the some NVIDIA exclusive technologies as Xfromer. Which one should I chose?

[prisma moss]

I would buy a used 3080 12gb at that point

Aint worth buy that b or 50 junk

Unless what that guy is saying is true about b580 being faster than 5050, should compare it to 3080 then

[civic nova]

Hi all, totally an end user type person so baby steps are appreciated... I had SD on my PC for a while, then when I had a SSD Issue the repair shop totally scrambled the entire thing and broke all my links. Now when I try to use SD I get the following error. stderr:
Python was not found; run without arguments to install from the Microsoft Store, or disable this shortcut from Settings > Apps > Advanced app settings > App execution aliases. Suggestions? I have the full Stable Diffusion folder and files on my F Drive but even trying to run the BAT file directly I get the same error.

I found the python.exe file it is located in sub folders of the Stable Diffusion folder: Stable Diffusion\stable-diffusion-webui\venv\Scripts

[vocal burrow]

What do where python, python --version and py --list-paths say ? (run those in cmd)

[civic nova]

python command sends me to install windows url python version says not found

All I find is python 3.13 but I think I read somewhere that I need like 3.10 or 3.11

[vocal burrow]

can you screenshot all of it please

[civic nova]

That what you need?

[vocal burrow]

ok so you don t have python installed. Maybe you used to, now you dont anymore

a couple of things

Dont ever modify the webui.bat, it will make your life way harder to update things

webui-user.bat is made for user to be customized and should have all you need.

[prisma moss]

this forge or summin?

[civic nova]

so which pythod to I need and where to find it please

[vocal burrow]

(i m getting to it dont worry)

[civic nova]

NP

[prisma moss]

https://tenor.com/view/tigger-tigger-leaving-tigger-sad-tigger-meme-meme-gif-1861164940938639699

[vocal burrow]

it s kay it comes frome a good place i get it

[vocal burrow]

Secondly, the venv folder is kinda like a "virtual machine" for your python (which you don t have anymore). That s where python will copy paste all of its stuff to be able to work without messing your whole system. It s keeping all its stuff neatly tied to one place.

But as mentioned you mentioned you don t have python anymore. So you can safely delete it, please do.

[civic nova]

Delete entire Venv folder?

[prisma moss]

you should do that ye

[vocal burrow]

yes

Thirdly, you need to reinstall python but like you said you need a specific version.

[civic nova]

ok dleted

deleted

[vocal burrow]

So finally, do you remember if that is Automatic1111's stable-diffusion-webui or some flavor of forge's stable-diffusion-webui ?

[civic nova]

As far as I know it is just the basic one I got a few years ago and I think there was an update.

[prisma moss]

if it was a few years ago is prolly a1111

how long ago we talkin

[civic nova]

Anyway to do a version check?

[vocal burrow]

ok so most likely A111 according to what you re saying + the format of webui-user.bat

[prisma moss]

what does it say in webui.py

[civic nova]

checking

[vocal burrow]

you can open a cmd in stable-diffusion-webui folder and run git remote show origin

[prisma moss]

https://tenor.com/view/write-it-down-take-notes-master-at-work-gif-15660117309056202885

[civic nova]

I am showing a Webui cpython-310 pyc and a Webui.py open py with note pad?

[prisma moss]

try the origin first

[civic nova]

ok as stated I am basically a end user these days, last time I did any real command stuf was with DOS 7.0 LOL Yes I am O_L_D

[prisma moss]

type cmd here

enter

[prisma moss]

run

[vocal burrow]

it s okay commands should be the same, it s still cd xxx to change directory. Or you can do as Kurton mentioned and type cmd then press enter in the top bar and it should open a cmd automatically at this folder

[civic nova]

ok sorry about this not as geeky as I use ot be 40 years ago.

yes I am pushing 70