🤝｜tech-support

[lucid yoke]

its not consumer friendly ill give you that.

[neon pecan]

its bullshit. it says amd but the dumb thing asks for torch which is nvidia but apparently it can be used for amd but when I try to download it the annoying thing says it cant find a satisfying version

Why tf does it need all this stupid shit???

it wants torch but doesnt have the working version but the amd tutorial says to use pythong 1.12 or whatever but apparently it needs torch which doesnt work for that version of python so it's just a cycle of stupidity

it's literally impossible. so wtf is with these instructions telling me to do things the software CANNOT do???

[lucid yoke]

it has worked for many other people so far so im sure theres something silly going on and getting upset doesnt really help the situation

[neon pecan]

BECAUSE THE INSTRUCTIONS TOLD ME TO GET A VERSION OF PYTHONG THAT TORCH CANT WORK WITH AND THE DUMB THING ASKS FOR TORCH BECAUSE IT DOESNT HAVE THE CORRECT VERSION OF PYTHON!!!!

Well ive been working at this for two hours. so i'm starting to get angry due to this stupid cycle

I just want to start using the ai already, im already sick of this stupid file crap

but cant because it's literally impossible due to the conflicting requirements

I have a literal headache...

[lucid yoke]

i understand that but this isnt really helping, i really recommend taking a break and waiting a bit even if you dont have the patience for it

[ornate elk]

A 9060xt dont need and shouldn't use zluda. Go with out ROCm guides from the pinned messages

[neon pecan]

awe I got to install everything again?

[ornate elk]

That reply was to someone else above, but What's your gpu?

[neon pecan]

oh thank goodness lol

[neon pecan]

radeon rx 9060 xt

I was following your instructions for zluda for amd. But despite following your steps I got stuck on setup because it keeps wanting torch

[ornate elk]

Then its the same for you, don't use zluda

The rdna4 GPUs are supported by rocm and should only use that

Follow the Forge Neo with ROCm Guide for example

[neon pecan]

already doing that

[neon pecan]

is the anime or realism models required? Cause I prefer it to copy styles from reference images

[ornate elk]

You can use any model

[neon pecan]

what model would you recommend I use for what im after?

I came from adobe firefly and I like how it uses different styles and even mimics the art from reference images

so im not looking for a single style

or more precise im not looking for a model that makes a single specific artstyle

[ornate elk]

You could try z-image-turbo then, or Flux Klein

Those aren't biased and can replicate a lot of styles

[neon pecan]

so are they about the same or is one better than the other?

[ornate elk]

If you want something for Anime then go for Illustrious based or Anime based models

[ornate elk]

Z-image-turbo is faster

[neon pecan]

but less efficiant?

[ornate elk]

No, they are just different

They perform sometimes better and sometimes worse then the other

[neon pecan]

so it wont matter which one I get?

[ornate elk]

If you want to generate new images then start with z-image-turbo
If you want to apply or change styles of an already made image then go for flux klein in img2img

[neon pecan]

z image turbo from hugging face?

or do I download it somewhere else?

[ornate elk]

For models get them from Civitai.com

Some models also need additional files like a text encoder and vae to work
You can get these info here:
https://github.com/Haoming02/sd-webui-forge-classic/wiki/Download-Models

[neon pecan]

thank God, im trying to get it from hugging face and its a huge pain

[neon pecan]

im asking cause the preview got me wondering. it can imitate 2d art right?

and it goes in the model folder right?

[ornate elk]

Models/stable-diffusion

[ornate elk]

Yes it can, z image turbo and flux Klein follow natural language prompts

[neon pecan]

IT WANTS TORCH AGAIN

I KEEP GETTING THE SAME ERROR

ERROR: Could not find a version that satisfies the requirement torch==2.11.0+cu130 (from versions: none)
ERROR: No matching distribution found for torch==2.11.0+cu130

[neon pecan]

Why do I keep getting this error?

this is the third time in a row

why is it doing this?

[ornate elk]

Can you show the complete cmd log?

[neon pecan]

Python 3.11.9 (tags/v3.11.9:de54cf5, Apr 2 2024, 10:00:00) [MSC v.1938 32 bit (Intel)]
Version: neo
Installing torch and torchvision
Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/cu130
ERROR: Could not find a version that satisfies the requirement torch==2.11.0+cu130 (from versions: none)
ERROR: No matching distribution found for torch==2.11.0+cu130
Traceback (most recent call last):
File "F:\ai\ROCm ai\sd-webui-forge-neo\launch.py", line 56, in <module>
main()
File "F:\ai\ROCm ai\sd-webui-forge-neo\launch.py", line 43, in main
prepare_environment()
File "F:\ai\ROCm ai\sd-webui-forge-neo\modules\launch_utils.py", line 316, in prepare_environment
run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True)
File "F:\ai\ROCm ai\sd-webui-forge-neo\modules\launch_utils.py", line 86, in run
raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't install torch.
Command: "F:\ai\ROCm ai\sd-webui-forge-neo\venv\Scripts\python.exe" -m pip install torch==2.11.0+cu130 torchvision==0.26.0+cu130 --extra-index-url https://download.pytorch.org/whl/cu130
Error code: 1
Press any key to continue . . .

it just keeps bugging me about torch

[ornate elk]

You have the wrong python version

Uninstall 3.11 and install 3.13.12

Then delete the venv folder and then relaunch. Then do the rocm steps again (after step 7)

[neon pecan]

how do I uninstall it?

[ornate elk]

Like any other software

From the windows apps or system control

[neon pecan]

I looked there, it's not on the list

I figured maybe it needs a special way to uninstall

[ornate elk]

Normaly not.
Can you open up a cmd and run
python --version

That should show the installed versions

[neon pecan]

ok I found them this time. dunno why it wasnt listed the first time. my bad

[neon pecan]

I can't. I couldn't install torch because it cant find any that satisfies it

[neon pecan]

(venv) F:\ai\ROCm ai\sd-webui-forge-neo>pip install torch torchvision torchaudio --index-url https://rocm.nightlies.amd.com/v2/gfx120X-all/
Looking in indexes: https://rocm.nightlies.amd.com/v2/gfx120X-all/
ERROR: Could not find a version that satisfies the requirement torch (from versions: none)
ERROR: No matching distribution found for torch

torch is causing me problems again

[ornate elk]

When you relaunch the webui-user.bat, can you share the log?

[neon pecan]

venv "F:\ai\ROCm ai\sd-webui-forge-neo\venv\Scripts\Python.exe"
Python 3.13.12 (tags/v3.13.12:1cbe481, Feb 3 2026, 18:12:42) [MSC v.1944 32 bit (Intel)]
Version: neo
Installing torch and torchvision
Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/cu130
ERROR: Could not find a version that satisfies the requirement torch==2.11.0+cu130 (from versions: none)
ERROR: No matching distribution found for torch==2.11.0+cu130
Traceback (most recent call last):
File "F:\ai\ROCm ai\sd-webui-forge-neo\launch.py", line 56, in <module>
main()
~~~~^^
File "F:\ai\ROCm ai\sd-webui-forge-neo\launch.py", line 43, in main
prepare_environment()
~~~~~~~~~~~~~~~~~~~^^
File "F:\ai\ROCm ai\sd-webui-forge-neo\modules\launch_utils.py", line 316, in prepare_environment
run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True)
~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\ai\ROCm ai\sd-webui-forge-neo\modules\launch_utils.py", line 86, in run
raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't install torch.
Command: "F:\ai\ROCm ai\sd-webui-forge-neo\venv\Scripts\python.exe" -m pip install torch==2.11.0+cu130 torchvision==0.26.0+cu130 --extra-index-url https://download.pytorch.org/whl/cu130
Error code: 1
Press any key to continue . . .

[ornate elk]

Have you run the uninstall commands before?
And can you open up a cmd and run
Pip cache purge
And then redo the rocm steps

[neon pecan]

(venv) F:\ai\ROCm ai\sd-webui-forge-neo>pip uninstall torch torchvision torchaudio rocm rocm-sdk-core rocm-sdk-devel rocm-sdk-libraries-custom -y
WARNING: Skipping torch as it is not installed.
WARNING: Skipping torchvision as it is not installed.
WARNING: Skipping torchaudio as it is not installed.
WARNING: Skipping rocm as it is not installed.
WARNING: Skipping rocm-sdk-core as it is not installed.
WARNING: Skipping rocm-sdk-devel as it is not installed.
WARNING: Skipping rocm-sdk-libraries-custom as it is not installed.

(venv) F:\ai\ROCm ai\sd-webui-forge-neo>pip install torch torchvision torchaudio --index-url https://rocm.nightlies.amd.com/v2/gfx120X-all/
Looking in indexes: https://rocm.nightlies.amd.com/v2/gfx120X-all/
ERROR: Could not find a version that satisfies the requirement torch (from versions: none)
ERROR: No matching distribution found for torch

(venv) F:\ai\ROCm ai\sd-webui-forge-neo>

it really wants me to use torch

[neon pecan]

I give up. its been seven hours and almost 5am. Let me know if there's a solution to this obnoxious torch error

[ornate elk]

Its very strange that you get that error.
I have to retest the install steps later.
I'll let you know

@neon pecan Do you have any antivirus software installed?

[neon pecan]

malewarebyte (free version)

[ornate elk]

hmm i think that should be fine, i then would recommend deleting the whole rocm ai folder and redo the whole setup

[knotty turret]

Comfyui have any setting to delete output eveything i done yesterday got deleted there only things i did today

[supple osprey]

hello guys

while installing comfyui on ubuntu, I got this error

it is run on venv

[lucid yoke]

disk quota exceeded? looks like you dont have enough space allocated for it?

[supple osprey]

how can i change it?

[lucid yoke]

im a windows user , i cant tell you that but is there by any chance your disk is full?

[supple osprey]

no , it is freshly installed os

Imma ask Claude then

[neon pecan]

ERROR: Could not find a version that satisfies the requirement torch==2.11.0+cu130 (from versions: none)
ERROR: No matching distribution found for torch==2.11.0+cu130
Traceback (most recent call last):
File "F:\ai\ROCm ai\sd-webui-forge-neo\launch.py", line 56, in <module>
main()
~~~~^^
File "F:\ai\ROCm ai\sd-webui-forge-neo\launch.py", line 43, in main
prepare_environment()
~~~~~~~~~~~~~~~~~~~^^
File "F:\ai\ROCm ai\sd-webui-forge-neo\modules\launch_utils.py", line 316, in prepare_environment
run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True)
~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\ai\ROCm ai\sd-webui-forge-neo\modules\launch_utils.py", line 86, in run
raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't install torch.
Command: "F:\ai\ROCm ai\sd-webui-forge-neo\venv\Scripts\python.exe" -m pip install torch==2.11.0+cu130 torchvision==0.26.0+cu130 --extra-index-url https://download.pytorch.org/whl/cu130
Error code: 1
Press any key to continue . . .

no dice. Still having torch problems

[ornate elk]

have you installed python from the offical website?

[neon pecan]

python.org yes

[ornate elk]

and have you re git cloned the wwebui? also remove the spaces in folder names just to be sure

[neon pecan]

I deleted the whole file and started from the beginning so yes

[ornate elk]

can you try in a folder that doesnt have a space in its name?

python is a bit picky

also we dont know if your antivirus blocked something

[neon pecan]

the folder where the web ui file goes?

[neon pecan]

same error. cant get torch that's satisfying

ERROR: Could not find a version that satisfies the requirement torch==2.11.0+cu130 (from versions: none)
ERROR: No matching distribution found for torch==2.11.0+cu130
Traceback (most recent call last):
File "F:\ai\ROCm\sd-webui-forge-neo\launch.py", line 56, in <module>
main()
~~~~^^
File "F:\ai\ROCm\sd-webui-forge-neo\launch.py", line 43, in main
prepare_environment()
~~~~~~~~~~~~~~~~~~~^^
File "F:\ai\ROCm\sd-webui-forge-neo\modules\launch_utils.py", line 316, in prepare_environment
run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True)
~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\ai\ROCm\sd-webui-forge-neo\modules\launch_utils.py", line 86, in run
raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't install torch.
Command: "F:\ai\ROCm\sd-webui-forge-neo\venv\Scripts\python.exe" -m pip install torch==2.11.0+cu130 torchvision==0.26.0+cu130 --extra-index-url https://download.pytorch.org/whl/cu130
Error code: 1

[ornate elk]

thats so strange, can you open up a cmd and run:
python --version
and then
where python

screenshot boath pls

[ornate elk]

ah there is the issue

its using the microsoft store python

instead of the other one

you have to go in your windows system environment variables and set the python higher than Microsoft.Windows.Apps

[neon pecan]

I know what you're talking about but I forgot how to get to it

[ornate elk]

click on windows logo and type Environment

[neon pecan]

I see python 1.13 but im not seeing anything for the microsoft store version

[ornate elk]

can you screenshot?

[ornate elk]

wrong window

you have to go into the bottom path

of system

[neon pecan]

I looked there, not a trace of python

[ornate elk]

can you show it?

it has to be in there

also this is strange:

thats a wrong folder name

so your python is corrupt probably

it has to look like this:

[neon pecan]

how is it corrupted? it's a brand new installation

[ornate elk]

yea idk

but the folder name is wrong

[neon pecan]

not a trace of python

[ornate elk]

here, check this with the edit button

[neon pecan]

all I see of python so far

[ornate elk]

seems like its missing there

you have to add it, but i would recommend reinstalling it because the folder name looks so off

[neon pecan]

well I did install it on a different harddrive. Im trying to save desktop space so I put it in it's own folder. I dunno if that has anything to do with it considering it put a file on my desktop anyway

[lucid yoke]

Before downloading it, please post the URL where your getting it from

[ornate elk]

yea thats not so good, better install it in its default place (appdata)

it doesnt take up much space

[neon pecan]

alright then. I still have the exe so i'll just put it on desktop and uninstall then reinstall it fresh

[ornate elk]

make sure you check "add python to path"

[neon pecan]

I do that every time I reinstalled it

[neon pecan]

I got it from python.org

[neon pecan]

well it looks like its correct now. though so some reason the location loops. but when I edit it all the python stuff is there

[ornate elk]

okay

[neon pecan]

I assume its in its correct place now?

I see it in appdata

and it looks like its above windows apps

[neon pecan]

should I try again?

the paths look correct

[ornate elk]

remove the paths of 311

[ornate elk]

yea paths look okay

still strange folder name of 313-32

[neon pecan]

I guess not all computers are the same

I didnt change its name. I got my exe from the source.

so I dont know

[neon pecan]

Installing torch and torchvision
Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/cu130
ERROR: Could not find a version that satisfies the requirement torch==2.11.0+cu130 (from versions: none)
ERROR: No matching distribution found for torch==2.11.0+cu130
Traceback (most recent call last):
File "F:\ai\ROCm\sd-webui-forge-neo\launch.py", line 56, in <module>
main()
~~~~^^
File "F:\ai\ROCm\sd-webui-forge-neo\launch.py", line 43, in main
prepare_environment()
~~~~~~~~~~~~~~~~~~~^^
File "F:\ai\ROCm\sd-webui-forge-neo\modules\launch_utils.py", line 316, in prepare_environment
run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True)
~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\ai\ROCm\sd-webui-forge-neo\modules\launch_utils.py", line 86, in run
raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't install torch.
Command: "F:\ai\ROCm\sd-webui-forge-neo\venv\Scripts\python.exe" -m pip install torch==2.11.0+cu130 torchvision==0.26.0+cu130 --extra-index-url https://download.pytorch.org/whl/cu130
Error code: 1
Press any key to continue . . .

it didnt even take a second. it instantly gave me the error.

This torch crap is really getting old

[ornate elk]

have you delete the venv folder?

[neon pecan]

no

[ornate elk]

you always have to after chaning python stuff

then redo the steps of rocm

[neon pecan]

IT GAVE ME THE ERROR AGAIN!!!! 😤

it keeps bothering me about torch!

(venv) F:\ai\ROCm\sd-webui-forge-neo>pip uninstall torch torchvision torchaudio rocm rocm-sdk-core rocm-sdk-devel rocm-sdk-libraries-custom -y
WARNING: Skipping torch as it is not installed.
WARNING: Skipping torchvision as it is not installed.
WARNING: Skipping torchaudio as it is not installed.
WARNING: Skipping rocm as it is not installed.
WARNING: Skipping rocm-sdk-core as it is not installed.
WARNING: Skipping rocm-sdk-devel as it is not installed.
WARNING: Skipping rocm-sdk-libraries-custom as it is not installed.

(venv) F:\ai\ROCm\sd-webui-forge-neo>pip install torch torchvision torchaudio --index-url https://rocm.nightlies.amd.com/v2/gfx120X-all/
Looking in indexes: https://rocm.nightlies.amd.com/v2/gfx120X-all/
ERROR: Could not find a version that satisfies the requirement torch (from versions: none)
ERROR: No matching distribution found for torch

(venv) F:\ai\ROCm\sd-webui-forge-neo>

[neon pecan]

it's bugging me about torch....AGAIN

[ornate elk]

Probably because your python is still messed up

I also saw miniconda in your path

That also messes up python stuff

[neon pecan]

remove it from path then?

[ornate elk]

You can also try to edit the webui-user.bat and there set the python path at PYTHON=

Then delete the venv folder again

[neon pecan]

which should I do?

both?

[neon pecan]

ERROR: Could not find a version that satisfies the requirement torch==2.11.0+cu130 (from versions: none)
ERROR: No matching distribution found for torch==2.11.0+cu130

same error

[ornate elk]

Uninstall everything you have from python then go to its folder path in appdata local programms python
And there delete everything thats in there

Then uninstall miniconda too if you dont need it

Then reboot the PC

Then reinstall python 3.13.12

Delete the venv folder and edit the webui-user.bat there remove the python path
Then save and launch

[neon pecan]

it wont let me uninstall it 😑

failed to run pre_uninstall

[ornate elk]

Reboot the PC and try again if that doesnt help delete the folder

[neon pecan]

I dont know where the miniconda folder is even at

[ornate elk]

Its under C/users/yourusername/miniconda3

[neon pecan]

ERROR: Could not find a version that satisfies the requirement torch==2.11.0+cu130 (from versions: none)
ERROR: No matching distribution found for torch==2.11.0+cu130

same error....again

bugging me for thousandth time about torch

[ornate elk]

Can you show the complete cmd log?

[neon pecan]

ERROR: Could not find a version that satisfies the requirement torch==2.11.0+cu130 (from versions: none)
ERROR: No matching distribution found for torch==2.11.0+cu130
Traceback (most recent call last):
File "F:\ai\ROCm\sd-webui-forge-neo\launch.py", line 56, in <module>
main()
~~~~^^
File "F:\ai\ROCm\sd-webui-forge-neo\launch.py", line 43, in main
prepare_environment()
~~~~~~~~~~~~~~~~~~~^^
File "F:\ai\ROCm\sd-webui-forge-neo\modules\launch_utils.py", line 316, in prepare_environment
run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True)
~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\ai\ROCm\sd-webui-forge-neo\modules\launch_utils.py", line 86, in run
raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't install torch.
Command: "F:\ai\ROCm\sd-webui-forge-neo\venv\Scripts\python.exe" -m pip install torch==2.11.0+cu130 torchvision==0.26.0+cu130 --extra-index-url https://download.pytorch.org/whl/cu130
Error code: 1

just the same error as usual.

[ornate elk]

I mean the whole thing

[neon pecan]

"wahhh cant find a satisfying version" error code 1 blah blah blah

[neon pecan]

that's all it gives me

venv "F:\ai\ROCm\sd-webui-forge-neo\venv\Scripts\Python.exe"
Python 3.13.12 (tags/v3.13.12:1cbe481, Feb 3 2026, 18:12:42) [MSC v.1944 32 bit (Intel)]
Version: neo
Installing torch and torchvision
Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/cu130
ERROR: Could not find a version that satisfies the requirement torch==2.11.0+cu130 (from versions: none)
ERROR: No matching distribution found for torch==2.11.0+cu130
Traceback (most recent call last):
File "F:\ai\ROCm\sd-webui-forge-neo\launch.py", line 56, in <module>
main()
~~~~^^
File "F:\ai\ROCm\sd-webui-forge-neo\launch.py", line 43, in main
prepare_environment()
~~~~~~~~~~~~~~~~~~~^^
File "F:\ai\ROCm\sd-webui-forge-neo\modules\launch_utils.py", line 316, in prepare_environment
run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True)
~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "F:\ai\ROCm\sd-webui-forge-neo\modules\launch_utils.py", line 86, in run
raise RuntimeError("\n".join(error_bits))
RuntimeError: Couldn't install torch.
Command: "F:\ai\ROCm\sd-webui-forge-neo\venv\Scripts\python.exe" -m pip install torch==2.11.0+cu130 torchvision==0.26.0+cu130 --extra-index-url https://download.pytorch.org/whl/cu130
Error code: 1
Press any key to continue . . .

[ornate elk]

I finally found the issue....

[neon pecan]

oh thank goodness

[ornate elk]

You installed the 32bit version of python ... XD

[neon pecan]

wait what???

THAT'S why?!

[ornate elk]

Yep

[neon pecan]

oh ffs lmao

[ornate elk]

It won't work with that

[neon pecan]

well at least it's simple

[ornate elk]

Yes xD and now it makes sense that your python folder was called 313-32

xD

[neon pecan]

ah

[ornate elk]

And in the cmd log above you can see 32bit Intel as version

[neon pecan]

I figured it didnt matter cause my system is 64 bit and it'd just be simpler to get 32 since I can easily run it.

[ornate elk]

Always go for 64bit!

32bit has limited abilities to talk to RAM and other stuff

[neon pecan]

64 bit installing now

yes 32 was uninstalled and I clicked add to path

should I delete the venv file too?

[ornate elk]

Yep

[neon pecan]

finally its downloading stuff instead of bugging me about torch

[ornate elk]

Good 😁 and then you can proceed with the rocm steps and it should work

[ornate elk]

I'm off now its late. Let me know if it worked

[neon pecan]

well its currently stuck at 1.9gb - 1.9gb

typical installation delays

its flashing too

[neon pecan]

I finally got it

but now I got to figure out how to load the model cause when I did a test image it said failed to load model

nevermind it was a one time thing. I tried to open it a second time and it wont let me open. I have to reinstall just to use it. Even when I did use it I was disappointed. It's not at all what I heard. I was told it can use multiple references, it can't and can only use one. I did a test render only for it to stop 3 seconds in just to tell me it failed to load the model and I have no idea what to do about it. I was hoping for an adobe firefly replacement with no restrictions or money but I cant even render one image and i'm lost again. I could only open it once. I tried to open it again and it wont let me

[neon pecan]

  File "F:\ai\ROCm\sd-webui-forge-neo\venv\Lib\site-packages\torch\utils\_contextlib.py", line 124, in decorate_context
    return func(*args, **kwargs)
  File "F:\ai\ROCm\sd-webui-forge-neo\modules\sd_models.py", line 357, in forge_model_reload
    raise BufferError("Failed to load diffusion model...") from None
BufferError: Failed to load diffusion model...

[ornate elk]

Which model have you tried?

[neon pecan]

it's that one you recommended me

Z something

Z image turbo

but it doesn't matter because it wont let me open it. I have to reinstall it just to open stable diffusion

[prisma moss]

Do you perhaps know how to 'blacklist' a node from updating ever? I've done a lot of custom work on impact-pack and i need it not ever be disturbed, if comfy/nodes decide to update themselves

[prisma moss]

Do you perhaps know how to 'blacklist' a node from updating ever? I've done a lot of custom work on impact-pack and i need it not ever be disturbed, if comfy/nodes decide to update themselves

i may be repeating myself im not seeing the messages i send

[neon pecan]

Traceback (most recent call last):
File "F:\ai\ROCm\sd-webui-forge-neo\launch.py", line 56, in <module>
main()
~~~~^^
File "F:\ai\ROCm\sd-webui-forge-neo\launch.py", line 43, in main
prepare_environment()
~~~~~~~~~~~~~~~~~~~^^
File "F:\ai\ROCm\sd-webui-forge-neo\modules\launch_utils.py", line 331, in prepare_environment
raise SystemError("Please update your GPU driver to support cu130 ; or manually install older PyTorch")
SystemError: Please update your GPU driver to support cu130 ; or manually install older PyTorch

I can't reopen stable diffusion

[ornate elk]

You have to do the rocm steps again to get the correct torch version

After step 7 of the guide

[ornate elk]

You can set the folder permission of the impact-pack extension to read only

[prisma moss]

Ty

[neon pecan]

so I do have to keep reinstalling just to run it?

[ornate elk]

No but you said you reinstalled forge neo

If you ever delete the venv folder you need to relaunch and do the steps again

[neon pecan]

I reinstalled the entire thing. I found if I delete the venv folder and just go back a few steps it'll let me open it because if I try to open it again I get another error

[neon pecan]

by the way I did some digging and I found out in order to get the experience I want (which is something similar to adobe firefly) I apparently need to somehow combine stable diffusion and comfyui?

how do you combine two ai platforms?

[ornate elk]

Comfyui is an webui for itself.
It can do the same thing as forge neo and more

You can have boath setup independent

[neon pecan]

but apparently combining the two will give me the experience im after. unless it doesnt make a difference. I was looking around on the internet and it said combining them gives me an experience identical to adobe firefly just without adobes heavy restrictions and credits

[ornate elk]

You can't combine them, they are seperate tools

You can have boath installed at the same time

A guide for comfyui is also in the pinned messages and can be setup in 5 minutes

[neon pecan]

then I wonder what it meant when it said "combine"

but anyway I should get to bed, it's 5am. ill have to figure out how to get the model working tomorrow

or later today I should say

I heard stable diffusion allows multiple picture references but it only allowed one.

[ornate elk]

That depends on the webui.
Comfyui supports multiple input images

[paper compass]

Hi all
If I buy a Pro license to make instrumental tracks I can sing to, will I be able to download stems (meaning individual .wav files for each instrument), or just mixes?

[lucid yoke]

So far all i could find is that the output is a single file in either wav or mp3 in a mixed format

[paper compass]

Mmmph. If it's a single file, that's a mix. Stems would be a separate file for each instrument; sometimes more than one, especially for the drums.

That's disappointing; the AI seems to do a pretty good job of creating instrumentals according to prompts, but I can't use it if it can't give me separate .wav files for each instrument

[lucid yoke]

Thats unfortunate, iirc theres some software that can split the output but considering it's very new and I haven't messed with it yet i cant give you a guarantee that it would work

Stable audio i mean ^

[paper compass]

That approach would inevitably result in really bad quality stems

There are other AI solutions out there that can do stems, but they don't make really good music like Stable Diffusion does

Soundverse, for instance, can do stems

[lucid yoke]

Ah shame, I'm quite hyped about the new models but it's always not there yet.
Hmm a alternative but time consuming method could be generating the instruments separately but I'll test that out when I'm home

[paper compass]

Not having paid for an account so far, I can't experiment with creating a mix and then trying to get the AI to pick it apart into stems, but it might be possible

[lucid yoke]

I think by locking the seed and llm modified prompt it might be workable? Ill let you know but im using the local options

[paper compass]

Thank you so much, you're very helpful

[lucid yoke]

No worries! Im quite interested aswell. I need a few hours as I'm at work still haha

[paper compass]

If you're interested in hearing what I've been doing with human musicians these last few years, have a listen: https://thefamilybear.bandcamp.com

[lucid yoke]

I'll check it out once I'm on the train 👌

[paper compass]

I'll keep this tab open as long as necessary; I'm at home

Heh, I've even got a song called BEAUTIFUL about a guy who is in love with an AI girlfriend

[gritty flint]

Hi, sorry if this is the wrong forum for this, but I'm wondering what's the "mainstream" software is for loading model and exposing API nowdays?
I think Forge (fork from automatic1111 iirc) has API, but I notice on their github that latest release was 2 years ago. Maybe its still usable? Many are talking about Comfy UI, but not sure if that has an API.

I don't care much about the user interface, I'm just looking for local hosting with a rest api

[knotty turret]

guys anyonne can help me my workflow was working fine yesterday but today the image not even generate in the end

no image in the end on output folder too

[lucid yoke]

ive been testing multiple tracks but infortuneatly it keeps mixing in different instruments or even using whole different instruments so i dont think it will work for your usecase

[lucid yoke]

comfyUI has the most extensive API options for the entire workflow

[gritty flint]

Ok I have not looked into comfyui at all really, but will do , cheers

[neon pecan]

multiple? that's promising

[rose apex]

@ornate elk The file seems to have expired and I'm unable to build it again either. What to do?

[knotty turret]

Yeah sure

I think there np saying what need to do here

Its fine

[neon pecan]

so how do I fix my model not working?

[prisma moss]

@vocal burrow

[lucid yoke]

Okay youve been warned before, if your a real human feel free to appeal

[paper compass]

I reported him

[neon pecan]

anyone know how I can learn comfyui? Im having a hard time finding the nodes I need

[ornate elk]

thanks for letting me know, i updated the steps now.

[neon pecan]

why is ai so difficult? stable diffusion model loading is broken, comfyui is a confusing mess and I dont know how to do anything

and google is giving me nothing I want

the information I want is impossible to find

[ornate elk]

which info?

[neon pecan]

for comfyui, how to use text prompts and reference images for one render

I heard its possible but cant figure out how

stable diffusion is out because it wont load my model so its useless to me

I tried comfyui but im getting nowhere

[ornate elk]

use the flux klein templates

[neon pecan]

where do I even get that?

[ornate elk]

can oyu try an other model?

[ornate elk]

on the templates button on the left

[neon pecan]

it just says it cant load the model. it cant be found or whatever

[ornate elk]

and which model?

[neon pecan]

the z model thing you recommended. I cant remember the name

[neon pecan]

nope, that's not it

[ornate elk]

you can try this model, it should work:
https://civitai.red/models/2173571/z-image-turbobase-aio?modelVersionId=2448013

needs a civitai account to download

[neon pecan]

I cant find out cause it says models dont exist

I put them in the correct folder but it cant find them and gives me an error

thus I gave up and tried comfyui

[ornate elk]

your giving me to less info to help, make screenshots, show me logs or something

[neon pecan]

alright im ditching comfyui. I was clearly lied to. Im starting to think adobe firefly is one of a kind

and there's no info on how to use it anywhere

how people figure out this node crap is beyond me. it makes no sense and is super limited even though I was told multiple references and text prompts were possible

[ornate elk]

it is possible:

[neon pecan]

I would but it wont even open it now. It just gives me three sentences and sits there

[neon pecan]

but how???? HOW????? I searched for nodes, no results. I tried google searching, no results. There's no information on this ANYWHERE

and now stable diffusion won't even open so I cant even use that now

[ornate elk]

its in templates on the left, then filter for flux 2 klein, then click on the first template

[neon pecan]

venv "F:\ai\ROCm\sd-webui-forge-neo\venv\Scripts\Python.exe"
Python 3.13.12 (tags/v3.13.12:1cbe481, Feb 3 2026, 18:22:25) [MSC v.1944 64 bit (AMD64)]
Version: neo
stuck with this in the cmd

[ornate elk]

hmm try close every other cmd end relaunch forge neo

[neon pecan]

NOW I CANT REOPEN COMFYUI!!! What is going on????

Why wont they let me reopen the software??

F:\ai\comfyui\ComfyUI_windows_portable>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
setup plugin alembic.autogenerate.schemas
setup plugin alembic.autogenerate.tables
setup plugin alembic.autogenerate.types
setup plugin alembic.autogenerate.constraints
setup plugin alembic.autogenerate.defaults
setup plugin alembic.autogenerate.comments
Fatal error in launcher: Unable to create process using '"D:\a\ComfyUI\python_embeded\python.exe" "F:\ai\comfyui\ComfyUI_windows_portable\python_embeded\Scripts\offload-arch.exe" ': The system cannot find the file specified.

[WARNING] offload-arch failed with return code 1

[neon pecan]

I tried, same results

[ornate elk]

maybe your antivirus software is blocking them?

[neon pecan]

it doesnt open any warnings

and if it was my software they wouldnt of worked the first time

[ornate elk]

then reboot the PC, thats should fix it

[neon pecan]

for some reason I have to reinstall just to get stable diffusion to open

and comfyui just works when it feels like it. its gives me the python error and then it just randomly loads and opens

after spamming it

I never knew ai would be so broken and difficult just to get open

[ornate elk]

normally it isnt that broken, but idk why it is for you

is your F drive an SSD?

[neon pecan]

I dont know either. comfyui is random and stable diffusion lets me open it once and never works again until I reinstall it

[neon pecan]

external harddrive, 9 tb

[ornate elk]

yea thats the issue probably, never install webuis on an external drive and also not if its an hdd

[neon pecan]

I dont know if its a hdd or sdd. I have no clue

[ornate elk]

you can later store your models on the hdd, but never use them to launch software from there

[ornate elk]

if its 9tb it wont be an sdd probably

[neon pecan]

how do I find out?

[ornate elk]

i would advice to move comfyui portable to an other drive thats an ssd on your system
and then also reinstall forge neo on that drive

[ornate elk]

get a tool like crystal disk info, that can also be used to check the drives health

[neon pecan]

how much space do the ai files take?

[ornate elk]

the webui itself around 7gb, the models are larger
but you can move the models onto other drives and let them load from there

[neon pecan]

aaaaaand that's why they're not on the desktop

that is too much memory for my liking. my desktop memory is nearly full enough as it is

putting the ai on the desktop will take what little space I have left

it took forever just ot get it back to over 9gb

[ornate elk]

no other internal drives?

[neon pecan]

I got two and the ai is on one of them

[ornate elk]

but above you said F is External

[neon pecan]

to be honest. I have no idea if its internal or external. All I know is it's a metal box, it got put in a slider inside the tower, plugged in, bam extra hard drive

that's all I know. I have no idea

I guess its internal cause its inside the tower

[ornate elk]

yes then its an internal drive

[neon pecan]

ok then

im using the ai on the internal drive.

[ornate elk]

because its an hdd its sometimes slow when launching stuff like forge neo, so it loads for a long time

maybe yu can get a bit more space on your ssd drive

tools like WizTree can show you whats taking up the most space on each drive

[neon pecan]

I already have it and it doesnt help much. it also doesnt help that its files I cant delete due to being used

so they're staying on the f drive

F:\ai\comfyui\ComfyUI_windows_portable>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
setup plugin alembic.autogenerate.schemas
setup plugin alembic.autogenerate.tables
setup plugin alembic.autogenerate.types
setup plugin alembic.autogenerate.constraints
setup plugin alembic.autogenerate.defaults
setup plugin alembic.autogenerate.comments
Fatal error in launcher: Unable to create process using '"D:\a\ComfyUI\python_embeded\python.exe" "F:\ai\comfyui\ComfyUI_windows_portable\python_embeded\Scripts\offload-arch.exe" ': The system cannot find the file specified.

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

still cant get in

[ornate elk]

is your F drive a bit older?

[neon pecan]

I would say it is. but it doesnt have problems running or anything

I guess four years old at least?

[ornate elk]

maybe worth to check its health and rpm speed with crystal disk info:
https://crystalmark.info/en/software/crystaldiskinfo/

[neon pecan]

and comfy ui decided to work this time

stable diffusion still wont do anything. the cmd does the same thing. same three messages

never did that before until today

[ornate elk]

its just loading, it will rpobabyl take a few mins to start

[neon pecan]

ill let it run while I try to figure out comfyui

Dunno why they cant just have a system. get image, get text, let it connect to one node and render. why does it need all this rocket science just to get two nodes to work together?

I just want to use reference images and text prompts to render an image. why is it so hard to accomplish?

[ornate elk]

the templates try to make it easier for the users, but yea comfyui is not easy to use

[neon pecan]

it'd help if the nodes I want exist. I search for the ones I see in images but the nodes arent there

[lucid yoke]

it shouldnt be this hard but you unfortunately have run into some uncommon problems

[neon pecan]

like the one you showed me, it doesnt exist

[neon pecan]

ooooooooooooof course I have.....it always feels like im the only one with "unique" issues while everyone else is like "never happened to me"

[neon pecan]

launch.py: error: unrecognized arguments: --use-zluda

this is a new error from stable diffusion

[ornate elk]

your not using zluda with forge neo (not supported)

your gpu should not use zluda

[neon pecan]

weird. Im not using zluda

I havent touched stable diffusion since I gave up on it cause it wouldnt load the model

[ornate elk]

idk what your trying to do

you probably launched some other webui i guess

[neon pecan]

im trying to start stable diffusion

[neon pecan]

I figured it out

[rose apex]

9060 XT Just recently installed Comfyui and it's crashing after 2-3 txt to img runs

[ornate elk]

which launch rags do you use?

[rose apex]

@echo off
echo Starting ComfyUI with ROCm experimental flags...

REM Go to ComfyUI folder
cd /d "%~dp0"

REM Activate venv
call venv\Scripts\activate

REM Set environment variables
set FLASH_ATTENTION_TRITON_AMD_ENABLE=TRUE
set TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1
set PYTORCH_HIP_ALLOC_CONF=expandable_segments:True

REM Launch ComfyUI
python main.py --auto-launch --use-pytorch-cross-attention --disable-smart-memory --bf16-vae --enable-manager --enable-manager-legacy-ui
REM --cache-none --use-sage-attention --use-pytorch-cross-attention --use-flash-attention
pause

The last REM line is just so that I can copy and paste the commands into the main.py line

[ornate elk]

try add:--supports-fp8-compute --enable-dynamic-vram
and as flag add this too:
set COMFYUI_ENABLE_MIOPEN=1

[rose apex]

Ok

[ornate elk]

and then try to install sage attention

then you can try that too

[rose apex]

I tried the set enable Miopen but it's taking a really long time, waited around 20 mins in the 2nd image to image Ksampler node.
If I don't use Miopen, I get this error during the Ksampler node or VAE node.

[ornate elk]

lol OSError: exception: access violation reading 0x00000000000000E0

[paper compass]

Does anyone have an answer for me on whether or not I can get Audio Diffusion to generate separate stem tracks for each instrument, or just mixes?

[lucid yoke]

just mixes so far as i could test & find. But in #✍🏼｜rules-and-tos under 6. theres a URL to the official sability AI support to be extra certain.

[prisma moss]

dynamic vram is enabled by default btw

[ornate elk]

i dont think so for AMD

[prisma moss]

True idk bout that

[rose apex]

What to do 😩 am I cooked?

Actually, I tried the bat file you made in your guide and it seems to be working fine. So the problem must be in the bat file only

[neon pecan]

how do I link multiple references and text prompts into one output in comfyui?

[quaint rivet]

Use separate CLIP Text Encode nodes for each prompt then merge them with Conditioning Combine/Average. For references use multiple IPAdapter or Reference nodes and adjust their weights before sending everything into the sampler.

[neon pecan]

are those custom nodes or are they already available?

[quaint rivet]

Yes

[neon pecan]

yes are they custom or yes they're already available?

[lucid yoke]

@quaint rivet i asked you to stop trying to drag people into dm's for no reason. For this I'm going to rule 6 you.
Its all you have been doing

[lucid yoke]

They should be existing nodes

But for what your trying to do, what model are you using?

[neon pecan]

I have yet to find a model that suits my needs and or isnt on huggingface because I cant figure out how to download models from there

Do I use a single text node to describe the plot?

[lucid yoke]

Okay what exactly are you trying to do

Different models do require a different workflow

[neon pecan]

well basically recreate the adobe firefly experience. It would allow at least six references and in the text prompt you describe what you want and it'd make it, random artstyles too which I didnt mind.

I just enjoy messing around and creating things like single images and or comics. Just for my amusement. Get some laughs. Just having fun basically

im trying to recreate that experience. stable diffusion only allows one reference image and no text prompts but comfyui has more potential

[lucid yoke]

Well six is gonna be pushing it, 3-4 is gonna be better with flux klein but not ideal regardless

[neon pecan]

I dont always use six.

just characters and scenario and tell it what to make

[lucid yoke]

Hmm realistic or anime style? Because depending on your needs some models are better/worse

[neon pecan]

not exactly anime. just cartoony. I'm hoping a model exists where it has no specific style and just copies what the references has or what it's prompt to make.

I wouldnt mind being able to mix two styles

adobe firefly pulled that off and it was pretty cool

[lucid yoke]

Hmm well I'm afraid theres no perfect omni model imo
I would say however anima/Illustrious has the broadest cartoon/anime styles with a TON of loras
And mixing styles is easy by prompting for it

Huggingface is only a good place for models if you know what your doing.
civitAI is a much better place as you get visual examples

[neon pecan]

how DO lora's work anyway? I am aware they apparently help train ai. How, i'm not sure. But apparently they do something. but I dont know how you can include one

[lucid yoke]

Let me hop on my pc and show you
I can do a big image dump or show you in vc if you want

But if its image dump im moving channels to not clog up the tech support lol

But basically

[neon pecan]

I would actually appreciate you showing me. I find it easier to learn by actually seeing how its done than describing it.

it's easier for me to understand that way

so if you can show me in vc that would be great

[lucid yoke]

bet, im in diffuser lounge

i got some time before i have to do stuff

[neon pecan]

any chance you can show me the process of nodes? It doesnt need to be anything huge. just enough to give me an idea how to accomplish what im trying to do. Reference images, scenario and text prompt describing what'll happen. I just dont know exactly what im supposed to do

and what if I want to use text prompts to identify the characters in a reference image as well as use a text prompt to describe what the image is supposed to be and look like

and how come you have menu's mine doesnt have? I dont have a generate menu on the top of my screen

so does it just edit the image or is it capable of creating something entirely new? I want to be able to make original images

if youve been talking this whole time I cant hear you. your box is flashing green but im not hearing anything

[lucid yoke]

can you hear me now?

[neon pecan]

no. I think it might be my end. something to do with audio output. im looking into it right now

I cant remember how it's done

[lucid yoke]

o

im gonna be busy for a bit but if you fixed your audio lmk

[neon pecan]

something to do with the output settings. I cant find it

[lucid yoke]

hmm settings > within discord

[neon pecan]

There's a setting I need to change to get discord audio to work but for the life of me cant recall where the output setting is.

[neon pecan]

I know. but it has input settings, no output settings

[lucid yoke]

it does?

[neon pecan]

guess I was doing it right. strange I still couldn't hear anything

well found the output option a little too late lol

[lucid yoke]

hmm ill condense the gist of it.
with IP adapter you can do certain style transfers with images like on firefly but its gonna increase the generation times by a lot
Loras: you can combine multiple style ones to create a whole new image or even combine characters if you play with it well
with flux kontext: you can edit images. move characters over etc but new images with references i noticed it copies the refence image a bit too much

are they serious

@sonic bane saw that

[neon pecan]

saw what?

[lucid yoke]

" support scam"

he realized a bit too late i was a guide/mod though

[neon pecan]

just sounds like firefly with extra steps. I just make comic panels or single images for fun. I have yet to figure out how to get the generation working so I have no clue what my gen time even is

the nodes still got me confused and leaving me with questions

its good to know the names of the nodes ill need to attach them to but im left with two more questions.

1. am I able to attach the images to prompts so I can type in the characters name so the ai can identify them?
2. Even if I do get the nodes set up I dont know if they all need to connect to one node for it to render. If so what in the world do I connect them to? I noticed nodes cant have more than one node attached to a certain slot at a time.

[lucid yoke]

1. no
   if you use flux klein you can say place the man from image 1 into image 2 but not " place XiXi into the image"

[neon pecan]

so it wont be able to address characters properly?

[lucid yoke]

they combine eventually into one node. theres multiple ways of combining nodes

[neon pecan]

hm, well that's one disadvantage it has compared to firefly which can include names in dialogue given to it

[neon pecan]

oh joy....more complicated and confusing node work

[lucid yoke]

well firefly runs on much larger hardware then we currently fit into a home pc

[neon pecan]

well so much for creating the firefly experience

well maybe I can just do what I did in firefly and just tell it in the prompt how to address a character.

[lucid yoke]

well ComfyUI is one of the hardest UI's to play around it. this just directly generates so im sure if your prompt involves a bit of fluff like "adress this woman as xixi, Xixi is now sitting on a bench using the style of image 2" is not gonna work nice
but if you use existing characters it might have in its dataset or lora's for said character then you use the lora trigger for that like "make danny devito hold a samurai sword" and should be working just fine.

[neon pecan]

so I cant just get two images of goku and vegeta, put in the prompt how they would say it like "image 1 says "hey vegeta" and "image two says "go away kakarot"?

it wont be able to do that?

[lucid yoke]

oh you want text? some models cant do text and some can do it basically, flux can maybe do some short text so sure

but youd prompt it as add a speech bubble to image 1 saying "hey vegeta"

in flux klein

[neon pecan]

well then I guess its still kinda like firefly

[neon pecan]

where's the IPAdapters?

[lucid yoke]

Ip adapter?
Thats mostly for sdxl based models, flux sorta does it natively with image references

I think its a node set ill check

[neon pecan]

I was following advice from kimmi so I was looking for it. Experimenting trying to figure this stuff out. turns out for what I want to accomplish I need alot of nodes and a language model. this is crazy advanced stuff

how do you guys even work with this? its a confusing mess.

[lucid yoke]

well normally people dont use ComfyUI as their first enviorment

[neon pecan]

well not like I had anywhere else to go. I was also excited so I dove in headfirst, brute forcing my way to learn but I had no idea that there'd be nothing to go off on.

[lucid yoke]

there should be a ip adapter workflow template

but youd need to download ip adapter models seperately but ive never used ipadapter so i wont be able to help you from there

[neon pecan]

well when I searched for the node I got nothing

oh

jeez this is going to take then thousand years to get mine set up

[neon pecan]

where do I even begin to learn this? I got so many questions and not enough answers and when I get an answer I get more questions

[idle olive]

does DWpose work for sdxl models, i cant get it to work, it doesnt follow the pose i want to to at all, any help would be apreciated

[ornate elk]

Yes it works with controlnet openpose or union models

[desert lodge]

rtx 3060 12g or rtx 3070 8g or rtx 5060 8g

which would generate faster images?

[desert lodge]

What is this

[prisma moss]

the answer to your question

[desert lodge]

So the 3070 is faster?

There is no 5060 in this graph

[prisma moss]

well i wouldnt take either

illd take 5060 16gb

[desert lodge]

What if you only have to pick between. Those

[prisma moss]

idk both are trash tier

illd pick 12gb 3060 over 3070 8gb

[ornate elk]

The 5060 can use newer tech to improve speed and vram usage

Sage attention v3 for example

[idle olive]

For some reason I can't even get normal openpose to work at all, not sure what I've done :/ I've got all the correct model thingies for it

It did work previous to installing dwpose

[ornate elk]

which webui do you use?

[idle olive]

And I've tried reinstalling Stable Diffusion WebUI Forge

[ornate elk]

forge is outdated and dwpose extension too

you should use forge neo instead, it has dwpose integrated

[idle olive]

What would u suggest to use that's better n all that

[ornate elk]

a setup guide for forge neo is in the pinned messages of this channel

[idle olive]

I tried neo but it doesn't have the 4x hires upscale option, just latent etc, is there an add on for that stuff or is it somewhere else in it?

[ornate elk]

it has that option too, but you have to download the upscale models yourself as the dev removed some of them

[idle olive]

Ah k, good to know, I'll tinker with neo this arvo and let u know how I go

[ornate elk]

upscale models can be found here: (.pth files)
https://openmodeldb.info/
they go into the models/ESRGAN folder

[idle olive]

I also use stability matrix to launch it, does that have any negative affect at all?

[ornate elk]

i dont recommend stability matrix as it just ads a layer of problems

we have simple setup guides on the first link of the pinned messages in this channel for each webui

they also include the best performance options for each gpu type

[idle olive]

Ok sweet, thanks

[ornate elk]

no problem, let me know how it goes!

[idle olive]

Will do!

[idle olive]

Hey, so ive gotten to the point of adding the cuda stream text to the .bat, then go to open it again and have it install stuff, i got this error ERROR: No matching distribution found for numpy==2.3.5, whats this mean?

[idle olive]

dw, i gots it workin 🙂

[idle olive]

wow, neo works so quick and good holyyy, why have i been using forge this whole time 😂

still cant get pose thingy to work tho

[idle olive]

no matter what different model/checkpoint, this dont work :/

[lucid yoke]

Hey silly question but what controlnet model are you using

Flux uses a different controlnet model then illustrious for example

[idle olive]

um, so far ive tried:
control-lora-openposeXL2-rank256
openpose_s6000
diffusion_pytorch_model

using DWpose as the preproc

[lucid yoke]

I think i see the problem, what checkpoint type?

[idle olive]

prefectIllustriousXL_v8.safetensors

[lucid yoke]

Gotcha, gimme a sec your using the wrong controlnet model

Lemme look which one i use. I got it from civitAI sec

[idle olive]

ok cool, thx for the hel

p

[lucid yoke]

I use these for others then openpose
I see they removed the open pose one
https://civitai.com/models/929685/noobai-xl-controlnet?modelVersionId=1731092

This one seems to be positive for openpose so i recommend you try this one
https://civitai.com/models/1359846/illustrious-xl-controlnet-openpose?modelVersionId=1536174

[lucid yoke]

I always forget to hit the reply button lol, no sweat

[idle olive]

aight ill try em out

does noobai ones work with my version?

ive tried the 2nd link u sent already, and it doesnt work

[lucid yoke]

Hmmm? What does it say

[lucid yoke]

It works for me

[idle olive]

it doesnt make the character do the pose

[lucid yoke]

Lemme get on my pc to recreate gimme a sec ill download the other version of open pose

[idle olive]

it creates the pose wire thing in the right hand box, but when generating, my character does a random pose, but yea that would be apreciated also

[lucid yoke]

i found the one i use i made a typo in my search earlier but ill still try the one i posted earlier since that one should work better https://civitai.com/models/962537/noobai-xl-controlnet-openpose

[idle olive]

almost done downloading it and will try

[lucid yoke]

however if your prompt says like handstand while the pose is sitting it will do some funky stuff

[idle olive]

yea ive got nothing to do with posing, just the character i want and clothes

didnt work

[lucid yoke]

hmm both models work for me

can you show your controlnet settings/

[lucid yoke]

you set the model on eps blur?

set it to openpose

lol

eps blur does this

[idle olive]

RuntimeError: mat1 and mat2 shapes cannot be multiplied (462x2048 and 768x320)

[lucid yoke]

what resolution is your output image?

shouldnt matter though hmm

[idle olive]

1024x576

[lucid yoke]

thats a funky 16:9, can you try 1344x768?

[idle olive]

which openpose one do i use?

[lucid yoke]

since you didnt rename it, illustriousXL_V10

[idle olive]

should i have?

[lucid yoke]

nah doesnt hurt but since i have many models i like them having names

[idle olive]

ok tru

didnt work

[lucid yoke]

??/

what did you get

[idle olive]

just an image with her face smack in the middle

[lucid yoke]

and the prompt says laying on side? tbh it is a complex pose

try a simple pose just in case aswell

[idle olive]

i dont have a pose in the prompt

[lucid yoke]

try guiding it a little

[idle olive]

ill find a basic pose image rq

its not follwing the pose

its as basic of a pose as i can get lol

legit just standing up with her hands outward

[lucid yoke]

okay just incase can you stream what your settings are in diffuser lounge? (vc)

i think thats faster for both of us

like controlnet wise, everything is properly set

[ornate elk]

That mostly means something is not compatible.
Like a wrong controlnet model for a different model type

[lucid yoke]

Wrong controlnet model + Hires fix was overwriting the controlnet pose for some reason

[ornate elk]

Ah okay, so it works now?

[lucid yoke]

yepp, hopped on vc with his stream on to see what was causing it

[maiden pendant]

where i can download
openpose-editor for forgeneo webui ?

[ornate elk]

There is a guide how to integrate it as every openpose editor is outdated and wasn't integrated in forge neo:
https://github.com/Haoming02/sd-webui-forge-classic/discussions/650#discussioncomment-15639548