🤝｜tech-support

[prisma moss]

should

[atomic zenith]

where do I enable tiled VAE?

1216x1216 => 1980x1980

[ornate elk]

If your on forge neo, then its under the Never OOM menu in txt2img

[atomic zenith]

found it

[atomic zenith]

so only 10 to 15 steps? Should I avoid using hires fix with controlnet right after the main generation?

[ornate elk]

Nah thats okay, and yea just 10-15 is enough

Tiled VAE will prevent the oom and freezes

[atomic zenith]

I hope so, I have a pretty good GPU

one would assume I wouldnt have these issues

[prisma moss]

cuz u had weights at 0.5

meaning u were only allowing 8gb of stuff to load to ur gpu

[atomic zenith]

I admit I have no clue what I was doing ^^'

but seriously, what should I have the blocks on GPU set as, before it was set to 0 by default

[prisma moss]

then leave it at the default value

whatever it is

[atomic zenith]

do I enable the first option as well or only the second?

[prisma moss]

try without it first at all

[atomic zenith]

it froze again

[prisma moss]

then try the enabled for vae thing

[atomic zenith]

This was not a problem before wtf is going on D:

[prisma moss]

and did u reload the ui after applying any of these changes

[atomic zenith]

yes

[prisma moss]

what does the console say when u freeze

if anything

[atomic zenith]

no it just freezes

[prisma moss]

time for a fresh install

[atomic zenith]

It was a fresh install, kept the old one, reinstalled a brand new copy after that

I dont use automatic1111 I use stability matrix

[prisma moss]

do u have studio drivers

[atomic zenith]

It should be the studio drivers, last time I checked thats what I had

Gonna try the old stable diffusion webui, see if it still happens on there

if it does then maybe something is wrong with my PC

[ornate elk]

Did it worked with tiled VAE enabled?

[atomic zenith]

it froze

[prisma moss]

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

[daring reef]

anyone knows?

[ornate elk]

which launch args do you use?

[atomic zenith]

hmm?

for that one none at all, before the update I tried --reserve-vram 3

before that I tried --reserve-vram 2

[ornate elk]

for rtx50 use these laucnh args:
--cuda-stream --sage --fast-fp16

[atomic zenith]

but will that prevent the freezes?

[ornate elk]

it should improve the performance

[atomic zenith]

Im working with the old forge, so far its working

the only performance Im concerned about right now is not having my computer require a manual reset every time I generate a pieceD:

I will give it a shot once I test one more generation on old forge

[atomic zenith]

you mean this?

[ornate elk]

i would recommend a fresh manual install of forge neo with the guides from the pinned messages

[atomic zenith]

If i run into anymore issues Ill definitely do that. It worked with one image

But thats just one, I shouldnt feel this anxious about generating an image with a pc like mine ya know? ^^'

[ornate elk]

yea definitly not xD

[atomic zenith]

Yeah Im probably gonna see if it works better on its own

got another crash, if I can help it Id like to not have to repeat that

like the sixth crash

[prisma moss]

dont use --cuda-stream

[atomic zenith]

what does it do?

[prisma moss]

basically nothing it says it does on paper

[atomic zenith]

So it cant be on an external hard drive or anything of the sort?

will it let me choose where all the files go?

lora are very large, I try to keep my main smaller drive free of lora

So I cloned it where it asked me to, I know Im getting technical but AI is very technical so... it asked me to add --xformers at the end of the set COMMANDLINE_ARGS=

Do I put a space between them or no?

so like

set COMMANDLINE_ARGS=--xformers

or

set COMMANDLINE_ARGS= --xformers

[prisma moss]

The latter

[prisma moss]

It can but you would need to specify paths

[atomic zenith]

Like I did with C:\AI?

[prisma moss]

Idk at the top of my head

[atomic zenith]

so like this then?

[prisma moss]

U dont need medvram

Shouldnt need at least

[atomic zenith]

Im just following instructions, it doesnt tell me what to add if you have a 16GB graphics card

oh

now that I look at it, just dont add anything

huh...

so like that then?

what about everything else?

[prisma moss]

Such as

[atomic zenith]

File "<string>", line 24, in <module>
ModuleNotFoundError: No module named 'torch'
[end of output]

note: This error originates from a subprocess, and is likely not a problem with pip.
ERROR: Failed to build 'xformers' when getting requirements to build wheel

[prisma moss]

Did u run the update.bat after installing

[atomic zenith]

D:

looks like im redoing it

I didnt see an update.bat in there though

there is no update.bat in there

[atomic zenith]

thats whats going on

I was using the wrong instructions xD

[atomic zenith]

ok its up

It was up for a while apparently

[atomic zenith]

Ill try it but first can you tell me where what folder thats in?

[prisma moss]

main folder

[atomic zenith]

I dont have one

[prisma moss]

then we r not on the same ui

[atomic zenith]

I think its Neo

[atomic zenith]

is there a launch augmentation that lets me only use gpu vram?

[sudden pike]

yo

i installed Khoya_ss for ROCm (RX 7800xt on windows) but my Ram explodes to full

[atomic zenith]

I think my GPU information is leaking into my normal ram memory somehow

[sudden pike]

i had one "training" moment where it sat down at 13gb vram and no shared RAM

but no matter what i try now. it runs fill really fast and then it learns slow af (obviously)

here my settings

here a example of "ram explodes" lol

[prisma moss]

That happens when u run out of vram

[tidal cargo]

Hey i've got serious problems to get Automatic1111 running again. I've installed it several times and used it for a long time. But after reinstalling windows (twice now to make sure) something is always missing. For example: Cant build clip, taming is missing, mdir is missing...

I've installed Python 3.10.6, up to date git and nvidia driver, nothing else yet

[vocal burrow]

known issue, short version, don t install automatic1111 it hasn t been updated in two years and it s broken by now.

cf pinned guide for alternatives

[tidal cargo]

So better to just completely switch to comfy?

[vocal burrow]

or some version of forge, yes

[tidal cargo]

Okay thank you

[atomic zenith]

That's the thing I was keeping a good eye on my VRAM, the amount of memory being used was primarily coming out of my system.Ram

[atomic zenith]

Is everything highlighted in yellow being used? Or are those just showing the launch augment codes that you can add to the top?

[prisma moss]

the latter

[frank swift]

anybody know how to fix this error? any advice would be appreciated

[ornate elk]

What's your GPU?

Auto1111 is broken and outdated

[frank swift]

cant recall it from the top of my head but its fairly less than average

im on a laptop from 2009

[ornate elk]

That won't work at all for image generation

Only on very very slow CPU mode maybe

[frank swift]

@ornate elk

[ornate elk]

@frank swift

[frank swift]

huh thats weird i cant see that reply

i cant click on the reply youre linking

[ornate elk]

lol

maybe restart discord

but yea that 2009 laptop cant be used for ai stuff at all

[frank swift]

aw

so im guessing that my stable difussion isnt working cause my laptop dosnt have enough gpu?

[ornate elk]

yes, and it doesnt have an dedicated gpu at all

the intel igpu is not powerfull enough for these tasks

[frank swift]

a shame

welp tyvm then

[ornate elk]

no problem

[balmy monolith]

I'm having some instabilities in comfy with a particular node, as it's a bit outdated, it requires a pytorch library that is much older than what I have and as such, it's running on cpu. I've read a few hints that I can install a virtual environment on a per-node basis, which might be the answer. Any suggestions how I can install a virtual Environment on this node? I've extensively searched the internet and there is almost nothing of substance which describes how to do it.

If I install the required pytorch, I break a LOT of other stuff

[sudden pike]

@ornate elk Can you help me with that?

i tried multiple settings but no matter what i do. it runs full

[ornate elk]

Which model type do you train and what's your gpu?

[daring crypt]

Hey man
Are you able to fix it now?

[ornate ruin]

Good day to you

Trying to make insightface work for me inside Forge Neo

Using python 3.13.12 as it's the recommended python version for Forge Neo

But I'm having issues installing Insightface I believe.

Could the too new python version be the cause?

[sudden pike]

sdxl and RX7800xt

also no matter what res i take

it just runs full

16gb vram and 48gb shared memory

[abstract cloud]

Hello guys i am using these models i am getting bad results what settings should i use ?

[white stirrup]

i get this when opening the comfyUI app however nothing shows up in the console and it seems to last forever despite me having a very powerful PC and high speed internet

[ornate elk]

hmm, cant help much with kohya :/ its also a bit outdated. Maybe OneTrainer would work better

[ornate elk]

select z-image on the top left

[abstract cloud]

U mean on ui preset?

Answer me bro

[ornate elk]

yes

[abstract cloud]

But there is no z image

I have sd sd 1.5 sdxl

Qwen

And lumina

And flux

[ornate elk]

ah okay, then use lumina

z image is based on that

[abstract cloud]

But my text enoder is qwen is it ok ?

[ornate elk]

yea thats fine

[abstract cloud]

Ok thanks buddy

[ornate elk]

you can see examples of the models here from the forge neo devs wiki:
https://github.com/Haoming02/sd-webui-forge-classic/wiki/Inference-References

[tepid crypt]

any recommendation for nsfw image editing tools through natural language? like chat gpt and similar
have tried already qwen image edit but seems like my pc can't handle it so gave up on it, is there any new way that doesn't requires as much as those heavy models or any tool online that allows nsfw and is relatively free?

[abstract cloud]

Thanks

[tawny saddle]

there is nu such thing as free mate

that said tho

Chatgpt offers or will, or is planning or it has been said that it will roll out adult mode for image gen

wont be free tho and i imagine u need to age verify

[tepid crypt]

people, do you know if is possible to use Anima preview to do image edit (img2img)? also wanted to use natlang with it, tried a few times but doesn't seem to work, not sure if my configuration isn't the proper one or i need another model, i'm using preview 1

[ornate elk]

follow the updated guides from the pinned messages

best would be Forge Neo as auto1111 is outdated

[grizzled mural]

Hey. How do I remove metadata/prompt in downloaded png files? I know I can do this by converting them to jpg, but is there any way to keep them as png?

[neon bone]

Auto1111 outdated? Damn I really fell behind...

[ornate elk]

this too lcan do it:
https://exifcleaner.com/

[prisma moss]

If forge/a1111 there are built in options for that

[prisma moss]

@ornate elk do u know for a bot/setting which can autoblock discord invites?

[supple osprey]

hello everyone

while installing Stable Diffusion , I came upon to the following error , and couldn't fix it myself. Please be kind enough to help me

[prisma moss]

install forge not a1111

[supple osprey]

why?

I see why

[supple osprey]

thank you

[supple osprey]

same error

[supple osprey]

YAAYY! it worked

[sudden pike]

@ornate elk do you know if its possible to use ADAMW_8Bit in Onetrainer with Zluda and an 7800xt?

or is TheROCK version better?

[celest mountain]

hi guys. i have a question. For in painting. Do you guys still use Stable diffussion or comfyUI?

I don't know where to start with comfyui. I just want to inpaint a couple of anime images.

i heard there's like a seperate checkpoint that works for inpainting and not just like any models would do.

idk about using other models for image generation since i really like Fabricated XL style. since if i do use another inpainting model, it would generate a different style

[ornate elk]

Adamw 8bit requires bitsandbytes to be installed first to make it work.
ROCm/TheRockis better than zluda

OneTrainer with Zluda is a bit outdated

[exotic kiln]

Anyone knows a good comfyui custom node that will let me save the image to a specific location with a specific name? Was node had something like that, but it breaks with every update.

[exotic kiln]

I also want it to save the metadata in the picture

[ornate elk]

by default save image is already saving the mtadata

[exotic kiln]

Yes, but i want to save it to a specific path

[sudden pike]

Thanks

[wispy steeple]

I formatted my PC and reinstalled stable diffusion, using the same models I used before but I keep gettting errors like these, any idea how I can fix it?

[wispy steeple]

No i couldn’t figure it out

Tried deleting and installing multiple times, tried to update python, tried different models , just tried forage neo, it does sooome what work but it gives way worse results than before and a lot longer generation time (i do use rtx 2070)

[prisma moss]

@ornate elk the guy above me legit?

[prisma moss]

seems more like a browser issue to me, not allowing insecure connections

[ornate elk]

was a support scammer

[prisma moss]

https://tenor.com/view/out-disappear-bye-vanished-gif-4932063

[ornate elk]

is banned now

[ornate elk]

forge neo should work much faster when you add the correct launch args.
make sure you followed the updated setup guides from the pinned messages of this channel

[wispy steeple]

Yeah i figured

[wispy steeple]

I did add it though it is noticeably slower than the normal forge for me

[prisma moss]

yep

[ornate elk]

hmm, then you can try setup Forge Classic, its nearer to forge then neo

[prisma moss]

same thing, i know what hes talking about lol

i still havent moved away from vanilla forge to classic/neo and i tried believe me

[wispy steeple]

The weird thing classic forge was working just fine

[prisma moss]

classic or vanilla, not the same thing

[wispy steeple]

After i formatted my pc few days ago something bricked

[wispy steeple]

Vanilla*

[prisma moss]

both classic and neo suck at cleanup inference memory or whatever the step is

it hangs so fucking long for no reason

so it makes hrfixing/adetailer miserable

[wispy steeple]

Yeah for me neo was pretty slow taking almost twice as much time to render

[prisma moss]

¯_(ツ)_/¯

[wispy steeple]

:/ though I have no idea how to get the vanila forge to work again

[prisma moss]

Try asking gemini or smth

[wispy steeple]

I tried that already sadly didn’t solve the problem

[ornate elk]

whats the problem?

can you share cmd logs etc?

[wispy steeple]

Sadly im not home now but it seem to have happened to others here too

The same error errortype: non type

[ornate elk]

feel free to share the info when you have time, its definitly fixable

[wispy steeple]

Alrighty

[ornate elk]

maybe this one helps

[chilly trail]

Nvidia drivers up to date and vc_redist.64.exe is installed.

This is a clean install, just to clarify.

[prisma moss]

did u run the updater

[chilly trail]

Which updater? It's the portable version.

[prisma moss]

Mb

[ornate elk]

What's your GPU?

[chilly trail]

Sry for late response. RTX 4080

[ornate elk]

can you make a new folder on the S drive call it AI, then download the latest zip again and rerun the bat?

[chilly trail]

Sure

[chilly trail]

what is this witchcraft

How the F does changing the folder help?

[ornate elk]

the main drives are system created so it handles permissions different, thats why a user created folder is mandatory

[chilly trail]

Well, weird. I tried to run it from the same DIR as my old install

Obv deleted everything inside the folder before.

thanks for your help

[ornate elk]

no problem

[tribal needle]

why do i keep getting this error when i try to dowload stable diffusion (web ui) 😭

[ornate elk]

because your not using the updates guides from the pinned messages of this channel

auto111 is outdated and broken

use forge neo instead

and use the guides from this channel

[tribal needle]

Oh yeah sorry, I was following some guides from internet so-
I just didn't know why for me it wont work

[ornate elk]

because the internet guides are old and not updated with a fix for this new issue of auto1111

[tribal needle]

Omg thx, sorry for not lookin around and just askin🔥

[ornate elk]

hey not problem at all!

always good to ask here before asking a ai xD

[tribal needle]

Was lwk asking chatgpt-

[ornate elk]

whats your gpu btw?

[tribal needle]

I think Nvidia?

Yep Nvidia

[ornate elk]

and which one?

you can see the model in task manager

[tribal needle]

Geforce rtx 5060
Is it good...

[ornate elk]

with 8gb vram? its okay

[tribal needle]

I mean
Idk theres like
16
But then also an 8 and another 8

My pc is in French so idk hows it called in English

[ornate elk]

dedicated vram and shared vram

[tribal needle]

Ye ye

That

[tribal needle]

Ok so uh, how do I actually use forge neo?
I installed like a model from civit ai

Is there any instructions

[ornate elk]

nope xD

[tribal needle]

Oh wow

[ornate elk]

make sure you go with an sdxl or illustrious model first

these are easy to use

[tribal needle]

I installed like, a model and placed it on the model file, stable diffusion

[ornate elk]

then all you need to do is to select XL or SDXL in the preset top left

if its an other model type you need an other preset and more files

[tribal needle]

sdxl?

[tribal needle]

So does the base model count for sum?

[ornate elk]

yep sdxl/illustrious mostly just works with one file

[tribal needle]

Oop- why do I get always an error, connexion errored out

[ornate elk]

make sure you dont close the cmd (its the actual programm), the browser ist just the ui to control it

[tribal needle]

OH ok ok imma see

Imma restart it

And is the VAE important?

[ornate elk]

yes its always good to have one for each model type

here is the vae for sdxl:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main

[tribal needle]

It's like, a quality upgrade?

Ok uhm lwk, wich file do I have to install?

[stone stump]

note: This error originates from a subprocess, and is likely not a problem with pip.
ERROR: Failed to build 'https://github.com/openai/CLIP/archive/d50d76daa670286dd6cacf3bcd80b5e4823fc8e1.zip' when getting requirements to build wheel

what is happening rn

[ornate elk]

download the sdx.vae.saftensor and put it into models/vae

[ornate elk]

whats your gpu?
auto1111 is outdated and broken

use the updated guides from the pinned messages of this channel instead
Go for Forge Neo

[chilly trail]

GGUF files should go into model/unet, correct? Because they are not showing up when I'm using "Load diffusion" node.

I have restarted comfyui.

[ornate elk]

you need the gguf loader node

[chilly trail]

Yea me dumb dumb. I saw unet and thought load diffusion would be able to load it.

[stone stump]

3070ti

[stone stump]

I'll try this later thanks

[ornate elk]

np, yea let me know if you have any questions

[chilly trail]

Is it just me or is it impossible to download qwen lightning lora from huggingface? I get failed download no matter what version I try.

[tribal needle]

What do I do when I get, failed to recognize the model?
(Also thx so much for the help🫡)

[tribal needle]

Oh nvm I got it

[floral cedar]

Hi everyone,

I’m currently working on creating an AI influencer using ComfyUI (SDXL).

I already managed to get very good and consistent results for the face using PuLID (the identity is almost perfectly preserved across generations).

My current issue is with generating the full body.
When I try to add ControlNet (pose) and generate a full-body image, I start losing consistency between the face and the body, or the proportions / realism don’t match properly.

What I’m trying to achieve is:

• keep a 100% consistent face identity
• generate different full-body poses
• control proportions (slim waist, wider hips, etc.)
• maintain realistic skin / lighting consistency between face and body

My current setup:

• PuLID
• IPAdapter
• ControlNet (OpenPose / DWPose)
• SDXL checkpoint (RealVisXL / Juggernaut)

If anyone has a simple and stable workflow (not too many custom nodes) that works for consistent characters / AI influencers, I’d really appreciate it 🙏

Thanks!

[blissful shell]

RuntimeError: CUDA error: no kernel image is available for execution on the device CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

I already updated the pytorch but this still this happened.

[ornate elk]

Which webui and gpu do you use?

Providing a complete cmd log can help too

[prisma moss]

u r cooked basically

[blissful shell]

actually nevermind, didnt know I'm suppose to use neoforge now

[balmy marsh]

Hey, got a real project question. Working with a product design company that does packaging for major UK retailers. Their team works in Adobe Illustrator doing 2D CADs and currently uses Adobe Dimension for 3D visuals, but they find it way too slow for iterating on client feedback. Anyone know a faster tool or workflow for going from 2D Illustrator packaging designs to realistic 3D visuals? Open to anything, doesn't have to stay in the Adobe ecosystem.

[steep granite]

Hi I am trying to decide if to get 64GB RAM for my RTX 5060 Ti to do Q8 Wan 2.2 nsfw videos in ComfyUi or if to get 96GB RAM my GPU is only 16GB VRAM so wondering how important this system ram is, I do want to do more than 5 second clip maybe combine it to 15 seconds with VACE etc

[prisma moss]

illd get 96

[prisma moss]

@light elm 👆

[celest delta]

did something happen with an update of forge neo or something?

everything was fine yesterday, and today all my gen looks like this :

regardless of settings

[celest delta]

@ornate elk hope yyou can help me whenever you can, no rush though, you've always been helpful

[celest delta]

FIXED IT! Yup it was an update

vae.py was changed, downloaded the older one, and now it works

[sudden pike]

@ornate elk
U know if ADAMW_8Bit works for AMD ROCm?

i try to make it run cuz of the benefits but yeah, no...

[ornate elk]

Yes it can work but only with bitsandbytes installed.
You can check my lora setup guides for the steps.

Python 3.12.10 is required

Its listed here under Step 6.
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Lora-Trainer-Setup-Guides#amd-sd-scripts-with-rocmtherock

[sudden pike]

wanna use Onetrainer

i have to reinstall it cuz i have Python 3.11

or can i just upgrade python lol

[sudden pike]

can i just "upgrade" pyhton itself

or will it break my other stuff too? xD

also have problems that Onetrainer cannot find ROCm

have 6.4

[ornate elk]

I need to update the OneTrainer guide with the latest ROCm instructions

[sharp mountain]

help

CLIP

it won't install, and I've tried everything

[vocal burrow]

auto1111 I guess ?

[sharp mountain]

yes auto 1111

go im give you any dask code

Can you help me?

[vocal burrow]

auto1111 is basically dead by now, it s not maintained anymore. I would advise against using it.

Instead use one of the many forge (forge neo, forge classic, etc)

it s basically the same interface so you should be able to transfer easily

cf pinned message for a guide on how to install them.

[sudden pike]

ah ok thx

[sharp mountain]

hmmm

I'll try, but I could use some help, because I'm good at it.

[vocal burrow]

post in here if you hit any roadblock

[sharp mountain]

could you please give a link to classic

I want to try to install

[fair oxide]

you should familiarize yourself with comfyui

comfyui is the cutting edge tool

[sharp mountain]

go voice am give stream

[fair oxide]

there are video tutorials on it

[sharp mountain]

okay

thanks

[fair oxide]

about 2 years ago when i was using auto1111, i didnt think i'd get used to comfyui, but i did, its not too difficult, also a lot has changed since then, now you have most of the workflows of all the popular models inside it as template that you can simply click and use

@sharp mountain go at your own pace https://youtu.be/HkoRkNLWQzY

[sharp mountain]

thanks

[brittle lantern]

@ornate elk I have a build question so 5070 ti in UK is £812 cheapest and 9070 XT is £550 ranges as usual do I overspend on 5070 TI than normal prices it should be or go ROCM on windows with 9070 XT genuinely no clue what to do here

[ornate elk]

that depends.
If you also Game, boath Cards have the same overall performance. its a bout a 5% difference on game to game base.
For AI stuff nvidia is still better in compatiblity/usability with ai software.
But with AMDs efforts on ROCM thats improving right now. ROCm support on Windows is now on a point where you get the same performance as on Linux. There are a few hiccups here and there but overall you can use all of the webuis and even lora trainer tools. They still adjust ROCm so memory management and vae stuff will get better.
If you also consider how windows goes into enshitification mode, going for AMD in the Long term makes an switch to Linux in the future easier.
Also I dont think a 5070ti is worth 812.. Ngreedia cant get enough it seems xD
But if your like an Apple User who just like if "It just works" I dont want to have anything to tinker, then nvidia is the goto for ai software compatiblity.

[brittle lantern]

Yk what u saying Apple user like it just works isn’t even it just works for me either 🤣

[brittle lantern]

🥀

[ornate elk]

asus prime oc would be the best quality after saphire

[ornate elk]

crazy xD

a friend bough it for 720 last year

[tawny saddle]

i have a pretty taskmanager

tasty aero glass mm

btw guess who got a 5060ti >:3

gonna be way better than 512MBs of vram of apu from 5600G cpu

[prisma moss]

hopefully not the 8gb version

[tawny saddle]

is the 16GB one :3

[prisma moss]

ait good

[tawny saddle]

ait as in

aint ovO ?

[prisma moss]

as in aight

[tawny saddle]

ah oki xd

my lil well worked and used APU can finally take a break xd

[zenith marsh]

Could anyone help me with the installation of stable diffusion

[prisma moss]

pinned msgs

[zenith marsh]

I did all the steps but I keep getting this error

ERROR: Failed to build https://github.com/openai/CLIP/archive/d50d76daa670286dd6cacf3bcd80b5e4823fc8el.zip' when gettin g requirements to build wheel

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

[prisma moss]

https://tenor.com/view/fade-away-oooooooooooo-aga-emoji-crumble-gif-20008708

[tawny saddle]

sounds like u could use a snickers

u get a lil error prone when ur hungry

im having to reinsall delicious stable myself, so that i may use my new card

was hoping that it would just work with the one install i had for new stable buh then i rememberd, it was compiled, or some of its compentents were for AMD

and uhm, das a problem if i wan use my card-

[tawny saddle]

i actually got the same error xd
tell u wut imma do tho, imma just try installing 1.10.0 mate

feck 1.10.1 issues 👀

Gemini said dis too

[tawny saddle]

think i may have fixed the issue

now my issue seems to be my card is TOO NEW

:(.

xd

attempting to install CUDA 12.8 or sum

[tawny saddle]

crikey mate, i reckon im stumped

it feckin tryna make me login to github after my attempts ot tryna manually install shi

now imma try wut ever tf forge is and see if that works

after i transplant this venv and see if that does anything

[ornate elk]

You should install forge neo

Guide is in the pinned messages

[tawny saddle]

yas im bouta :v
how is it?
will my shi work like it posed to? all my ancient ahh loras and models

will seeds still be the same and work for diff versions of stable?

like, if an image is made with 456789, will that same seed give me the smae thing output on neo forge

,and will a lora compiled in 2022 work like it does on my current cpu version

[ornate elk]

Nearly the same output but not 100%

[tawny saddle]

interesting

[ornate elk]

Loras will work without problems

[tawny saddle]

das gud..

feckin, shi changing mate, changing too fast imo

here we go againnnnnnnnnnnnn

round 32, fight!

gemini said this posed to be easier to install than waiting for a heap of stuff to do wut ever in termnial

heres to that

[ornate elk]

Yea its easy if you use my setup guides ^^

[tawny saddle]

where the uhm

thingy for neo

[ornate elk]

Pinned messages of this channel first link

There are all the guides

[tawny saddle]

do we have to link loras and models

or cant we just toss'em in

like i been doing since the gud ole days

[ornate elk]

you an just toss them in

[tawny saddle]

yayy

this lookin a lot like regualr stable diff install so far 👀

the way gemini made it sound was like, i wasnt expecting to see a terminal

[tawny saddle]

something is doin'

dargonit, i knew it wouldnt be as easy as update gpu driver

i saw sum1 have this issue awhile ago..or rather, i think it was with torch tho

i just dunt get it

why is it feckin hell to install stable when im using a nivida GPU, and easy as feck to do it on an AMD cpu

a processor that stable wasnt even designed for

goofy ahh shi mate

[ornate elk]

wrong python version

[tawny saddle]

wha

i been using 3.10.6 since the oblivion days

wut am posed to use

oh it literally says it

ahggggggggggg

does 3.14 work

[ornate elk]

nope

[tawny saddle]

still thinks i have old

[ornate elk]

delete the venv folder and relaunch

[tawny saddle]

FUVJ '

this is getting old :3333333333333

greattttttt

time to download the shit againnn, whoopie fuckin do

love doing that

this shit should be plug and fuckin play

this shit easy on a fuckin CHip it wasnt even designed to run on

but nah mate

i go buy a card for nearly 1000

and this shit like, nah

:з

this MF better support my CSS

[ornate elk]

dont forget to install adetailer, forge couple, boorutag autocomplete extensions

[tawny saddle]

ive deleted it twice

[ornate elk]

uninstall 3.10.6

then reinstall 3.13.12

[tawny saddle]

bruh ima bout to purge my whole drive and all to zeros

[ornate elk]

check "add python to path"

[tawny saddle]

alt deleted it all

round fucking 68

[ornate elk]

i hope you uninstalled and not just deleted the python folders

[tawny saddle]

i looked in apps and all i saw was 3.12, no sign of regualr 3.10 anywhere except where it was in C:

i must have never "installed" it

[ornate elk]

hmm ok

[tawny saddle]

so far it seems it thinks im on wut ever it was

3.13

its doin this rn

[ornate elk]

what was your gpu? i now have the same issue

[celest delta]

5080

[ornate elk]

did you used tiled vae at that time?

[celest delta]

yeah

as i said, i just changed back the very last thing that was updated, the vae.py, and it was fixed

[ornate elk]

hmm okay

[ornate elk]

which launch args do you used and which vae was affected (sdxl, sdxl_fp16 ?)?

[celest delta]

--cuda-stream --sage --fast-fp16
no idea for vae? regardless if i was using vae or not the effect was the same until i reverted back to previous vae.py from the forge neo github

[ornate elk]

so even if you didnt choosed a vae in the top dropdown you got these results?

[celest delta]

yeah, i have a z-image checkpoint that has everything inlcuded and didn't need a vae, and same result with it

hope it helps

[tawny saddle]

guess what finally decided to open into the webui in browser

i haven't tested it yet, but I was bouta fall asleep waiting, I closed it to move my loras and models, so idek if 1, it'll reopen without any issues and/or 2, do anything after it opens but, least it made it that far aye

[ornate elk]

it should work normaly

[ornate elk]

thanks it helps alot

[ornate elk]

I messaged the dev and some other users also reported the same behaviour. we will get a fix soon

[ornate elk]

Its fixed now

[tawny saddle]

so, thanks chu CS for da help, it all seems to be goin pretty gud now, really didnt expect it

[ornate elk]

no problem, make sure to add
Git pull to the webui-user.bat in an empty row to always get the latest updates like the fix from 15 minutes ago

[tawny saddle]

dang, it takes my card like

nearly less than 2 secs to make an image

at like p much 2k res, thats wut im talking about '

dang that sure beats waiting 4mins for an image thats like about 512 res

and breh, my card chillin

it aint even crack a sweat mate, at like 36C

xd

[ornate elk]

welcome to the future of image gen 😄

with 16gb vram you shouldnt get any problems at all

you can even try out larger models like z-image-turbo, qwen or Anima, but they need mroe files to work, a special vae and text encoder for each of them
the infos model links and usage examples can be found here:
https://github.com/Haoming02/sd-webui-forge-classic/wiki

[abstract cloud]

Hello guys can anyone help i want to train a z image lora how can i do it with kaggle online gpu

Because google colab hai low system ram

[tawny saddle]

ngl

stable is so much more fun now that i can p much watch the noise turn into an image xd

its fun to watch the morph, and also how fast it be

its also nice that now that i am using a CUDA GPU, i dunt see errors everywhere and things about xformers everytime i start it xd

such a clean start uvu

[viscid oxide]

Which version is the best to use on my Ford ranger? installer.comma.ai/BluePilotDev/bp-5.0 ?

[late tree]

Is there anything that can add shading to a flat color image?

[tawny saddle]

when it says make share true, in launch()
wuts it mean, I remember just having to type "--share" into the command arguments to enable share link

wuts this "launch()" biz

[ornate elk]

yes you can just add --share

[tawny saddle]

oh oki noice 💖

[hidden night]

Guys help I get an error on ComfyUI setup I can't even troubleshoot

[ornate elk]

What's your GPU and what error do you get?

And which install method did you used?

[brittle lantern]

@ornate elk would wanna run on my 6600 xt again forge Ui but any new optimisations or extensions or nothing for forge or older cards anymore

[ornate elk]

There are some improvements as ROCm now supports rdna2 GPUs too

I have to edit my guides today with the latest steps, then you can install it

Forge with Zluda still works.
But with ROCm you can use Forge Neo

[brittle lantern]

Wait I thought ROCm was RX 9000 thing on windows?

[brittle lantern]

So I’m reading this

And still asked

[brittle lantern]

This question

Cool Kat

Good one

[ornate elk]

Nope

Works on rdna3 and rdna4 nativly and now rdna2 too

[brittle lantern]

Woah so when u plan to update it (no rush just questioning) and how big of a improvement is it for me from Zluda to ROCm

[ornate elk]

Idk how it is for a 6600xt 8GB but for the newer cards It was a big improvement

[ornate elk]

@brittle lantern I updated it. Feel free to test the AMD Forge Neo with ROCm guide from here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

Let me know if it works and if its faster

[brittle lantern]

So which one am I following on the ROCm part the 6800-6900?

Or only the top part

[ornate elk]

The 6800-6900

Should work for 6600xt too

[chilly knot]

I set a fixed paging file and now when generating in Wan, my computer reboots with a blue screen🙁

How i need set it

[ornate elk]

whats your gpu and how much ram and pagefile min max sizes?

[chilly knot]

5070 ti 32 gb ram i set 20 and 20 pagefile for test

cose i have out of memory problem

[ornate elk]

set the pagefile to 16gb min and 32gb max

[prisma moss]

No, you should set ur pagefile to system managed, so it will scale to whatever accordingly

[prisma moss]

@light elm

[hidden night]

5070 ti

[ornate elk]

and whats the issue?

any log?

[glad swan]

I keep getting this error. I just had to reinstall Stable Diffusion after getting some new hardware for my computer. I have a GeForce RTX 5080. Below this message is the error

I have Python 3.10.6 as well. This is Automatic11111

I also went as far as uninstalling all of Stable Diffusion, Python, Git and everything else associated with it and I still get this error.

[ornate elk]

Auto1111 is outdated and not recommended for rtx50 GPUs.
Uninstall python 3.10.6 and delete auto1111.
Then follow the pinned guides of this channel for the Forge Neo Guide. It has the same UI but is more updated than Auto1111 and faster too

[glad swan]

What version should I move onto then?

if I am not going to use auto1111

[ornate elk]

Forge Neo like I said above

[glad swan]

will that still work with the models and loras used in auto1111?

and will it also work with my embedding file from 1111?

[ornate elk]

Yep its based on forge which is based on Auto1111

Everything will work

And new stuff too

[glad swan]

also on a side note. will the quality of images look far more improved on forge neo? as in such as better hands and less distortion?

[ornate elk]

That is something the webuis dont set.
It depends on the models and settings you use

[glad swan]

that's ok. as long as i can still do AI with how i did with auto111, then its good

[ornate elk]

Yea sure and you have support for newer models which have better quality

For anime stuff go for Illustrious models.

For realism there is, flux, ZIT or qwen

[glad swan]

should i still remain on Python 3.10.6?

[ornate elk]

Also upscaling with hires fix improves the quality a lot when used correctly

[ornate elk]

No, uninstall that.
Forge Neo needs 3.13.12

[glad swan]

is that the latest?

[ornate elk]

The latest supported version

Python 3.14 won't work

[glad swan]

for 3.13.12 should i tick on both PATH and admin privileges?

or just PATH?

[ornate elk]

Just path

Not as Admin (thats important)

[glad swan]

in the process of installing it. if it works, i'll let you know, if i get errors, i'll let you know

[ornate elk]

sure let me know

[glad swan]

just got a pop up and then i see that stable diffusion has loaded

also, is it possible to create your own preset?

as in, the UI preset

[ornate elk]

you cant create one but you can edit the ones available

[glad swan]

so if i am on the SD preset and i change sampling method to DPM++ SDE, will that automatically save it?

[ornate elk]

nope you have to edit the presets in the settings

[glad swan]

Also, does this new version have improved inpainting?

[glad swan]

the image results in my pics are looking different, and not in the way i want them to. I am using the same prompts as the prompts i used in automatic1111, same CFGs, same model and yet it is looking different in ways i do not like

[ornate elk]

do you have an example you could share?

the webuis dont alter the quality, sure the output cant be 100% the same but the quality isnt affected

[glad swan]

what would i share, the pics or the prompts?

I think i may have fixed it. I went into settings > compatiblity and enabled (use old karras), then I went into the SD settings, changed the schedulars to "karras", the samplers to DPM++ 2M, all CFGs to 7, the height and width to 768 and the steps to 20. The high res is still at 32 though as I do not use that setting

[ornate elk]

yea hires setting has to be adjusted in the presets, hires steps 10 is enough

[tawny saddle]

wuts hires do

[glad swan]

Hey. If I wanted to retrain my pics, how would I do that in Forge Neo? There is no train button.

[prisma moss]

does another latent pass over the image

[tawny saddle]

oh :o.

[ornate elk]

It can improve the quality a lot and also increases the resolution

[ornate elk]

Most people don't train embeddings anymore but instead make Loras. They are better.
To train a lora you need a seperate tool like Ai-Toolkit or OneTrainer

[lyric vector]

I need some help I try to run the user.bat but it keeps telling me to update pip but I already updated pip so when I go to the scripts folder to open up python I try to update pip on that but it keeps telling pip is invalid or whatever\

idk how to fix i

it*

it just keeps giving me a invalid syntax error on pip

[lyric vector]

Nvm I just read I can click run.bat and it just downloads everything for me

Ok nvm I did that but im sitll getting errors left and right

[lyric vector]

Well I almost got close to installing but im get error code 128 and dont know how to fix that ive went to youtube and google and nothing is working

[ornate elk]

You have to use the updated guides from the pinned messages of this channel

Normal installations won't work without the fix.
If your on an rtx GPU for for Forge Neo

[lyric vector]

Yeah so apprently now I need 3.10.11 I had 3.10.6 installed last night and I completely forgot about xformers

[prisma moss]

@ornate elk

[frosty thorn]

Guys I am a rookie, and I have a question. I am using SD Forge, and for some reason the images generated every time have the exact same face. I am using juggernautXL_ragnarokBy.safetensors and the seed is set to -1, maybe because I was trying to use ControlNet? But I turn it off later, can any expert tell me what could be the problem and how could I fix it? Because I used to be able to generate random faces.

[lyric vector]

Can anyone help me here im getting error code 1 and dont know what to do now @ornate elk

[ornate elk]

Use the guides from the pinned messages of this channel

Then it will work. Because they are not outdated like any YouTube video or other old guides

[ornate elk]

reinstall using the guides from the pinned messages, make a new folder for it and make sure your gpu drivers are updated

[lyric vector]

I followed the first pinned support link to the letter and im still getting error code 1

I have everything installed correctly

Im gonna try the redownloadomg torch thing

Thats didnt work either

I followed everything the guide saod and nothing

[ornate elk]

Wrong python version and whats your GPU?

[ornate elk]

The command for installing the manager is wrong

[onyx ledge]

Is there a repository of extensions that are compatible with Forge Neo?

[weak oak]

When trying to generate any image i am getting this error, i am aware that it is because i am using an AMD graphics card (9070xt 16gb) but no matter where i look i cannot find any clear solution to resolve this error, ive tried clicking the link within the error message, installing HIP software and changing the launch options of comfyui, I have managed to get stable diffusion to work before on an Nvidia card, but other than that, i am very new to this, i dont understand any of the command line commands that im seeing everywhere or where to put them so please explain that kind of thing to me like im 5 if possible. i just want to get started as quickly as possible

[ornate elk]

hey, dont use stability matrix for AMD

its not optimised and breaks often

better is to use the manual guides for AMD.
You can find them in the pinnes messages of this channel

Go for Forge Neo with ROCm or ComfyUI with ROCm

[wraith vapor]

Hellooooo Good Evening everyone I hope you have a Great Day so Far :>

Im a New Rookie in Stable Diffusion and Tried Comfy UI just as Automatic 1111 And I see no Future for me.... I really Tried everything but I never get what I want.... I dont even Get the workflows,.... I managed to Learn some Knowledge depending GPU, Detailers, Loras, Promting and Upscaler and more but I never get Sucsess or am Able to Create the Anime Pictures I Want T_T If theres Any Cutie arround here that wouldnt mind give me Some HeadPats of Wisdom and maybe some Help in a Private chat Id be super duper thankful for that ❤️

Stay Healthy Yall!

PS.... I do this like literally 1 Month every secound im not at School.... T_T

[short nymph]

Good day. About half a year ago, I noticed something strange in SD. In an attempt to mix 2 characters and 3-4 different styles in a prompt, I made a simple mistake while writing. Instead of 0.8, I wrote 0,8 in the weight of one of the LoRA styles. How does this work, and is it worth using? Or is it harmful to the hardware?

[short nymph]

For some reason I don't understand, with a weight error it comes out beautifully, but with the correct prompt it gives something terrible.

[lyric vector]

I need help for something minor so I have models I put in the loras folder but only one shows up how do I get them all to show up

[ornate elk]

dont use 0,8 always use 0.8

[ornate elk]

recheck if they are really loras

[lyric vector]

ill double check but im pretty sure they where

[ornate elk]

also check their base type, they onl will show up if a compatible checkpoint is selected first

only loras with the same base as the checkpoint will work with it

[lyric vector]

all but one was lora

one of em was some type of file I never heard of before

lycoris is what it was called

[ornate elk]

works the same as a lora, its just a different type

[short nymph]

the fact is that when I specify the correct weight, something like a screamer comes out in some horror movie of category B. I don't understand how I can achieve the same excellent result with an incorrect weight.

[ornate elk]

it depends on how much loras you use at the same time

some of them break

some only work then on 0.5 or less

dont use multiple loras on 1.0
at least lower each to 0.8

[short nymph]

that is, if there is a weight error, some kind of auto-balancer is turned on, which gives rise to a beautiful picture?

[ornate elk]

there is no balancer

you have to manually test what works and what doesnt

[short nymph]

I just want to understand why with an incorrect weight value, the image turns out great...

[ornate elk]

because it skips using the lora at all probably, in your log you see that it cant handle the string 0,8

also whats your gpu? using --lowvram --no-half slows down gen time a lot

[short nymph]

NVIDIA GeForce RTX 3060 Laptop GPU 6gb

[ornate elk]

then you should only use --xformers --medvram --no-half-vae --theme dark --api
and remove the rest

[short nymph]

webui bat

[ornate elk]

dont know what they do, i would remove them

[short nymph]

I found this on DTF
set CUDA_MODULE_LOADING=LAZY
This variable activates the mode of delayed loading of unnecessary CUDA modules. Aimed at saving video memory.

set NUMEXPR_MAX_THREADS=16
where 16 is the number of threads of your processor.

[ornate elk]

ah okay

[tribal oak]

what cmd do i use, i copied it from the forum

[ornate elk]

python_embeded\python -m pip install -r manager_requirements.txt

[cosmic furnace]

@ornate elk About TheRock any good-decent updates? Or still needed tiled vae & use same resolution everywhere or it bumps precaching ? No other alternatives? Ive heard zluda or something similar to it is coming back soon

[cosmic furnace]

anima not working on zluda forge ? or im missing something? o.0 why it says its a qwen type model on hugging ? or am i missing something ?

[ornate elk]

Forge with Zluda doesn't support any newer models

[ornate elk]

Not much but yes zluda gets more updated. But still no pytorch support

[cosmic furnace]

zluda had updates? o.o

can i run forge neo on zluda ?)

[ornate elk]

No

[ornate elk]

Its always in development

[cosmic furnace]

Well TheRock kinda still ahead then -_- . for at least running on forge neo =\ . Although they become kinda slow with updates & progress y ? As i remember u had hopes they will do better (faster) ? 🙂

[ornate elk]

yea progress is steady but nothing major rn

[weak oak]

apparently this is common... does anyone know how to fix it

[ornate elk]

dont use the AMD build in ai stuff, use the recommended guides from here instead:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides
Go with ComfyUI with ROCm instead

[cosmic furnace]

even though i write uninstall before. hm . Is this latest Therock ? ---- Attempting uninstall: rocm-sdk-core
Found existing installation: rocm-sdk-core 7.12.0a20260219
Uninstalling rocm-sdk-core-7.12.0a20260219:
Successfully uninstalled rocm-sdk-core-7.12.0a20260219
Attempting uninstall: rocm
Found existing installation: rocm 7.12.0a20260219
Uninstalling rocm-7.12.0a20260219:
Successfully uninstalled rocm-7.12.0a20260219
Successfully installed rocm-7.10.0a20251120 rocm-sdk-core-7.10.0a20251120 rocm-sdk-libraries-gfx110X-dgpu-7.10.0a20251120 torch-2.9.1+rocm7.10.0a20251120 torchvision-0.24.0+rocm7.10.0a20251120

[ornate elk]

you installed the latest -dgpu version for gfx1100

its from last year november

latest gfx1100-all is from today, better use that

[cosmic furnace]

pip install --index-url https://d2awnip2yjpvqn.cloudfront.net/v2/gfx110X-dgpu/ torch torchvision

isnt it like this ? o.0

oh i got it... i need the ones for 9**x series ones newer

pip install torch torchvision torchaudio --index-url https://rocm.nightlies.amd.com/v2/gfx120X-all/ this ones ? y ?

yeah 2025 in version was suspicious )

[cosmic furnace]

@ornate elk AcceleratorError: HIP error: device kernel image is invalid Search for hipErrorInvalidImage' in https://rocm.docs.amd.com/projects/HIP/en/latest/index.html for more information. HIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing AMD_SERIALIZE_KERNEL=3 Compile with TORCH_USE_HIP_DSA` to enable device-side assertions.

🤨

File "B:\SD-FORGE-NEO\Forge-NEO\sd-webui-forge-neo\venv\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "B:\SD-FORGE-NEO\Forge-NEO\sd-webui-forge-neo\backend\operations.py", line 369, in forward
return torch.nn.functional.embedding(x, weight, self.padding_idx, self.max_norm, self.norm_type, self.scale_grad_by_freq, self.sparse)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "B:\SD-FORGE-NEO\Forge-NEO\sd-webui-forge-neo\venv\Lib\site-packages\torch\nn\functional.py", line 2567, in embedding
return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

did i messed up when updating or whuht ?

[ornate elk]

When do you get that error?

[cosmic furnace]

when i press start generating button =\ on any model -_-

[ornate elk]

Oh

Probably a broken release

The one from 25.03.2026 works good

[cosmic furnace]

what do i do ? i cant roll down to a specific release ? hmgm

PyTorch Version: 2.10.0+rocm7.13.0a20260329 memory_management.py :: INFO
AMD Arch: gfx1100

03-29 -__-

mb fix in a few days ? +)

[ornate elk]

You can install a specific release but I can't tell you how rn

[near plinth]

Hello,my name is Yuki.
As an AI and full-stack developer with over 10 years of experience, Currently, I’m based in Japan, where I work as a PM on a globally distributed team.
I recently developed and launched a new AI agent system, and I now have some availability to take on new projects or collaborate with a team.
I’m excited to be active in this community and contribute meaningful value to the server and its members.

[prisma moss]

Bro has more experience than AI has been out lol

[lethal tiger]

@ornate elk Hey, I just wanted to say, thank you so much for helping me, and us, download and set up Stable Diffusion. I was able to put together SwarmUI and Forge Neo, and I can decide from there which one I'd like to commit more to. But I really appreciate you explaining everything, I was really stressed out when I switched from my old computer to my new one and Automatic1111 was no longer working. Now, I have Automatic1111 again, basically just under a different name, and SwarmUI which is essentially ComfyUI accessible in the back and with an easier interface in the front to understand and use. So, thank you sincerely, from the bottom of my heart.

[ornate elk]

hey, thanks for the kind words, glad everything worked out 🙂
With Forge Neo and SwarmUI you can do pretty much everything ^^ Always good to have multiple options

[near plinth]

thanks

[cosmic furnace]

@ornate elk so how do i install this specific release?

[ornate elk]

For GFX1100-all it would be these 3 commands:
venv\Scripts\activate.bat
pip uninstall torch torchvision torchaudio rocm rocm-sdk-core rocm-sdk-devel rocm-sdk-libraries-custom -y
python -m pip install torch==2.10.0+rocm7.13.0a20260325 torchaudio==2.10.0+rocm7.13.0a20260325 torchvision==0.25.0+rocm7.13.0a20260325 --index-url https://rocm.nightlies.amd.com/v2/gfx110X-all/

[hushed otter]

I only know how to do math with pascal in secondary school like ten years ago. Im trying to get stable diffusion running on my 4050 6gb and i think i have been having problems every step of the way. Right now im getting an error that is telling me a repository i have to clone is invalid causing it to exit and giving me code 128 the url in question is https://github.com/Stability-AI/stablediffusion.git

i have no idea what url i need to update it to and even if i do get it how do i edit it into the script

[ornate elk]

Hey, you have to use the updated setup guides from the pinned messages of this channel

Go with Auto1111, Forge or Forge classic

[hushed otter]

ty

[cosmic furnace]

Indeed everything working like this... i hope its failed night update or smthing . not 7k series cut out from 9k updates -_-

[ornate elk]

I'm confused why you thought you need gfx12 here?
Because 7 series is gfx11

They publish each new rocm+torch for each gfx type seperately

[tacit basin]

What methods are available to generate pixel images?

The size is approximately 16x16

[serene comet]

hello, can someone help?

[ornate elk]

hey, whats ypur gpu?
auto1111 is outdated, better setup Forge Neo

guide is in the pinned messages

[serene comet]

1660 ti nivida

[ornate elk]

oh then go with forge classic instead

guide is also in the pinned messages

[serene comet]

thanks

[ornate elk]

no problem, feel free to ask if you have any questions

[tawny saddle]

I have a question

how was ur day :3

[ornate elk]

it was pretty okay

now finally 4 days freetime

[serene comet]

hello how to extract promots from image in stable diffusion? I don't have ''Interrogate CLIP''

[ornate elk]

try load the image into the PNG-Info tab

[serene comet]

it's not png.

even if it's , I get (none)

[ornate elk]

then it has no meta data in it

[serene comet]

is there a way to fix it?

[ornate elk]

nope, you would have to use other tools to analyze it and create a prompt, like chatgpt etc

[serene comet]

Is there a way for local?

[ornate elk]

not really

i would suggest you install the Booru tag autocomplete extension

it helps with prompting anime stuff a lot

[tawny saddle]

Yay :3

[serene comet]

how ?

and is there a way to fix this?

gpu : 1660 ti nivida

cpu : i5 9600 k

[tawny saddle]

check out my taskmanager, its glossy uvu

[serene comet]

niec I wanted to do it but I couldn't, youtube wasn't useful, what is your way?

[tawny saddle]

:3

[tawny saddle]

is pretty easy..ish
I installed a few things that edited the windows shell, and sum registery things, I'll need to go over wut exactly i did to explain im detail on how xd, I think it was like two other things other than DMWBlurGlass,

just so that I don't accidentally cause u to brick ur OS xd

[cosmic furnace]

Rly? But in launch guide u mentioned 7x series the rock doesnt get updated unlike 9x ones) so its better to use 9x

[serene comet]

thank you for information

[ornate elk]

You should increase the Windows Pagefile. That can help

[ornate elk]

No, I just showed gfx110x-dgpu and -all and dgpu doesn't get updated anymore

[serene comet]

I gave 24 gb for page file
should I give more ?
if I give more, will that make it faster?

[ornate elk]

16gb min and 32gb max, and make sure its only enabled for the C drive

[serene comet]

should I put stable diffusion on C drive?

[ornate elk]

can help for model loading speed, but doesnt imporve generation speed

[serene comet]

thanks

[ornate elk]

no problem

[serene comet]

hello the problem didn't get solved

same issue

[ornate elk]

can you describe it more?

[serene comet]

now it says 100 mb

[ornate elk]

yes because the illustrious models are around 6gb and your gpu has only 6gb of vram

can you show your settings like resolution etc?

make sure you close all other browser tabs and games etc to free up vram

[jovial vault]

Good morning
I do kept having Server error 403, any one else?

It’s disconnected from API and internal apis

Nothing urgent just write when I remember it Sunday enjoy Sunday guys!

[celest delta]

is there a way with forge neo to associate a simple tag to an image and it being recognised without the need of a .safetensor? LIke i want to see the artstyle of an artist with a picture, click on it, and the tag i associated to it just appears, same as if i was clicking on a LORA

[prisma moss]

No, but theres a website where you can broswe artists and see an example image

[prisma moss]

.

[celest delta]

i know, just wanted to know if there was a way to do it all directly in forge without having to check everytime on another place :p

thanks anyway

[pseudo narwhal]

hello good afternoon. im trying to run (sd-webui-controlnet) on my SD. but this extensions seems not to show up no matter what i do. its the only one that doesnt show up in my feed after installing it. And i was wondering if any of you guys had run throught the same problem. (the other 3 extensions were just to test if they show up in the text generator settings) Thank you for taking ur time to read this, have a great day!

[ornate elk]

hey, can you show the complete cmd log?

[pseudo narwhal]

ive sent it to your DM to not flood the chat with code

[near plinth]

Is there anyone looking for an AI and Full stack developer?

[chrome forge]

gang ive been having this problem for so long it wont fix im using AMD gpu

[ornate elk]

What's your gpu?

Wouldn't recommend using sdnext for amd. There are better options

Check the pinned messages of this channel for the updated setup guides.
Go with Forge Neo if your GPU is rdna2 or better

[chrome forge]

RX 6700XT

[chrome forge]

Send me a dm, I'm so tired rn I need to sleep

[ornate elk]

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

Go for AMD Forge Neo with ROCm

[chrome forge]

Can it run SDXL models?

[ornate elk]

Yes it can

[chrome forge]

so i looked into a bit but it says that its mainly for linux will there be issues if i download it on windows?

[chrome forge]

also it only supports gfx1100, gfx1101, gfx1102, gfx1103, gfx1151, and gfx1200, gfx1201 GPUs. but my gpu is gfx1031

[ornate elk]

My guides from the link above are all for windows

[ornate elk]

Yours is supported too. I need to update the list there. The steps for gfx103x are included

[jade narwhal]

heeeeeelp, so I'm new to stable diffusion and I got automatic1111 working just fine, now all it comes back with is black images and then crashes, I have a RTX 3060 with 12GB VRAM, I've tried the --disable-nan-check --precision full --no-half --medvram --opt-split-attention set OPTIMIZED_TURBO=true but it solves nothing

[ornate elk]

Hey, remove --disable-nan-check --precision full --no-half --medvram and add --no-half-vae

Then make sure the checkpoint you try to use is actually a checkpoint (2-6GB) and not a lora (mostly smaller than 1gb)

[jade narwhal]

trying now

still black

coming back with this error

NansException: A tensor with NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

the model sd-v1-4 works fine, but all others that worked just fine the other day are not

[ornate elk]

What are the other models?

How large are the files?

[jade narwhal]

perfectillusriousXL_40 and it's almost 7 gigs, so it's def a model

[ornate elk]

Then make sure your gpu driver is updated. And you have whitelisted the webui in any browser adblocker.
And if nothing works you should use Forge Neo webui instead of Auto1111 because Auto1111 is outdated since over a year now

[jade narwhal]

got it

[ornate elk]

And don't use SD 1.4 its outdated since 4 years

Newer models can be found on civitai.com

[jade narwhal]

got it, I'll redo everything with Forge Neo, might as well

thank you

[ornate elk]

No problem and use the setup guide from the pinned messages for it

[jade narwhal]

ah, perfect

[ornate elk]

It includes the correct launch args for each GPU type

[jade narwhal]

epic, much appreciated

[ornate elk]

Your welcome, and feel free to ask if any step isnt clear

[chrome forge]

hey man at the end i downloaded comfy ui with zluda when i try to run an sdxl model i get this error

[jade narwhal]

new problem, the webui-user.bat isn't finding Python, I switched to 3.13.12 and uninstalled the other version, I installed it to a path with no spaced, I tried putting the path to the EXE into it's settings, can't find it

nvm, figured it out, app aliases

[jade narwhal]

forge neo set up and my images are back 👍 thanks for the recommendation!!

[ornate elk]

Ah perfect

[jade narwhal]

also the image quality is much better, less glitches

[ornate elk]

Good to hear but normaly the webui shouldn't affect the quality. But probably other default settings were just better now

[jade narwhal]

maybe

[ornate elk]

The folder doesnt look like Comfyui-Zluda for me.
Can't help much with it but can recommend Forge with Zluda if you want it easier

For Comfyui-zluda problems you need to open up a issue on their github

[smoky schooner]

Hello i can make short videos with stable diffusion forge Neo?

[ornate elk]

Hey, yes it can do this with Wan 2.2. but requires a lot of vram

Most people use ComfyUI for video stuff

[smoky schooner]

where can download? I have 5070 12gb

and what if the vram is full?

[smoky schooner]

Why did u cry on my message? @prisma moss

[prisma moss]

y not

[smoky schooner]

instead of crying, could you help too

[prisma moss]

in short, 12gb is gonna be a big bottleneck

there are smaller models but still

[smoky schooner]

But i saw this youtube video and he has 6gb vram and make video

https://youtu.be/55jzTHIhv5k?is=b7MkXRYTU6Jk2fWA

[cobalt agate]

12Gb will work using Wan 2.2. You can only run 81 frmae batches, however. There are some workflows that stitch those clips together for longer output.

[smoky schooner]

I got this error

where can i dowload this? Wan2.2-T2V-A14B-4steps-lora-rank64-Seko-V2.0

[ornate elk]

From huggingface if you google that lora name

[tribal oak]

i pasted the cmd u sent me

[ornate elk]

ok then try the steps from here:
https://github.com/Comfy-Org/ComfyUI-Manager

[chrome forge]

okay so i downloaded Forge with zluda when i try to generate this is the error i get

can you help me with this?

[chrome forge]

my goal is to train a lora and make images using that lora

i have tried comfy ui + zluda (this actually ran but it didnt generate any images),sdnext with zluda, just sdnext, the A1111 ,DirectML setups (A1111 / SD.Next)and now forge with zluda ive run to all kind of issues i looked into the rcom but not 100% sure cus it says its for linux. ive been trying this straight for last two days.

[ornate elk]

Can you show the complete cmd log?

[ornate elk]

Your folder name is incorrect and a wrong launch arg got used

Please follow the exact steps from the guide from the pinned messages

[chrome forge]

so delete this and restart?

[chrome forge]

is this version of amd software important? im running adrenaline version 26.3.1

[young totem]

what's the hardware requirements to run stabilityai/stable-diffusion-3.5-large?

[ornate elk]

You still need HIP SDK yes

[ornate elk]

12GB vram

[jade narwhal]

hey, so I'm struggling to figure out how to do something, I want to take this character and have her in different images, I can't find a good way to do it, I've tried messing with the built in img2img, I've tried controlnet, I've tried ipadaptor, nothing is giving me what I want, any suggestions?

[chrome forge]

okay so i have comfy ui with zluda up and running and theres no issues

now how do i train a style lora? do i need to download somthing else?

[ornate elk]

For training loras you need other tools like OneTrainer

[chrome forge]

i tried downloading it but it said it need a different version of HIP SDK

[ornate elk]

Which one do you have?

[chrome forge]

6.4.2

im pretty sure it asked a 5. smth

[ornate elk]

That will work with OneTrainer too

You have to use the zluda fork of OneTrainer

That supports 6.4.2

[chrome forge]

also im using Python 3.13.0

[ornate elk]

Oh okay

What was your gpu again?

[chrome forge]

rx6700xt

[ornate elk]

Okay, then you shouldn't look into self training of loras

It needs a lot of vram to train and takes very long.
But there are enough already trained loras out there on civitai anyway

So just try them

[chrome forge]

oh, this is a very specific style and i really cant find any loras online with this style

[ornate elk]

Hmm idk but maybe flux Klein 4b can transfer the style on other images

[chrome forge]

can you tell me more about it?

[ornate elk]

It's a model used to edit images.
You can give it multiple input images and edit them
Here is some info and examples:
https://blog.comfy.org/p/flux2-klein-4b-fast-local-image-editing

Idk how well it works on 12GB vram but there are also gguf versions of it that are smaller

[chrome forge]

so there is really no other options?

[ornate elk]

There was also IP adapter style transfer. Its a bit more complex to get working

Easiest way is to ask someone with a 4090 to train the lora for you or to train it online on web services like civitai

[chrome forge]

but im broke 💔

[chrome forge]

i have 125 credits on civitai lets see if that would help

[light maple]

I'm running into an issue trying to launch Fooocus in Stability Matrix. I've installed it via the UI there but when I go to launch even a newly installed fork I get this error (last part of t he stack trace)

  File "C:\Stable Diffusion\Data\Packages\Fooocus - mashb1t's 1-Up Edition\venv\lib\site-packages\jinja2\utils.py", line 477, in get
    return self[key]
  File "C:\Stable Diffusion\Data\Packages\Fooocus - mashb1t's 1-Up Edition\venv\lib\site-packages\jinja2\utils.py", line 515, in __getitem__
    rv = self._mapping[key]
TypeError: unhashable type: 'dict'
Traceback (most recent call last):
  File "C:\Stable Diffusion\Data\Packages\Fooocus - mashb1t's 1-Up Edition\launch.py", line 152, in <module>
    from webui import *
  File "C:\Stable Diffusion\Data\Packages\Fooocus - mashb1t's 1-Up Edition\webui.py", line 1138, in <module>
    shared.gradio_root.launch(
  File "C:\Stable Diffusion\Data\Packages\Fooocus - mashb1t's 1-Up Edition\venv\lib\site-packages\gradio\blocks.py", line 1971, in launch
    raise ValueError(
ValueError: When localhost is not accessible, a shareable link must be created. Please set share=True or check your proxy settings to allow access to localhost.

Any ideas on how I can fix this?

[chrome forge]

it needs 500 credits now where will i find a guy with 4090 wlling to train a lora for free 💔

[ornate elk]

You should try setup RuinedFooocus instead its more updated

[jaunty vale]

can someone assist me with an image im trying to generate please? dm me if you can

[serene comet]

hello, sorry for late, these are settings

can I make it to only cpu??

[ornate elk]

whats was the issue and whats your gpu?

[lunar granite]

hello

I'm looking for a long term partnership to work with me

[serene comet]

this was the issue

gpu : nivida 1660 ti

[ornate elk]

and how long does an 1024x1024 image takes?

[serene comet]

5 mins or less before the issue happens

[ornate elk]

have oyu increased the windows pagefile?

[serene comet]

I made it as you told me

can I make it through cpu?

[ornate elk]

it says 16gb but it should be 16gb min and 32gb max

[serene comet]

these are my settings for page file

[ornate elk]

looks good and you restarted the PC once after that?

[serene comet]

yes

[ornate elk]

what yre your launch args of forge classic?

[serene comet]

the defualt

[ornate elk]

add --cuda-stream --xformers --medvram

[serene comet]

don't I already have this?

unless if you mean I only put these

[ornate elk]

nope

you have to add the after the line that says COMMANDLINE_ARGS=

[serene comet]

like this?

[ornate elk]

save and relaunch

[ornate elk]

ah ok remove --medvram

then save and relaunch

[lean brook]

can anyone guide how to ad the image generation bot to my server??

[agile sun]

I want to generate on the CPU while downloading, I got something like this

[vocal burrow]

check pinned guide

auto1111 has been broken for nearly a year by now. Don't use outdated tutorials

[agile sun]

I don't really understand, I wanted to download a1111 to create images, so how am I supposed to download it if this one doesn't work according to you

or you can help me go through the whole downloading process, I can share my screen with you on a voice chat and follow your instructions

[vocal burrow]

you can use one of the many alternative; forge, forge neo, forge classic, swarmui, comfyui, sdnext, etc

But no, don't waste your time on getting A1111 to work, it's outdated and broken.

[agile sun]

A year ago it worked for me, I don't understand why it doesn't now. As for other interfaces, they are not on the CPU, and some are unknown. For example, comfui is a type of interface that I really don't like, and forge is only supposedly for graphics cards. I prefer not to download these little-known interfaces, you never know if it's not a virus a1111 I trust I can definitely somehow run a1111, it's local, not some server that goes down and doesn't send information

[vocal burrow]

I'm pretty sure webui-forge can be run in cpu only mode with --always-cpu --skip-torch-cuda-test

Also all the suggested softwares in the pinned guide are also for local generation.

And finally A1111 doesn't work anymore because clip removed this file from their github

[agile sun]

Now I understand why a1111 killed the good UI, it was, and what this forge looks like

I decided to download Forge to see how it looks. Should I run it using cmd or not? does this forge start in the same way as A1111

I will run Forge for the first time

[vocal burrow]

yes just run forge classic with the --always-cpu --skip-torch-cuda-test arguments

[agile sun]

--use-cpu all --precision full --no-half --skip-torch-cuda-test wouldn't it be better to use it like this won't it work

[vocal burrow]

different program, different arguments

[agile sun]

Should I use what you wrote in the chat, is what I wrote. sorry for the mistakes in English, I am writing with a translator

[vocal burrow]

use what i wrote

[agile sun]

ok run after saving

ok i run

It is in the process of starting, I hope it will work

[vocal burrow]

if it doesn't work with "forge classic", try using "forge" with arguments --use-cpu all --skip-torch-cuda-test I just tested it and it works

this is the link for "forge" https://github.com/lllyasviel/stable-diffusion-webui-forge

[agile sun]

I downloaded Forge from this site and it is in the process of downloading, I hope it will work

[agile sun]

.

[agile sun]

what should I do now

[vocal burrow]

what exactly did you do ?
Running forge on "cpy only mode" should be as simple as
1/ open cmd in whatever folder you want forge to be installed
2/ run git clone https://github.com/lllyasviel/stable-diffusion-webui-forge.git
3/ put --use-cpu all --skip-torch-cuda-test in the set COMMANDLINE_ARGS= line of webui-user.bat
4/ run webui-user.bat

[agile sun]

I followed the instructions above, I even asked you if it was correct, and I have what was sent I will write above and I have to download again what difference does it make, you also gave the download link before with the same start parameters and it's what was sent in cmd

[agile sun]

Above you have our message on this channel, and from the previous ones that I downloaded from your link, and now with the help of Git I have

[vocal burrow]

ok weird it's working on my linux, switched to windows to give it a try and yeah it's (even more) broken nowadays.

Sigh ok ok. Let's go back to auto1111 but use the dev branch this time.

Let's try this :
1/ open cmd in whatever folder you want forge to be installed
2/ in the cmd, run
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git -b dev
3/ With a text editor, put --use-cpu all --skip-torch-cuda-test in the set COMMANDLINE_ARGS= line of webui-user.bat
4/ run webui-user.bat

[agile sun]

I followed your instructions, now it is downloading components for me

I hope that this is for the CPU libraries

[rose apex]

Current pytorch version: 2.10.0+rocm7.12.0a20260208
is it slower than than 2.9 + rocm 7.2.1? @ornate elk
Also, how do i update to 7.2.1 if my current one is slower

[agile sun]

WHY INSTAL torch-2.7.0+cu128 for nvidia after all, I ran everything above on the CPU

if it doesn't work again could you please go on a voice chat i'll show you my screen and follow your instructions so you can walk me through it step by step while we're on the call, my friend speaks both english and polish so there won't be any translation issues, we'll get this sorted out and everything will be fine

[ornate elk]

you can always upgrade by running these 3 commands inside the webuis main folder:
venv\Scripts\activate.bat
pip uninstall torch torchvision torchaudio rocm rocm-sdk-core rocm-sdk-devel rocm-sdk-libraries-custom -y
pip install --index-url https://rocm.nightlies.amd.com/v2/gfx110X-all/ torch torchaudio torchvision

im on 0325 but its not much faster but maybe more stable

[smoky schooner]

How do i fix this error in ComfyUi

i want to use ultralyticsDetectorProvider but i got this error

[smoky schooner]

How?

[ornate elk]

if they can't help in this chat its probably scam

[prisma moss]

heelp

EDIT: SOLVED

tryna install classic (official instructions)

[vocal burrow]

yeah I'm done

[vocal burrow]

Can't read the error it's the bottom right part and cut

[smoky schooner]

This previous person wrote down how should i fix it, i need to do coding in ComfyUi and instal pip ultralytics and then he blocked me lol. Now, should I do what he wrote or not?

[vocal burrow]

Honestly, They've been trying to "help" people in dms quite a few time. Conviniently out of sight for us to monitor.

Called them out on that and they just kept doing it

Out of curiosity what did they suggest exactly ?

(because yes this server being big we have some scammers)

Also regardless we can't help you without seeing the error

which is cut in your screenshot

[vocal burrow]

looks very llm but not dangerous

[vocal burrow]

and that's your error

[smoky schooner]

Do you know how should i fix this?

I want to fix the face in the image i can’t do it without this ultralitycs right?

[vocal burrow]

pretty close to what Anime mentioned

what's in your comfyui folder ?

[smoky schooner]

The anime guy was a good person and try to help me,why he blocked me🥲

[vocal burrow]

not sure if it's because he blocked you or because I gave him the boot