🤝｜tech-support

[timber oak]

12gb

i lost all my stuff bc i had to reset my pc

[ornate elk]

oh okay

[timber oak]

i had a massive workflow with detailer and regional prompter, control net and upscaler all in one

😭

but maybe its good i lost my 2.5 tb of ai models

lol

[ornate elk]

haha true xD

more space for new models

[timber oak]

are these settings fine for pony?

or is there a better way?

[dusky osprey]

Hey! Thanks for answering. This is my workflow, I'm using all the fp8 models so I'm not sure why is crashing

[timber oak]

my old model had a base generator and a refiner, aka 2 samplers

but i dont understand how it works anymore

[dusky osprey]

the workflow is not made by me! Sorry if im missing some obvious stuff

[timber oak]

most of it was pony haha

[ornate elk]

yea but i would go with 30 steps and euler a or dpm++ 2m not sde sampler

[timber oak]

oh

why not sde?

i remember that euler gave me really bad results back then : (

so im a bit biased

[ornate elk]

takes longer the output was not that good and for anime models euler a is cleaner.

[timber oak]

the fastest method by far was unipc but with pony it gave me horrible results

[ornate elk]

for realism it might be good going with sde or dpm++ 2m

[ornate elk]

have you checked your vram and ram usage in taskamaner while genning?
how much ram do you have?

[timber oak]

so the results will be more unstable if i use dpm++3m sde?

[dusky osprey]

Ive 16ram and yea with flux it just spikes and all the PC freezes. I know 16ram is kinda ass but i dont fully understand why I can't even generate a 512x512 image? 🤔 If you say is the ram I can upgrade it soon

I haven't checked the vram per se but should be a problem even with --low-vram?

[ornate elk]

yea the RAM is the issue here. 16gb is to less. Either grade up soon to 32 or 64 but for now you can quickly fix the issue by enlarging your Windows Pagefile:
Here is the Guide how to increase it:
https://www.tomshardware.com/news/how-to-manage-virtual-memory-pagefile-windows-10,36929.html
Enable it only for C and make sure its not enabled for any other drive.
Set it to 16000 Min and 24000 Max for C
Then apply and reboot the PC.

Then the PC shouldnt freeze anymore

[dusky osprey]

Oookay. Thank you so much! I'll try to upgrade to 32gb soon. My motherboard only has two sticks sadly. I'll give it a try with the windows pagefile too

[ornate elk]

np, make sure the RAM has the same specs as your current one or if you buy a new kit with 2x16gb and your on Intel or AM4 then go for 3600MHz, for AM5 go for 6000MHz

[ornate elk]

not unstable, just not that great for anime, but for realism

[timber oak]

okay thank you

whats the difference to _gpu?

[ornate elk]

that lets the sampler run on gpu

[timber oak]

which one is better for me?

[ornate elk]

wait, it shouldnt make a big difference

compare the speeds of 2 images of each of them, it should behave the same

[timber oak]

okay thank you

[sharp shore]

I thought and fought with that thing last night, and no matter what I do I get an error. Comfy is now saying that pitorch is compiled without cuda or something like that. I tried reinstalling pitorch with cuda support, but it doesn't seem to work

[ornate elk]

What's your GPU?

Make sure your GPU driver is updated

[sharp shore]

5070, just installed earlier this week so the drivers are also updated

[ornate elk]

Hmm then you probably should reinstall Comfyui, use its portable Version

[sharp shore]

And that is just the interface right? Nothing to do with capability?

Sorry, again, still trying to understand this stuff haha

[ornate elk]

Its one of multiple webuis with a different ui and usage

With a 5070 12Gb you could also setup Forge Neo for example

[sharp shore]

Forge Neo?

[ornate elk]

yea its fork of forge with more features and updates

runs very well on rtx30/40/50 as it uses the latest torch and cuda

[sharp shore]

can I get a link? i couldn't find it. all i'm finding is like.. minecraft crap lol

[ornate elk]

You can find it on my install Guides:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[sharp shore]

Thank you

[ornate elk]

No problem! There is also written which launch args you need for each GPU type for the best performance

[bitter storm]

im trying to change aspect like the shoes or the pose on a picture but it always look like shit. i tried 3 different videos with inpainting. i use automatic1111, is there any good guide that even i can understand? 😄

[sharp shore]

@ornate elk I followed your instructions and it said there was no model. So I put sd3.5_medium.safetensors into \sd-webui-forge-neo\models\Stable-diffusion but now I'm getting AttributeError: 'SdModelData' object has no attribute 'sd_model'

any idea what I'm missing?

[ornate elk]

Hm it should support sd3.5 but try am other model to test if that works like sdxl or flux

[sharp shore]

I will have to do that on my new computer. Which I'll be building tonight.

[ornate elk]

Oh alright!

[sharp shore]

Yeah, I reset my motherboard back to factory default so it's not seeing the graphics card now. And my CPU does not have integrated graphics. So it's either build a new pc, or find somewhere local to sell me an LGA 1151 processor that happened to have integrated graphics!

And I was going to build a new pc anyway

[ornate elk]

Ohhh damn

But you have a new CPU now to build?

Or new Mainboard + CPU?

[cloud geyser]

LGA 1151 is just the socket, it doesnt mean all CPU that use that socket has an integrated (embedded) GPU inside CPU

[sharp shore]

Yes, correct. My old PC had an i7-9700KF which doesn't have integrated graphics. Prior to that CPU, I had an i5-9600K which did have integrated graphics.

[cloud geyser]

K and KF differs

just for that

[paper dagger]

K = Onboard GPU. KF = no onboard GPU

[rocky root]

Can anyone help

I cant get Forge Webui to work

ComfyUI works

I made sure to open update.bat and let it download all the dependencies

[ornate elk]

hey, wahst your gpu?

[rocky root]

RTX 5080

[ornate elk]

Ah then thats the Problem. Forge doesnt support rtx50xx nativly.
You need to use Forge Neo which ia more updated.
Here can the Guide be found:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[rocky root]

I installed python

And it says this

Whenever I try and open Forge neo

[ornate elk]

pythion issue

[rocky root]

How do I fix it?

I installed python

[ornate elk]

you need to run the installer again, select modify, select next, and check "add python to environmentr variables"

[rocky root]

Using The windows 64 bit installer

Lemme try

Gotta restart pc real quick

Ok now its installing all the stuff

Oof im gonna run out of hotspot

Goddamn it

And I ran out ugh

[ornate elk]

oh damn

[proud fiber]

is it possible for batch converting to use subfolders? (recursive folders?) Im running blackwell version

[untold agate]

Hello @ornate elk , I have a problem with a workflow in ComfyUI.
One of the jobs in the workflow gives me the error attached.

Unfortunately I now have the only option to run it on a Macbook Air M2, so no NVIDIA for me.

Is there anything that I can do to switch from CUDA to mps? Or is there any other solution you would recommend?

Thanks in advance

[ornate elk]

oh cant help much with mac, but your out of luck if something needs cuda specifically

some tasks can be outsourced to cpu, but not everything

[untold agate]

Thanks @ornate elk I've heard there should be a way to use MPS instead of CUDA but I have no clue how to verify that

[ornate elk]

okay, sadly idk too ^^

[untold agate]

thanks anyway @ornate elk

[ornate elk]

but for mac, checkout https://drawthings.ai/
i think they use MPS

its the most optimised SD tool for apple

[sharp shore]

Is anyone available right now? I want to get this working with my fresh install haha

[sharp shore]

Never mind! I'm all set

[solid vector]

Hi guys i need help on using control net. i put the image on control net and i add some prompts and after generating it came out not exactly the image i wanting for.

it showed this error after trying it multiples times

[lucid yoke]

Are you using illustrious, if so are you using the illustrious specific controlnets?

Though im not sure if a1111/sd-web-ui supports that since its kinda outdated

[solid vector]

im using this model. i think is illustrious

let me check where i got it

yep is illustrious

[solid vector]

illustrious specific controlnets?

where do i get that

[river stirrup]

How do I get my chatbot i made in python to stop saying "You are a loser" or "I am not programmed to emulate the emotion of sadness. Im sorry" when I ask it what it means by "The IBM PC XR was painted red"

[lucid yoke]

CivitAI has em, but again i dont think your UI supports different controlnets

[solid vector]

oh

[lucid yoke]

? Just edit the code or base prompt

[lucid yoke]

Hmm wait it might

But yeah civitAI has them if you look for illustrious controlnets

[solid vector]

ok ok i'll try look for one thx fo the tip.

[lucid yoke]

Dw and goodluck. If you can't find them lmk I'm on my pc in a bit so i can link it

[solid vector]

i found one i search illustrious controlnet and it lead me Illustrious-XL ControlNet Openpose and when i downloaded it the file is "illustriousXL_v10.safetensors" where do i exactly put it?

[lucid yoke]

In the controlnets folder not sure where sd-web-ui stores that but probably under models somewhere

Also odd file name hmm

I'll check what i use in a sec

[lucid yoke]

https://civitai.com/models/929685/noobai-xl-controlnet
I used this one

[solid vector]

alr im gonna give it a try

[solid vector]

omg it work thx for the help

it looks exactly identical from the other image

[lucid yoke]

your welcome

[vast crest]

Hi, just curious. I saw these new models have come out: Chroma. I have and use Sd-Webui-Forge-Neo. Are they compatible with it?

[sharp shore]

does Forge Neo support Wan 2.2? Or will I need Comfy for that?

[lucid yoke]

Quick google search shows someone using it in forge, so you can give it a spin

Iirc chroma is very vram hungry

[lucid yoke]

Ive seen people on reddit using it in forge neo, you have to use the right vae so be mindful of that

[ornate elk]

Forge neo supports wan, qwen, chroma, flux and sd3

And of course the basics like sdxl, pony, illustrious, noob, 1.5

[ornate elk]

Yes see my message above

[vast crest]

so the lora chromas go into the Lora folders and the chromas checkpoints go into the Stable-diffusion folder?

[ornate elk]

That's correct

For loras make subfolders

To better organise them

[vast crest]

When I need to use Chroma checkpoints, which of these options should I choose?

[ornate elk]

Flux as chroma is based on it

[vast crest]

ahhhhh,ok,thx 🙂

[vast crest]

anyway

Google Cola hasn't been working for me for a few days now. I can't train my Lora. Can you help me?

[ornate elk]

Nope can't help with Google collab

[vast crest]

what do you use for training?

[ornate elk]

OneTrainer

[vast crest]

Anyway, However, I can't use Chroma because it gives me this error: AttributeError: 'SdModelData' object has no attribute 'sd_model'

[ornate elk]

which version did you tried?

[vast crest]

this chroma_v10HD

[ornate elk]

it worked with the qwen-image-Q4_K_M.gguf for me

[peak sigil]

I have tried many times but i have to ask this here after doing everything with zluda im trying to force webui-user.bat to use python 3.10 but it keeps using 3.12 no matter what i do

[ornate elk]

Which webui do you try to setup?

And what's your GPU?

[peak sigil]

Sorry for late ish reply im using sd next with zluda and i have an rx 580 i solved the python bs by removing my newest update from path and reinstalling sdnext now the the other issue im having is its saying modulenotfounderror: no module named optimum.onnxruntime
Plus at the start it says skipping onnxruntime installation

[ornate elk]

Have you already installed hip SDK 5.7?

I wouldn't recommend using sdnext for zluda.
Better use Auto1111 or Forge with Zluda

[peak sigil]

Oh sorry i meant sd zluda im following your tutorial for auto1111 and yes i have rocm hip 5.7

[ornate elk]

ah okay then make sure you have --skip-ort in the launch args

[peak sigil]

Yep there

[ornate elk]

can you show the full cmd log?

[peak sigil]

1 sec

venv "C:\FISH\SD-Zluda\stable-diffusion-webui-amdgpu\venv\Scripts\Python.exe"
WARNING: ZLUDA works best with SD.Next. Please consider migrating to SD.Next.
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-amd-43-g1ad6edf1
Commit hash: 1ad6edf170c2c4307e0d2400f760a149e621dc38
ROCm: agents=['gfx803']
ROCm: version=5.7, using agent gfx803
ZLUDA support: experimental
ZLUDA load: path='C:\ZLUDA' nightly=False
Skipping onnxruntime installation.
version check failed 'commit'
W1012 23:45:35.540373 8268 venv\Lib\site-packages\torch\distributed\elastic\multiprocessing\redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.
C:\FISH\SD-Zluda\stable-diffusion-webui-amdgpu\venv\lib\site-packages\timm\models\layers_init_.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers
warnings.warn(f"Importing from {name} is deprecated, please import via timm.layers", FutureWarning)
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.

C:\FISH\SD-Zluda\stable-diffusion-webui-amdgpu\venv\lib\site-packages\pytorch_lightning\utilities\distributed.py:258: LightningDeprecationWarning: pytorch_lightning.utilities.distributed.rank_zero_only has been deprecated in v1.8.1 and will be removed in v2.0.0. You can import it from pytorch_lightning.utilities instead.
rank_zero_deprecation(
Launching Web UI with arguments: --use-zluda --update-check --skip-ort --skip-python-version-check
Traceback (most recent call last):
File "C:\FISH\SD-Zluda\stable-diffusion-webui-amdgpu\launch.py", line 48, in <module>
main()
File "C:\FISH\SD-Zluda\stable-diffusion-webui-amdgpu\launch.py", line 44, in main
start()
File "C:\FISH\SD-Zluda\stable-diffusion-webui-amdgpu\modules\launch_utils.py", line 714, in start
import webui
File "C:\FISH\SD-Zluda\stable-diffusion-webui-amdgpu\webui.py", line 13, in <module>
initialize.imports()
File "C:\FISH\SD-Zluda\stable-diffusion-webui-amdgpu\modules\initialize.py", line 36, in imports
shared_init.initialize()
File "C:\FISH\SD-Zluda\stable-diffusion-webui-amdgpu\modules\shared_init.py", line 18, in initialize
from modules import options, shared_options
File "C:\FISH\SD-Zluda\stable-diffusion-webui-amdgpu\modules\shared_options.py", line 9, in <module>
from modules.onnx_impl.execution_providers import get_default_execution_provider, available_execution_providers
File "C:\FISH\SD-Zluda\stable-diffusion-webui-amdgpu\modules\onnx_impl_init_.py", line 6, in <module>
import optimum.onnxruntime
ModuleNotFoundError: No module named 'optimum.onnxruntime'
Press any key to continue . . .

2k character limit is why i split it into 2

[ornate elk]

ah okay, thats a new issue since a few days, to fix that:
edit the requirements.txt file and set optimum==1.27.0

then save, delete the venv and relaunch

[peak sigil]

delete the venv folder?

ok thankyouuu ill reply when it finishes

same thing

first time after installing it gives the expected error of ImportError: cannot import name 'broadcast_to' but then after i relaunch it gives that onnxruntime error

[ornate elk]

Hmm then try setup forge with Zluda instead

You already have done all the zluda steps so you only nees to git clone the forge repo from my guide and edit the webui-user.bat and run

[dusky osprey]

Hello! I had the chance to try zluda on a similar PC to mine, same gpu, but this one has 32gb RAM.
I made zluda run again but now I've a new issue

I tried two different FLUX worklflows but I get those errors

This is one of the workflows I was using

I'm using lowvram on the args and I think I've everything on FP8 for lowvram too

[peak sigil]

Uhhh i used forge and got the same error

[dusky osprey]

Damn. I've AMD, you too? Maybe is something related

[peak sigil]

Nope i was talking about the error on my last attempt here sorry for confusion

[dusky osprey]

Oh mbmb

[random night]

do we have any torch 2.5.1 for gfx 110x dgpu?

[ornate elk]

Why torch 2.5.1 ?

So far we only have 2.7 and upwards

[ornate elk]

Seems like your zluda files got probably flagged and removed by the antivirus

You have to whitelist the zluda folder and then delete the venv folder and launch the install-n.bat again

[random night]

well, because the requirements.txt in spark tts has that
einops==0.8.1
einx==0.3.0
numpy==2.2.3
omegaconf==2.3.0
packaging==24.2
safetensors==0.5.2
soundfile==0.12.1
soxr==0.5.0.post1
torch==2.5.1
torchaudio==2.5.1
tqdm==4.66.5
transformers==4.46.2
gradio==5.18.0

[ornate elk]

Then try a newer version

It will work with it just fine probably

[random night]

I have with no success, i'm busy right now, so later i'll show you the log

[ornate elk]

Alright

[ornate elk]

The bug has been fixed by the dev now.
You need to git pull to update or git clone the repo again

[solid vector]

hi guys i need some help i think theres a error. after i downloaded a lora and put it in the lora file. i open the lora tab and all my loras are not there

this is the file i put before it broke

i tried re opening the web ui and nothing work

[random night]

something like this

(sparktts) C:\AI\Paimon\Spark-TTS>pip show torch
Name: torch
Version: 2.10.0a0+rocm7.10.0a20251013
Summary: Tensors and Dynamic neural networks in Python with strong GPU acceleration
Home-page: https://pytorch.org
Author:
Author-email: PyTorch Team [e-mail]
License: BSD-3-Clause
Location: C:\Users\nhung\anaconda3\envs\sparktts\Lib\site-packages
Requires: filelock, fsspec, jinja2, networkx, rocm, setuptools, sympy, typing-extensions
Required-by: torchaudio

(sparktts) C:\AI\Paimon\Spark-TTS>
here if you wonder anything

[ornate elk]

Is the model you selected an sdxl model or 1.5 ?

[solid vector]

https://civitai.com/models/261433/line-art-sdxl

i got it from here

heres all the loras in my lora folder. the other 2 are flux and pony. i decided to download the SDXL after i relized that im been using wrong model for my img2img prompts when the other 2 didn't work

because i relized they are the wrong models

[ornate elk]

Okay, quick tip make subfolders in the lora folder for 1.5, sdxl, flux etc

And then restart the webui and paste a complete cmd log in here to check

[solid vector]

Ok I'll try it tmrw.

[dusky osprey]

Same issue here

I added zluda to the whiteilst of windowsdefender, deleted the venv folder and launched the install-n.bat.
When I run the prompt it took a bit of time and then crashed like that again

[ornate elk]

Hmm then try reinstalling it completely, I remember someone had the same issue a few days ago and whitelisting + reinstall worked

Also what's your GPU? You might dont need zluda

[dusky osprey]

Gotcha, I'll try reinstalling the whole thing. Do you think that whitelisting the WHOLE comfy folder may work better?
I've Radeon 9060XT. I read that Zluda worked "better" but lmk if this is not the case for mine

Thanks

[ornate elk]

Dont whitelisted the whole folder because custom nodes are not checked against malware.

So whatever you download after Comfyui or the manager wouldn't get checked

The best for your card would be the new native version of Comfyui that supports ROCm out of the box

[dusky osprey]

Thanks, I didn't know that

[dusky osprey]

Ok. I'm a total noob. Do you think is worth deleting all the zluda stuff and changing for the ROCm? I've read just a little about it but I don't know anything or what changes exactly.
My main objective is to use FLUX locally and combine them with different models or stuff like that

[ornate elk]

Its very easy.
Delete comfyui-zluda.
Dont uninstall HIP SDK 6.4 as its needed and supports your GPU.
Python and git dont hurt so no need to uninstall them.

Then download the latest Comfyui_windows_portable_amd
ZIP from here under assets:
https://github.com/comfyanonymous/ComfyUI/releases/tag/v0.3.64

Then unpack in a new folder called AI or whatever.
Then run the update comfyui stable from the update folder.
And finally run the amd portable bat

[dusky osprey]

Gotcha! Thanks mate. I'll do it rn

[quick fiber]

Hey. Im having issues with pytorch and not sure why it wont let me download it

    main()
  File "C:\Users\..\Documents\WebUI\stable-diffusion-webui-directml\launch.py", line 39, in main
    prepare_environment()
  File "C:\Users\..\Documents\WebUI\stable-diffusion-webui-directml\modules\launch_utils.py", line 614, 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

Any ideas on how to fix, im using a RX 9070

Ive downloaded AUTOMATIC1111 stable-diffusion-webui

[ornate elk]

Hey, you followed an old guide

[quick fiber]

oop

i do remember u tbf

i had problems earlier this year and finally got a new pc

How shall i redo this

[ornate elk]

follow one of my setup guides from here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides
the Auto1111 with Zluda
or Forge with Zluda
or "Any webui with TheRock" (experimental)

[quick fiber]

ok will do auto1111 with zluda

[ornate elk]

alright

[split heron]

Hi! I have been having troubles with creating a LORA. Regardless of tutorial/settings/trainer/images I use, I seemingly can't make a LORA that has any influence whatsoever. I am on a low-spec machine (6GB vram, 16GB ram) and have tried both SD1.5 and SDXL tutorials. I have tried a LORA that someone else made and it has an effect. I have tried via Forge, Kohya-ss, and OneTrainer.

I was wondering if anyone had like a minimal working example of training data and settings (e.g. json, maybe aimed more at a low-spec pc so it can be quicker for me to test) which I could test to confirm whether there is something deeper wrong with how the training programs are installed? Or maybe someone has an idea for a really fast diagnostic test (to generate a LORA that certainly should have an effect).

[rain stag]

HI, i was hoping to find someone to guide me with fixing a specific error code "RuntimeError: Your device does not support the current version of Torch/CUDA! Consider download another version:" im not all that techy, ive tried reinstalling cuda but line code isnt working for me

[ornate elk]

hey, whats your gpu?

[ornate elk]

okay make sure your gpu driver is updated

[rain stag]

all drivers are up to date

[rain stag]

it still showing same error

[ornate elk]

can you show the full cmd log?

[ornate elk]

hmm, try set up Forge Neo then

its more updated and better for rtx40 gpus

guide is in the pinned messages first link

[rain stag]

is there major difference

[ornate elk]

same ui, but more updated and better performance

[rain stag]

im kinda stuck at step 5, where is webui-user.bat

[ornate elk]

bottom file

[rain stag]

idk where it is

it didnt install where i wanted

[ornate elk]

you didnt do the first steps correct

cmd is not opened inside the folder

[rain stag]

so i wasnt supposed open it in the AI folder

[ornate elk]

in the AI folder type that into the explorer bar and hit enter

[ornate elk]

it is supposed to

but your cmd was not opened there

not correct

correct

[rain stag]

i think i found it but im not seeing bat

[ornate elk]

Bottom one

Windows 11 hides file extensions by default

You should enable it sometime

In explorer under View -> more options -> Show file extension names

[rain stag]

thank you i did not know that

[rain stag]

i have python, im not sure why it wasnt found

[ornate elk]

Run the python installer again then click, modify, next, and check "add python to environment variable's"

Then relaunch the webui-user.bat

[rain stag]

i think its working now, its intalling. thank you so much

is couple net better this version ?

[ornate elk]

NP, and do you mean controlnet?

Should work normaly

[rocky root]

Can anyone help

Im getting this weird ass error

And all I did was reboot comfyui

How in the fuck does this even happen

The pc isnt even connected to the internet

All I did was restart comfyui to get a new lora to work

Ughhh

Dont have any more hotspot on my phone

What the actual fuck

I dont get how this even happened

[rain stag]

i meant Latent Couple but i just noticed refiner isnt on menu bar. what would be install for that

[tribal anvil]

I installed windows 11 now I can't generate at all my laptop keeps needing to restart is there anything I can do?

I've been getting these everyday if it has anything to do with it

[cosmic furnace]

Forge Neo doesnt run with zluda or whuht? o.0

[cosmic furnace]

If not. What are best alternatives? Id like to test Qwen

[ornate elk]

It runs with ROCm/TheRock

[ornate elk]

There is the extension Forge-couple that works very good.
Refiner is not needed.

[cosmic furnace]

Is that rock good enough? what yud recommend to use now ? btw ? For best performance & is the most versatile and can use all checkpoints -Like qwen-flux-chroma ?

[ornate elk]

Yea TheRock is the fastest we have right now.
On Forge Neo you can use chroma, flux, qwen and wan with it.
There is just one bug currently that crash's the driver when not using Tiled VAE.

[cosmic furnace]

So yud recommend to go Rock + Forge neo ? It crashes because very new it seems? Lets hope upcoming updates fix it )

[ornate elk]

Yea it works great when you check Tiled VAE at the bottom of txt2img.
AMD is currently working on a bug fix for the VAE issue.

You can test TheRock, if you checkout my github guides. Its the "any webui with theRock"

There my example setup is ForgeNeo

[solid vector]

i did what you said

but

nothing showed up

[ornate elk]

Which model is selected?

What does the cmd shows?

[solid vector]

nothing. it showed nothing when i click on lora tab so i didn't choose any model. and this is what the cmd shows

also sorry for the late reply temporary blackout

nvm it fixed by itself. i just press the refresh button and then it showed an error which it didn't happen before. and the cmd shows a error.
and when i reopen

it got fixed

[ornate elk]

They won't show up because the 1.5 EMA pruned model is loaded

You need to switch the checkpoint (model) to something that is compatible with the loras

For sdxl loras a sdxl based model etc

[solid vector]

Ohhh ok ok

Thx for the knowledge

[ornate elk]

No problem 🙂 on civitai you always find a small box with infos of the loras or model and there you see the "Base" where you see with what its compatible

[tribal anvil]

I've been told comfyui is better for low end than webui forge is that true?

[ornate elk]

I would say the opposite. They performs similar but with forge you can test and know faster what you can or can't do with your GPU

[dusky osprey]

Did it and everything worked perfect. Ur the goat mate

[ornate elk]

Nice, no problem 🙂

[quick fiber]

@ornate elk i got a grey output from a generation but in the cmd everything looks fine. any reason for that?

[ornate elk]

probably an issue of the checkpoint

make sure its not a lora

[quick fiber]

ah makes sense. what ones cant i use?

[ornate elk]

loras are loras, they dont go into the models/stable-diffusion folder
they go into models/lora

[quick fiber]

i see

[ornate elk]

you need checkpoints (models) to generate images

loras are like addons on top of a model, boath need to be the same base type

[quick fiber]

okay

[quick fiber]

so something like this wouldnt work?

[ornate elk]

you would first need a compatible model to use it with

like:
https://civitai.com/models/376130?modelVersionId=2268916

[quick fiber]

ahh

okay

[ornate elk]

see boath have the same Base Model type (Illustrious)

so they are compatible

[quick fiber]

okok

[quick fiber]

even without using a lora

[ornate elk]

Can you show your txt2img settings and model?

[ornate elk]

flux is not supported in auto1111

[quick fiber]

rip

[ornate elk]

you would need Forge or Forge Neo for that

[quick fiber]

😭

[ornate elk]

but its not hard to install

[quick fiber]

yeah

will try it tomo

[ornate elk]

what was your gpu again?

[quick fiber]

9070

[ornate elk]

xt or non xt?

[quick fiber]

non

[ornate elk]

okay

yea flux will work but not that fast as sdxl etc

[quick fiber]

i see i see

[ornate elk]

Forge works with Zluda, and Forge Neo uses TheRock, boath support flux, but forge neo is more updatet

[quick fiber]

i see i see

i can have both at the same time i assume

[ornate elk]

yep

i have every webui and just link the same models folder to all

[chrome tangle]

Does anyone have any 50 series cards? I haven't genned for a month but i am getting cuda errors suddenly. Did Nvidia or Python release an update that fucked things up?

and my friend who has another 50 series reports the same

[ornate elk]

Hey, in which webui do you get errors?

And is your GPU driver updated?

[chrome tangle]

It is. Driver is updated. Cuda version 13.

and both invoke and forge

[ornate elk]

Forge Neo? Or Forge?

[chrome tangle]

Standard forge. And there is a forge neo you sya?

[ornate elk]

Yea Forge Neo is for newer nvidia GPUs

As it supports the latest torch and cuda versions

[chrome tangle]

Oohs...

[ornate elk]

You find a setup guide on the link in the pinned messages

[chrome tangle]

Thankies!

I will try it

[ornate elk]

Sure, let me know. But it should work very well

It also supports new models like chroma, wan, qwen etc, and even txt2video

[chrome tangle]

I am seeing that, that is exciting as hell D:

[lucid yoke]

SwarmUI is also a solid alternative 👀

[chrome tangle]

i get two errors normally

with normal forge

"NVIDIA GeForce RTX 5080 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90."

i haven't genned in a month, so i was just wondering if updates fucked a few things

[ornate elk]

Did you changed your GPU in that time?

[chrome tangle]

No, had the 5080 for about a year now

[ornate elk]

Ah okay, thats strange then

As forge didnt got any meaningful updates

[chrome tangle]

thats why am thinking an update to nvidia, cuda or torch has messed with things

[ornate elk]

Yea could be the case

But anyway wouldn't recommend normal forge for rtx50 as it won't likely get any native support for them or improvements.

But idk which torch version invoke ai uses. But maybe they need to updated their dependencies too

[chrome tangle]

@ornate elk When i get paid i need to buy you a game on steam or something, your guide fixed it for me

thank you

[valid holly]

Can somebody tell me what i am doing wrong?
i want to change the expression. it is not my first time doing this it worked fine yesterday.
i bet i just fumbled a basic setting but i cant find it...

[valid holly]

Okay i managed to fix it but i have no clue how i even did that.
so if somebody sees my error pls still point it out... 😄

[rain stag]

i was hoping to get some assistance on issue of getting the installed extensions to appear on menu/Ui.

[ornate elk]

Tagcomplete will show up when typing words.
Refiner is not needed and I guess not compatible with forge neo.
Dreambooth is outdated and breaks the whole setup.
The dataset tag editor is also outdated and recommends a standalone version instead.
The two shot extension is also out dated and regional promoter doesnt support forge. You should install Forge couple instead.

To fix your webui delete all of thesr extension but not the webui - tagcomplete one.
Then delete the venv folder and relaunch the webui-user.bat.

[rain stag]

Where do I get forge couple ? What’s good way make check points, I was told dream boat was the way to go

[ornate elk]

You have to mask only the face and not more and then set the Inpaint Area to "only masked"
Then set the resolution to 1024x1024 and denois to 0.5 or higher.
Then it will work.

[ornate elk]

To install forge couple:
Click on extensions, click on "install from URL"
Copy and paste it's URL:
https://github.com/Haoming02/sd-forge-couple
Then click install.

Dreambooth was the way 2 years ago.
Now everyone uses Khoya_ss or OneTrainer to train models and loras.

Boath are standalone tools, not extensions

[sly pike]

please help me.
how can I fix this problem? I have 5060ti gpu. tried lots of ways to fix ts but nothing works(I might do smth wrong)

[ornate elk]

Forge doesnt support rtx50 cards natively.
You need to use Forge Neo which is more updated.
A setup guide is on the first link of the pinned messages here.

[silent stream]

Where is that setting to only display compatabile lora with the loaded Checkpoint?

For Extra Network

I am trying to see if I can separate Pony Loras and Illustrious Loras

[ornate elk]

Its active by default. The issue is that boath pony and Illustrious are based on sdxl. So they all show up on any of these models

[silent stream]

Darn

[silent stream]

No way to separate them then?

I have been searching "illustrious" so it only shows illustrious based model along with any other search words but it is still annoying

Since I used file structure to separate them

[ornate elk]

Yea subfolders will work

Thats the best way to organise them

[silent stream]

😭

[quick fiber]

@ornate elk RuntimeError: PyTorch is not able to access CUDA

@echo off
set TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1

:: set PYTHON=
:: set GIT=
:: set VENV_DIR=

set COMMANDLINE_ARGS=

:: --xformers --sage --uv
:: --pin-shared-memory --cuda-malloc --cuda-stream
:: --skip-python-version-check --skip-torch-cuda-test --skip-version-check --skip-prepare-environment --skip-install
:: --skip-python-version-check --cuda-stream --attention-pytorch

call webui.bat

this is in the thingy

i even moved the --skip-python-version-check --cuda-stream --attention-pytorch around and no fixy

do i need to download cuda or is it something else

[ornate elk]

Can you show the full cmd log?

[quick fiber]

venv "C:\Users\..\Documents\TheRock\sd-webui-forge-neo\venv\Scripts\Python.exe"
Python 3.11.9 (tags/v3.11.9:de54cf5, Apr  2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
Version: neo
Traceback (most recent call last):
  File "C:\Users\..\Documents\TheRock\sd-webui-forge-neo\launch.py", line 52, in <module>
    main()
  File "C:\Users\.\Documents\TheRock\sd-webui-forge-neo\launch.py", line 41, in main
    prepare_environment()
  File "C:\Users\.\Documents\TheRock\sd-webui-forge-neo\modules\launch_utils.py", line 315, in prepare_environment
    raise RuntimeError("PyTorch is not able to access CUDA")
RuntimeError: PyTorch is not able to access CUDA

Press any key to continue . . .

[ornate elk]

You installed it in documents folder. That can make permission issues

Redo the steps but create a new folder on C and dont copy this one over

[quick fiber]

ok

[quick fiber]

i had the other one in my doc's folder so i thought it would work

venv "C:\AI\TheRock\sd-webui-forge-neo\venv\Scripts\Python.exe"
Python 3.11.9 (tags/v3.11.9:de54cf5, Apr  2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
Version: neo
Traceback (most recent call last):
  File "C:\AI\TheRock\sd-webui-forge-neo\launch.py", line 52, in <module>
    main()
  File "C:\AI\TheRock\sd-webui-forge-neo\launch.py", line 41, in main
    prepare_environment()
  File "C:\AI\TheRock\sd-webui-forge-neo\modules\launch_utils.py", line 315, in prepare_environment
    raise RuntimeError("PyTorch is not able to access CUDA")
RuntimeError: PyTorch is not able to access CUDA
Press any key to continue . . .

same issue

[ornate elk]

Ohhh I know the issue now, I need to update the guide.
Give me 15-20 mins

[quick fiber]

okok

[ornate elk]

so the guide is now updated with the correct commands for the rx 90xx gpus

[quick fiber]

does it auto update

[ornate elk]

you have to reopen my guide and do the setup of the rock again

[quick fiber]

yeah

so since mine doesnt open the browser

can i continue still to the setup of therock?

[ornate elk]

yep start there

[quick fiber]

cool beasn

[near dome]

Bro can u help me?

I tried to download that fck Stable Diffusion but portable

And it is not wrk

[ornate elk]

whats your gpu?

and which webui

[near dome]

By Serpotapov

[near dome]

4060

On laptop

16 RAM

[ornate elk]

ah okay, I wouldnt recommend that version its outdated.
Better setup Forge or Forge Neo.
You find boath setup Guides here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[near dome]

Which one is better?

[ornate elk]

Forge Neo has more updates and better performance

[near dome]

First one?

Im dumb, sorry

[near dome]

So iv chosen this))

[ornate elk]

thats the correct one

[near dome]

Nice

[near dome]

And whats then

I got it

Just launch?

[ornate elk]

yep click trough the installer and leave it its default settings, Git is just needed to download the webui stuff

[near dome]

Idk There were a lot of settings, I just clicked "next"

[ornate elk]

this can be closed

[near dome]

Next step?

[ornate elk]

install python

like the guides says

3.11.9 64bit

[near dome]

Where is the guide? Maybe i could nt distrust you

I am trying to install this soft for the last 2 hwrs🤣🥲

[ornate elk]

a linked it a few mins ago:
#🤝｜tech-support message

[quick fiber]

venv "C:\AI\TheRock\sd-webui-forge-neo\venv\Scripts\Python.exe"
Python 3.11.9 (tags/v3.11.9:de54cf5, Apr  2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
Version: neo
Installing packaging
Installing clip
Installing nunchaku
Failed to install nunchaku; Please manually install it
Installing gradio

wats nunchaku?

[ornate elk]

some performance tool that helps in performance for qwen and wan, but is currently only usable with nvidia gpus

[near dome]

So I got pyth and Git

The Third step?)

[ornate elk]

Here are the Guides, follow the Forge Neo setup for nNvidia there:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[quick fiber]

uh

doesnt open

[ornate elk]

ohh damn, it seems that it got an update that broke some stuff.
I got the same issue now, but yesterday it worked normaly.
I will open up an issue on the Forge Neo repo and hopefully the dev finds a fix.
I will let you know, sry

[near dome]

Bro there is no webui-user.bat

[ornate elk]

at the bottom

[near dome]

.bat is lost

[ornate elk]

yea this, windows 11 just hides file extension names by default because microsoft is stupid

[near dome]

Oke:)

[ornate elk]

you can enable file extensions in the explorer by clicking on top at "view" -> "show" -> "show file extension names"

[near dome]

I did edit

Turned off my Win Defender

Then..?

After the Installation which can take a while, is done, the Webui will autolaunch in your Default Browser.

But nothing happens

correct, isn it

[ornate elk]

can you show the cmd log?

[near dome]

which

sec

[ornate elk]

the thing you have to launch

black window

[near dome]

ist gone

already

[ornate elk]

relaunch

[near dome]

It is not even launching

And now it is working

Sry buddy

[ornate elk]

np ^^

[near dome]

3.5 GB yea?

[quick fiber]

alrighty

[near dome]

and it stopped

[ornate elk]

just wait a bit

[near dome]

:3

[near dome]

it is like 10 minutes

[ornate elk]

make sure you didnt clicked into the cmd, that pauses it, press space to unpause it

[near dome]

I didn’t

It is the same problem I had last time..

[ornate elk]

ho much RAM do you have?

[near dome]

16

[ornate elk]

ah that makes it takes longer to install probably

you should increase the windows pagefile then to get better performance

[near dome]

How long can it take?

[ornate elk]

probably max 30-40 minutes

[ornate elk]

for now you can quickly fix the issue by enlarging your Windows Pagefile:
Here is the Guide how to increase it:
https://www.tomshardware.com/news/how-to-manage-virtual-memory-pagefile-windows-10,36929.html
Enable it only for C and make sure its not enabled for any other drive.
Set it to 16000 Min and 24000 Max for C
Then apply and reboot the PC.

You also need to do this step if you want to use larger image models in the webui later.

[near dome]

Could this be because I didn't run the file as administrator?

[ornate elk]

nope, you should never run the bat as admin

that causes issues

[near dome]

Ill wait half of hr more

[near dome]

Then try

[ornate elk]

yep alright

[near dome]

Now it snot even starting

[near dome]

did

Waiting)

[near dome]

Thanks!

[timber gate]

Yo can someone help me with installing forge ui neo? I’ve been having issues with my original automatic1111 forge ever since I upgraded to a 5070ti and I keep getting no kernal error msg every time I get it launched and running.

keep getting this

[ornate elk]

Hey, follow the setup guide for Forge Neo from first link of the pinned messages, and make sure your GPU driver is updated.

[timber gate]

i have to be on cuda 13?

[ornate elk]

Nope cuda should not be installed manualy

Just update your GPU driver with the nvidia app

[timber gate]

Then launch it?

Only other thing I’m having issues downloading is sageattention, I installed everything pretty easily a chat gpt, sent me all the cmds to paste in powershell

Bear w me I’m still kinda new to all ts

Only thing I manually installed was python 3.11.9

[ornate elk]

Please don't use chatgpt for the install. The guide covers every step and sage attention doesnt need to be installed manually If your not using an rtx50 series

[timber gate]

I am using the 50 series

[ornate elk]

Ah alright then

If sage attention 2 doesnt work you can just use flash attention

[ornate elk]

Hey, the repo got updated and it should be fixed now.
You either need need to add
Git pull in an empty line of the webui-user.bat or you need to git clone the repo again.

[quick fiber]

like dis?

actually ill just git clone it again

[ornate elk]

Have you done the therock steps?

[quick fiber]

yup

[ornate elk]

Hmmm

I'm home in 30 mins, then I will test it with my setup

[quick fiber]

Let me try something one sec

[quick fiber]

didnt work 😭

[vague prawn]

Hi' i've installed the stable diffusion webui for amd with Zluda, any way to save my text2img settings so that i don't have to choose them every time i restart the UI?

[upbeat trail]

Anyone know how I can run NovelAI Diffusion V2?

Preferably with ComyUI. I don't see any tutorials for it

[ornate elk]

You can edit the ui_config.json or click the blue white arroe to apply the last used settings

[upbeat trail]

https://huggingface.co/NovelAI/nai-anime-v2/tree/main

[ornate elk]

Why do you want such old model to work with?

[lucid yoke]

"This is our old, more capable SD1.5 based anime model. "
The base sd 1.5 workflow in comfyUI should work

Kinda nutty that 1.5 is still being used today

[ornate elk]

If you want to use it then just download the 2gb model And put it into models/checkpoints

And use the 1.5 workflow like euro said

[ornate elk]

True xD

But I'm still sad we have no sdxl equivalent to offset noise

I know there are loras but in 1.5 days I merged offset into the models and got nice dark images

[lucid yoke]

True dark images with the newer models are rare indeed

[vague prawn]

Where are these arrows?

[ornate elk]

It seems like you installed a different overlay

[vague prawn]

i honestly don't know, just followed the Zluda guide on github

can i change it?

[ornate elk]

Can you share which guide you followed?

The normal one doesnt include this ui

[ornate elk]

i clean installed it and it works for me now

[quick fiber]

hmm

[quick fiber]

what did u put in the web thing

[quick fiber]

Already up to date.
venv "C:\AI\TheRock\sd-webui-forge-neo\venv\Scripts\Python.exe"
Python 3.11.9 (tags/v3.11.9:de54cf5, Apr  2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
Version: neo
Traceback (most recent call last):
  File "C:\AI\TheRock\sd-webui-forge-neo\launch.py", line 52, in <module>
    main()
  File "C:\AI\TheRock\sd-webui-forge-neo\launch.py", line 41, in main
    prepare_environment()
  File "C:\AI\TheRock\sd-webui-forge-neo\modules\launch_utils.py", line 315, in prepare_environment
    raise RuntimeError("PyTorch is not able to access CUDA")
RuntimeError: PyTorch is not able to access CUDA
Press any key to continue . . .

still get that

[quick fiber]

smh pytorch smh

[ornate elk]

and you have HIP SDK 6.4 installed?

[quick fiber]

should do

how do i check

[ornate elk]

and did the uninstall torch commands and pressed y two times and installed the new torch?

[quick fiber]

yeah

[ornate elk]

in the system control pannel

[quick fiber]

?

im gonna see if redownloaded it works but i had to do it for the previous auto1111 i had so idk

C:\Windows\System32>hipcc --version
HIP version: 6.4.50101-9a6572ae7
clang version 20.0.0git ([e-mail]:Compute-Mirrors/llvm-project 33ab2c2f7838239f1e2e5c06432bbb8d887e8cb2)
Target: x86_64-pc-windows-msvc
Thread model: posix
InstalledDir: C:\Program Files\AMD\ROCm\6.4\bin

i do have it

yet no worky

[ornate elk]

Hmm strange

What's your torch version in venv/lib/site-packages folder?

@quick fiber

[quick fiber]

helo

how does one check

[ornate elk]

Go inside the folder path I wrote to see the torch folder

[ornate elk]

Ah okay so the install didnt worked correctly

[quick fiber]

i see

[ornate elk]

When you do these steps can you screenshot it?

[quick fiber]

ok

[vague prawn]

Could you pls send me the link to the correct one? I have amd GPU

Maybe it was some strange fork

I don't have the link right now

[ornate elk]

Sure here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides
Follow the Auto1111 or Forge with Zluda Guide

[vague prawn]

Thanks, later I'll try to reinstall

[ornate elk]

Looks good

Afk now

[dreamy lodge]

RuntimeError: mixed dtype (CPU): expect parameter to have scalar type of Float

Anybody know what could be throwing this error?

Heres the log

[ornate elk]

your HIP SDK is not the latest version and your torch version is broken

[timber gate]

would u be willing to join me on call so i dont mess anything up, ill stream it.

[ornate elk]

dont have time for that sry, can only help in chat

[quick fiber]

wait tf it works

[ornate elk]

perfect ^^

[quick fiber]

well

it goes to the web

ill try models

[ornate elk]

make sure you enable "tiled vae" at the bottom of txt2img

[quick fiber]

where?

[ornate elk]

at the "never oom" option

[quick fiber]

tyty

[ornate elk]

but only enable tiled vae, not the other one

[ornate elk]

you need a model

and select the correct version on the top left

[quick fiber]

oh wait

it dipeed

[ornate elk]

yep but flux needs 3 files more

vae, and 2 text encoders

and the 23gb model wont work

[quick fiber]

https://tenor.com/view/head-explode-mind-blown-gif-11038567854390045170

[ornate elk]

better start with something easy

[quick fiber]

okay

[ornate elk]

like an illustrious model

[quick fiber]

how do i get a VAE / text encoder

[ornate elk]

from civitai or huggingface

[quick fiber]

ok

[ornate elk]

for sdxl/illustrious get this vae:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/blob/main/sdxl.vae.safetensors

[quick fiber]

okay dokes

where would i put it

[ornate elk]

models/vae

[quick fiber]

alrighty

[ornate elk]

and then get for example this model:
https://civitai.com/models/1986798/cyberrealistic-cyberillustrious-catalyst

and put it into models/stable-diffusion

also make subfolders for illustrious, pony, flux, sdxl

so you dont mix them up

[quick fiber]

okok

suprised it doesnt have them already

[ornate elk]

true xD but no webui has them

[quick fiber]

🫡

[quick fiber]

ggs

[ornate elk]

try pause and unpause the download, sometimes it stuck on slow mode

[near dome]

Bro, why are the models looking crooked and LoRa not working? How do I set it up?

[quick fiber]

what ui do i need for illus

[vocal burrow]

xl

[ornate elk]

Loras and models work only if they are based on the same base model type

[quick fiber]

still dont work

[ornate elk]

Can you show the full cmd log?

[quick fiber]

maybe cos i put it in the sd folder cos it doesnt see the illustrous folder?

[ornate elk]

Ah it still tries to load the flux model on startup

Delete that and relaunch

[quick fiber]

okay doeky

okay its doing stuff

i assume itll take a bit before it creates anyhting

[ornate elk]

Yea first time it can take a bit

[quick fiber]

i feel the pc being sloww

..

whats it with the grey omfg

and a regen has broke it

[near dome]

How can i check it

[ornate elk]

Relaunch, then make sure to enable the tiled VAE option.
Also select Euler a as sampler

Then generate again

[ornate elk]

Every model and lora on civitai shows its "Base" in the info box on the right of their pages.

[quick fiber]

Pc did not like that

Screen go black

I think it crashed..

[ornate elk]

What was your GPU again? And how much ram?

[quick fiber]

32gb

[quick fiber]

I assume so. Display went black

[quick fiber]

Ddr5

[ornate elk]

And the GPU model?

[ornate elk]

Probably a driver crash

[quick fiber]

9070

Non xt

[ornate elk]

Ah okay

16GB vram

[quick fiber]

Mhm

[ornate elk]

Make sure you dont run anything else in the background, like Wallpaper engine or games, twitch etc

These will cause issue when generating at the same time

[quick fiber]

Ok

[ornate elk]

And then try again. It could crash again like stated in my guide, but mostly after the third time it will then work normaly

[quick fiber]

It's stuck at 95% eta 3sec

Discord won't even open 😭

[near dome]

It will work only if Base Model?

Whats the best model

[ornate elk]

For what? Anime, realism? Semi realism?

[near dome]

Realism

Incl. people

[ornate elk]

You can test this one then:
https://civitai.com/models/312530?modelVersionId=2152184

[vague prawn]

hi, i've downloaded the controlnet extension but the webui won't load anymore. Using amd with zluda. if i remove the extension from t he folder, it starts again

Blocked here for 10 minutes

[quick fiber]

I think it bricked my connection to the wifi..

I'm.connrcted to the wifi yet task manager says 0kbps in and out

[ornate elk]

Have you tried restarting the PC?

[quick fiber]

Yup

Deleted the drivers in device manager too

[ornate elk]

Does browsing works?

[quick fiber]

Nope

[ornate elk]

Have you changed anything before it broke?

[quick fiber]

Nope

[quick fiber]

It was after this

[ornate elk]

Does your smartphone is connected to WiFi right now and not to 5g or LTE?

[quick fiber]

Wifi

[ornate elk]

Using the GPU or a GPU driver crash can't kill the WiFi

Thats a very strange connection xD

[quick fiber]

Yuh

I've tried to connect to my hotspot now and it says no Internet secured

[ornate elk]

Hmm you can try reset the network connection In the Windows settings

Its under troubleshooting I guess

Or use a second device and go to the website of your Mainboard and there download the latest WiFi driver.
Or plug in ethernet and then redownload the WiFi driver.

[quick fiber]

I'll restart one more time

[ornate elk]

Make sure to click restart and not shutdown

[quick fiber]

Still the same

[ornate elk]

Try open up a cmd and type
ipconfig

Then you should see if it got an IPv4 Adress

[quick fiber]

I do

[ornate elk]

Then type
ping 8.8.8.8

[quick fiber]

4 sent 4 recievef

[ornate elk]

So internet seems to work

[quick fiber]

Oop it just disconnected

No Internet now

No wifi networks found..

[ornate elk]

Maybe unstable location or router update?

[quick fiber]

Uni wifi

Working on everything but my pc

[ornate elk]

Oh so maybe its just overloaded or unstable

[quick fiber]

Mayb

However my hotspot doesn't connect to it

Or well it does but says like no Internet whatsoever

[ornate elk]

Make sure to disable WiFi on the phone when enabling the hotspot

[quick fiber]

I have

[ornate elk]

If not it tries to route it again to the WiFi network

[quick fiber]

A video said I might need to insert my ip manually?

[ornate elk]

Probably not

But idk why the hotspot doesn't work. Thats not easy to troubleshoot from remote xD

After connecting to the hotspot you could try to restart the PC so that the first connection it gets is the hot spot again

That makes sure its In the same IP range

[quick fiber]

I see

[ornate elk]

Also if your uni blocks some websites you should change your PCs DNS server to 1.1.1.1 or 9.9.9.9 or 8.8.8.8

[quick fiber]

It's at 8.8.8.8 I think

[ornate elk]

I need to go off now, its getting late.
Let me know if it works tomorrow or if you have any questions.

[quick fiber]

Okok

Cya

fixd it

u woukdnt guess what it was

vpn

it was off yet still bricked it

[valid holly]

I just found out that even tho stable diffusion is not actively generating images it still claims its V-ram like a hungry beast, resulting in my text model crashing when trying to respond...

is there a way to make stable diffusion only eat V-ram when it needs it?
i was planning on mainly texting and then every once in a while generating one image. is that possible or do i have to sacrifice something to make more headroom for V-ram?

[grim cape]

How much VRAM do you have?

[valid holly]

24 i think (4090)

[grim cape]

Damn. I’ve never had any issues with that and I’m on a 4070 Super

Half your VRAM

[valid holly]

maybe bcs i try running this Cydonia-24B-v4j-Q8_0.gguf
i think i just have to go lower.
its just so wrong in my head when there is a better version 😄

[late tree]

My old installation of SD Next stopped working for some reason so I tried to reinstall it again, but for some reason the command line gets to a point and says "Press Any Key To Continue" and when I press a key it just closes, with nothing in the logs indicating any problem...

[lucid yoke]

Theres comfy nodes that unloads a model after generating but if you also load a llm that big thats also in vram it will compete for the space and run one in normal ram making it hella slow

[ornate elk]

Oh wow, nope wouldn't have guessed that

[ornate elk]

What's your GPU?

[late tree]

Radeon RX 6800 XT

Install worked fine before but I hadn't used it for a while, went to try it again and it didn't work so I tried a fresh install and it just stops. No errors in the logs

[ornate elk]

Are you on HIP SDK 6.2 or 6.4 ?

[late tree]

How do you check that?

[ornate elk]

Open up a cmd and run:
hipcc --version

[late tree]

Exactly that? Does it matter what directory the cmd is in?

Cause just opening cmd and using that just says 'perl' is not recognized as an internal or external command, operable program or batch file.

[ornate elk]

Okay strange

Then check for hip SDK in your apps or system control panel

[late tree]

I should have it installed, did before at least.

HIP SDK Core? Two of them 5.5.0 and 5.7.0

[ornate elk]

Oh they are very old

Uninstall everything you have from HIP SDK

Should be like 4-8 programms

[late tree]

Uninstalled those and installed the 6.4 one, though it is still coming up "Press a key to continue" and then closing with no error message. Might need to restart before it works

[ornate elk]

yes PC restart and the deletion of the venv and .zluda folder in the webuis folder is required

[late tree]

I'll have to do the restart later, need to sleep some before work... Yay...

[ornate elk]

alright 👍

[near dome]

Isnt this

https://github.com/cruelpleasure/sd-webui-forge-NEO?tab=readme-ov-file

Forge NEO the best?

Inpaint works bad

SDXL 1.0 is a little worse

Or am I doing some shit?

[ornate elk]

Idk, what issue you exactly have xD

[near dome]

How can I use Loras that based in Illustrious?

[ornate elk]

by downloading an illustrious based checkpoint (model)

[near dome]

Can I get a link? Illustrious XL is worse

It doing random shit

[ornate elk]

anime or realism?

[near dome]

Real

[ornate elk]

https://civitai.com/models/1125067?modelVersionId=2239173

[near dome]

Marked as NSFW 😅

[ornate elk]

oh

[near dome]

Is there opportunity to do good AI porn?🤔

I would test it

[ornate elk]

not allowed on this server

but try the model i linked it can do nsfw stuff

[near dome]

And it goes on Forge Neo?

[ornate elk]

yea in models/stable-diffusion folder

[near dome]

I downloaded it earlier, it works also bad…

[ornate elk]

you can show examples or get prompting help in #📝｜prompting-help

mostly its not the models issue

[near dome]

Sd / Xl ?

[ornate elk]

internet issue?

[near dome]

No

Soft issue

[ornate elk]

whats your gpu again?

and which settings did you tried?

mostly the crash comes from "out of vram" when using to high resolution or settings

[near dome]

4060

[ornate elk]

with 8gb vram?

[near dome]

Yes

[near dome]

look what I got

https://github.com/cruelpleasure/sd-webui-forge-NEO?tab=readme-ov-file

[ornate elk]

seems very bugged

[ornate elk]

this link is the wrong forge neo version

[dense echo]

So I downloaded comfyui from the official website and installed the image to video template

What other steps do I need to complete to start generating image to video

Or can I start right away?

[stray jacinth]

Make sure to disable WiFi on the phone when enabling the hotspot

[tawny saddle]

cs1o to da rescue

can chu send me a link to the AMD gpu version to da newest stable pwease

[dense echo]

Did I do something wrong here?

nothing is happening

[tawny saddle]

potato

:3

[tawny saddle]

xdd

[tawny saddle]

wut to do's

i dunt get it, cuz it updated

imma try reinstall bc i stopped it a few times in, to move it to another drive that wasnt an SCSi thing

and i dunno, maybe that broke it or sum

[tawny saddle]

didnt work :(.

its giving me this, but this is supposed to be the AMD version on stable...
it shouldnt need cuda

hhhhhhhhhhh, so it looks like the version i was tryin was sum called 1.10.1, which is apprently a merge of both why it dunt still dunt work even being a merge of both amd and nividia, beats me, but now im trying wut i assume is 1.10.0, and AMD, so, maybe now it work aye

seems to be working nows

ish downloading da modelu, next imma test if its actually using me gpu

[tawny saddle]

kinda confuzzled

it works, in the most liberal usuage of the phrase, but, i feel like i should be able to use --medvram, but it VIDEO_MEMORY_MANAGMENT_INTERNAL's on me using that

but i do believe my lil laptop with a processor not as capable as this desktop ran using that arg, i mean sure i typically kept the res like 512x512, but 620x750 aint that much bigger init

kinda stuck using --lowvram, ig its fine, it works, cant really complain, but ig AMD5600G with Radeon Graphics is kinda meeting its match aye

kinda weird that i can play games at like 1600x900 res, BeamNG and MSFS2020/2024 being such games, but it cant generate images that high without crashing 🤔 ..ig maybe generating images and rendering things is a lil diff under the bonnet

[lucid yoke]

Amd falls behind on consumer ai, you sadly need to emulate using zluda or directML

[lucid yoke]

Pinned messages in this channel

[tawny saddle]

i got directml uvu

i assume

xd, all ik is asked for it, i set it and it dunt ask for it

where do i put upscalers

ig in here aye

[lucid yoke]

Yeah

[lucid yoke]

Still, zluda is the recommended method, the cs1o guide will probably speed up gens

[tawny saddle]

ill look at it, cs1o may as well be stabe diff but as a person, cuz know so much about it, cs1o might be secret dev 👀 xd

[tawny saddle]

so uhm

where is that guide

[lucid yoke]

Pinned messages

#🤝｜tech-support message

He updates them on that site

[tawny saddle]

ahh, thanks chu

how come, larger models take longer to generate images xd

i mean, from an obvious perspective, its bc its bigger, but, why it be that way

like, if its loaded already, id imagine that its done, and all u gotta do is just use it, so, when its already loaded, and u finished waiting for that, why would it continue to take heaps to make an image wit it when its loaded and gud to go 🤔

[lucid yoke]

Hmm well if the model is larger then your vram capacity it offloads to normal ram

That is drastically slower

And nowadays like flux, chroma etc have large text encoders (that use a llm) so its no longer 8gb and your done

Its 12gb + 5gb + 7gb for example

But if it nicely fits its quite fast still

[tawny saddle]

ahh i see

[lucid yoke]

I used to run a 3070 before and when i ran sdxl it took 20s, (older hardware and loras taking up space)

And flux took a whole minute

Now? Flux takes 15s and sdxl 5s, 15s with 10 loras

[tawny saddle]

;o i really need an dedicated gpu

my lil 5600G apu is getting left behind

[lucid yoke]

Hmm it honestly depends, if you plan to do image generation a lot it might be a good investment but if you dont do it a lot and it still works for your other needs such as gaming i wouldn't upgrade

[tawny saddle]

fair xd

u been really helpful, answered all my curious questions :3 💜

would a 3070 still be pretty gud for stable, say at like 1080 res stuff

[lucid yoke]

Hmm honestly, only a 3090 is a bit relevant nowadays

3070 can run flux schnell and sdxl but out of luck for the newer models

[tawny saddle]

oh ;o

[lucid yoke]

But you can generate at 1mp yes

Iirc theres some people who rent cloud gpu's to gen without too much of a cost/hassle or use cloud generators if the prices are not too bad

[tawny saddle]

eh, id jus like to sum times make images, for fun, and it not take 15mins to 30 xd, and i dunt plan to train or do large batches

i mean, 8gbs of vram is better than my 512mbs of vram aye xd

[lucid yoke]

Oh yeah definitely

512mb is really the reason why it takes so long oof

Not knowing the rest of your computer specs but the newer entry level hardware might be better (iirc a 5070 a better deal then a 3070)

[tawny saddle]

yas :3 xd
currently using 5600G apu, and its 512mb vram, and i have 16GB ram, and its using a ~21GB pagefile on an pny ssd xd

so yas, it do be slow xd

[lucid yoke]

If you want no hassle consumer ai, get nvidia

[tawny saddle]

yas

so ive heard, or read rather- xd

AMD is def a hassle wit ai

bc when u think of an graphics card, most ppl name nividia

so, there isnt really a lot of support for AMD things, at least thats how i think of it

it got tired

xd

[ornate elk]

Upscalers go in models/ESRGAN

[tawny saddle]

cs1o has returned :o. xd

[ornate elk]

And yep you should get a dedicated GPU sometime ^^

[tawny saddle]

Yas :3

[dense valley]

This error message keep popping out. Sorry im new to SD.

[lucid yoke]

error on their end, nothing you can do

i dont recognize the UI but at worst you can contact them directly

if this keeps happing later

[vocal burrow]

probably some third party website

Probably some url squatter such as stablediffusion.com or stablediffusionweb.com (very likely according to the white theme I can see there).... Real owners behind those websites are known as darkmagic.cc... Seems like they now moved to "Black Technology LTD"... very creative....

Those are not official ones and have been known for their unreliability and shady practices around here.

The fact that their refund policy can be summed up as "no lol" is telling.

The only official website is https://stability.ai/ and stuff linked from there such as https://dreamstudio.stability.ai/

Not to say that it is the only platfom to be trusted, but I strongly advocate you to do your researched before giving money to similar sounding / confusing / suspcious websites.

[dense valley]

Thanks for the info ill make sure to subscribe to official website

Yes this is stablediffusionweb 😢

[vocal burrow]

oof sorry about that, you can try mailing them about the issue and/or ask for refund but I wouldn't hold my breath about that.

[vague prawn]

any way to make controlnet work on zluda with rx 7900xtx amd card? I've installed from git link, but the webui doesn't start. Only way to make it start is deleting the venv folder and reinstalling everything, but it will only work for that session, then again, webui not starting

[haughty pilot]

hey, what's the recommended UI for rtx 5090? forge?

[ornate elk]

You need to install the controlnet extension not the original controlnet

Then it works

[ornate elk]

Forge Neo, As it supports rtx50, not like forge

[vague prawn]

Isn't it this one? https://github.com/Mikubill/sd-webui-controlnet

where is the original

[ornate elk]

Yes this one

[vague prawn]

i have this

[ornate elk]

Which Webui do you use?

[vague prawn]

zluda

amd gpu

It blocks here

and do nothing

[ornate elk]

Can you show the full cmd log?

[vague prawn]

sure

[haughty pilot]

do you mean reforge?

[vague prawn]

Ok i think i've solved

--disable-nan-check-ort i had this in the launch options, removing it make it work

i don't even know what was that for. Is it something important?

[ornate elk]

Nope Forge Neo, a setup guide for it is in the pinned messages

[ornate elk]

It hides the error and doesnt fix anything

[vague prawn]

ok but now it starts and controlnet works :/ what happened here

[ornate elk]

Mostly some wrong usage or models you use with Controlnet

I have no issues with it on my 7900xtx with Zluda

[vague prawn]

i'll see if it works from now on, thanks for the help

[ornate elk]

Alright np

[haughty pilot]

thanks, are all extensions from forge supported in forge neo?

[ornate elk]

Not the inbuilt ones from forge, but most of the others

[cosmic furnace]

@ornate elk I have a question about TheRock.

Do I need to install a separate ROCm distribution (like a standard installer) if I'm using TheRock-forge neo? Or does TheRock manage the dependencies itself? 🤔

Also, I heard that ROCm 7 for Windows is available now. Is that correct, or did I misunderstand the release status?

[sick merlin]

Forge Neo was working for me for like 50 or so generations but ever since then it's only been generating black images

And I'm not sure what the issue could be, I've reinstalled the whole repo like 10 times and it still has the same results

[sick merlin]

is it always best to be on the latest Nvidia drivers for Stable Diffusion? I'm on a version few months back because newer drivers made the games I play laggy

[ornate elk]

Hey, yes you need to install HIP SDK 6.4 for TheRock and zluda.
Then using my setup guide for forge neo will install rocm 7

[cosmic furnace]

Btw, I'm choosing between Python 3.11.9 64bit and 3.12.10. I'm guessing most ROCm alternatives (the non-TheRock options) still prefer the older 3.11.9 version, right? Or will they be fine with the newer 3.12.10?

On a separate note, it looks like TheRock is superior to ZLUDA in terms of optimization, and maybe stability too. Is there any real benefit to installing 3.11.9 instead of just going with the latest 3.12.10? Kinda on the fence about this. 🤔

[timber gate]

Can someone walk me through the installation process step by step. Im reading the github page and just overthinking and confusing myself. I just did a hard wipe on my pc because of some issues while gaming plus i needed to its been 3 years. I have a 5070ti rn.

[ornate elk]

You can go for 3.12.10 64bit too. Most stuff should already support it. And yep TheRock is better in performance because it doesnt need a translation layer

[ornate elk]

I have a bit of time rn

[timber gate]

want me to stream?

[ornate elk]

that or screenshots

[timber gate]

im in the lounge

[ornate elk]

Wan 2.2 Vae: https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/tree/main/split_files/vae

Wan Video GGUF Versions: https://huggingface.co/calcuis/wan-gguf/tree/main

[sick merlin]

every image I generate using forge neo comes out black

does anyone know what the reason could be?

[cosmic furnace]

set COMMANDLINE_ARGS= --xformers --sage do i need to have this on forge neo btw? And how to make sure ive installed zi correct forge neo. I remember recently installing it , but i guess theres only one of it?) huh ?

[ornate elk]

seems like the dev broke something:
https://github.com/Haoming02/sd-webui-forge-classic/issues/281

[sick merlin]

ahhhhh thank you so much

I spent all night trying to mess around with it thinking it was an issue on my end

because it was literally working then all of a sudden it stopped

[cosmic furnace]

At this step it errors - Edit the webui-user.bat (right click, edit)
At the second line below echo off you need to add: set TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1
At the line COMMANDLINE_ARGS=
For Forge / ReForge / ForgeNeo add: --skip-python-version-check --cuda-stream --attention-pytorch
For Automatic1111 add --skip-python-version-check --opt-sdp-attention (opt-sdp-attention is very fast but unstable. Its up to you if you want to add it).

---

venv "B:\SD-FORGE-NEO\sd-webui-forge-neo\venv\Scripts\Python.exe"
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: neo
Traceback (most recent call last):
File "B:\SD-FORGE-NEO\sd-webui-forge-neo\launch.py", line 52, in <module>
main()
File "B:\SD-FORGE-NEO\sd-webui-forge-neo\launch.py", line 41, in main
prepare_environment()
File "B:\SD-FORGE-NEO\sd-webui-forge-neo\modules\launch_utils.py", line 315, in prepare_environment
raise RuntimeError("PyTorch is not able to access CUDA")
RuntimeError: PyTorch is not able to access CUDA

Mb something wrong in webuiuser ? -
@echo off
set TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1

:: set PYTHON=
:: set GIT=
:: set VENV_DIR=

set COMMANDLINE_ARGS= --skip-python-version-check --cuda-stream --attention-pytorch

:: --xformers --sage --uv
:: --pin-shared-memory --cuda-malloc --cuda-stream
:: --skip-python-version-check --skip-torch-cuda-test --skip-version-check --skip-prepare-environment --skip-install

call webui.bat

[cosmic furnace]

@ornate elk Boink xD . Sorry.This windows is fresh with everything. But pretty sure hipsdk & python installed correctly o.0 . hmgm?

[ornate elk]

make sure you installed the correct torch version for gfx1100

[cosmic furnace]

Oh. I thought i shud do this first- Setup the prefered Webui:
And this then -
Setup TheRock:
While in the webuis folder click in the File Explorer bar (not searchbar) and type cmd then press enter.
4. Now run these commands one by one:
venv\Scripts\activate.bat