🤝｜tech-support

[jagged crane]

which model should I use for Furry training? I'm using v1-5-pruned-emaonly

[cosmic furnace]

File "G:\SD\SDForge_ZLUD\NEWST\stable-diffusion-webui-amdgpu-forge\backend\memory_management.py", line 133, in <module>
total_vram = get_total_memory(get_torch_device()) / (1024 * 1024)
^^^^^^^^^^^^^^^^^^
File "G:\SD\SDForge_ZLUD\NEWST\stable-diffusion-webui-amdgpu-forge\backend\memory_management.py", line 100, in get_torch_device
return torch.device(torch.cuda.current_device())
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "G:\SD\SDForge_ZLUD\NEWST\stable-diffusion-webui-amdgpu-forge\venv\Lib\site-packages\torch\cuda_init_.py", line 1069, in current_device
lazy_init()
File "G:\SD\SDForge_ZLUD\NEWST\stable-diffusion-webui-amdgpu-forge\venv\Lib\site-packages\torch\cuda_init.py", line 403, in _lazy_init
raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

Fresh new installment of zluda forge rezults in same error o.0

[chrome rampart]

hello all and best wishes. Hope somebody can help why I get (see below) when i start up comfy? It seems to increase FETCH ComfyRegistry Data: 10/118
FETCH ComfyRegistry Data: 15/118
FETCH ComfyRegistry Data: 20/118
FETCH ComfyRegistry Data: 25/118
FETCH ComfyRegistry Data: 30/118
FETCH ComfyRegistry Data: 35/118
FETCH ComfyRegistry Data: 40/118
FETCH ComfyRegistry Data: 45/118
FETCH ComfyRegistry Data: 50/118
FETCH ComfyRegistry Data: 55/118
FETCH ComfyRegistry Data: 60/118
FETCH ComfyRegistry Data: 65/118
FETCH ComfyRegistry Data: 70/118
FETCH ComfyRegistry Data: 75/118
FETCH ComfyRegistry Data: 80/118
FETCH ComfyRegistry Data: 85/118
FETCH ComfyRegistry Data: 90/118
FETCH ComfyRegistry Data: 95/118
FETCH ComfyRegistry Data: 100/118
FETCH ComfyRegistry Data: 105/118
FETCH ComfyRegistry Data: 110/118
FETCH ComfyRegistry Data: 115/118
FETCH ComfyRegistry Data [DONE]

[ornate elk]

Its from ComfyUI-Manager, it fetches all custom nodes so if you install one they are updated.

[ornate elk]

I guess in the full log you have the same issue ?
Can you show it again?

[chrome rampart]

hi cs. a ok so this is normal ?

[ornate elk]

Yes its like this for everyone

[chrome rampart]

ok super thanks like always

[cosmic furnace]

this is after reinstalling rocm . although it was workin on rocm the rock forge neo. so it seems like problem is with zluda

auto update(download) of forge doest want to download zluda folder btw. errors now

also i was trying to update both zluda forge & forge neo same time that day. cud messed up trying to install therock torch files for zluda forge. but deleting venv shud solved it so =\

gemini advices to instal lower version torch - venv\Scripts\activate.bat
pip install torch==2.1.2 torchvision==0.16.2 --extra-index-url https://download.pytorch.org/whl/cu118
2.1.2 one
2.download zluda manually
3. Solve the "Invalid folder name" issue.
The ROCm agent enumerator failed: [WinError 267] error still persists.
Even though you renamed the main folder, you still have a nested folder called ...\NEWST\stable-diffusion-webui-amdgpu-forge....
Try moving all contents from stable-diffusion-webui-amdgpu-forge directly to G:\SD\SDForge_ZLUD. The shorter the path, the less likely this error is to occur.

😵‍💫

[ornate elk]

remove C:\Zluda from the envirnoment path settings.
then open up a cmd and run this:
pip uninstall rocm rocm-sdk-core rocm-sdk-libraries-gfx110X-dgpu -y

then delete the venv and the .zluda folder and relaunch forge zluda

[ornate elk]

delete or ban xD

[full silo]

why ?

[ornate elk]

wrong channel and we have enough of these spam ai dev bs here daily, you joined today just to spam this, i guess in much more servers too

[full silo]

nope

As AI developer , I wanna be powerful member of this server.

To contribute to this server and all users

[ornate elk]

okay, then first thing is to read the rules xD
Rule 5 prohibits Self Advertising.
Also Tech-support channel is for getting support about ai stuff not to self advertise.

[full silo]

yeah, that's right

But How can i help all users and this server ?

[ornate elk]

if someone has a question about local or online technical ai stuff they will mostly ask here, just check in sometimes and if you know the answer feel free to help here

if you have a project your working on, that benefits the users, then you can create an entry in #1092446741984444416

[cosmic furnace]

cmd anywhere or zluda -forge one ? o.0

i guess where zluda forge is) but ...

[ornate elk]

Anywhere

[cosmic furnace]

ModuleNotFoundError: No module named 'optimum.onnxruntime'

oh wow. same old error

[misty marten]

i have a Rx 6600, this is the log

[cosmic furnace]

everything seems to work fine now ^^ thx.
wth was that btw ? 🤔

Did i by accident installed new torch for neo forge everywhere ?

[ornate elk]

Yep you probably installed rocm outside the venv, that breaks paths and stuff that relies on rocm

[ornate elk]

The Issue is that you installed Comfyui on Desktop and that is snyced with OneDrive.
That breaks Comfyui

Install or move comfyui to C:\Ai\ for example.
Don't use Desktop, Downloads, Documents, Programm Files.

[glad pasture]

Hello and happy new year. I’ve just installed Stable Diffusion. I followed a YouTube video from a guy, but I’m getting a really awful result and I honestly don’t understand why I’m not getting the same outcome. Could you tell me what might be wrong? Sorry for the beginner question.

I should have gotten this result.

[cosmic furnace]

seems like classic vae problem

[glad pasture]

Could you give me a link that explains how to fix this issue? I’ll try to look it up on my side as well. Thanks.

[ornate elk]

Yes its a VAE problem.
Your model is 1.5 based but you added the sdxl vae

They are not compatible

Just unselect the sdxl vae in the settings and apply and reload the ui.

[glad pasture]

Thanks for your feedback. What I don’t understand is that I’m doing the exact same thing as in the video, but something is missing. I’m also set to 512 × 512.

I will try that

[ornate elk]

At first I thought your using an sdxl model because of the sdxl VAE in your output meta data below the imagr

Disabling it will fix the output but for 1.5 based models

[glad pasture]

I unchecked all the boxes, reloaded the UI, and I still don’t understand.

[ornate elk]

The VAE is still aet

Either in the VAE settings Or in a dropdown at the top

Also you have to press Apply before reloading

[glad pasture]

how thanks

i didnt apply the setting, My apologies !

[ornate elk]

Ah ok xD

[glad pasture]

Thanks for explaining all of this. I’m still a bit nervous about going on Discord since I’m not very social, but I always find nice people. Thanks again.

[ornate elk]

No problem, feel free to ask here again or in #📝｜prompting-help if its related to the output or usage.
A lot of helpful people are in here 🙂

[misty marten]

The first error was resolved, now this one is appearing

I downloaded this workflow, but I have no idea which node this error is.

[night brook]

Hey guys, I'm using automatic1111 with a radeon 6700xt (12gb) and ubuntu, since I upgraded to Quokka, stablediffusion will crash the desktop I need to restart my x server) when I try to create an image.

Comfyui seems to work though

[ornate elk]

Hey, have you tried to reinstall it?
Use the dev branch if so.
The Main branch is outdated and broken.
Or better install Forge classic or ReForge or Forge Neo,

[night brook]

Thanks, I'll git pull the latest.

Forge classic or ReForge or Forge Neo,... any suggestions among them? Really I don't feel like downloading 20+gb for pick and choose time.

[cosmic furnace]

ValueError: Failed to recognize model type! and now it wont start because of this error o.0

[ornate elk]

Complete log pls xD

[cosmic furnace]

nvm i tried pressing it few more times different models. all of 5 ended up same error

and i ended up in blue screen of death. but after it everything launched it seems o.0 . for now

[ornate elk]

Forge classic would be enough as with 12GB vram your not going to use larger models that forge neo supports

[teal ruin]

Hi everyone! I’m Donis from Mexico. I’m starting an open‑source project called MANGA Studio: an offline‑first suite to turn scripts and audio series (YouTube, podcasts, etc.) into anime‑style episodes using local AI models (no cloud, no subscriptions).
I’m not a programmer; I have a detailed technical spec of the full pipeline and I’m looking for collaborators (Python / ML / architecture) to build a small MVP.
GitHub repo: https://github.com/adonismaikel-glitch/manga-studio
Any feedback or help is very welcome!

[hot thorn]

is it better to do (illustrious checkpoint)
prompt
BREAK
style loras
BREAK
character
BREAK
place/pose
or
prompt
BREAK
Character
BREAK
Style loras
BREAK
pose/place
or even keep the style loras at the bottom

[cosmic furnace]

some say tiled vae may result in lost of quality. is that true? like some pixels, visible stitches, or smthing

[cosmic furnace]

https://discord.com/channels/1002292111942635562/1011743094309396631 ull most likely gonna be sent to this channel

[chrome rampart]

Hello all. I am getting msg Legacy ComfyUI-Manager data backup exists. Please verify and remove when no longer needed. See terminal for details. in comfy but no clue what to do an somebody advice?

[rigid notch]

Hello everyone, I had some issues with ReActor node in ComfyUI and I followed Gemini's advice to reinstall onnx. The thing is : it kinda broke python 3.12 in the python-embeded file (it weighted 0 ko after that...) and now I obviously cannot start comfyui.

I tried to download comfy again, to copy a clean python-embeded folder, but the actual has python 3.13 and if I try to use is all my previous workflows are broken...

Is there a way to repair this ?

[night brook]

Hi all, I figured I'd put everyhing I faced so far. Maybe someone has a similar situation.

Issue: Complete X Server crash (kicked to login) when generating in Stable Diffusion / Forge. Interestingly, ComfyUI works perfectly fine on the same machine.

System Specs:

• GPU: RX 6700 XT
• CPU: Ryzen 2700
• Ram: 64 GB
• OS: Ubuntu 25.10
• Kernel: Linux 6.17.0-8-generic #8-Ubuntu SMP PREEMPT_DYNAMIC Fri Nov 14 21:44:46 UTC 2025 x86_64 GNU/Linux
• ROCm Tried both Versions 5.4 and 6.2
• Interface: X11 (Xorg)

What I’ve Checked:

• dmesg / journalctl: No entries at the time of crash. It's a "silent" failure.
• Xorg Logs: No (EE) errors logged; the log simply ends or shows a new session after crash to login.
• VRAM: Monitored via nvtop. Crash occurs even when VRAM isn't fully pinned.
• Environment: Already tried HSA_OVERRIDE_GFX_VERSION=10.3.0 and --medvram flags.

The Mystery: Since ComfyUI is stable, I suspect this is related to how Forge/A1111 handles memory/power spikes or VAE decoding compared to Comfy. I suspect the driver and power delivery strategies changed during the upgrade. Things worked before.

Has anyone seen Forge/A1111 cause an AMD "Ring Timeout" that doesn't reach dmesg? Are there specific PYTORCH_HIP_ALLOC_CONF settings or kernel parameters (like amdgpu.gpu_recovery=1) that help stabilize this?

[ornate elk]

You need to use tiled vae. And I guess rocm 6.4 is better for your GPU then 6.2 but idk on linux

[night brook]

I think I was, I will try again and crash again. 😉

first I'll update rocm.

Also, thanks again for all the help!

[night brook]

Launching Web UI with arguments: --medvram
Arg --medvram is removed in Forge.
???

[ornate elk]

Yep its removed

But you dont need it for forge

[night brook]

Ah, ok, so I am trying to launch now... it looks like updating to rocm6.4 broke some dependencies.

[ornate elk]

Maybe you need to delete the venv folder and relaunch

So it rebuilds

[night brook]

I can do that, but it will revert to rocm 5.7

I modified webui.sh's installer let's see how it goes.

[night brook]

Ok, trying now. Fixed everything and re-downlaoded a lot

And it crashes even faster now!

[ornate elk]

Ohh

[night brook]

I ran furmark, my gpu is stable

[night brook]

Ok, I tried python 3.10, 11, 12, 13. rocm 5.7, 6.2.4 and 6.4, I'm gonna try cpu and cry.

I am getting much better at comfyui though

[misty marten]

comfyui-zluda error

RuntimeError: cuDNN Frontend error: [cudnn_frontend] Error: No execution plans support the graph.

[atomic cape]

anyone in here use the desktop version of comfy for amd that was just released a few days ago? got a friend whos installed with desktop exe for windows and is having some ram issues when vae decoding node starts. ive gotten him to test disable pinned memory flag, force fp16, replace vae decoding with tiled vae decoding, etc and still has issues with crashing when hes only using the basic starter zimage template. he has 32gigs vram and 32gigs ram, so shouldnt be having this issue. hes using desktop version for AMD (which ik is new) and is just having grey image output (indecating ram issues) on hardware that should have 0 issues w this

[prisma moss]

check if he has page file enabled

[ornate elk]

Which GPU does he have?

[atomic cape]

alr so after lots of testing with different launch commands, swapping nodes, etc, i ended up telling him to just nuke the desktop exe version of comfy...

we then just instead installed with this script, and it seems to be working MUCH better now. hes able to generate images fine with the default zit workflow, and im now ab to give him my workflow with upscaling etc to test.

he has a AMD Radeon 7900xtx, so using this script seemed to install everything better for his specific hardware:

https://github.com/aqarooni02/Comfyui-AMD-Windows-Install-Script

thank you sm to everyone that repsonded to help. ill be back in here if something else goes horribly wrong 🙂

[proper needle]

"Hi everyone, I'm fairly new to ComfyUI and I'm really struggling with the IDM-VTON installation. I keep getting 'config.json not found' errors, so I assume I'm messing up the folder structure or missing some files.

I have an RTX 3060 (12GB). Could someone please share the correct download links and a quick step-by-step on exactly where to put the model files? I've been stuck on this for a while. Thanks!"

[chrome rampart]

Hello all. I am getting msg Legacy ComfyUI-Manager data backup exists. Please verify and remove when no longer needed. See terminal for details. in comfy but no clue what to do an somebody advice?

[jagged yew]

The 'config.json not found' error happens because you likely only downloaded the .safetensors checkpoint but missed the model config files from the HuggingFace repo. You need to clone the entire repo into ComfyUI/models/diffusers/idm-vton (not just the checkpoints folder).

If you get tired of debugging JSON errors for standard stuff (Flux, LLMs), I built a portable app called V6rge that auto-installs everything for you. It doesn't have IDM-VTON yet, but it handles standard Image/Text/Voice gen without the headache.

Release: https://github.com/Dedsec-b/v6rge-releases-/releases/latest

[jagged yew]

Glad the script sorted it! AMD on Windows is always a battle.

If you ever want a break from managing Python scripts and terminals, I built V6rge as a 'Zero-Code' alternative. It’s a modern Dashboard UI (like a native app) where you just click 'Download' on models and start generating.

It handles the environment setup automatically in the background, so you never see a command line.

Note: It’s optimized for CUDA right now (so your 7900XTX is still better off with Comfy for raw speed), but it’s great for anyone who wants a 'Mac-like' experience for local AI.

https://github.com/Dedsec-b/v6rge-releases-/releases/tag/v0.1.1

[gloomy bolt]

hello, I'm still having some issues with the CivitAI Browser+ extension https://github.com/anxety-solo/sd-civitai-browser-plus.git, i'm trying to use the update model preview function but it's not working.

[unreal talon]

hi im trying to install the reactor extension for my SD webui, i followed the instruction and everything is set up but it still doesn't show up/
instruction here:https://codeberg.org/Gourieff/sd-webui-reactor
i need help

[ornate elk]

Which webui do you use?

Can you relaunch and copy the complete cmd log and post it?

[vocal burrow]

1.6gb in one single executable, zero source, zero information, AI looking readme.md.
I would not recommend anyone to touch this.

[unreal talon]

nvm i got it working, for some reason i didn't have microsoft c++ (i could've sworn i had it) like i had the folder of the 2022 version but it was empty?? my guess is that windows deleted it when i swapped to windows 11?? but it's working now

[ornate elk]

Ah alright!

[distant bear]

how can I bypass this error?

[ornate elk]

What's your GPU?

[distant bear]

ok nvm, I fixed that by deleting venv directory

[distant bear]

Nvidia GeForce RTX 3060 12GB OC

[ornate elk]

I checked the log and noticed two things.
You installed Automatic1111, which is outdated since over a year by now and won't get updates anymore.
And you added --skip-torch-cuda-test which makes it run on CPU mode instead of your GPU.

Best would be to check the pinned messages of this channel for the updated guides and then install Forge Neo as it supports more models and still gets updated

[distant bear]

now I removed --skip-torch-cuda-test

and it works

[ornate elk]

Alright, sure it still works, just not updated anymore.

If you get some slower speeds (over 1min) on xl models, add --medvram-sdxl to the launch args

[unreal talon]

@ornate elk now for mov2mov i installed the extension but it's not working

i get like this type of error

[ornate elk]

Mov2mov is broken. But there was a fix on github for a file that has to be replaced to make it work

But idk if thats still the case.

[unreal talon]

omg i've been trying to look for smtg, i need the fix

[supple dawn]

How can i solve this?

[ornate elk]

What's your GPU?

[lucid yoke]

What app is that and cmd log?

[supple dawn]

5070 12gb

[supple dawn]

Forge Webuii

https://youtu.be/zqgKj9yexMY?si=HSVPECax1DX-ZkXf

i dowloaded from here

[ornate elk]

Then you should setup Forge Neo, as it has better support for 50 cards.
Checkout the pinned messages for a setup guide.
Auto1111 is outdated.

[supple dawn]

This?

Ahh i found Forge neo

Thanks! i will try to dowload tomorrow i go to sleep now

[ornate elk]

Alright, feel free to ask here if you have any questions but the setup should work easily

[sick merlin]

should I get the same results from using Forge Classic than I would using Forge Neo?

I'm running a 12gb vram gpu

just looking to generate images

[ornate elk]

Yes, its using the same stuff in the background

[sick merlin]

I just did a fresh install and I'm only generating black images

nvm adding a negative prompt seems to have fixed the issue

[ornate elk]

Yep thats needed for 1.5 and XL based stuff

[steep gate]

In A1111 I used XYZ Plot to generate 3 prompt S/R for X, and about 30 checkpoints for Y. The grid I get is 276X4000, which is wayyyy too skinny to read the checkpoint names on the left. How do I change the grid size?

[ornate ruin]

I've waited far too long but I'm finally going to do it..
Upgrading from a1111 to forge neo

Before I just follow the next best how to install forge ui neo guide on yt

Do I need to factory reset my PC or anything? Or is deleting the stable diffusion folders enough?

Afair the already set up python for a1111 might cause problems or so, I'm not fully aware

-thanks

[ornate elk]

You will just need to uninstall python 3.10 and install python 3.11.9
Then you can follow the Forge Neo setup guide from the pinned messages of this channel

[ornate ruin]

Many thanks

[errant orbit]

hey guys need help about my current specification, can i create a video that almost same as grok imagine using RTX 3060 12gb?

[gloomy bolt]

@ornate elk hi, I'm still having some issues with the CivitAI Browser+ extension https://github.com/anxety-solo/sd-civitai-browser-plus.git, it's not scanning the lora or checkpoint file when I try any of its functions, and it can only detect the TI (in the embeddings folder)

[ornate elk]

Hey, where are your models placed?

[gloomy bolt]

\stable-diffusion-webui\models\Stable-diffusion
and
\stable-diffusion-webui\models\Lora

\stable-diffusion-webui\embeddings
^
while my TIs that are being scanned are in this folder

[ornate elk]

Hmm thats strange

Can't help much rn, im on vacation

[gloomy bolt]

oh i see, np then enjoy your vaca

[ornate elk]

For me it works but I use --model-ref and you use forge-ref-a1111-home
Maybe thats why

The difference is that I point to a models folder directly

[gloomy bolt]

i see, if it's no problem I can run your method then

[ornate elk]

Nope you just have to edit your path to /models at the end and change the arg

[gloomy bolt]

--forge-ref-a1111-home "C:\Users\asdas\stable-diffusion-webui\models"

(?) will that mean it'll ignore the TI's instead tho unless i put a seperate argument for it

[ornate elk]

You have to move the tis into the models folder then

And you need to use --model-ref

[gloomy bolt]

mk got it, let me test it out in a bit and ilyk how it works. Ty for helping on ur vaca tho

[gloomy bolt]

it works now ty 🔥

[hot thorn]

anybody knows how to train an embedding?

[hot thorn]

@ornate elk help?

[ornate elk]

Sry can't help with that

[solar hamlet]

I cannot get a workflow file to open in comfyui....this is a nightmare...how do yall get thm to open

trying to open them in firefox--not working

Only the code opens up

I've hunted for multiple tutorials on this and they all direct me to do things that don't exist

[calm cairn]

So been a while
I got a new computer! Video card is 5090, feeling like trying it out
But ofc on a fresh install I gotta put git back on, and then SD
so I figured why not try out SD forge neo?
But I keep getting an error message so I thought I'd check here to see if anyone knows what to do--it says it "failed wheel build for install"

[ornate elk]

Which launch args did you used?

[peak olive]

I'm running (I'm pretty sure) version 1.5 of stable diffusion webUI and I'm trying to load a checkpoint, but every time I do it just refuses and the CMD spits out a massive list of errors

the checkpoint says it's for 1.5 and not SDXL so I'm kinda confused.

it also comes with a VAE which I tried to load the checkpoint with, and without. Both unsuccessful.

[calm cairn]

first I did git clone directory of https://github.com/Haoming02/sd-webui-forge-classic/tree/neo

did it 2 different ways

then i ran webui.bat

[ornate elk]

Please follow the updated setup guide for forge neo from the pinned messages

[calm cairn]

ah, in this tech-support channel?

k ty

[ornate elk]

No problem, should work without issues

[ornate elk]

And what's the checkpoint you try to load?

[peak olive]

"ChilloutMix"

[ornate elk]

Ah ok yea that should work

Can you relaunch the webui-user.bat and share the cmd log?

[peak olive]

sure! Is there a text file of the log?

[ornate elk]

Nope its copy and paste from the cmd

[ornate elk]

When did you installed the webui and what's your GPU?

[calm cairn]

so I followed the guide in the message link and now it says "python was not found" which is awfully weird

[peak olive]

my GPU is a Geeforce RTX 3060

[ornate elk]

Rerun the python installer. Click Modify, clikc next, select "add python to system environment variable's"
Then relaunch the webui-user.bat

[peak olive]

that "Created January 9, 2026" is when I installed the webUI

[calm cairn]

by "modify" do you mean "reinstall python" or "launch python"?

[ornate elk]

Launch the python 3.11.9 installer exe if you have it installed rn, if not reinstall

[calm cairn]

ok i was using install manager, got regular installer now

[ornate elk]

Ah okay, the guide you followed was very outdated. As Auto1111 is not updated anymore and your python version should be higher.
Best would be to reinstall it using the updated setup guides from the pinned messages of this channel.
For your GPU I would also recommend using Forge Neo for the best performance instead of Auto1111.

[calm cairn]

got a bit further, but now back to "failed to build installable wheels for some pyproject.toml based projects", which is similar to earlier error message

[peak olive]

oh my god yesss W community thank you bro (love the PFP by the way)

[ornate elk]

No problem and thanks 😁

[ornate elk]

Can you show me the full cmd log?

[calm cairn]

copy and paste here?

is long, but sure

[ornate elk]

Yea but the first warning indicates the issue here xD

Wrong python version

[calm cairn]

oh, what do I do? I just updated today

[ornate elk]

If you dont need python 3.14 for anything else then uninstall it. Its not compatible with any webui stuff

You need 3.11.9 64bit

After Installing that, delete the venv folder and relaunch the webui-user.bat

[calm cairn]

since i had both installed, i uninstalled just 3.14 and will try that

[ornate elk]

Yep then just delete the venv and relaunch. It can work

[calm cairn]

building venv again now, but looking good

and we are in! thank you for your help

[ornate elk]

Perfect, no problem 🙂

[calm cairn]

ah i hope you are still here

i added this in command line args

--ckpt-dir E:\Life\Projects\ART\Stable_Diffusion\Models --lora-dir E:\Life\Projects\ART\Stable_Diffusion\LoRA --theme dark

and it did in fact implement dark mode but did not find models and loras, what did I do wrong here?

[ornate elk]

Can you show the cmd log again? There we see what happens

[calm cairn]

sure, im testing with "" marks right now so i should have a log shortly

changing
--ckpt-dir E:\Life\Projects\ART\Stable_Diffusion\Models
to
--ckpt-dir "E:\Life\Projects\ART\Stable_Diffusion\Models"
solved the problem, thanks for your attention anyway

[ornate elk]

Alright np

[calm cairn]

apparently this card screeches while being used...hope that doesnt continue

oh its from my surge protector

hmm, do I need to reduce load somehow?

(well its an uninterruptible power supply)

oh and all the pictures are pure black too...i thought i was out of the woods

i have these args
--cuda-stream --sage --fast-fp16
so I'm going to toggle different ones off to see if this reduces power demand

I'll have to ask my computer hardware guy tomorrow in the day, until then
I think I had black images like 2 years ago, do you know what causes that?

[calm cairn]

UPDATE: SD 1.5 models CAN generate pictures, Pony and Illustrious models only produce black images

[ruby yoke]

are you using the clip skip node?

[ornate elk]

All of them won't be black if you use negative prompts

Its like that for 1.5 and sdxl based stuff

[ornate elk]

You can keep all of them

[calm cairn]

fell asleep, am back

first test: move clip skip from 2 (default) to 1

still black

2nd test: add negative prompt

success

weird, I could gen on old computer setup without using negative prompt

but I'm glad you knew the answer and it didn't require massive retooling

[unreal linden]

Hey guys; I've got a fresh install of stable diffusion forge but for the life of me I can't get xformers working at all. Not sure what I'm doing wrong but it's not as straight forward as it was installing it with normal automatic1111 stable diffusion

not sure what I'm doing wrong, been at this for hours lol

[ornate elk]

Hey, what's your GPU?

[unreal linden]

4070 super

[ornate elk]

Then you should not use Forge but Forge Neo instead

More updated, better performance, more model support

A setup guide can be found in the pinned messages of this channel

[unreal linden]

I... wut. HOW CAN IT BE EVEN FASTER

[ornate elk]

It uses newer torch and supports flash and sage attention aswell as fast fp16

[unreal linden]

ooh la la

I'm getting something around double speed from SD to SD Forge I can't even imagine it getting any faster than this

I look forward to it, even if I have to install this stuff all over again 😅

[hot thorn]

@ornate elk or anyone with a lot of SD experience. Is spamming "detailed this and detailed that" in my prompt (im on illu checkpoint if that matters), cause drop in quality? Should I just keep my prompt as basic as possible to avoid dip in quality?

[honest harness]

Hey so i had reinstalled my pc without deleting stable diffusion and now it says this after i installed python back

[ornate elk]

Delete the venv folder and relaunch the webui-user.bat

[honest harness]

it was a wrong python version

i downloaded 3.14 and i needed 3.10

let it download all it needs to then it gave me this error

@ornate elk

[ornate elk]

Yea auto is broken anyway. You would need a reinstall using the updated guides from the pinned messages.
But if you have a GPU with 12Gb or more VRAM go for Forge Neo instead

[honest harness]

got through that to here and im on a 4090

i had reinstalled my pc and i had an issue where the user ssid was different i believe so i cloned the stable diffusion with the git and i think thats why i got this

[ornate elk]

You have to follow the updated guides from the pinned messages.
Auto1111 can't be installed anymore with the normal git clone. The command changed a bit. But with a 4090 you should definitly not use it anymore.
Use Forge Neo as you there can also use Qwen, Z-image, Wan and wan video too and it has better performance than auto1111

[stoic latch]

First time trying to instal 1111 and stable, what did I do wrong?

[ornate elk]

Nothing, the normal install is broken. Follow the updated setup guides from the pinned messages of this channel.
Also if you have 12GB or more VRAM, go with Forge Neo instead

[unreal linden]

Do I need a particular version of python to run forge neo or can I update to the latest?

[stoic latch]

i swear i already downloaded this one from this website?

[ornate elk]

You will need 3.11.9 64bit

3.12.10 can work too but nothing newer or older

[ornate elk]

You dont need that then

If you have GIT already installed proceed with the next steps

[stoic latch]

But it gives me that error 128 thingo

When the Git pops up and I log in with browser it doesn’t load the site or anything

[ornate elk]

Thats because you have used the wrong git cloned command for the install

Its an issue of Auto1111 not GIT

[stoic latch]

how do I use the right command for the install?

[unreal linden]

is there a tutorial somewhere for generating videos with forge neo using wan?

Also thanks everybody for all the help I've recieved so far, it's been a major game changer

[ornate elk]

Its in the updated guides

[arctic goblet]

hello, ive downloaded some checkpoint mergers from civitai, do i still need to download the checkpoint?

[ornate elk]

Hey, nope checkpoint merge works the same as a checkpoint

[distant bear]

I'm using VPN

[vocal burrow]

most likely you don t have python installed, make sure that it s a 3.12 or 3.13 version and to tick the "add to PATH" checkbox during install

[distant bear]

I installed python 3.13, but it doesn't help

[vocal burrow]

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

[distant bear]

Python 3.10.11

[vocal burrow]

please send the whole thing. or screenshot it

[distant bear]

how to switch python versions?

[vocal burrow]

edit system's PATH and make sure that python 3.13 's path is at the top

cf #🤝｜tech-support message

[distant bear]

ok thanks

[distant bear]

it's freezing...

-# not really

-# just download is so slow af

I got a new problem

[ornate elk]

For comfyui better use the portable Version from the github

[gusty reef]

So, when I am generating an image, BOTH the gpu and cpu are spiking to close to 100%. Is this normal? I could have sworn a month ago the cpu remain decently low while the gpu would spike to a high amount.

[ornate elk]

Not normal, what's your CPU, GPU, ram amount and webui?

And make sure enough SSD space

[gusty reef]

CPU: Intel(R) Core(TM) Ultra 9 275HX
GPU: NVIDIA GeForce TRX 5070 Ti
RAM: 32G
VRAM: 12G
WebUI: ComfyUI Desktop

When viewing the task manager, the python script appears to be taking up most of the CPU, also it might be related to closing comfy, as after I manually shut down python script in task manager and restarted comfy its now spiking to around 40-60% cpu usage

[distant bear]

guys, do y'all know how to show AI generated video info with ComfyUI?

-# ig wrong channel

[ornate elk]

Do you use any workflow when is spikes?

[gusty reef]

Just the main workflow I tend to use when generating an image.

[ornate elk]

Is comfy installed on an SSD?

[gusty reef]

Yep comfy is installed on my SSD

[ornate elk]

Okay thats very strange. Do you see any GPU vram overflow into shared vram?

Also maybe update your Comfyui to the latest version

[tacit river]

yoo boss! cs1o

i wanted to ask if automatic1111 can be installed on rtx 50 series gpus now

Last time I tried; I couldnt and I just switched to forge neo

[ornate elk]

Hey, yes it can but your better of by using Forge Neo.
But if you want to to setup Auto1111, check the updated setup guides on the pinned messages of this channel

[gusty reef]

Ya, I've updated to the latest version. However, are you sure that spikes in cpu are not normal/common during image generation?

[stoic latch]

what does this mean?

[tacit river]

Ohh gots it!

i was looking thru the options and some were blank; like textual inversion

I learned that it’s only available on other models (?) like automatic1111

[vocal burrow]

old copy pasta but it should stll apply

How to survive NansException, beginners guide :
1/ First thing, double check that the model you are using is indeed a proper stable-diffusion model and not a Lora (spoiler if it's under 2gb it's probably a Lora)
2/ Did you try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion in your SD Webpage as the error message recommend ? Did you click the big "Apply and Restart" button at the top of the page ? If it didn't fix the error then you can turn it back off.
3/ Did you enable some sort of speed optimization using command line arguments ? No ? Then maybe you should. Edit your webui-user.bat, find the COMMANDLINE_ARGS= line, what's in there ? For Nvidia 3000/4000 series gpu you probably want to give --opt-sdp-attention or --xformersa try. For other Nvidia cards --xformers will be the only option. While we're here to prevent any "ouf of memory" error, if you have a GPU with 8gb you'll probably want to use --medvram-sdxl, or --medvram for GPU with under 8gb of vram, or --lowvram for GPU with less than 4gb of vram. Try with those it might be enough to fix your issue.
4/ Despites all that some cards will still refuse to work correctly (notoriously the 1600 series cards) for those you will have to add --no-half to the command line args in your webui-user.bat

ps : Do NOT use --disable-nan-check command line argument as it will only HIDE the error instead of fixing anything and generate complete black or white output. The only proper usage I can find for it is if you're generating many many long batches of images and you don't want your whole pipeline to break and stop dead in its track at the first error.

[vocal burrow]

automatic1111 is not a model.
a model is a file holding a neural graph, stable diffusion 1.5, SDXL, Z-image, etc
automatic1111's stable-diffusion-webui aka automatic1111 is a program using such models to generate images.

[tacit river]

Ohhh

I see

Thank you for correcting me dawg!!!

[stoic latch]

whats the upcast setting??

[stoic latch]

wow im blind

[bold orbit]

so im trying to use 2 key words to run 2 lora at the same time as i was told you guy but its mixing them into the same guy. how do i fix that?

[stoic latch]

is there a way to make it not genereate so slowly

[bold orbit]

better gpu

[stoic latch]

is 6gb slow 🙁

I also wanna know if theres a way to generate higher quality photos, mine are looking like the ai from a few years ago

[ornate elk]

Which model do you use?

And which launch args did you added to the webui-user.bat

[stoic latch]

The launch args made the thing not open properly I got rid of them and how do I check the model? I’ll redo the args properly though

[stoic latch]

is this right? when i did it before it wasnt opening

[vocal burrow]

missing a space there, it is --xformers --medvram not --xformers--medvram

[stoic latch]

klike this?

[vocal burrow]

no still no space in between --xformers and --medvram

[stoic latch]

oh shit wait thats not one command im tripping

still keep getting this crap, alraedy did the 32 thing

[ornate elk]

What's your model?

How large is the file?

[low stirrup]

hi. i just installed stable difussion locally, and with my first try its asking me to pay for credits? O_o ("comfyui")?

[ornate elk]

Then you probably used a workflow which needs a cloud API to run the stuff

Try load a different workflow

[rose apex]

Can you please help me? I was previously using comfyui with the manual install method and 25.12. drivers. I was getting 2.8 it/s average with rocm7.1 (9060 XT).

But I just switched to Comfyui desktop app and also did a clean install of the 25.20.01.17 pytorch drivers, I am getting much lower speed. It now takes 3-4s/ its.

All with the same workflow.

[ornate elk]

In would recommend using the 25.12.1 drivers and the latest ROCm/TheRock pytorch

[rose apex]

I wanted to use wan and someone said wan gives grey noise output with 25.12 drivers but works fine in 25.20 so that's why I tried that.

Also, should I revert back to manually installing comfyui? And delete the desktop app

Edit: it seems by default comfyui desktop was using split attention instead of the pytorch cross attention. After enabling it in settings it's back to the same speed now.

[hot thorn]

is it true that spamming detailed this and ultra-detailed that very often messes up with the generations as a whole?
also, using Face_focus worsens small things like fingers and toes?

[rose apex]

Use lazy embeddings instead of using any quality prompts. Works well for me

[hot thorn]

should i get rid of any and all things like "detailed hands" etc?

[rose apex]

Yes. Use the lazy hand embedding in negative prompt

[ornate elk]

Ah nice. Wan on 25.12.1 should work as for me it worked. But I'm on a 7900XTX

Also the manual install with venv is the best as you then can install sage attention

Which helps a lot for WAN

[rose apex]

Wait sage attention works for AMD? I thought it was nvidia only

[ornate elk]

Nope not nvidia only anymore xD

It works for rdna3 and 4

[cosmic wedge]

hi all...i am experiencing a lot of trouble getting literally any iteration of SD running, ive tried automatic1111, comfyui, stability diffusion and i keep running into similar errors or pbolems

[prisma moss]

also studio drivers

[prisma moss]

such as

[prisma moss]

a1111 or forgeui

[stoic latch]

thats true actually

[stoic latch]

a1111

[prisma moss]

Install forge instead

[rose apex]

That's nice. How do I install it? I mean do I just use the node and it'll work normally?

[rose apex]

The only nice thing about A111 is it's Hires fix tbh. It's better than Forge somehow idk.

[prisma moss]

it works the same, i use it

[rose apex]

I think you're right. I think it was SD.next which had worse hires fix than a1111. I forgot, been years since I used any of them

Once you shift to comfyui, I feel it's impossible to go back

[prisma moss]

i tried all of the forges (neo,classic,and reforge)

they are all broken, or have shit not working with them

some cant use adetailer, some cant hrfix properly, some are slower, than vanilla forge (which is 'abandoned'),

As for comfy ive got no reason to switch

[ornate elk]

I will make a guide for manually Comfy install + sage attention when I'm back from vacation

[ornate elk]

Hmm can you try this model to test:
https://civitai.com/models/4384/dreamshaper
Then share the complete cmd log here

[bold orbit]

whats a Upscaler?

[waxen ivy]

Please be aware that Civitai is not available in some countries. It's blocked for the UK.

[ornate elk]

Good point, for that, use a VPN or this site without vpn:
https://civarchive.com/

[rose apex]

Yeah please do. Your guides are very helpful and somehow the only working ones in the internet

[rose apex]

It helps to increase your image's resolution

[bold orbit]

how do i use it and where?

[rose apex]

What do you use to generate

[bold orbit]

unstable

[rose apex]

What even is that

[bold orbit]

sorry brain fart, stable diffision

[rose apex]

No I mean do you use webuis like forge, automatic1111 etc. or do you use comfyui

[prisma moss]

Unstable diffusion 😂

[rose apex]

Just to be sure, you mean sage attention in windows right?

[ebon orbit]

Hello i just installed automatic 1111 and im getting error code 128

and its saying repository not found for some reason

[ornate elk]

Hey, normal install is broken, what's your GPU?

[ebon orbit]

its 5070

[ebon orbit]

Ah i see how do i install it properly?

[ornate elk]

Okay, then checkout the pinned messages of this channel. For the updates Guides.
You want to install Forge Neo as its made for newer GPUs like the rtx50 series

Auto1111 is outdated since over a year now

Forge Neo has the same UI and is updated

[ebon orbit]

Gotcha

Forge neo is working for me now but when i try to generate a image its just black

[ornate elk]

Make sure to use a negative prompt too

Then it won't be black

Its needed for 1.5, and sdxl based models

[ebon orbit]

It's my first time using something like this so im not sure what im doing yet haha

[ornate elk]

No problem ^^ just type cute cat in the positive prompt field and blurry into the negative prompt field

If you use and model files thats around 6GB you need to select XL At the top left as preset

[cosmic furnace]

Any good updates on wan or the rock ?

[wet elk]

New with stable diffusion, have used it for 3 days, its a absolutely incredible tool, really grateful to whoever made it - I have a problem though, when I try to install animatedifff it dosent work no matter what i do or what gemini tells me I should do to fix it - is there some bug or can someone help a new guy

I get this message: *** Error loading script: animatediff.py
and
ModuleNotFoundError: No module named 'ldm.modules.attention'

[cosmic furnace]

https://civitai.com/models/1981116?modelVersionId=2512098 get insta blue screen when trying to use this one on the rock + forge neo -_- or this error - Failed to recognize model type

Error code Memory management o.0

[crude musk]

hi,i would be thankful if any1 would help me with installing stable diffusion.Im trying for 4 hours but i get multiple weird errors (im new to generating ai photos/vids) but i feel like its kinda my passion lmao

[carmine ibex]

https://github.com/nunchaku-ai/comfyui-nunchaku/pull/728 how do I install a pull thingy? My nunchaku doesn't work after an update but this is supposed to fix it

[ornate elk]

Hey what's your GPU?
Check the pinned messages of this channel for the updated setup guides

[rose apex]

Man wan2.2 is really slow on AMD. For a Q6 model with 6 steps, it's taking me almost 25 mins for a 3s 16fps 720x1280 video. 12-14 mins for 5s 480x720. Am I missing some performance optimisation settings?

[ornate elk]

Yes you have to use the smaller Q5 if your on a XTX, Q6 is to large

And you need to use the lightning loras

And sage attention

[ornate elk]

What's the model, can't open it rn

[rose apex]

Does q5 generate faster? I didn't use lightning lora because the checkpoint I downloaded from civit {WAN 2.2 Enhanced NSFW | camera prompt adherence (Lightning Edition) I2V} said the lightning loras are included in the model and shouldn't be used. Maybe I should try once with them anyways. I have 9060 XT 16 GB. I thought Q6 would be fine because it barely fits in.

Also, I don't know how to install Sage attention, waiting for your guide.

[rose apex]

It's the Tastysinv8 wan2.2 i2v a-14B model

[ornate elk]

A little bit. Someone said 3-4 minutes Q4 and Q6

[rose apex]

I see. I will give it a try then

[ornate elk]

For 16GB you have to go much smaller when using Wan 14b

Q5 is the largest that fits in 24GB

Because it loads high and low together

[analog finch]

@ornate elk Long time, do you have a tip for how to uninstall my ZLUDA/ROCK setups for SD? I switched from my amd 9070 xt to a rtx 5080, also 16gb vram

Can I just delete folders or should I run something in cmd

[ornate elk]

Uninstall hip SDK if installed, uninstall AMD Adrenalin.
Delete C:\Zluda folder if you have that.
Then open up a cmd and run
Pip cache purge

[analog finch]

hip sdk seems very familiar but I couldnt find anything from searches. Is there a cmd to find it?

but thanks!

Im sure I fixed Hip SDK 6.4 or something with numbers but I dont find it or ROCm

[ornate elk]

Under programms

If its not there its gone

[wet elk]

New with SD, can anyone please help me fix how I can make video, i've tried to install both animatteddiff and SVD and both dosent work in stabilitymatrix program 🙁

Thank you - but what is that - a place where you can get help? oO

Im not complaining at all love the program tried it for 2 days its just cant get it to install

Im not sure what you mean, I can pay for help sure

Thank you for trying to help me - im not super IT savy, i have downloaded StabiltyMatrix and been adviced to run Stable Diffusion WebUI Forge - Classic, when I install extensions such as Animatediff I get a errormessage, i believe it says cant find lmd file, i can check, I dont know what you mean by wallet

[ornate elk]

What's your GPU?

AnimateDiff is outdated since over a year now

It doesnt work in forge webuis

[wet elk]

Thank you so much for clearing that up, what do you recommend using instead if you want to try video or gif making

GPU: NVIDIA GeForce RTX3070 but im buying a new one soon

[ornate elk]

ComfyUI is the recommended webui for video stuff. But if you have 8GB of vram its not worth to even try

For image generation use Forge classic or neo

[wet elk]

Ok, thanks alot for clearing that up! I am using both

What GPU would you recommend for it, just out of curiousity

And again really appreciate the help

[ornate elk]

No problem, the recommendation for video stuff is: get as much vram as possible.
The other limit is your price you want to pay.

16GB vram should be the minimum

[wet elk]

Thank you so much @ornate elk!

[ornate elk]

Your welcome, also if that Santiago guy messages you privately, ignore and block. It seems very fishy if they ask for wallets...

[wet elk]

YEa... he actually sent a link, thats what got me a little suspecious

Like "complain here"

As if I had a complaint.

[ornate elk]

Ah okay, yea thats the scam, thanks, I'll ban him rn

You can report him to discord too

[crude musk]

gtx1650

but i just wanna try n have fun lmao even tho its shitty

[ornate elk]

Then you can setup Forge from the guides

[finite lotus]

Hello, I know this doesn't belong here, but my League of Legends won't start i get a vanguard problem i think.I tried everything asked chatgpt,youtube tut and nothing works.Someone help me pls?

[ornate elk]

Any antivirus software installed?

Did League worked before?

[ornate elk]

What's the size?

[cosmic furnace]

CUDA is NVIDIA's proprietary intellectual property. They don't let anyone else in. Recently, enthusiasts created ZLUDA, a "translator" that allows CUDA code to run on AMD. So what did NVIDIA do?She updated her license agreement, prohibiting the use of her libraries through such compatibility layers. It's a battle for the market.

so this is why probably zluda stoped developing... fracking nvidia... how low can they be

[cosmic furnace]

🩻 Known issues

[cosmic furnace]

On a vide i can see a dude with an 6gb vram launches & uses this wan 2.2 . Sooo. i cant believe my 20gb vram & 32 ram cant handle it 😄

[analog finch]

I had similar issue and I had to close vanguard and restart pc and it usually worked after. Just a desync error from riot's side

@ornate elk Should I just run Automatic1111 with 16gb vram 5080?

[finite lotus]

How can i close it? I can’t close it bc “run in the backqround”

[finite lotus]

I don’t have antivirus and yes worked before

[analog finch]

next to you language on the windows bar you have the little up arrow

just right click vanguard there and click exit vanguard

if it persist I would recommend a full reinstall, also had to do that once

you load into a game and it blacks out and you get a vanguard error code yes?

[finite lotus]

Nope i get this

but I managed to delete vanguard by putting Windows into safe mode. And re install and still don’t work

[analog finch]

right, that one.

Why did you need safe mode to delete vanguard?

If you reinstall vanguard and you have issue on launcher, reinstall Riot games too

If repair client existed id say that, but I havent found it

[finite lotus]

it wouldn't let me delete because running in the background

[finite lotus]

I reinstalled everything

[analog finch]

Yeah you can close it from the arrow, or click CTRL + LShift + Escape

and find it

[finite lotus]

Okay i try it

[analog finch]

If you close riot games, then click arrow and "Exit Vanguard

when you open riot games it will say vanguard is required, restart pc

I found the settings for repair client 😛

[finite lotus]

Where do I close Riot Games in the task manager? In the processes or in the startup?

[analog finch]

Just use the arrow to close Riot games, its simpler

[finite lotus]

Ohh got it

[analog finch]

Task manager is mostly for "leftover" applications on crashes/freezes

[hot thorn]

@ornate elk u there man?

[analog finch]

He's usually on in a few hrs

[finite lotus]

Okay i exited both like u said

I have to restart pc

[analog finch]

Yes. vanguard should relaunch properly

[finite lotus]

Don’t working the vanguard doesn’t relaunching

[analog finch]

You started pc and its not running?

[finite lotus]

Yes i restarted and open lol and i got this again

[analog finch]

In the arrow, is vanguard running?

[finite lotus]

Yes

[analog finch]

Right click Riot Client > Settings > League of Legends > Repair Client

if not you may need some other steps

[finite lotus]

I repaired

[analog finch]

Im trying to give you the easiest steps to troubleshoot first.
In Task Manager (CTRL - LSHIFT - ESC) you can close "vgc.exe" "vgtray.exe" "RiotClientServices.exe" and restart pc. (check this later)

cmd > services.msc > find vgc > right click > restart

Find riot client.exe and run it as administrator

Do you have TPM 2.0 and Secure Boot in BIOS?

[finite lotus]

I just don't know what the fuck this is in my old pc never have something like this i buyed a 5070 and i have problems like this😩

[finite lotus]

Thanks for help i try it

[analog finch]

Its not related to the gpu sadly 😛 I also had this issue

For me it was something weird like close vanguard, restart pc, then restart pc again without closing and it worked

I think, but Im not gonna send you on 5 restarts yet 😂

I sat for 2 hours getting it to work and servers were up on riots side

but its worked no problems since i fixed

[analog finch]

You on Win10 or Win11

[finite lotus]

Win 10

[analog finch]

So you dont have TPM 2.0 and Secure Boot to upgrade win11 right

[finite lotus]

Idk what’s that? I’m not that smart😂

[analog finch]

Ok nevermind the error code you got is related to Services, Cache or Network, not TPM

[finite lotus]

I close vanguar.exe riot.exe in the task manager

[analog finch]

"vgc.exe" "vgtray.exe" "RiotClientServices.exe"

If you gonna restart dont forget these

[finite lotus]

Yes i closed vgc.exe and TiotClientServices.exe

I don’t see vgtray.exe

[analog finch]

It probably closes together with VGC then

[finite lotus]

Okay and now i should restart?

[analog finch]

You can also go into CMD and type

sc stop vgc
sc start vgc

[finite lotus]

Before restart?

[analog finch]

Launch CMD as admin

yes

[finite lotus]

Okay

Like this?

[analog finch]

yeah

it will restart vanguard and clean up glitches/conflicts/locked files and then do a clean restart

[finite lotus]

it says the service is not started

Now i should write the sc start vgc

[analog finch]

sc start vgc and restart

it seems there is no conflict with the application tho

since closing it manually works

[analog finch]

Im unsure if you should restart now or just uninstall vanguard again before restarting

while keeping riot client

btw windows defender can also be blocking it. If you can disable windows defender real-time temporary you can run Riot Client as Admin, run CMD as admin and do "ipconfig /flushdns"

you could try the cmd flushdns without disabling defender and see if it helps

im not IT guy,but its one of the things I tried

For me its mostly trial and error

[finite lotus]

I can’t uninstall vanguard only in safe mode

[analog finch]

That sounds like a problem on its own. Why does it say that

[analog finch]

What about from the arrow if you right click vanguard there?

[finite lotus]

rage bait me this sh*t fr man😩

[analog finch]

Vanguard is new virus lmao

[finite lotus]

If i uninstall there nothing happens

[analog finch]

You can run CMD as admin and type "rd /s /q "C:\Program Files\Riot Vanguard

if its in your C:

And then also write "sc delete vgc"

[finite lotus]

Okay but what if i can delete it? I deleted it in safe mode before

and then when I reinstalled it it was bad again

[analog finch]

You repaired Riot Client and checked if it ran and closed correctly so it might work. Im just a samaritan sadly, not an IT pro. But if it wouldnt work after that its Antivirus / Network issue

But I dont understand why you need safe mode when I can just click uninstall from the Arrow

Does running CMD as Admin and typing "ipconfig /flushdns" work

[finite lotus]

The chatgpt say i can’t delete it bc “the vanguard stuck” run in the background and in safe mode nothing runs in the background

And that’s why i can delete it in safe mode

[analog finch]

but it runs in the background for me too tho 🤔

[finite lotus]

Weird something is so fucked up about in my vanguard

[analog finch]

sc delete vgc

does that work?

rd /s /q "C:\Program Files\Riot Vanguard
if not you can unlock the files with this

[finite lotus]

Worked lol

[analog finch]

you flushed dns, repaired riot client and uninstalled vanguard. I would give a restart a shot

you may need to do 2 to install vanguard again

[finite lotus]

But i can see it in the folder

[analog finch]

Can you delete the folder?

rd /s /q "C:\Program Files\Riot Vanguard
run in cmd admin

[finite lotus]

Nope it says i can’t delete bc running

[analog finch]

ok sc stop vgc

and the other command, then delete

[finite lotus]

I stoped vgc and i still can’t delete

[analog finch]

rd /s /q "C:\Program Files\Riot Vanguard

in CMD

[finite lotus]

Access denied

“I cannot access the file because it is being used by another process.”

[analog finch]

Did you run CMD as admin?

[finite lotus]

Yes

[analog finch]

You may have to go Safe Mode again and use CMD there

and do
"sc delete vgc"
rd /s /q "C:\Program Files\Riot Vanguard"

and if still denied in safe mode try

takeown /F "C:\Program Files\Riot Vanguard" /R /D Y

icacls "C:\Program Files\Riot Vanguard" /grant Administrators:F /T

rd /s /q "C:\Program Files\Riot Vanguard"

[finite lotus]

And this will delete vgc right? And after? Should be working if i reinstall?

[analog finch]

It will remove all existing files on Vanguard

Keep nothing from previous

[finite lotus]

Before in safe mode i delete vgc from the from control panel that was not good?

[analog finch]

often files remain, a clean reinstall is often better

it would be a lot easier if you didnt need safe mode

[finite lotus]

Okay i restarted in safe mode now

[analog finch]

CMD Admin > sc delete vgc > rd /s /q "C:\Program Files\Riot Vanguard"

[analog finch]

Whats it say

[finite lotus]

the command is not correct

[analog finch]

oh no do it in sequence

first sc delete vgc

then the 2nd one to remove files and subfiles

[finite lotus]

Okay

Wtf It says the specified file is not installed.

[analog finch]

1. takeown /F "C:\Program Files\Riot Vanguard" /R /D Y
2. icacls "C:\Program Files\Riot Vanguard" /grant Administrators:F /T
3. rd /s /q "C:\Program Files\Riot Vanguard"

Try this sequence

[finite lotus]

Nothings work

[analog finch]

takeown /F "C:\Program Files\Riot Vanguard" /R /D Y

whole line

[ornate elk]

With a rtx50 best is to use Forge Neo

[ornate elk]

Yes

[analog finch]

Yeah I followed the guides and found it, thank you

Everything downloaded but the webui wont launch

[ornate elk]

Oh any error?

[analog finch]

Not that I can see no, but out of the 7 memory_management.py, 6 has blue info and 1 has a yellow warning

Should I do a clean python download? was there a difference for the one I downloaded when AMD

[ornate elk]

You need python 3.11.9

Uninstall every other version

Then relaunch the webui-user.bat

And show the CMD output if it doesnt work

[analog finch]

I have 3.11.9

This is the entire cmd

[ornate elk]

Can you put a model into models/stable-diffusion and relaunch?

And then copy the whole cmd log

There is more when you scroll upwards

[analog finch]

Have to redownload a model, deleted all 😛 2 sec

Do you have some CMD expertise for @finite lotus?
He can only uninstall Vanguard in Safe Mode. He ran sc delete vgc and it said it wasnt installed in Safe Mode and in Normal Mode its always running in the back so he cant uninstall. Hes done multiple reinstall and restarts, He flushed DNS.

I tried helping but I have no idea tbh

@finite lotus Did u fix yet?

[finite lotus]

I’m out now I need a break to calm down lmao but thanks for the lot of help. I think i deleted everything in safe mod not with the cmd. I deleted the folder and in the comtol panel

And i turn of safe mod start lol and i got this now

I have to dowload vanguard again so i hope i delete everything

Later i try to dowload vanguard again

[analog finch]

ok me too, check for vanguard files now that its not installed

Never saw on old one

[ornate elk]

That warning is because your running a webui localy and that means http instead of https

Nothing to worry about

[analog finch]

Alright, ill try a gen and see if it works now. would you recommend this hint from the launch?

Hint: your device supports --cuda-malloc for potential speed improvements

[ornate elk]

You can add it but it doesnt speed stuff up

Reforge dev said its miliseconds difference

[analog finch]

Im generating black images hmm

[ornate elk]

Use a negative prompt

That fixes that

[analog finch]

fair enough, just picked a quick model and he didnt use a negative

[hot thorn]

any chance you give me some guidance with prompt @ornate elk ?
i pretty much self taught myself my prompt structure, and i know there's a lot of redundancy etc
no matter what I add, remove or change im always not happy about the results, my neg prompt is fine.
even now, when i stripped most things from my prompt and all the ultra-detailed hands etc (chat gpt told me it's just more detail, not better quality) it just looks worse.

for majority of the time i used this concept
general prompt (applies to all characters), aka quality tags, detailed this, ultra-detailed that, texture, reflection, lighting, midriff, hands, feet etc
BREAK
Character lora and description
BREAK
setting/pose etc
BREAK
Single Style lora and few extra tags

at this point I have no idea if it's good or not, some thing might be good about it, some terrible.
I want this pos prompt setup to be usable with all characters, since in place of character lora and setting/pose i run dynamic prompts, so it just takes 1 entry from each and generates it, which i leave for long sessions
I want to know if I should remove all "detailed" and "ultra-detailed" stuff or not, are chunks very important or not, do I need to count tokens for good results? if I do, then how do i count them
and anything else you can tell me.
I'm so desperate now, with no one to help me on high level that I've been even thinking off giving up AI for a few weeks and come back to it when i'm no longer stressed by it, or just stop entirely if I can't get this to work the way i want it to

if its gonna take too long here u can DM me

[cosmic furnace]

wrong channel ...

https://discord.com/channels/1002292111942635562/1011743094309396631

[ornate elk]

Yep that goes into #📝｜prompting-help not here.

And sry but I dont have the time for that. Still on vacation and not in a mood to fix or enhance prompts. Also your thinking to much about that prompting stuff.
Dont get stressed, just have fun. Maybe take a break. Its a hobby.

[hot thorn]

prompting help is dead

[ornate elk]

Your prompt idea looks good btw. If it works for you use it.
I never use Break as I dont care about that.
For multiple characters I use Forge couple that has a specific way to prompt.
And if you use Illustrious models keep the quality tags to known booru tags

[rancid nest]

Is chroma-HD bad? I'm definitly a casual image generation user, but I tried chroma-hd-fp8 and have very good results. Now I want to fine tune the images and learn inpainting and image to image, but I'm reading that chroma-hd is bad for that. What model suggestions for improving images I've made in chroma-hd without losing the aesthetic I've acheived?

[hot thorn]

I started using lazy embeddings recently, it packs quite few quality tags and saves space.
is there a chance you will help me out in few days or a week or something? I could take a break from AI for a bit this way. Maybe go through all thousands of pics i made in the past few months (good ones i kept), and have some base to work with.

[ornate elk]

Not with prompt building, I just don't have time for such stuff. More interested in technical issues or new techniques.

[cosmic furnace]

So what about wan not working on forge neo ? -_-

[ornate elk]

It worked for me

But I used the standard gguf files

Q4 should work okay

[cosmic furnace]

gguf u mean kinda ... light version for low vram ? or nah

[ornate elk]

Its smaller

To fit in 24gb

High and low have to be around 10GB each

[cosmic furnace]

yahhh i used high and low 13 gb each . ish?

but they all are? kinda?

[ornate elk]

Yea fp8 is to large :/

[cosmic furnace]

o___o

thats why i see comments like this?

[ornate elk]

Yes, as always, the model needs to fit into the GPU vram.

[cosmic furnace]

but what dya mean too large if ya got 24 gb vram xD u cant go larger -_-

dont tell me most popular on civitai dual high-low wan modelis for... 28 gb vram ? o.0

[ornate elk]

Yes or ram only with --novram on comfyui

[cosmic furnace]

but... its dasiwa lighting wan ... with skull ... even 2 of them xD

[cosmic furnace]

comfy -_______-. novram with 32gb .... 28gb will barely fit 😄 . Unless pagefile... dang i wanted to upgrade to 64 ram . but now not sure... u know)

[finite lotus]

I did and i got this error again

[ornate elk]

Yea me to but ram prices are to high rn.

[cosmic furnace]

lol 20gb each high and low 😄

[ornate elk]

You need Q4 or ROCm 7.2 maybe improves stuff in the future

[cosmic furnace]

U wanna say ur in the same boat? the ones who didnt buy those desired additional rams in time ? 🙂

ppl cant say if itll go lower or higher for sure... most likely lil bit higher =\ but shud calm kinda in a ... year or so ? 🤔

[cosmic furnace]

Thx for clearing this out. THis makes things alot clearer +). GOsh 2 files for 13 gb. that both need to "fit" -_-

no Q4 on civitai ? 🤨

[ornate elk]

Select gguf as model type

On the filter above

If not use these:
https://huggingface.co/QuantStack/Wan2.2-I2V-A14B-GGUF/tree/main

[cosmic furnace]

i cant believe dude with 6gb vram had 48-64 ram =\ . but i think he wasnt using zi model from civitai y. although. i thought there pretty much low vram-ram ppl there o.0 . DOnt tell me. for nvidia it goes easier )

[ornate elk]

Yea amd still struggles with memory management. But I hope with ROCm 7.2 in a few weeks it will be much better

[cosmic furnace]

it says gguf but still both high & low is 14.35 gb o.0

[ornate elk]

Q8 is to large

[cosmic furnace]

for me even Q6 is large 🙁

lol Q4 might barely fit

as u said wan is very unoptimized it seems -___-

[cosmic furnace]

this is weird

after updating therock. pretty sure i sued venv activate this time

but yet again had same problem zluda not working until - remove C:\Zluda from the envirnoment path settings.
then open up a cmd and run this:
pip uninstall rocm rocm-sdk-core rocm-sdk-libraries-gfx110X-dgpu -y

[ornate elk]

Probably you were not in the venv again

Forge Neo got a huge update rn. The dev has changed the whole memory management to be as efficient as comfyuis

[hot thorn]

@ornate elk can you maybe recommend someone to me who could help me with prompting? im seriously feeling lost ;<
everything I try to do seriously i research and then do everything how i see fit and it works, but SD has me seriously stomped ;<

[wet elk]

Anyone ever tried getting this error when trying to start ComfyUI "Checkpoint files will always be loaded safely."

It just ends there - is it beacuse it is conflicting with my SD classic UI's checkpoints?

[rose apex]

Tell me about what kind of results did you expect and what did you get instead

[hot thorn]

just not good enough results. not enough detail, not very well done hands or feet, sometimes messed up torso etc

[rose apex]

That's not an error

[rose apex]

Your resolution might be too low. What resolution are you using for generation? And what's your GPU? Also, what checkpoint are you using

[hot thorn]

rtx 3060 gaming x trio 12gb
1080x1080 upscaled by 1.5

[rose apex]

What's your checkpoint

[hot thorn]

i mostly use wai nsfw illustrious v13, tho over the past few days i tried versions 13 to 16, and non nsfw 12 to 16

[rose apex]

You use civit ai right? Did you try to use the prompts from their example images of this checkpoint?

[hot thorn]

dude my prompt is way longer

i dont use prompts to just generate 1 or 2 images

i have a system that allows me to use the base prompt for all characters

• dynamic prompts for characters and poses

especially so I can make dynamic prompts work with adetailer

[rose apex]

Oh you're using Forge?

[hot thorn]

reForge

[rose apex]

Not sure I understand what you mean by this

[hot thorn]

by system i mean prompt structure

general prompt
BREAK
character lora
BREAK
pose
BREAK
style lora

[rose apex]

I doubt you need to use BREAK for stuff like this

BREAK from what I know is used with Forge couple where you are trying to generate two or more different characters in one picture

[hot thorn]

ye nah...

[rose apex]

How about you try Oneobession2.3D checkpoint. Try the prompts of the example images. Try 1024x1536 resolution without any style or character lora

[left cloak]

Greetings, I am a need of some help. When installing a clip during the first launch of stable diffusion a1111, it fails with an error: Cannot import 'setuptools.build_meta'.

[hot thorn]

checkpoint is not the problem man, i ran my prompt across 30 checkpoints the past week

[rose apex]

If that gives you satisfyingly accurate limbs and body parts then you should try adding Stabilizer lora (0.75 or lower strength) for illustrious and see if it improves the details.

[hot thorn]

i need something to properly fix my prompt

[rose apex]

Show me a example of your prompt

[hot thorn]

cant here, im doing nsfw

[rose apex]

Dm me ig

[hot thorn]

need to friend u, u have ur stuff blocked

[verbal loom]

Hi, I am learning to use inpaint in Automatic1111 WebUI, the shortcut to pan an image is spacebar + drag, but spacebar just scrolls to the bottom of the browser window, I want to zoom in and be able to pan around but spacebar is not working for this, is there a setting that I can change to fix this?

[plain nymph]

I read in #💬｜general-chat from Aryetis that we dont do NSFW here, does that mean I cant ask for help on the matter? sorry in advance will delete if its against rules

[ornate elk]

Yes you can't ask for nsfw help here but for sfw

[plain nymph]

Ok, sorry for that

Would you suggest any T2V models on comfyui that has the best details of anatomy in any case?

[ornate elk]

Which clip?

[ornate elk]

WAN 2.2

[plain nymph]

Or should I run some models in SD? same stuff right

[ornate elk]

Idk what you want to do

[ornate elk]

What's your GPU?

Auto1111 is outdated

[plain nymph]

ComfyUI has same models as me getting SD to import on my own is what I meant

so no point

[verbal loom]

rtx 3070 8GB, Forge A1111

[ornate elk]

Ah okay, your on Forge

Make sure to whitelist the webui in the Browser adblocker

[verbal loom]

I use Firefox with UBlock Origin add blocker, I disabled it for WebUI, but still can not use spacbar to pan image

[ornate elk]

Hmm okay strange, can't test rn on my end

[verbal loom]

Thanks, I might try this in ComfyUI instead

[ornate elk]

Or forge classic

Its more updated than forge

Many improvements added and less bloated

[verbal loom]

Thanks for the info, did not know about other versions of Forge, I will look into it.

[ornate elk]

No problem, install guides are in the pinned messages of this channel

[willow abyss]

What would be the most beginner friendly way to do a style transfer? Preferably using ComfyUI as I have that set up but also open to other options
I'm keep seeing IpAdapter, is this something I need? The github repo I found says it's in maintenance mode

[quick fiber]

Do u do comfyui help here?

[unreal linden]

Okay so I was using PNG Info and loading in old pics that I had generated on previous iterations of SD and noticed something particularly interesting about 'Skip Early Config' which was in the Override settings; although it drastically effects the appearance of the images, they still maintain extremely high quality while generation time is cut in half. Is there a similar implementation of this available for Forge Neo?

[analog finch]

Installation Queue: [ComfyUI-Manager] Skipped fixing the 'comfyui-frontend-package' dependency because the ComfyUI is outdated.
Downloading nodes from workflows keeps "freezing" and it wont cleanly download missing nodes

[analog finch]

Found a workflow that atleast downloaded correctly but testrun popped this up

ComfyUI is the most useless help discord 😂 sorry for the bother

@ornate elk I need you brother

[ornate elk]

You need sage attention for that node

[pulsar rose]

can someone help me with Kohyaa_ss. Im getting an error when i try to train a SDXL LoRA. Please DM if you are familiar with it

[quick fiber]

I'm trying to use Wan 2.2s img2vid but I get a grey output. Am I doing something very wrong

[left cloak]

Greetings, I have encountered a problem. Can someone help?

[vocal burrow]

don t use automatic1111 it s broken, cf pinned message for some guides about other clients

[rose apex]

I always had problems with installing custom nodes in the desktop app. Sometimes even after installing the missing nodes, it says missing the next time I open comfyui. The manual install comfyui don't have these issues

[rose apex]

Which sage attention version do you use with your 7900 XTX?

[analog finch]

Thanks for the reply, is Manual ComfyUI any different?

[ornate elk]

The 1.0.6 one

[misty sphinx]

Can i ask questions here about if i should upgrade my system or not? or is this more of a advanced help support?

[unreal talon]

does img2vid still work?

[unreal talon]

controlnet doesnt show up for me even though i installed it

[calm cairn]

I'm looking for a couple good Illustrious models
IllustriousToonMix has been pretty good but I would like to know any others people like

[rose apex]

It's different as in more customizable ig. You can choose different python versions and stuff. Not much different than the desktop app but less headache

[rose apex]

oneObsession_1323D has lots of baked in artstyles and produces really accurate hands. illustrijEVO_lvl3 has a neat style but produce bad hands without embeddings. unholyDesireMixSinister_v70 and Hassaku XL 1.3 is also pretty good

[analog finch]

I appreciate that, is manual download just the zip file from github or something?

@ornate elk Im doing manual download of ComfyUI. I did "pip install comfy-cli" into a premade ComfyUI folder but "comfy install" said "warn: The recent workspace None is not a valid ComfyUI path."

How do I git clone the repo

[ornate elk]

you dont need the cli

[analog finch]

That was the instructions for manual install 😮

[ornate elk]

i will make an guide for it soon

[analog finch]

Xtreme above said he had no issues with custom nodes

But desktop app is super bugged

How do I move forward with manual tho

[ornate elk]

the portable version is the easiest if you want everything to work

the manual is for more advanced stuff

[analog finch]

Advanced how? In nodes and stuff?

[ornate elk]

in manual you can install sage attention for amd and other stuff

[analog finch]

I just want custom workflow + nodes to work

And install correctly

[ornate elk]

then try the portable version

[analog finch]

Alright, thank you. Do I have to do anything about the pip comfycli or leave be

[ornate elk]

you can run pip uninstall comfyuicli

[analog finch]

I did ^^ @ornate elk May I ask when I should look to put a model in Diffusion_Models and when they go into checkpoint? Different loaders load them different so I was curious

[ornate elk]

it depends on the models, mostly the comfyui examples state where it hasto go:
https://comfyanonymous.github.io/ComfyUI_examples/

[analog finch]

I appreciate that, slowly learning shit

I watched a vid on the entire process of a workflow to understand

How do I download missing nodes in the portable? manual install?

[ornate elk]

install the comfyui manager

and with that you can easily install any missing node

[analog finch]

Oh so it just runs better than desktop

Is there a cmd to install it?

[ornate elk]

Getting ComfyUI-Manager:

1. After the comfyui installation is done go into ComfyUI/custom_nodes folder
2. Click in the File Explorer bar (not searchbar) and type cmd then press enter.
3. Then you copy and paste this command:
   git clone https://github.com/ltdrdata/ComfyUI-Manager.git
4. Press enter and after its done you can close the cmd.
5. Relaunch comfyui.

[analog finch]

Appreciate that, youre so knowledgable and fast

ComfyUI discord could never

Are you AI

[ornate elk]

haha nope ^^

[analog finch]

Thats what chatgpt told me aswell

ComfyUI-Manager: installing dependencies. (GitPython)

WARNING: The script pygmentize.exe is installed in 'C:\SD ComfyUI\python_embeded\Scripts' which is not on PATH.

It downloaded dependencies but this popped up, disregard?

[ornate elk]

can be ignored

[analog finch]

C:\SD ComfyUI>echo If you see this and ComfyUI did not start try updating your Nvidia Drivers to the latest. If you get a c10.dll error you need to install vc redist that you can find: https://aka.ms/vc14/vc_redist.x64.exe
If you see this and ComfyUI did not start try updating your Nvidia Drivers to the latest. If you get a c10.dll error you need to install vc redist that you can find: https://aka.ms/vc14/vc_redist.x64.exe

C:\SD ComfyUI>pause
Press any key to continue . . .

[ornate elk]

your on nvidia or amd?

[analog finch]

Nvidia, and they are updated

[ornate elk]

hmm ok

whats your gpu?

[analog finch]

rx 5080.
[W122 12:13:42.000000000 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator ())

[ornate elk]

ok

very strange and you downloaded the latest portable?

[analog finch]

Yeah

https://github.com/Comfy-Org/ComfyUI?tab=readme-ov-file#manual-install-windows-linux

[ornate elk]

im talking about the portable

not manual

[analog finch]

Yeah it has portable section

I downloaded from there with 7zip

Yeah I am running it

[ornate elk]

ok

[analog finch]

Wait now it started

3rd times the charm?

It worked fine before comfyui manager

Its downloading ComfyRegistry Data

[ornate elk]

ok then its working

[analog finch]

[DEPRECATION WARNING] Detected import of deprecated legacy API: /scripts/ui.js. This is likely caused by a custom node extension using outdated APIs.

Is this from custom workflows or what? Its a fresh install

[ornate elk]

its from nanager and can be ignored

[analog finch]

I restarted it and it started ComfyRegistry again 🤔

It opened but no manager

[ornate elk]

it will always do this

[ornate elk]

screenshot the ui

[analog finch]

Sure once it opens

Takes a minute to launch comfyregistry

I have manager now

When it launched it said missing UnetLoaderGGUF but there is no download button, so I have to look into those custom nodes myself?

I can relaunch and send a ss, i clicked it off

[ornate elk]

yes you can click on manager and on custom nodes manager

there select install missing nodes

[analog finch]

Ah sweet, this is the manager I remember from my first try with comfy

Maybe I never had it on desktop and thats why I had issues

[ornate elk]

it should be included in desktop

[analog finch]

Does it matter what sageattention I use?

Can I download it through manager? I forgot how I got it 1st time

This github is telling me to get Triton Windows. How do I install it into python environment

[ornate elk]

you have to manually install it + triton

inside the venv or the portable one, boath works

on the comfyui discord ask Brksoul he made a guide, i will include that in the next days too

[analog finch]

Should I start with triton?

I did not understand the triton github installation for the life of me 😂

Peanut brain

[ornate elk]

cant help rn sry

[analog finch]

Nah its alright im just venting, youve been super helpful thank you

He just had a million possibilities and brain scrambled, ill get back to it

[analog finch]

RTX 50xx (Blackwell architecture, sm120)
This is officially supported by Triton. It only works with Triton >= 3.3, PyTorch >= 2.7, and CUDA >= 12.8 .
I have Pytorch 2.9.1?

[unreal talon]

controlnet isnt showing up after succesful installation in a1111 webui. anyone got a fix for this?

[rose apex]

What's your GPU? If it's NVIDIA then just use comfyui easy Install to install comfyui and it's dependencies and stuff. Really simple to do. https://github.com/Tavris1/ComfyUI-Easy-Install

[rose apex]

Don't use a11 lots of bugs. Use forge classic instead

[unreal talon]

been using since the beginning...

[rose apex]

I started with A11 as well but it's outdated and many things are broken. Use forge, same ui but it still receives update and less bugs

[unreal talon]

do oyu happen to have an easy install guide like the one above?

[rose apex]

Idk about an one click installer no. But there are guides in this channel's pinned messages.

[analog finch]

Nono I have portable comfyui and im on rx 5080

That was for Triton download so I can fix sageattention

[viral plinth]

Hi,
I’m looking for options to access an NVIDIA GPU remotely
for ComfyUI / AI work.

If anyone offers remote GPU access or knows good options, let me know.
Thanks.

[analog finch]

https://github.com/woct0rdho/triton-windows/releases/tag/v3.3.0-windows.post19
RTX 50xx (Blackwell architecture, sm120)
This is officially supported by Triton. It only works with Triton >= 3.3, PyTorch >= 2.7, and CUDA >= 12.8 .

I have Python 3.13.9
Pytorch 2.9.1+cu130
RTX 5080 gpu

Am I following this correctly? @ornate elk

[ornate elk]

follow the guide from the comfyui discord

brksoul made it, just ask if he shares it in the help channel

[rose apex]

I know, just saying comfyui easy install lets you install Sage attention and all other stuffs along with comfyui portable.

[analog finch]

Idk man how many different version of comfyui do I have to reinstall just to get nodes to work aiaiai

[analog finch]

BrknSoul didnt answer so i guess its waiting game

youtube videos for this shit should all be deleted, useless guides

[steel comet]

Heyooooo! I wanted to make a few posts on reddit about my problem but I couldnt so I joined this server. I keep trying to install Stable Diffusion but keep getting the same "Error Code 128". It also said before something about fatal and directory missing and I tried using ChatGPT to figure it out but to no avail? Anyone experienced this before?

[still basin]

Hello everyone! I’m trying to generate a consistent 2D-stylized character based on a real person using ComfyUI + SDXL.

Goal: produce at least 30 images of the same character for an ebook, with different poses and different facial expressions, while keeping the face recognizable.

Currently, i trained a realistic LoRA of the person, it works for realistic outputs, but it looks bad when I push into Ghibli/2D styles (the face becomes ugly/off-model or turns generic, looking like her "cousin").

My main issues are identity drift across images (face shape, eyes, nose/mouth change). Full-body / pose variation makes identity consistency worse. Increasing style strength (2D / Ghibli-like) makes the face more generic and less recognizable. Fixed seed does not maintain identity when pose/expression changes.

I planned to: Use OpenPose/ControlNet to lock body pose. Use FaceDetailer/Inpaint as a second pass to stabilize the face and control expressions. Possibly add InstantID (Since IPAdapter FaceID doesnt seems to work)

Questions:
For SDXL, which is more reliable for identity consistency across pose variation: InstantID or IPAdapter FaceID?
Any parameter ranges that consistently work (LoRA weight, CFG, steps, etc) for keeping identity under a strong 2D style?

[ornate elk]

Auto1111 is broken and outdated. Use the guides from the pinned messages of this channel for alternatives or if you really want Auto1111 to work.

Whats your GPU?

[analog finch]

@ornate elk I keep getting this

I fixed Triton 3.1.0, SageAttention 2.2 and Visual Studio Build Tools and path environments and everything grok told me

It's progress but grok is circling the same shit now without a fix

[analog finch]

I think I fixed it with correct paths 😂

[exotic river]

hmm

i am having issues with items from NVIDIA Computing directory but I am using ForgeNeo

3 different installs...out the box too...didn't really install anything...just install, requirements.txt and some safetensors

kriss, could you post fix ?

[ornate elk]

can you show a cmd log when you launch forge neo?

also have you used the updated guide from the pinned messages from here?

[exotic river]

no on the update on pinned stuff...let figure out how to do that part first and then I will post log error is still there...ty

[ornate elk]

ok, yea follow the first link of the pinned messages, there select the nvidia forge neo guide

[exotic river]

nice...all good...easiest install so far...ty 🙂

[ornate elk]

perfect, np 🙂

[wet elk]

For using ComfyUI I am considering getting this GPU:

Gigabyte GeForce RTX 5070 Ti EAGLE OC ICE SFF 16G 16GB

Can it make video realisticly?

[ornate elk]

it can

[wet elk]

Thanks for responding and thanks for helping me once again!

[exotic river]

that video card is the price of my whole PC, it better be able to do videos (and make me toast) =p

[wet elk]

Prices are completely crazy right now... and probably wont change sadly

[exotic river]

i do remember when $200 got you a nice video card

paid 290 for the one i have now and thats over my rule on video cards but would have liked to have spent around 500 at least

[gleaming plume]

hi anyone active?

[rose apex]

26.1.1 drivers are out and I installed it but my Rocm version still says 7.1 in comfyui. Maybe I did something wrong? I am trying to do a clean reinstall now

[ornate elk]

nope, 26.1.1 doesnt ships rocm

rocm 7.2 hast to be installed manually inside the venv of comfyui or forge neo

[rose apex]

Oh yeah lol I just googled it. They have some instructions on how to do it on their website

[ornate elk]

https://rocm.docs.amd.com/projects/radeon-ryzen/en/latest/docs/install/installrad/windows/install-pytorch.html

[rose apex]

Just to be clear, I need to do this inside my .venv folder right? Any other command I need to use before pip install?

Nvm it seems to be working. Thanks

[rose apex]

Man i don't mean to bother you but I ran both the commands to install Rocm 7.2 and torch but did I just downgraded from Torch2.10 to 2.9?
I thought I am already 2.9 so Idk why it says it uninstalled 2.10

[ornate elk]

stable is 2.9.1

2.10 is dev

[rose apex]

Should I uninstall 2.9 and get back to 2.10?

[ornate elk]

nope not needed

[heavy estuary]

did stable diffusion or anything involved with it, mine stopped working

[ornate elk]

any error?

[exotic river]

Running into some issues with Wan 2.2+2.1 (chkpt is 2.2, vae and lora are 2.1), I am not sure if it is due to my setup or not. I am running an 8gig nvidia card with 16gig of memory(this might not be enough. If it is, let me know and ill post error message...ty

does with both i2v and t2v

[cobalt agate]

Well, what is the issue? With an 8GB card and 16GB ram, you may not be able to run Wan 2.2. Try using the 5B model and lowering your frame count from 81 down to 25. You might be able to get one second out of it.

[exotic river]

sorry took so long...had to finish some RL stuff

pls ignore the different directory issue..it is all in the same directory C:\Apps4Me....

[ornate elk]

Thats not enough to run wan 2.2 14b

5b only works in Comfyui

[exotic river]

ty 🙂

[rose apex]

It could be placebo but Romc7.2 does seem like a 10-15% faster, I only tried wan2.2 and SDXL but more importantly it's incredibly stable. Previously when I did very high resolution generations, it usually crashed after 1 or 2 generations or sometimes even the first run crashed unless I restarted my pc but now it's completely gone. Not a single crash in 10+ runs in wan2.2

[exotic river]

what can i run currently. 8gig vram, 16gig ram (creating videos). Plan to upgrade to at least 32gig of memory. maybe more if cpu fan isn't too far into memory slots. video card will remain 8gig for the time being.

[tall pier]

hello

is there support for forge neo for macOS?

I heard its better than normal forge

[brave egret]

Hi 👋, wish ya all a nice weekend!

Since I couldn´t find a good extension for my SD to automatically add a watermark on all my images so they are fully automatically tagged without additional waiting time and saved in the same path as normally

I just wrote one myself with Gemini and works perfect

here are my first 2 test images (1 on normal size and the other with upscale)

Not sure if I´m allowed to post the script in here, so if anyone is interested in it, let me know

[exotic river]

nice 🙂

i had Gemini compile a report of my health issues (im sick, cough cough). 19 page indepth report (it can read the mri scans too which is just crazy)

you wrote in python?

[rose apex]

What's your GPU

[exotic river]

R5 5600

oops

2060 super 8gig

[sudden lantern]

Guys, is anyone else having trouble with KJNodes Set/Get nodes after the latest ComfyUI update? They are missing from the search list, and my existing workflows show a red frontend_only error on them.

I checked my custom_nodes folder and setgetnodes.js is definitely there. Is this a known compatibility issue with the new frontend? Should I just migrate to 'Use Everywhere'?

[rose apex]

Re install it. Also turn off Nodes 2.0 if it's enabled. Just re installing it fixed it for me though

[rose apex]

You can run SDXL and illustrious models ig. Set page size of your windows to 16 - 24 GB

[fast nova]

(Regarding slow upscale on windows&amd, moved from general chat)
These's console output:

[ornate elk]

can you shwo your webui-user.bat content?

[fast nova]

I wonder what makes 5 steps upscale even for 1.2 multiplies takes 2 minutes – what does "CPU Composite" means?

[fast nova]

set TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1
@echo off

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

set COMMANDLINE_ARGS=--cuda-stream --use-pytorch-cross-attention

:: --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

[ornate elk]

looks okay

cpu composite means that the cpu does the tiling math as its very fast

[fast nova]

I tried nmkdSiaxCX_200k and 4x-UltraSharp upscalers, they're ESRGAN

[fast nova]

Is it ok for it to take 2 minutes?

[ornate elk]

can you screenshot your webui? with the steps and resolution etc