🤝｜tech-support

[wicked widget]

https://tenor.com/view/king-of-the-hill-you-gif-5929698

[ornate elk]

I knew it from the profile PIC + borb = birb and you like birds xD

[fair oxide]

Ya xdd :3

fair fair

lost like, most of my loras tho xd, i rememerbed most of the ones i used more of the time, and my 3 models xd, no big deal tho, not remembering what i lost is like not ever having them so eh xd

better than having 220GBs of stuff i dunt use

[ornate elk]

True, helps a lot sometimes to have less

[stark coyote]

Hi, im new here and im using A1111 but im having trouble generating a good quality image and im tired of researching on how to do it, my goal is to create AI fanart of anime characters, can someone help me or give me a tip on how to do it?

[wicked widget]

use sdxl models

pony/illustrious

[ornate elk]

Use illustrious models

[stark coyote]

yea im using them

still no luck

[ornate elk]

And get the boorutag autocomplete extension

[wicked widget]

show us ur prompt

and other settings like cfg

[stark coyote]

yea i didnt get any of that sorry xD

[stark coyote]

give me a sec

[fair oxide]

was there another version of stable diff that was for AMD..or gpu mode 🤔

i ask bc

[ornate elk]

What was your GPU again?

[fair oxide]

AMD 5600G raedon graphics

[ornate elk]

Here Auto1111 with DirectML or Forge webui With directml:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[fair oxide]

ahh ;o
so is another version xd, oop

tinx xd

[stark coyote]

sorry only saw this msg rn, where can i get that?

[fair oxide]

🤔 two years ago, how long ago was 1.7.0 xd, that was the last version i was using, and it had direct-ml

[ornate elk]

Click on extensions tab, then on available, then on load from, search for booru autocompletion and install it.
Also search for adetailer. For face enhancing

Then restart the webui

[stark coyote]

may your day be blessed

[ornate elk]

Auto1111 is on 1.10 and Forge on 2.0

[fair oxide]

;o

1.10 was 2 years ago ovO

[ornate elk]

No

July 2024

[fair oxide]

ah

xD

[ornate elk]

I'm off now gn

[fair oxide]

goodnight~

tried to look online and a lot of the comments are jus, same xd

dunno if its the same issue u having but

xd :3

[mystic basin]

you have problems with flux?

with forge ui

?

[vast pulsar]

hey guys, am i doing anything wrong? i got flux from civitai and the vae/text encoder stuff from huggingface or wtv it's called. And yet all my generations are: black squares

[vast pulsar]

yes

[mystic basin]

can't help, i haven't used flux yet

[vast pulsar]

rip

[mystic basin]

i only got the needed files here https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/1050

the text encoder and vae

[vast pulsar]

yea, i supposedly got all the sutff

aka no clue what's going wrong

[mystic basin]

gonna see if it work for me

once the compiling is done, first time using forge

might take some time since i'm on a laptop rn with zluda

[vast pulsar]

hopefully you got better luck than i do lol

[mystic basin]

we'll see

[vast pulsar]

just in case i found a single file that has everything needed merged into it so ain't no way that's failing

but it's a 31 gb file so... ya know

[mystic basin]

good luck

once the compiling is done my pc gonna get nuked, my 12gb of vram won't be enough i think

the image size is quite low so it might work

[vast pulsar]

won't know till ya try

[mystic basin]

yeah, the first time is always long for I don't know what reason (it say on the guide it take longer the first time but doesn't say why)

gonna be back soon once the compiling is done to see if it work for me

[vast pulsar]

well another black square :)))))))) guess i'll wait for the 31 gb file

[mystic basin]

could it be the schedule type?

it was on automatic by default when i downloaded forge and the same goes for automatic1111 when i was using it before

[vast pulsar]

doubt it, never had issues with that in normal webui

never changed it

oh, you mean that it's not in automatic

well-- i can try ig

[mystic basin]

yeah, i don't know if it's that

[vast pulsar]

more likely to only solve it once i finish the big download with all the bullshit merged

[mystic basin]

could it be the seed? like could the seed have problem with flux but if you use another seed, it won't happen?

[vast pulsar]

no, seed is fine, i'm trying to recreate an already existing generation

[mystic basin]

then i don't know what could cause this

i don't have enough knowledge

[vast pulsar]

i'll just try this

[mystic basin]

you could try

[wicked widget]

move the clips from whatever folder to text_encoder folder

reboot ui

retry

[vast pulsar]

it's already in that folder

reddit suggests using euler and simple schedule so i'll try that ig

[mystic basin]

i mean multiple errors

[vast pulsar]

ok i was told to just forget about flux 1 dev and use the many versions with vae baked in as they "work out of the box" we'll see

[mystic basin]

did it work?

i had pip but it disappeared and it doesn't want to get downloaded now

[vast pulsar]

It'll take a while to find out, testing multiple stuff at once

[mystic basin]

alright

[midnight kiln]

Hey - does anyone know exactly what the order of operations is, in txt2img or img2img for example? Does it strictly follow from top to bottom? And if so - is there any way to change it around? Like, I'd want to run Adetailer after a tiled upscale, not before

[mystic basin]

yoooooo, i think i fixed my python error

uninstalling completely and reinstalling

might need to do the compiling again

[vast pulsar]

Works now, had to use that 1 file with everything but now it's ignoring loras so- one problem at a time lol

[mystic basin]

yeah

i have problems, doesn't seem to work

it seemed to work once, it froze my laptop

and every other time i get error

[upbeat igloo]

Hi there, I was wondering if anyone was able to help me get stable diffusion running. I'm using Linux Mint, and I'm trying to get it to work in venv but for some reason its not detecting my 4070ti gpu

Been scratching my head and trying all day to figure it out

[novel owl]

Heho i have a question stable video diffusion is compatible on AMD graphics card ?

[mystic basin]

i keep getting these weird error, can anyone help?

[midnight kiln]

yes, it can work. Slightly more setup and it's not as well optimized for

[novel owl]

Okay because i have compared m'y perf with m'y friend hé have à 4090 and m'y 7900xt have the same perf and a litle bit more it/s

With the same model and setings and prompt

[midnight kiln]

that sounds a bit off - but there are loads of things that can be tweaked that affect performance

I just gained like 50% by running the right runtime args after learning from an expert here

but I have an old card, so it was mor essential for me than most

[fair oxide]

i was makin borbs, and was oki, then suddenly i got a this

;o

[wicked widget]

its downloading

[upbeat igloo]

As a followup to this, i tried seeing if i could see the gpu in venv and it does show up with cuda enabled. Now to figure out why torch can't use this gpu...

[waxen pasture]

its very strange, i have Pony in Vae, like it recommends
but Lora doesnt allow me to use detail tweaker

[fringe ether]

Hello! I am currently trying to get stable diffusion working on my linux pc. I have installed all the dependancies, but whenever I try to launch webui.sh, it opens for about 2 seconds before closing right after

[waxen pasture]

what is vae approx?

[upbeat igloo]

Im going through the same thing. When I run it in venv the terminal stays open but then my other problems begin.

[fringe ether]

mine doesn't even stay open, so I can't see my issue

[upbeat igloo]

Try going through the terminal instead of the graphic interface

[fringe ether]

huh?

what do i launch for that

or how would i do that

[upbeat igloo]

What kind of linux are you using?

[fringe ether]

ubuntu

[upbeat igloo]

Open the terminal and run the file with the command ./webui.sh

[waxen pasture]

RuntimeError: Error(s) in loading state_dict for AutoencoderKL: Missing key(s) in state_dict: "encoder.conv_in

[fringe ether]

bash: ./wubui.sh: No such file or directory

[upbeat igloo]

Find the directory its in and do that

[fringe ether]

/home/sidonplays/stable-diffusion-webui

Thats the directory

ohhhh, okay. One sec

[waxen pasture]

does the refiner use vae?

[lucid yoke]

Yes, either the one your using or one you specified. You can also choose none

[fringe ether]

so... by the looks of it, the issue is that it couldn't install torch

[waxen pasture]

AssertionError: VAE found near the checkpoint doesn't exist: C:\Users\XXX\Downloads\stable-diffusion-webui-master\stable-diffusion-webui-master\models\VAE\Concept_PONY_XL.safetensors

[lucid yoke]

And what is the concept pony xl?

[upbeat igloo]

Yup same thing happened here, your version of python is up to date where as stable diffusion uses a way out of date version.

[lucid yoke]

Is that a vae or something you downloaded from civit?

[fringe ether]

how does one fix that?

[waxen pasture]

from civit

[lucid yoke]

Link it?

[fringe ether]

its a pony checkpoint

[waxen pasture]

i think its pony?

[lucid yoke]

?? Then it doesnt go in vae

[waxen pasture]

it recomends putting it in vae folder

eeh!?

[fringe ether]

Move the concept pony xl to the Stable diffusion folder

thats where checkpoint files go

[waxen pasture]

This checkpoint recommends a VAE, download and place it in the VAE folder.

[lucid yoke]

If its a checkpoint it needs to go in there

[waxen pasture]

all checkpoint files?

[fringe ether]

yeah

just drop the file into it, and it should work

[lucid yoke]

You need to download a seperate vae

Or use the SDXL base vae

[waxen pasture]

kk

[waxen pasture]

i think thats the one i downloaded

because its smaller

is the pruned model better?

its heavier

[lucid yoke]

Just use the base sdxl vae in the vae folder

Checkpoints in stable-diffusion or "checkpoints" folder

And your good to to

[upbeat igloo]

Well... Ive been sitting here for 12 hours fighting and struggling to get it to work with no success here. So far I've been following vague instructions about downloading different versions of python and starting it in a virtual environment. I'm not really sure of the answer myself.

[fringe ether]

If you ever find out, can you hmu by chance?

[waxen pasture]

then what? its in vae, but now i cant use it as a checkpoint

[upbeat igloo]

Sure.

[fringe ether]

merli

can you hop into a call, and I can help you out in a screenshare?

well, wait

what operating system

[waxen pasture]

windows

[fringe ether]

okay yeah, I can help

[waxen pasture]

but wait

where do you use vae?

loras have a section, checkpoints too, where does vae come in place?

brb

[fringe ether]

I have yet to put any vae's in lol

[waxen pasture]

eeh??

so they are useless?

oh

is it for refining or something

[fringe ether]

They aren't useless, but they aren't required either

[waxen pasture]

how foolish of me

i see the issue now

damned vaes

fuuuck, the pruned model is 6GB

[waxen pasture]

i got it to work!

now i heard you made these batches in low res, and tinker the good ones to obtain greater product
how do ya do that?

thank you by the way!

[vocal burrow]

using img2img on the lowres afterwards
or simply regenerating them with same parameters + high res fix + some other optiuonal stuff such as adetailer, etc...

[waxen pasture]

oh yeah? i see, i didnt know that, thanks

[vocal burrow]

vae are used to translate stuff from and into latent space. In common words they transform math stuff from the ai graph soup from prompt and into pixels.

what it means for you ? basically some models require their own custom vae to work correctly, and you also can change vae to slightly tune the output (mostly used for color correction, remove the grey veil, boost the contrast, etc)

[waxen pasture]

grey veil?

i think ive seen that

[vocal burrow]

https://aituts.com/vae/

(notice the purple stuff too)

[waxen pasture]

thanks!

[latent sky]

It seems there's an issue with the stability AI upscale fast API, and when you choose the output format as PNG, it timeout after 12 minutes. Has anyone else experienced this problem? It looks like it's been happening consistently for a week now
{
"errors": [
"Timeout"
],
"id": "xxx",
"name": "internal_status"
}

[next wyvern]

anyone knows how do you run this colab https://github.com/gutris1/segsmaker ?

[novel owl]

hello someone can help me to install stable video difusion on AMD graphics card please ?

[novel owl]

@ornate elk m'y lord XD
how can i install stable video diffusion ?

[lucid yoke]

The best video ai is hunyuan atm

Theres some other promising competitors tho

[ornate elk]

can only help for ltx video

svd is outdated

[novel owl]

Ah okay

[novel owl]

What is Itx ?

[ornate elk]

a video model

like svd

[upbeat igloo]

Would anyone familiar with Linux be willing to help me troubleshoot my issues? If needed I'd be more than willing to negotiate compensation for time.

[novel owl]

Yeah but how use it ?

[ornate elk]

in comfyui

it supports video stuff

[novel owl]

its compatible with amd ?

[ornate elk]

yep its a different webui and i also made a guide for it on github

its not as user friendly as Auto1111 or Forge but supports more things

[lucid yoke]

i never recommend comfy to new users unless they have a very spesific goal in mind and know how to read documentation 🗣️

[wicked widget]

no need to compensante

whats the iossue

first provide me with ram amount, gpu/vram amount

[upbeat igloo]

That is extremely kind of you and I appreciate it.
The ram is 48gb of DDR5
The GPU is a GeForce RTX 4070 ti super
And it has 16gb of ram on it

[wicked widget]

ok ty what seems to be the issue

[upbeat igloo]

Frankly, I'm not really sure what the problem is. I was trying everything yesterday from changing the user.bat to skip the torch cuda test, to installing python 3.10 and 3.11, to deleting and reinstalling, installing pulling as many drivers and cuda as I could, to running it from venv/bin/activate....
When I run it without starting it in venv it says it can't launch venv, when I do launch it in venv it says it can't use my gpu.
I'm totally perplexed lol.

Its saying Torch cant use the GPU at this point and I'm pretty sure that I have it starting python 3.10 so it should have the pytorch libraries right... I hope.

||Also if what I am saying doesn't make sense, forgive me I am a novice and am learning as I go, so to speak.||

[wicked widget]

which method did you use to install sd

[upbeat igloo]

I used $wget and it pulled everything from github i believe

Oh, I am on Mint, if you need to know

[regal mason]

I cleared out my "input" folder to clean it up, now I get this error. How do I restore the "Load Checkpoint" node? Doesn't show up in any node manager

[safe inlet]

python does not come with pytorch, if your using venv, activate it and get command for proper one for your system https://pytorch.org/get-started/locally/

[wicked widget]

with linux i find the easiest way to use SD is with Stability Matrix, try a fresh install with that. Otherwise we'll have to wait for someone with more knowledge to chime in

[vast pulsar]

figured it out: gave up on flux lmao

back to o'l reliable pony i went

[upbeat igloo]

I hate to ask, but could you walk me through that at some point that is convenient for you?

[wicked widget]

sure

[wicked widget]

what was ur issue i didnt see, maybe id have an idea

i use flux regurlarly with comfy

[vast pulsar]

by the end FLUX was basically ignoring loras regardless of using the... (i think it was called 16f LORA) thingy

that and the fact flux took like 3x the time for a render just made me say "Lmao fuck it"

[upbeat igloo]

Thanks, can I DM you?

[wicked widget]

would you be willing to try it on comfyui if i provide you my workflow?

also wahts ur vram

flux is demanding, i need to use an optimization node with my 12gb 3060

[vast pulsar]

8vram, which worked actually quite fast when i got a diferent version of flux

[wicked widget]

ah yea you'd need to use the quantized versions

like Q4

[vast pulsar]

it 8f or smth

i don't remember

it was+

[wicked widget]

flux on forge is a pita anyways

it is supported yes

but it runs way better onc omfy

[vast pulsar]

i did try confy once but it confused outta me

which is why i use webui

i'm too stupid for confy

[wicked widget]

if you ever want to try it again, ehres my pesonal workflow i use for comfyui and flux with lora

its currently set up to run the dev model but i can always help with making it run the quantized versions if needed 🙂

[vast pulsar]

i don't even know what workflow means bro but if i ever feel like smashing my head against a wall again i'll let ya know

[wicked widget]

bwahaha

it means its the setup

but yeah no pressure

^^

[vast pulsar]

yea, i saw some "node" stuff too, since no such thing exists on my "peasant app" aka webui i have no clue what it is

[wicked widget]

i.e youll open that in comfy and everything wiull be ready for you to start using

alternatively, look at SwamUI

SwarmUI

[lucid yoke]

its great

[wicked widget]

its comfy but with A1111 gui

I love it, i gradually swapped from A1111 > swarm > completely comfy

When im lazy i stil use swarm

[vast pulsar]

wait, in that case why would ya swap to comfy? isn't swarm like that but easy then?

unless there's a diference (?)

[lucid yoke]

theres an extra ui. some people dont like it

[vast pulsar]

wym extra?

[lucid yoke]

theres a comfy tab, and theres a generation tab

but if you arent a comfy user yet. i recommend swarm

but if you are a comfy user. you can still switch but custom extensions/nodes need a extra step most tutorials wont mention

[vast pulsar]

~well, if it easy for the "simple minded" such as myself then i could try it

[lucid yoke]

but theres a super cool discord for swarm that knows how to fix those issues etc

[vast pulsar]

ohhh do link me in case i do feel like staying up to 5 am tonight

aka set it up

[lucid yoke]

https://github.com/mcmonkeyprojects/SwarmUI
the link has a discord invite and all the information you need

[vast pulsar]

tought ya said discord

[lucid yoke]

tough if you use a AMD card iirc CS10 has a tutorial for amd

yeah

theres a discord invite in the github

[vast pulsar]

no, i got a mixed system, no clue if that's an issue(?)

nvidia gpu, amd cp

cpu*

[lucid yoke]

third row of text under status. (servers dont like direct invites lol)

oh you got a Nvidia card

its fine same

[vast pulsar]

yea, 3080

[lucid yoke]

just follow the github for an install

[vast pulsar]

alright, i'll try it out maybe this week

[lucid yoke]

bet!

its a community server. take it up with support or your creditcard company to get a refund. we cannot help you in this

it sounds rude but we are just hobbyists

[wicked widget]

do you know what a community server is?
we're just users helping users

[lucid yoke]

also comments might get deleted due vulgar language. hope you get the result you need

[tawny saddle]

was there a fight here or sum-

wuts an x token

ahh its a bitcoin scam xd, yes, lets all invest 100k in xcoin so the owner can cash in and not let other cashout, sounds like fun

xd

[lucid yoke]

Not really, just a upset guy who accidentally spent a lot of money. But i think the mods got him out as were just a community hub

[tawny saddle]

ah oop

[paper dagger]

Pump n dumps

[tawny saddle]

xd

[wicked widget]

keep the channel to tech support yall we have a gen chat

[paper dagger]

My bad

[midnight kiln]

What happened to the Re:actor extension in automatic1111 ? The repository has been removed panic

please tell me there's somewhere else to get it

[tawny saddle]

1 sec

prolly this nu?

https://github.com/Gourieff/sd-webui-reactor

[midnight kiln]

yes, exactly but ...

[tawny saddle]

ahh

oofta

[midnight kiln]

I mean, someone here surely has it installed? You better make a backup of it (and send me whiel you're at it 😛 )

[tawny saddle]

xd

wonder why they disabled it, maybe its stolen code or sum

the sfw one appears to be there tho

[midnight kiln]

But... but my smut!

and faceswaplabs causes my package dependencies to collape - great!

[tawny saddle]

xdd

anyone got an SMB or sum storage server yall dunt mind me pulling someting from

i wan the discordinstaller.exe u-u

i feel like peer2peer would be faster than waiting an hour on the website

oop nvm it decided to go finally xd

[knotty turret]

my stable diffusion not working idk what going on

inpaint not working

not VAE problem the model i using dont need it

forge with zluda

[ruby yoke]

wrong scheduler and sampler

[knotty turret]

everything i try

just give me gray

[ruby yoke]

can i ask how you know the model you're using does not need a VAE?

[knotty turret]

baked vae

that means vae inside model right?

[lucid yoke]

You still need a base vae

[knotty turret]

it was working before and never used vae with that model

[ruby yoke]

it does - but - which checkpoint are you using?

[knotty turret]

btw already test it with VAE

didnt work too

[lucid yoke]

Ah then something else is goin on

Had a similar problem in the past (vae related)

[knotty turret]

utopianpony

[ruby yoke]

are you loading anything at all that isn't pony derived?

[knotty turret]

find that my cmd says "ZLUDA device failed to pass basic"

trying to delete venv and .zula folder

[ruby yoke]

oh - okay - that's a different issue

[knotty turret]

see if works

trying this

[ruby yoke]

read through this https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu/issues/540

it's not your model - you've got to get zluda working

[knotty turret]

yeah i going to sleep now tomorrow try it

[ruby yoke]

and if you still have issues, you're going to want to talk to @ornate elk

[knotty turret]

this guy always save me

ik him already

[ruby yoke]

he's the AMD GPU expert

[knotty turret]

👍 ok good night

[noble idol]

@ornate elk do you have any experience with supermerger

[ornate elk]

Nope

[noble idol]

😔

i think i found my issue anyways, i just dont have a fix for it now

[knotty turret]

CS1o can u help me

already tried it but didnt work

[ornate elk]

Looks normal

What's the issue?

[knotty turret]

and my output are all giving me just GRAY image

and inpaint not working too

[knotty turret]

.

[final vector]

Can anyone with knowledge of Comfy-UI integration with Intel/IPEX Help me Here? #🧣｜comfy-ui message

(Trying to run Flux-Dev)

[ornate elk]

Which model do you use?

[limpid osprey]

Um, any good online models for cyberpunk story setting art?

[knotty turret]

right now trying utopianinpaintg model

its sdxl model

[lucid yoke]

Wrong channel! This is for tech support :-)

[ornate elk]

Don't use an Inpainting model for image generation
Try an other model.

[knotty turret]

mb i said the wrong model i using this

Its SD 1.5

[ornate elk]

Can you show me your webui-user.bat?

[ornate elk]

When did that issue started? Did it worked before normaly? What did you changed

[knotty turret]

started when tried to install hip sdk 6.2.4

like 10 days ago

but just tried stable diffusion yesterday

but now i using 6.1.2

[knotty turret]

i am missing something?

[ornate elk]

Looks good.
Do you have both hip sdks now installed? Or only one? Please check

Also what's your GPU?

[spare knoll]

I was working with this youtube videohttps://www.youtube.com/watch?v=onmqbI5XPH8

and downloadet this model: 768-v-ema.ckpt

do i have to install python and the other things in the same disk?

mybe thats the problem idk

[lucid yoke]

i think the bigger problem is:
its an old guide
your using a old model too

[spare knoll]

Ouh

[lucid yoke]

i recommend following the guide written by CS1o
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[spare knoll]

i dont found a new tutorial

[lucid yoke]

i believe the Forge-Webui is the best one in your case

[spare knoll]

ah thanks

[lucid yoke]

just make sure if you got a AMD card to use the AMD guide with ZLUDA (NOT DirectML)

[spare knoll]

what model do i have to use?$

[spare knoll]

im using amd

[lucid yoke]

what graphics card do you have?

[spare knoll]

rx 6800

[lucid yoke]

it depends on the Vram capacity on what im going to recommend sec

[spare knoll]

12gb

[lucid yoke]

ah then SDXL, do you like realistic or anime pictures?

or would you like to do both?

[spare knoll]

both xD

[lucid yoke]

for Realistic images i prefer to use a SDXL 1.0 model like https://civitai.green/models/139562/realvisxl-v50?modelVersionId=789646 (you need to use these settings in the model page to get the best results!)

[spare knoll]

but is it the same to install like in the video?

just never versions?

[lucid yoke]

for anime images. i recommend checking on the "models page" and then filter for illustrious

[lucid yoke]

not fully since your using AMD so you need ZLUDA or your speed is gonna be agonizingly slow

[spare knoll]

oke ill use zluda

[lucid yoke]

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides#amd-forge-webui-with-zluda

this one will jump to the proper guide

[spare knoll]

do i have to get the python in the same disk where i want to have the model

?

[lucid yoke]

i would recommend so

[spare knoll]

is this right so?

at the bar commandline_args

?

[lucid yoke]

maybe add a space after the =

i use NVIDIA so i didnt have to do that step

but looks good

[spare knoll]

oh oka

ill add a space

I get this error Couldn't launch python

[lucid yoke]

hmm

[spare knoll]

do i have to start it first?

[lucid yoke]

are you trying to launch python seperately?

[spare knoll]

nope

should I?

[lucid yoke]

no?

[spare knoll]

oke

[lucid yoke]

do you get any errors or?

[spare knoll]

i get exit code 9009

maybe i have to restart my computer

[lucid yoke]

Uninstall Python, install it again, don't forget to check the "PATH" box when installing, (found on github for people with similar errors)

[spare knoll]

i will restart my pc first

still dont work

[spare knoll]

i installed it new, ill do it again and screnshot it

[lucid yoke]

looks good to me so far

[spare knoll]

ill download it from microsoft store

[lucid yoke]

????

[spare knoll]

maybe it will work then

[lucid yoke]

i mean

maybe?

[spare knoll]

3.10 right?

[lucid yoke]

before you do. ill tag our expert (who wrote the guide) and if hes availible he might hop in to help

[spare knoll]

oke ill wait

[lucid yoke]

@ornate elk any idea what could be going wrong here. python not launching Code 9009

[spare knoll]

@lucid yoke I downloadet this. what do i have to do with it? i dont understand the guide 😭 : ZLUDA-windows-rocm6-amd64

[lucid yoke]

Setting up Zluda:
Important: If you already installed Automatic1111 or ComfyUI Zluda Version with my Guide, you can jump directly to Step 5. of the Zluda setup.

Important: If you want to use the latest ROCm HIP SDK 6.2.4 and your GPU is higher/newer than a RX6800, then:

1. Download and install AMD HIP SDK 6.2.4 from here
2. Download the latest Zluda files for ROCm 6 from here
3. Unzip as Folder and move it onto the C: Drive and rename the Folder to ZLUDA
4. Then restart your PC and skip to Step 4 below and proceed there with the Guide.

where do you get stuck?

[lucid yoke]

example image:
extract in> (file name)

[knotty turret]

i have only one

6.1.2 gpu 6600

[spare knoll]

yes but where do i have to put it?

after it is extracted`?

[lucid yoke]

the entire folder must be in the C drive

just C:

and rename the folder to ZLUDA

[spare knoll]

but my other things are in D drive,

[lucid yoke]

this must be on C

[spare knoll]

but still my other things like the webui is in d

there is no problem right?

[lucid yoke]

shouldent be.

[spare knoll]

ahh but now it works

i installed it from windows

[ornate elk]

delete the venv folder and relaunch

the windows store python isnt recommended and wont work with every webui

[spare knoll]

waht folder?

[ornate elk]

venv

inside the stable-diffusion-webui folder

[spare knoll]

alr deleated it

now im gona install python again

[knotty turret]

@ornate elk need more information?

[spare knoll]

alr and now?

[ornate elk]

relaunch the webui-user.bat

[spare knoll]

still the same

[ornate elk]

yep can you try delete the venv and the .zluda folder and the files config.json and ui_config.json ?
then relaunch the webui and then try to generate an image with a normal model

[ornate elk]

can you show me the cmd log?

[knotty turret]

ok

[spare knoll]

what do you mean?

[ornate elk]

the black window

with the error

[spare knoll]

its on german

but with the windows python it worked

the one on the store

[ornate elk]

yea but thats not supported

[spare knoll]

hmmm

what can i change then so it works?

[ornate elk]

can you dm me if you have time i can take closer look

[spare knoll]

sure

[lucid yoke]

CS1o being the g.o.a.t as useual

[knotty turret]

the model i show you before can be him?

chilloutmix ni pruned SD1.5

[ornate elk]

yep

[knotty turret]

Alright

[ornate elk]

but if you want to be sure try dreamshaper v8

[knotty turret]

Ok gonna do it

[knotty turret]

same

😫

[ornate elk]

can you screenshot me the forge ui?

when you get that gray output

[wicked widget]

is that a 1.5 or xl variant of dreamshaper? if xl res also looks abit low

@ornate elk

[knotty turret]

the web page?

idk if get it right

for a moment could see the blue fried image

but cant see it anymore

now just

[mystic basin]

ZLUDA device failed to pass basic operation test

could it be why you get grey image?

[mystic basin]

runtime error: invalid argument to reset peak_memory_stats

it could probably be a vram error when the image finish generating that causes the image to become grey (i could be wrong, i don't know much about these kind of stuff)

black square can be caused by certain model when using vae and text encoder (like for flux if the model you are using doesn't have the vae and text encoder integrated), the gray square could be a similar issue even though i'm pretty sure you don't need those for sd 1.5

[knotty turret]

yeah i waiting for CS1o aswer now

[mystic basin]

good idea

[ornate elk]

can you show me the extensions tab?

or the extensions folder
and can you send a full cmd log after a relaunch?

[olive lintel]

Is it easy to upgrade to SwarmUI 0.9.4 from Stable Swarm 0.6.x?
I'm using SD on vast.ai, so I need to use one of the preconfigured docker images to get the services in correct ports etc.

[ornate elk]

idk but you can ask this in #🐝｜swarm-ui too

[lucid yoke]

i mean using stability matrix, other dockers etc are generally not recommended for swarm but it should be as simple as using the update .bat file

[knotty turret]

sure

[ornate elk]

very strange

[ornate elk]

your vae gets loaded with pytorch, i think thats the issue but idk why

i would recommend you install forge clean into a new folder.
and please dont use () in a folder name or any other special characters

[midnight kiln]

Does anyone happen to have ReActor (face swap for automatic1111) installed? It was very suddenly pulled from github. It would be easy enough to just grab the folder and zip it if anyone wants to help me out

[ornate elk]

i have it

[midnight kiln]

begs

😛

[knotty turret]

that one right now its clean install

how do i unistall hip sdk

i think need install everything again

[ornate elk]

via system control

[knotty turret]

should i remove python too?

zluda too

delete all folders

[ornate elk]

nope

[ornate elk]

yes

[knotty turret]

Okk

😫 hope work this time

[upbeat igloo]

Should I uninstall SD and auto 1111 before I install stability matrix?

[wicked widget]

yea

[upbeat igloo]

ty, im gonna give it a shot now, wish me luck! 🙂

[wicked widget]

when you download the package you want to use, hit advanced settings and select cuda (assuming you got nvidia gpu)
otherwise id follow the amd install pins

[upbeat igloo]

oh this is weird, ive never seen this kind of file before

[upbeat igloo]

how do I actually execute this?

[wicked widget]

appimage? run it as an executable

[upbeat igloo]

chmod?

[wicked widget]

try right clicking

see what it shows

on my popos i can just right click and run

[upbeat igloo]

its asking to run it with "disk image writer"by default

[wicked widget]

https://docs.appimage.org/introduction/quickstart.html

you might have to chmod yea

[upbeat igloo]

wait im dumb... im still learning linux i went into the properties and changed it to be executable

[wicked widget]

im about 1 year into linux

still banging my head on it

dw

that being said, still prefer it over windows

[upbeat igloo]

that seemed to get it to open

[wicked widget]

sure do

gj]

[upbeat igloo]

||i hope i will feel the same way, I'm coming off win7 which i loved with all my heart||
okay ill let you know how it goes, and thanks again

[wicked widget]

np

[upbeat igloo]

what do

[wicked widget]

go to settings like it says

https://civitai.com/

create an account here

and get ur api key

[upbeat igloo]

got it

[wicked widget]

alternatively u can skip that and import manually but

this is for the built in model manager

[upbeat igloo]

my noviceness at anything ai is showing

[wicked widget]

its okay its a new a tool

still really young in its infancy, everyday improving

compred to 2 yrs ago when it first started

[upbeat igloo]

im genuinely curious, this civitai, is this an active service i need to be logged into to gen?

[wicked widget]

its just a model hub

dont need it logged in to use

can forgo the connection and import manually

when they ask you to download models at first launch u can just skip that part

[olive lintel]

Some models do require you to be logged in to download them. Likewise anything NSFW is censored unless you're logged in and have enabled it.

[upbeat igloo]

But it doesnt upload pics or parse what i am making does it?

also, thank you @wicked widget , my machine is now making AI images thanks to you. Now the next step is to learn to control it!

[olive lintel]

There's multiple parts to civitai. The ones that are the most popular are image and model galleries for which you don't need a login. Then it also has a (paid with some free credits) generation and training service which obviously means they can see what you generate (whether they care is a different thing)

[wicked widget]

glad to help, i like matrix because i like to use multiple uis such as comfyui, swarm, forge, and with matrix i can Symlink my models across them thus saving space, remind me whats your gpu and vram?

[upbeat igloo]

thanks for the info. I've got a long week of learning and reseach ahead of me!

[lucid yoke]

Better prompting and staying within the model image size ratios will help lots. On civit ai you can check other images on the checkpoint you use to see what other people do with their prompts

[upbeat igloo]

RTX 4070Ti super and 48 gb ddr5

[lucid yoke]

Id love to give more concrete examples but i gotta go

[olive lintel]

But beware that most prompts there are what's called "prompt salad" where people throw everything and the kitchen sink at the prompt and the only directed result they get is down to pure luck. IOW if you copy those large prompts, your results may be completely different.

[upbeat igloo]

yeah i played around with some of the free ai stuff online and discovered less is more

[wicked widget]

oh youll be able to use most models with ease, i recommend using IllustriousXL models, IllustriousXL is the base model that was finetuned off SDXL, the second iteration of SD models, Illustrious_miaomiao is one i personally use.
Youll want to be generating images at 1024 resolution with XL models. any aspect ratio that maintains 1024 should work well.

[upbeat igloo]

Noted!

1024x1024 or rather a 16:9 ratio?

[wicked widget]

1024 is 1:1

[upbeat igloo]

good to know

[wicked widget]

1:1 of 1024

2:3 of 1024 is 832x1216

[lucid yoke]

I like 3:2 and 8:5 ratios (and in reverse)

Best width/tall with good enough cohesion

16:9 gets odd sometimes

[wicked widget]

2:3 3:4 those are nice for me, aspect ratio helper extension will probably be needed if you struggle with ratios

[olive lintel]

I saved this list of resolutions from reddit a while ago

[wicked widget]

yup those are nice

good guideline to follow for XL

[upbeat igloo]

saved now as well

https://tenor.com/view/abeeo-write-that-down-gif-21386042

[wicked widget]

lolol youll be like that for awhile

a sponge soaking it all up

[upbeat igloo]

Well, I am going to call it a wrap since i have alot of how to videos to queue up. Thank you all so much!

[wicked widget]

np

[flat yarrow]

Hi I thought I'll throw this out there just in case someone knows what this is, but I think it's more of a hardware config issue than Comfy. Am doing some small batch runs of SDXL in ComfyUI and am running batches of around 200 image generations. Am using RTX 4060 8gb VRAM (models fit). I can generate around 80 images then my Screens go blank and I can't see anything do I reboot, reload comfy and set to generate 1 image. When running the workflow it gets to the KSampler and screen go off again, this is a brand new GPU card an I've only had it today and it did the same a the one I bought last week, so I know its not a faulty GPU. Temps are fine and the screens only go off when put under load, GPU work fine for normal use and I'm using the GPU to write this message. Am sure it's a config somewhere just don't know where to start looking to fix this issue, can anyone help.

[midnight kiln]

It sounds like a particularly bad driver crash. Go into even viewer and check what happened. That will give you some good clues. Under "windows logs" you will want to check both the system and application logs.

[flat yarrow]

@midnight kiln Faulting application name: python.exe, version: 3.11.9150.1013, time stamp: 0x660bda91
Faulting module name: ucrtbase.dll, version: 10.0.22621.3593, time stamp: 0x10c46e71
Exception code: 0xc0000409
Fault offset: 0x000000000007f6fe
Faulting process id: 0x0x15E8
Faulting application start time: 0x0x1DB728FBD2A6391
Faulting application path: E:\ComfyUI_windows_portable\python_embeded\python.exe

Looks like a python issue,

[midnight kiln]

That really just means it was the stable diffusion process that caused the issue - but it could be indirectly

I won't first suspect a bad python install

any mentions of GPU reset around the same time?

[flat yarrow]

Can't see anything around that, I'll have another look

[midnight kiln]

Bad GPU crashes can sometimes take down the whole system - and it can happen for other reasons than a faulty GPU.

if your system becomes super sluggish when it processes, that can also lead to starvation of essential system processes in the most extreme cases. Depends a lot on the CPU and setup - but setting the SD python process to "below normal" priority can help it from hogging everything. It doesn't really impact performance unless you actively ue the system for other things at the same time

[flat yarrow]

Yeah, I purchased this to put me on until I can get my hands on a 5090, my 2070 super failed and I got the 4060 to put me on until I get the 50 card. But this is the second 4060 I've installed so not a GPU issue

I got the --lowvram set

[midnight kiln]

And another common issue is RAM overload. SD can easily overflow so badly it fills all virtual memory and crashes the system that way. It's not the first time I forgot to set just one setting that suddenly turned the RAM usage baloon out of control, especially when doing very high res upscaling

[flat yarrow]

when I've been rendering batches of 200 I just leave the PC to get on with it, but I'm new to ComfyUI so I'm just figuring things out and this is not helping

[midnight kiln]

I'm sorry that I don't have time to lead you through troubleshooting right now, but hopefully you have some pointers to look at. It can be tricky to narrow down instability causes.

[flat yarrow]

OK thanks, any ways to control the RAM

[midnight kiln]

not AFAIK - other than being concious of what settings you use, and use GPU-Z to have a realtime readout of how much actual physical VRAM is in use

you don't really want to run outside of the VRAM if you can help it - as it is at least an order of magnitude slower

[celest delta]

my dynamic prompt extension suddenly doesn't work anymore... i didn't change anything, an hour ago it worked, and now it doesn't, i'm talking about this one :
https://github.com/adieyal/sd-dynamic-prompts

{something|other} thing doesn't separate anymore, it keeps the whole prompt

[midnight kiln]

First thing is to check for updates.

probably not the problem, but its the first thing to check

[flat yarrow]

@midnight kiln I'll run it again and watch GPU-Z and see what happens I'll put the log on

@midnight kiln I've updated Comfy first thing I did

[midnight kiln]

I've not yet dug into Comfy, so that part I know nothing about. I'm still learning automatic1111

[flat yarrow]

@midnight kiln I like the flexibility of ComfyUI and the control you have compared to A1111

[celest delta]

i already did, wasn't that

[flat yarrow]

@midnight kiln I've got a really long prompt and I've read that has a habit of causing problems, so the VRAM is a good pointer

[midnight kiln]

I've never seen promts cause RAM overflow - on automatic anyway. It's usually more settings in functions. But if you don't already have a very good idea of how much VRAM you are using in your typical gens, that is something that you simply have to now and control

resolution is typically the biggest killer

Use tiled upscaling methods for upscaling past resolution you can handle inside VRAM

[flat yarrow]

@midnight kiln I am upscaling, so I could run the workflow without the upscale and see if it completes the batch, maybe I'll need to break the work into 2 runs, but I'm off to check the VRAM as that's probably the issue It's only 8gb so am probs hammering it

[celest delta]

okay i found the issue... it's so weird... i had like 3 lora in the dynamic prompt, changing randomly, i did this a ton of time

but a specific lora compltely disabled the dynamic prompt for some reason

for the whole thing, even other dynamic prompts

it's a shame cause i like this lora

[flat yarrow]

@midnight kiln Just made it crash and the VRAM was 7750 at MAX on a 8GB card, I really don't know how much of a buffer is needed, but I didn't fill it.

[midnight kiln]

at that amount, it could already be overflowing - it won't necessarily go to 100% usage becaue of how data is loaded

but just use task manger to watch your normal RAM

and it will be very obvious...

if it overflows.

[flat yarrow]

@midnight kiln I'll do that now and see what happens

@midnight kiln OK this is NEW: VRAM 7125 - Sys Ram @28% ----- Monitors flashed off and back on again with the system crashed (had to reboot and in event viewer same error around Python and Comfy. Think I'm going to run Heaven Benchmark on my other installation of windows and see if the card crashes when that is run, just to isolate the issue around Comfy. BTW this install of windows in almost 14 days old so corruption maybe unlikely put possible,

[midnight kiln]

isolating the problem as much as possible is the key, yes

you can always use DISM and SFC to check windows just to be sure

are you running a CPU or mem overclock?

because what you describe sounds like a general instability crash

[flat yarrow]

@midnight kiln I think the overflow is unlikely as it was @28% on sys ram, Done that DISM and SFC

@midnight kiln No over clocking stock settings at moment

[midnight kiln]

yes, 28% obviously no issue. you have to fill the RAM + fill all virtual memory too. It takes some doing

[flat yarrow]

wouldn't hurt doing the DISM and SFC again though, when I but this GPU in I flushed out the old Nvidia drivers and installed everything again so GC side am sure OK

I'm thinking it's something around the Comfy environment whether it's user error or a config / quirk

I use Topaz Gigapixel and that crashed the system on the previous GPU so am seeing if that can do the same thing on this one and crash the system. Doing that now...

@midnight kiln It turned the screens off but I was able to shut the PC down when I ran Topaz Gigapixel to render an image, so it's something around rendering / processing the image. The issue isn't solely ComfyUI and a config. To me it appears to be a wider issue with the system config at a guess, but everything is at defaults settings.

[gusty stirrup]

Heyy, this is probably a dumb question but am I able to get the WebUI (For Forge) on my IPad's browser?

It would still obv be running off my PC.

I tried doing the usualy local hosting with my computer's IP and the port but i get stuck in infinite loading.

Is there something like an extension i have to download or am I just being dumb?

[midnight kiln]

Does anyone know if you can continue expanding an answer-section with more content? I know you can just enter a blank prompt - but that also jumps to the next section so you can't swipe. I like to filter through a few answers and keep the best stuff

[knotty turret]

@ornate elk Worked now🔥🔥 🗣️🗣️

[ornate elk]

Perfect!

[ornate elk]

You have to add --listen to the webui-user.bat

Then you can access it trough the PCs IP+Port

[gusty stirrup]

How would i do that? I launched it through the cmd in the directory with that but it still didnt work

[slim fern]

does anyone know why i get complete random picture generation using flux based model?

something like this....

[tawny saddle]

i kinda like that one tbh

[lucid yoke]

Cfg and how many steps?

[slim fern]

cfg1 and 20 steps

[lucid yoke]

Decrease it

[slim fern]

i guess its to do with the sampler....

[tawny saddle]

looks like the cover art of an scifi, mystery movie from like 2001 or sum

[lucid yoke]

And sampler probably

[slim fern]

Sampling: Euler + simple/beta (for general)，DPM2M + SGM uniform (for skin texture)，DEIS + DDIM uniform (for casual realistic look)
That is recommended one. But how to apply two sampler...

havent use the webui for ages..

[ornate elk]

You right click and edit the webui-user.bat
At the line commandline_args=
You add --listen

[ornate elk]

What's your webui version?

[slim fern]

ok its fixed...need set the Schedule type correct...

but the generated image is still a bit blurred somehow

[ornate elk]

For local image generation or accessing the PCs webui instance?

[restive apex]

local

like on the phone

[ornate elk]

Thats currently only possible on iphones. There is no such app for android

[next wyvern]

anyone knows how to use current google colab?

[wicked widget]

weylus

for windows

alternatively can do what my bestie did for me and set up a tunnel

we using cloudflare. can access the ui on my phone off data

[stone coral]

Hello, I used to generate stuff as a hobby before and I wanna do it again with the new models, can anyone tell me what is the best web ui to use right now?

I used to use auto1111

[wicked widget]

if you want to keep using a1111 since its a familiar tool you can look at installing a1111-forge or swarmUI. sdxl models like Pony and Illustrious are popular and recommended gen at 1024res

[uncut jewel]

Hey!
Does anyone know why this kind of VisualStudio issue happen on a simple FluxTextEncode ?!

[restive apex]

guys can anyone help me with making fastsd cpu work? heres the repo btw https://nolowiz.com/how-to-install-and-run-fastsd-cpu-on-android-temux-step-by-step-guide/ it keeps on giving the same error, i executed the scripts all in order too. also the tut i used: https://github.com/rupeshs/fastsdcpu

why is the seconf lik not working

sorry for the mistakes im just a bit tired

[gusty stirrup]

Tysm!! Its working now

[fair oxide]

How do I get multiple distinct characters in a single image?

[lucid yoke]

Luck mostly, ai tends to mix their features. Iirc regional prompting can help with it but its not perfect

[olive lintel]

Stupid question: How do I download additional themes to SwarmUI? (Specifically, default gradio theme - The normal SwarmUI layout is completely unusable to me)

[knotty turret]

i think its better if u ask this there

[olive lintel]

Alas, I found the answer and it's basically "You don't", at least in the sense of themes I was thinking of

[tawny saddle]

automatic1111 u can use jus css styles that websites use xd

dunno if it works the same for swarm

[olive lintel]

only to the extent that you can change colors

[knotty turret]

@ornate elk As said before my forge zluda is working fine now can u try to help me with comfy ui zluda too

On this step 6 webui didnt open on my browser

[vale veldt]

hi, im on Forge. I reinstalled and Im missing the Lora tab now, no longer next to generation and textual inversion. How can I get it back?

[lucid yoke]

Theres some colour themes what?

Or do you mean a complete ui overhaul theme?

[olive lintel]

Moving sub-frames to different places and changing their size

[lucid yoke]

Newest update just did that (mostly) lolhttps://github.com/mcmonkeyprojects/SwarmUI/discussions/1#discussioncomment-12004159

Only different places

[olive lintel]

Unfortunately it didn't. The sub-frames are still fixed to their old places (the real problem) and you can just changing which sub-frame things are (except the prompt and generated image). My main issue is that I really really want to change the prompt size and get the generated image the hell away from taking space in the center of the screen.

[lucid yoke]

But you can resize/move certain bars already iirc

[lucid yoke]

Did you try the gear icon in the batch preview? And select auto focus new image?

Prompt box is something youd have to ask mcmonkey for in a feature request as it looks like it could be a real QOL

If you dont autofocus new image mine useually are very very small

You can just unfocus it

[olive lintel]

Won't help for the real issues which is making space for the actually important things (to me) and making those central. Basically making it look more like A1111 in some ways.

I did however get controlnet union inpainting working in A1111 / Reforge so I have no pressing need for swarmui (particularly as swarmui requires the exact same kludge workaround for that inpainting)

[lucid yoke]

Well aslong your happy with reforge i suppose its good

[olive lintel]

It's the least bad solution for now

[ornate elk]

Okay can you still try do the other steps after step 6.
Like replacing the zluda files

[restive apex]

help

ok

[tawny saddle]

chu welcome

[ornate elk]

Dont answer the scammer that messaged you. He is a fake support
Don't give away any personal data or payment.

[tawny saddle]

why there so many support scammers here

did a sussy bot get like installed here or sum

[ornate elk]

Yes they are mostly bots that automatically scan the chat for keys, login data and words like "help"
Then they auto reply

[restive apex]

help

help

help

help

lol

[restive apex]

im gonna give fake info

like my name being rily guhlible and my ssn being 80082

[ornate elk]

Okay xD
Also do you still need help?

[restive apex]

yeah idk how to run the thing

however...

i found an error

I'll show it after i do a chore my grandma gave me

[ornate elk]

Sure, show the cmd log and maybe tell what GPU you have

[restive apex]

UNISOC Tiger T606 processor

infinix shit 8

cd fastsdcpu
chmod +x install.sh

command returned "No such file or directory"

./install.sh --disable-gui

command results into error
ERROR: Could not find a version that satisfies the requirement mediapipe==0.10.9 (from versions: 0.10.13, 0.10.14, 0.10.15, 0.10.18)
ERROR: No matching distribution found for mediapipe==0.10.9

[ornate elk]

Oh okay, can't help with Linux on mobile sry

[restive apex]

okok no worries!!!!!!

[ornate elk]

But make sure you have the correct python version installed

[restive apex]

whats the cmd

[ornate elk]

Cmd is the terminal of windows

[restive apex]

i know but whats tge command used to update python

[slow phoenix]

hello everyone, am struggling with comfyUI Reposer workflow, it seems like it doesn't recognize the path to the IPAdapter Model and Charger Clip Vision also turns red. Is anyone able to help me ? thanks

[flat yarrow]

@midnight kiln An update for you as you were so kind to help with some pointers. I passed the issue to the Penguin to have a look: It turns out to be that common issue of 'Ran Out OF VRAM' - I was able to repeat the issue in Linux and because Linux just handles everything better, my screens didn't go off and I could see what the issue was from the CLI output, which I couldn't see in Windows. So I have my answer, it didn't help me out doing up-scaling in Latent space through the KSampler and doing batches with 8gb of vram. So I'm going to switch the upscale to Pixel space and see how I fair there, maybe enough drop in VRAM use that I'm able to perform the upscale, plus I'll drop my batch size down. I found a node that has 'Clean VRAM' when I did a search in Comfy, I popped that in the workflow as it couldn't hurt. But to be honest I don't know exactly what it does and will it help me with my issue. Any pointers would be great, plus any doc's (I'll be lucky). Anyway Thanks so much for your help and your pointer's regarding your thought...Cheers! and Thank You!

[ornate elk]

Batch Upscaling with 8gb of vram?
Then you also should use the tiled VAE node at the end

That can prevent an out of memory or crashes

[tawny saddle]

ovO how u have terminal on da phone

[ornate elk]

Termux

[tawny saddle]

oh .o.

[reef swallow]

hello, does anyone know if a board that does not have enough RAM can affect the performance of the final generated picture?? that is, it should not be as the promts are?

[ornate elk]

To less ram doesn't affect the quality but can effect the speed an image is generated

When you only have 8 or 16gb RAM you should increase the Windows Pagefile

[reef swallow]

I have a 3GB video card and 48GB RAM

[ornate elk]

3gb vram is pretty bad for image generation

As the vram amount is the most important part

What's your GPU?

[reef swallow]

so I have a 3gb gtx 1060, a xeon e5 2680v4 and 48 gb ram, so I want to upgrade because I always get an insufficient vram error and I thought it affects the generated image, not just the time

[ornate elk]

Yep then you would need a better GPU with more vram

Out of vram error interrupts the generation

[reef swallow]

ok thanks for the help, when I generate simple images it is ok but when I use Regional Prompter or other extensions some images are generated strangely

and I didn't know if it was because of the gpu or stable

[flat yarrow]

@ornate elk Thanks for that input, I was looking at that tiled VAE node but couldn't hook it up to the decoder for some reason and sure am missing a link node. I was using the TiledKSamplerProvider from the impact pack, I piped the input but couldn't output it.

[ornate elk]

Do you use forge webui or Auto1111?

[reef swallow]

forge

is good ?

[ornate elk]

Yep

Its better for GPUs with low vram

[reef swallow]

ok ,thx👍

[flat yarrow]

You mentioned 'hello, does anyone know if a board that does not have enough RAM can affect the performance of the final generated picture?? that is, it should not be as the promts are?' I've had this while testing the GPU in another PCIe slot an x8 card in an x4 slot (I was testing if the GPU was faulty) that produced degraded images due to the bandwith of the lanes. This may not be your issue but it's worth knowing your x Factor on your MB

[reef swallow]

I checked but that's not the problem, I think I simply have a weak gpu

[ornate elk]

Try forge couple extension if you want 2 different characters in one image.

Should be simpler than regional prompter

And if you use 1.5 based models (2gb) it might work

And enable tiled VAE at the never oom setting at the bottom

[flat yarrow]

Well I least you know, I had a RTX2070s and that failed on me so I got the RTX 4060 to put me on for a few months while I'm learning ComfyUI, I intend to get the RTX 5090 when all the craziness dies down. I've run into the same issues as you with low vram and its fun learning the work arounds. You could use an online service like Think Diffusion or Colab to to your Image output when you need more juice. I'm playing with different services at the moment just to see whats out there and if it's any good without the hefty price tag.

[flat yarrow]

Whilst am here, Does anyone use RunPod? They have API calls so your not charged for idle time with remote access, (as you'll know) you can use ComfyUI locally and via the API outsource the rendering online for processing. Does any one know how to set this up?

[dusty herald]

Upscaled picture (400%) downloads in 1024 x 1024 resolution 🤔

[knotty turret]

Ok gonna do it

[knotty turret]

Same cmd log after ZLUDA steps

[midnight kiln]

great to hear you found the root of the problem. Getting a good feel for where your limits are in terms of what you can do inside VRAM is a great help - but there are also methods like tiled upscaling that you can use to make a 4K upscale that would otherwise just not be possible. If batches work like in automatic1111 you can have them go in sequence, nut also paralell (for those cases where you do low-res stuff so 2 can fit in VRAM at once (speed benefits). But for everything else you should make sure to run them sequentially. Lastly you can set a very large virtual memory size minimum in windows as a last line of defense to not crash if you mess something up badly in the settings 😛

[granite sonnet]

I'm trying to install the newest stable suffusion that'll give me access to Illustrouis and if possible flux. I have no idea where to go. Update I tried forge but this popped up

[ornate elk]

Dont install it in documents folder

Make a new folder on C

Name it ai and there put forge in

[ornate elk]

Which hip SDK version do you use?

[granite sonnet]

Tried it and it still have that same message

[ornate elk]

What's your GPU?

[granite sonnet]

XFX Speedster SWFT 309 Radeon RX 6700 XT 12 GB Video Card

[ornate elk]

Oh then you downloaded the wrong webui version

You need one for AMD

You can find the setup guide for Forge AMD Zluda on the first link of the pinned messages

[clever glacier]

I get this message on Re-Forge after upgrading to a 5080 card

[ornate elk]

Yep because torch first needs to support it

Should be a matter of days

[ornate elk]

Oh the update for Blackwell got implemented one hour ago.
So git pull and then delete the venv folder.

Then relaunch

[clever glacier]

Trying that now

[granite sonnet]

Is there video I follow

[clever glacier]

That worked!

[ornate elk]

There is no video. Just my guides:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Webui-Installation-Guides

[granite sonnet]

Gotcha! Thank you!

[knotty turret]

6.1.2

[ornate elk]

How do your environment path entries look?

And do you have the correct zluda 3.8.4 rocm 6 files for 6.1

[granite sonnet]

What does this mean?

Open the .zip file and drag and drop all files of the library folder into the library folder. Not into the old_library folder

[ornate elk]

You have to download the library files for your gfx version

Then you can replace the files as described there

[granite sonnet]

something this? I'm new to this pc life

[ornate elk]

One folder has to be named library

And one of them old_library

Both need to contain the same files.
But in the library folder you drop the files from the zip you downloaded

[granite sonnet]

So I need to download the zip gfx 1031? Since I have an RX 6700 XT?

And put that zip into the library?

[ornate elk]

Yep

[ornate elk]

The files from the zip

Not the zip itself

[knotty turret]

and my system path too

[ornate elk]

Looks correct

Open up a cmd and type
Python --version
And then
Where python

[ornate elk]

@granite sonnet did you got it working?

[granite sonnet]

No lol I took a break to eat something

[ornate elk]

Oh okay xD

[granite sonnet]

Could we do a voice chat and I show my screen?

[ornate elk]

okay then open up a cmd and run
pip cache purge
then delete the venv folder and rerun the comfyui-start bat again

[knotty turret]

ok

[knotty turret]

need to be admin cmd?

[ornate elk]

nope

[knotty turret]

ok

[reef swallow]

hello, does anyone know if the NVIDIA Tesla K80 is good for stable diffusion?

[knotty turret]

i did the pip cache purge but seeing alot "using cached"

its normal?

[ornate elk]

did pip cache purge showed that it deletet some files?

[knotty turret]

yeah 13 files

[ornate elk]

hmm okay

[knotty turret]

cmd closed so fast that i need rerun again

and same

comfyui is not detecting my zluda?

[ornate elk]

when di you git cloned comfyui?

and did you cloned it from the guide?

[knotty turret]

after done with forge zluda

[knotty turret]

Yep

[ornate elk]

also please dont use () or spaces in folder names or any other special characters

[knotty turret]

oh

cant do it?

[ornate elk]

not good

[knotty turret]

ok removing it

[ornate elk]

u can use - or _

but nothing more

[knotty turret]

safe?

ok trying again

who knows

[ornate elk]

also git clone it again

to be sure

[knotty turret]

oh delete everything?

Ok

Using that one " git clone https://github.com/LeagueRaINi/ComfyUI.git && cd ComfyUI && git checkout nodes-cudnn-patch "

[granite sonnet]

[ornate elk]

wrong webui

thats your old one

not the one for AMD

better delete that

[granite sonnet]

I'm so confused. i unintalled the other ones before doing all this

[ornate elk]

first step of the guide is to make a folder on C called SD-Zluda

check if its there

and delete the old one from documents

if you need more help i also can help you quick in DM when you share the screen

[knotty turret]

https://tenor.com/view/rage-gif-24341837

[ornate elk]

strange

[knotty turret]

[ornate elk]

and forge and auto1111 work for you?

[analog hill]

is my radeon 7700xt not enough?? RuntimeError: Could not allocate tensor with 144179200 bytes. There is not enough GPU video memory available!

[analog hill]

im checking my gpu usage and its only using like 12%\

i dont understand

[knotty turret]

Yes

sorry for late answer

[tawny saddle]

wemember to wike and subscwibe

[lucid yoke]

Did you install it with zluda?

[sinful iron]

Hello tech support, I am at the end of my rope, so I was hoping to reach out here to see if what I'm doing is just a noob mistake I'm not seeing. I've downloaded and installed comfy ui, made a working workflow that I think has my lora in it properly, and gotten it to generate some images, but I don't think the lora is actually doing anything. When I check the console, I get hundreds of lines of "lora key not loaded: model.diffusion_model.output_blocks.8.0.skip_connection.weight" and similar. I'm using Illustrious-XL and a lora thats listed as using Illustrious. Am I just not putting my lora in the right place or something? (comfyUI/models/loras). I've been at it for hours with no luck, so any advice would be greatly appreciated.

[ornate elk]

What's in your webui-user.bat?

[ornate elk]

Hey, have you tried to use an other illust model with the lora?

And is your comfyui updated?

[sinful iron]

Comfy UI is updated (Its a fresh install, and I ran the update.bat, even the update with python dependencies one) And I had tried using SD 3.5 turbo with it, got the same errors.

[ornate elk]

SD 3.5 won't work with an illustrous lora

I mean have you tried an other illustrious based model like this:
https://civitai.com/models/376130?modelVersionId=1225516

[sinful iron]

You know, as you're saying it out loud, I remember realizing it was an illust lora, and I was dumb for even trying that. So no, I haven't used any other illust models with it.

[sinful iron]

Do you think it matters between 1.0 vs 2.0? I suppose it wouldn't hurt to try both

[ornate elk]

Try the one from my link it has a v3 and v4

The original illustrous behaves a bit different so wouldn't use it

[sinful iron]

I didn't even realize IL ment illust until you mentioned it T-T I am further out of my depth than I even realized I suppose XD

[ornate elk]

Also did you trained the lora by yourself or is it from Civitai.com?

[sinful iron]

Civitai

[ornate elk]

Okay can you share the link. Then I'm gonna try it later

[sinful iron]

I uhh think it can be used for nefarious purposes, but as long as its okay for me to link to, I will....

https://civitai.com/models/97479?modelVersionId=1323509

[ornate elk]

Thats not a lora

Type is checkpoint merge (model)

[sinful iron]

Please... please tell me its not that bad... I'm so lost I'm going to need a passport to get home

[ornate elk]

A lora is mostly always smaller than 1gb

So move the file to the checkpoints folder and use it as model

And then get a real lora if needed ^^

[sinful iron]

I'm honestly embarrassed for being this far off >.< So a checkpoint merge goes into like the load checkpoint like a version of stable diffusion would?

[ornate elk]

Exactly

Checkpoint merge is just a merge between multiple checkpoints.
But still a checkpoint

[sinful iron]

I mean, as you're telling me this it seems so obvious. Man, I made this waaaaay harder than it needed to be

[ornate elk]

Always check the info box on the models site

[sinful iron]

Thank you so much for taking the time, as awful as it sounds, I was working on that from 8pm until I posted.

[ornate elk]

Type and base is important

[ornate elk]

Oh no problem ^^ and your not the first one.
We had a couple of users who tried to use a Lora as checkpoint. Which doesn't work either

[sinful iron]

So just to make sure I've learned my lesson here, I'm always checking 'Type' to see if its a Lora or a checkpoint. If its a Lora, I then need to check the 'base' to make sure that the base model is whats in my checkpoint load?

[ornate elk]

Yep the base has to be the same so that it works.
But it doesn't mean you need the first sdxl or original illustrous model.
Every model that has illustrous as base will work

[sinful iron]

So as long as I'm using a model that based off the original base, everything should be compatible

[ornate elk]

Same base means it should work

[sinful iron]

I'd love to say something along the lines of 'Hours of youtube videos, why didn't anyone say that!?' but if I didn't know a Lora from a checkpoint, I don't think a tutorial was gonna save me XD

[ornate elk]

True xD
Also the file sizes are a good indicator.
1.5 based models are around 2-4gb
Sdxl/pony/illust are mostly 6gb
Flux and SD3.5 are mostly 8-23gb but have also smaller versions. Mostly in gguf format.

Best is to make subfolders in your checkpoints and lora folders and name them like the base.
So you never get confused on which works with which

[tawny saddle]

cs1 is so smort about stable

imagine all this time CS is actually the lead dev of stable

and dont want the pulblicity

xd

[covert veldt]

The search-and-recolor and search-and-replace stability.ai api end points appear to be busted. For every image I try to edit, no matter how innocuous or safe I request I keep getting 'Your request was flagged by our content moderation system, as a result your request was denied and you were not charged.' . This has been happening now for the last couple of days when either using my own python calls or the colab notebooks. Originally they worked fine but it's happening consistently now.

[tawny saddle]

whaa

i didnt know there was stuff on stable that kept stuff sfw, interesting

[covert veldt]

Yeah it's their online platform stuff at https://platform.stability.ai/

[tawny saddle]

ahhh

u should try jus installing it on ur own pc

freedom is better than paying uvu

[covert veldt]

Yeah - just some of the stuff looks like I can't do right now on my pc - mainly search and replace and search and recolor

[tawny saddle]

are those maybe downloadable extensions

[covert veldt]

Anything else I have the medium models that I mess with in Comfyui

can't see them

[tawny saddle]

i see :o.

aw

proprietary

[covert veldt]

I'm waiting for the 3.5 medium controlnet stuff - my machine can't cope with the large models

[tawny saddle]

i feel u xd, i jus need an actual card, and more ram, large models for me trigger a bugcheck

with stop code video_memory_management_internal