🤝｜tech-support

[noble belfry]

reading through the git, the dev recommends just upscaling after

[heady grove]

What mean upscaling after

[noble belfry]

it doesnt look like they have any intention of adding it

[heady grove]

I make my picture form text2img and send it to img2img

[noble belfry]

if this is helpful at all. i do agree with him, but also i like things to just do all the steps together

[heady grove]

[maiden coral]

you can do that in Comfy, but yeah, it's a mess

[heady grove]

I'm trying to use control net + ultimate sd upscale

But

How can I fix this

[maiden coral]

I don't have forge. Any errors there? What settings are you using?

post an image with your settings

[heady grove]

I opening control net and ultimate sd upscale

[heady grove]

Probably the model issues I think

[signal sigil]

@burnt lichen Hey man, since you seem to have a bit of experience using Ella, i thought i might ask you about it. I have successfully install the Ella extension on ComfyUi, but the T5 LLM seems to be ignoring my prompts. It also messes up anatomy by a lot. Is this normal? If it's because it's a new tech and need refinement that's fine, i just don't want to be missing out if im doing something wrong. Thx in advance.

[waxen pebble]

Hey all... i am looking to try to upgrade my gpu... but dont have the best cpu, and dont really wanna do a full overhaul

Will a 3700x bottleneck a 4090?

(only care about kohya training and a1111, and then other AI applications down the line)

[zenith sail]

I'm building a rig for generating / training going with a GTX 4090, looking at the 14900k cpu, could anyone give some advice on what cpu would be best?

[vocal burrow]

cpu don't matter that much when doing SD stuff. The gpu is doing all the work.

Assuming same hardware and a non bottlenecked ssd/m.2 buying a better gpu is mostly just gonna improve the model loading speed.

[vocal burrow]

using sdxl model with sd15 controlnet model

that's a no go

[solemn atlas]

what does that mean

[vocal burrow]

You're running out of vram/hitting the limit of your GPU. How to fix it (Nvidia edition) :
1/ If you're using a 2000/3000/4000 RTX series cards then you can try to use the --opt-sdp-attention command line argument or --xformers argument. To do so, Edit your webui-user.bat and append it to the set COMMANDLINE_ARGS= line. Performances between --opt-sdp-attention and --xformers can varry depending of your GPU and usecase. Please note, if you're using vladmandic's fork then you won't be able to use--xformers.
2/ Also throw --medvram in there if you've got a GPU with less than 8gb of vram, --lowvram if less than 4gb. (each argument must be seperated by a space). If you intend to use SDXL with a gpu that has less than 12gb add --medvram-sdxl or --lowvram-sdxl if you have less than 8gb. To alleviate Vram usage you can also try to turn on fp8 mode in the settings.
3/ If you're using --precision full, --no-half and don't know what they're doing. Then you probably don't need them and therefore they're increasing your vram usage for nothing. Remove them. Notoriously --no-half is necessary for GTX1600 cards.
4/ Make Batch size lower, use Batch count instead if you want to make multiple generations.
5/ Output at a lower resolution (and eventually upscale later on with hires.fix or img2img feature)
6/ Write less complex prompt
7/ Use lighter model (use fp16 pruned variant of your model if available)
8/ If you're using a VAE and your generation crash at the very end. You should try to install multidiffusion-upscaler-for-automatic1111 extension (available in the default extension list) and enable the TiledVAE features (cf screenshot)
https://media.discordapp.net/attachments/1002602742667280404/1111097170150563881/image.png?width=810&height=228

PS : keep in mind that SD1 models are trained with 512x512 images, SD2 models with 768x768 and SDXL 1024x1024. So stick close to those resolutions before doing any upscaling,

[maiden coral]

--no-half forces all models into f32 mode resulting in more resources usage

that's what I got to know today

so instead --no-half-vae

but I wouldn't use it.

Generate images in 512x512 or whatever is your limit. (I'd still choose lower because then you can upscale them easily and bigger images will result in more errors during generation)

[maiden coral]

What are your specs? GPU, VRAM, RAM

[vocal burrow]

yes but it's necessary for some cards like I mentioned.

[maiden coral]

since it converts all models to f32 => more VRAM consumption

[vocal burrow]

yes ?

[vocal burrow]

--xformers should be enough then

[solemn atlas]

I had --xformers but it said it didnt work

[maiden coral]

you must install it

pip install xformers

[solemn atlas]

from where

[maiden coral]

write it into CMD, also use Python 3.10.x

[vocal burrow]

no. like stated it's --xformers or --opt-sdp-attention not both at the same time.
Also until you have errors in vae there is no need for --no-half-vae.

[vocal burrow]

@solemn atlas please no.
1/ xformers will install itself by adding --xformers
2/ by doing what anon says it's gonna install xformers on your system and not in your SD's venv.

[solemn atlas]

oh okay, I tried to do it but it says this if i only use --opt-sdp-attention

[maiden coral]

I think you have to install it or it installed itself. I never used these 2 argumements together, then I switched to ComfyUI.

[solemn atlas]

thats when i use only --xformers

maybe thats a problem?

[charred verge]

i throw rock at computer 😱 🖕 computer stop work 😔 someone help 🎡

[maiden coral]

but if you have VRAM problems or ever get to them use --xformers

[vocal burrow]

delete your stable-diffusion-webui\venv folder and try again

[maiden coral]

it reduces VRAM usage for (only) Nvidia cards

[charred verge]

hellp

[solemn atlas]

okay I'll try

[vocal burrow]

Otherwise it's probably reactor being reactor again and causing issues

[charred verge]

https://www.thoughtco.com/rhyolite-rock-facts-geology-uses-4589452

[solemn atlas]

I'll see how it works if starting it again

[hardy canopy]

Is the Extra Networks tab broken in Forge? I can't get my models to show

[solemn atlas]

it doesnt work

[maiden coral]

What Python version are you using? Have you tried reinstall of Forge?

[solemn atlas]

where can i look this up

the version

[maiden coral]

cmd -> type "python"

[solemn atlas]

python 3.10.6

[maiden coral]

1. If you modified some files: try to clone the repository again and just move the models into the new folder
   or
2. uninstall all requirements if you installed incorrect versions

[vocal burrow]

what's the full log there ?

[vocal burrow]

add pydantic==1.10.11 in requirements_versions.txt

delete venv again,

should be fine

[solemn atlas]

should i put it here?

[vocal burrow]

no, in requirements_versions.txt

[solemn atlas]

just adding it right

[vocal burrow]

yes

[solemn atlas]

okay, I'll try again, thank you

it doesnt work sadly

[tiny abyss]

A tensor with all 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.
Time taken: 1.9 sec.

also this
A: 1.78 GB, R: 2.05 GB, Sys: 2.9/4 GB (71.3%)
What to do? i tried the method it says, but no luck

[ornate elk]

nope it should work
whats in your forge webui-user.bat?

[hardy canopy]

Humm. Nothing yet. I just installed forge a couple hours ago. Haven't customized it much yet. I did bring over a couple things from a1111. It must have boogered something.

[ornate elk]

okay, strange, did you tried my solution for a1111 ?

[ornate elk]

hey, whats your gpu ans whats inside your webui-user.bat?

[vocal burrow]

why are you still using 1.8.0 ?

latest version of Auto1111 is 1.9.3

[karmic crown]

it is posible that you did not properly set paths for a1111, so forge is looking at wrong place

chekout this one (not mine, but it works on his machine)

export COMMANDLINE_ARGS="--attention-pytorch --all-in-fp16 --use-cpu interrogate --no-download-sd-model --theme dark --forge-ref-a1111-home /Users/miguel/stable-diffusion-webui --ckpt-dir /Users/miguel/stable-diffusion-webui/models/Stable-diffusion --vae-dir /Users/miguel/stable-diffusion-webui/models/VAE --hypernetwork-dir /Users/miguel/stable-diffusion-webui/models/hypernetworks --embeddings-dir /Users/miguel/stable-diffusion-webui/embeddings --lora-dir /Users/miguel/stable-diffusion-webui/models/lora --controlnet-dir /Users/miguel/stable-diffusion-webui/extensions/sd-webui-controlnet/models"

you might figure out whats wrong with yours

[viral knot]

the better question is why a1111 at all :3

[karmic crown]

becuse it is much easier to understand it (for beginners)

[vocal burrow]

forge's last commit

[viral knot]

sigh 😦

[karmic crown]

when you learn to use a1111, its not a problem to understand comfy and what you did wrong

but if you never used sd before and node based interface, you will be f....

thats why i always recommend start with a1111, then use what ever you want

[viral knot]

i mean i also started with a1111, so that is fair indeed

[solemn atlas]

API  •  Github  •  Gradio  •  Startup profile  •  Reload UI
version: v1.8.0  •  python: 3.10.6  •  torch: 2.1.2+cu121  •  xformers: 0.0.23.post1  •  gradio: 3.41.2  •  checkpoint: ef49fbb25f

I'm using this

yes

[vocal burrow]

yes that's what I'm saying

1.8.0 is old

extensions you have installed are probably not compatible with it

so try updating your a1111.
1/ Go to your stable-diffusion-webui folder
2/ in the top bar of explorer (where the path is) type cmd and press enter
3/ and in the cmd window that just opened, type git pull

[karmic crown]

ImportError: cannot import name 'Undefined' from 'pydantic fields' (C:\Users\moonshine\stable-diffusion-webui\venv\libls ite-packages\pydantic\fields.py)

i am almost sure that was added in 1.9.0, but i am to tired to check the log now

try

venv\scripts\pip install albumentations==1.4.3

venv\scripts\pip install pydantic==1.10.15

from cmd in the same folder where your a1111 is

[vocal burrow]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/15564

[karmic crown]

i need to test my script tomorrow, to see if it is affected

https://github.com/pytorch/pytorch/releases/tag/v2.3.0

2.3.0 is released, i can upgrade my script now
and to leave intel at 2.1.2 while i can switch back arm to normal install

https://github.com/pytorch/pytorch/pull/116726

bye bye intel macs

[solemn atlas]

it says this

[karmic crown]

git restore requirements_versions.txt
to restore the original version
than pull again

[vocal burrow]

ah yes, run git checkout --theirs requirements_versions.txt

or git restore, sure

[karmic crown]

git restore requirements_versions.txt
is easier 🙂

[vocal burrow]

potato patato :p

[karmic crown]

less characters, easier to remember 🙂

@vocal burrow expect increase of intel mac users soon 🙃

"buy a new mac" 🙂

[vocal burrow]

because of pytorch ?

[karmic crown]

yes

no builds for x86 anymore

[vocal burrow]

yeah I saw that we discussed it already :p

[solemn atlas]

so i run this again on cmd

I tried but it doesnt do anything

it says updated 0 paths from index

[karmic crown]

my script already detects the platform x86 vs arm, since brew has different path
so i will just add a condition that if platform is x86, 2.1.2 should be used, otherwise the latest stable
actually i will make a PR to a1111 (that will be wiser i guess)

[vocal burrow]

screenshot it please

[karmic crown]

now do git pull

[vocal burrow]

like I said the only (unsolvable) problem is that stuff will break in the future if they (or any extension) use any pytorch 2.whatever feature.

[karmic crown]

the worst part is that 2.2.x does not work properly with latest macos 🙈

[vocal burrow]

just "think different" :p

but yeah that sucks

[karmic crown]

all A samplers are broken with 14.4.x+ and 2.2.x

they work with --use-cpu all 🙈

[vocal burrow]

I hope so, stuff could probably work on a samsung fridge with --use-cpu all if it has enough ram

nonetheless that's a strange decision on pytorch's end

I guess they think that every mac users have infinite supply of cash

[solemn atlas]

its like this but doesnt do anything else

[vocal burrow]

it did a lot of things

now delete venv and run webui-user.bat

[solemn atlas]

okay should i delate the cmd?

[vocal burrow]

you can close the cmd

[karmic crown]

@vocal burrow

if [[ "$(sysctl -n machdep.cpu.brand_string)" =~ ^.*"Intel".*$ ]]; then
    echo -e "Buy a new Mac\n"
fi

🙈

i need to go now, cu people

[past willow]

The biggest SD community and no one ever encountered or knows how the solve this problem?

[vocal burrow]

are the loras actually used what is shown in the text under the image inside the webpage ?

[solemn atlas]

doesnt work

[vocal burrow]

open cmd again from stable-diffusion-webui folder and run git show -l

[maiden coral]

if nothing works just do a clean install of all requirements and everything related to it, python must be installed with the checked option "ADD TO PATH" edit" path

[vocal burrow]

"add to path"

[solemn atlas]

rror: switch `l' requires a value

thats what I get

[vocal burrow]

git show -1

my bad

what does it says ?

screenshot the result

[solemn atlas]

it's this

[vocal burrow]

ok so you should be up-tod-date. yet it still boots up as a 1.8.0 ?

can you run webui-user.bat again and copy paste the log (and please no .rtf, just a plain .txt)

[solemn atlas]

its doesnt open SD

[vocal burrow]

why a .rtf again

sigh

ok so this one says 1.9.3 at least

open cmd again from stable-diffusion-webui,
run

venv\Scripts\activate
pip install albumentations==1.4.3
pip install pydantic==1.10.15

and then launch webui-user.bat

[vocal burrow]

.txt please

[solemn atlas]

I run now bat. and it worked

API  •  Github  •  Gradio  •  Startup profile  •  Reload UI
version: v1.9.3  •  python: 3.10.6  •  torch: 2.1.2+cu121  •  xformers: 0.0.23.post1  •  gradio: 3.41.2  •  checkpoint: ef49fbb25f

[vocal burrow]

yup

[solemn atlas]

its worked

THANK YOU SO MUCH

what does schedule type mean? Its new

[vocal burrow]

leave it on automatic

[solemn atlas]

okay

[rancid nest]

I was just checking out a conversation on these from this morning maybe? https://civitai.com/models/38784/controlnet-11-models

They don't work really for me in ComfyUI and blow out with an error. I stopped trying to use them awhile ago, but I find there really aren't any alternatives.

Just curious if anyone else cannot use these in ComfyUI.

[obsidian carbon]

Anyone know how to deal with a check point merge ? I found a model I reallllly wanna use and like but the issue is whenever I generate with it I get random colours that don’t show anything. The model on civitai doesn’t have any details yet I see others use it fine and I think I’m missing something. I’ll send below an example of what’s happening

like, it comes out like this

The checkpoint also doesn’t have a VAE with it so I’m confused on if I’m missing a specific vae. The checkpoint is called DarkSeal ((WARNING: 18+ NSFW)) but it doesn’t seem to wanna work with me at all ://

[onyx solar]

Without looking, I don't know what model version it is, but that doesn't look like a VAE issue. So first, have you tried a different CLIP Skip setting? Also, what resolution are you trying to generate?

[obsidian carbon]

It’s an XL model and it’s at clip settings 2 I think.

And it’s the latest version when you click download

[onyx solar]

What resolution are you trying to gen?

Also, just to be sure, you do need the SDXL VAE.

But if you're not generating a 1 megapixel image, you need to.

[ornate elk]

is it Pony based?

and did you used AnimateDiff extension in the past?

[gaunt mural]

hey i have problem with training my Dreambooth model can someone help

[noble belfry]

yes. especially if you look at it while its in the middle of a generation job.

[rocky adder]

You saved my a**
Thanks a lot again

[gaunt mural]

can someone pls help me

[ornate elk]

no problem 🙂

[sturdy forge]

Interesting...
Using unprompted on latest (A1111) dev branch atm and it seems like it's breaking Always save all generated images
one of the modules uses a mask, and it appears to be saving said mask and never the final image
Manual saving appears to work as intended

[mellow stone]

can you get models like checkpoint loras and control net from other folder in your drive ? i dont want to have duplicate files can it load from one specific folder? in stable diffusion ?

like i have comfi ui and stable diffusion and they use same models

like checkpoints can somehow import it to stable diiffusion with out copy paste it to stable diffusion

so i have 1 file each

[sturdy forge]

yeah, I always just create links

And then I can keep everything big in one place, etc.

[fair oxide]

I just yeeted the venv or whatever it's called and it started working again woooooo

[obsidian carbon]

Its pony based and I havent used AnimateDiff before - Its a "Merged checkpoint" also so I dont know what that means

[ornate elk]

Okay make sure you set clip skip to 2.
Then remove negative prompts to test if it works

[ornate elk]

Your still on an outdated version

[fair oxide]

Probably

It works tho

That's all I need

[ornate elk]

okay xD some extensions might wont work with that version

[fair oxide]

That might be the issue

I downloaded an extension and it yeeted my positive prompt box

Then it just decided it wouldn't launch

So I yeeted my extensions and the venv and it opened and was in perfect working order and the extensions work now too oddly enough

[ornate elk]

okay, good that it works now

[sturdy forge]

Not sure of your platform, but:

Windows (via PowerShell):

# Create an item in G:\gen\automatic1111\models\Stable-diffusion that is a link to the directory at G:\gen\models\stable-diffusion
New-Item -Type SymbolicLink -Target G:\gen\models\stable-diffusion -Path G:\gen\automatic1111\models\Stable-diffusion

You might need -Type Junction instead of -Type SymbolicLink... but I think a link should work fine

Linux:

# Create a file in /home/username/automatic1111/models/Stable-diffusion/ that is a link to the directory at /home/username/gen/models/stable-diffusion
ln -s /home/your-username/gen/models/stable-diffusion /home/your-username/gen/automatic1111/models/Stable-diffusion

You could also link individual items/files, but that can get annoying

[obsidian carbon]

Aight I’ll try that.

[rotund musk]

Hello, saw some kaggles for stable diffusion like this one, but is stable diffusion banned? what if like for example i had in mind to make a no ui kaggle of stable diffusion for kaggle, do i risk ban?

[noble belfry]

yes and no. they've banned webuis, with good reason, its not there to let randos mooch compute from them, its for data science. i mean you could probably sneak around it, but they're going to assume its done in worst faith, just to circumvent their policies.

so like, you're not asking if you should steal a car, you're asking if you can do it without getting noticed.

[upper goblet]

when i upscale this image its staying blurry. not really sure why. I have tried many methods now but I am going insane

[noble belfry]

are you using an upscaler or latent?

[upper goblet]

most of the time im using SD upscaler or ultimate upscaler

in img2img

[noble belfry]

is it specific to this image?

[upper goblet]

i was having issues with other images too i suppose

but they were similar to this

[noble belfry]

in your models folder, do you have any model files in your ESRGAN folder?

[upper goblet]

yes. 2 i just added one thinking it would help. 4x-UltraSharp

[noble belfry]

yes thats one i was going to recommend. it works best on low denoise levels and subtle upscaling sizes. like in steps of 1.0 to 1.25 to 1.50 not just instantly tripling image size

[upper goblet]

alright should i do it with SD or ultimate you think?

[noble belfry]

thats personally who i find it works best anyway. and by low denoise i mean less than 0.2

[upper goblet]

okok

looks just as blurry just a bigger res

going crazy

nothing works!!

[upper goblet]

after restarting about 6 times seems like im getting somewhere

[fading crypt]

So I'm trying to train a LORA as a learning exercise. Using Kohya_ss and the SD 1.5 model worked out just fine. Then I decided to switch to SDXL, and I'm getting some wild error messages that end with: '--config_file', './outputs/tmpfilelora.toml']' returned non-zero exit status 1 I know this isn't a Kohya specific discord, but has anyone run into this issue?

[fluid nexus]

Is it possible to run stable diffusion locally on Android?

[tacit crystal]

Can someone help me with this? I previously had an old version of SD, but after updating I started to have a couple of problems. One of those is that it doesn't let me select the loras folders, and it's as if "all" option was selected indefinitely without me being able to do anything to change this and select another folder of loras, so i had to scroll all the way to serch the lora I want to use, also functions that I should have due to the version I use, don't appear to me.

[sinful needle]

Help I disabled an override called "downcast alphas_cumprod: true" and now my outputs look cartoony and stupid. how do i reverse it?

[sinful needle]

never mind turns out i just need hires fix. blyaaaaaaaaaaaaaat

[stark mulch]

where do we go for help?

[fierce hollow]

Reinstalled Windows, please tell me what exactly I need to install to use the neural network (I'm talking about python, I saved the neural network itself)

[ornate elk]

For technical help your right here. For image questions there is #📝｜prompting-help

[stark mulch]

i mean with stable

like i got everything im just confused on the civati stuff

[ornate elk]

Hey, you need python 3.10.11 64bit and Git 64bit

[ornate elk]

Okay then you can ask here

[fierce hollow]

Okey...

[stark mulch]

okay so i was trying to use a style and how do i use it i followed the directions and stuff but idk how to use it and also how do i use sliders like a hair slider i strugled figuring it out'

[uncut jewel]

I opened my ComfyUI and now it's not working, anyone know what's happening with Torch and Cuda ?

[ornate elk]

It depends on the file. Was the style a .CSV ?
The hair slider is mostly a lora.

Here is a lora guide:
#🤝｜tech-support message

[stark mulch]

what about the style ui?

[ornate elk]

Styles can be saved in auto1111 under the generate button

[stark mulch]

how do i import one from civtai

[vocal burrow]

style is just a bunch of world that get copy pasted onto your prompt. It's basically a "text shortcut" for your prompt. So there's nothing to download. What civitai page are you reffering to ?

[stark mulch]

https://civitai.com/models/132426/art-styles-expansion-for-styleselectorxl this one im stuck on what to do i downloaded and thank you for helping me does this stuff get easier?

[stark mulch]

i cant find stable xl folder

[stark mulch]

i made one for it

i still cant even use it

[vocal burrow]

stable xl folder ?

[stark mulch]

yes in the direction theres supposed to be a stable xl folder

[vocal burrow]

there's no mention of a "stable xl"

[stark mulch]

Styler slector xl mb

[vocal burrow]

probably because you didn't install the suggested extension

[stark mulch]

broooooooo

u a life saver

[vocal burrow]

Otherwise you can probably just take the .csv from the archive and replace your styles.csv with it

[stark mulch]

wdym

[vocal burrow]

tke the .csv file from the civitai archive, rename it styles.csv and drop it into your stable-diffusion-webui

everything should show up in the usual styles selector

[stark mulch]

okay tysm!

[stark mulch]

you guys freaking angels is there a way i can repay u

[vocal burrow]

meh not really, it's fine.

help others if you can.

[stark mulch]

@vocal burrow

how does training work

[vocal burrow]

what do you mean ? what's the logic ? how does it work technically ? Or how to train something ?

[stark mulch]

all 3

LMAO

jk

how to train something

cause some of these require it

and idk how to do it

[vocal burrow]

"some of these" ?

[stark mulch]

https://civitai.com/models/137258/realorarealistic-skin-texture

requires 5k training

or steps

[vocal burrow]

you don't have to train anything to use a lora

[stark mulch]

oh

[vocal burrow]

loras are the results of training

[stark mulch]

ohhh

how do i use the strength

of a lora

Or edit it

[vocal burrow]

cf #🤝｜tech-support message

[stark mulch]

what about training what does it do?

[vocal burrow]

it's altering the weight of the models to introduce a new concept / modify the already existing weight associated to an already existing concept.

The result of training is either a fully updated model or some sort of patch that can be applied to a model (eg : lora)

[stark mulch]

so its not really important alright

[vocal burrow]

I mean, it allows the community to add a bunch of stuff to the model and make them better. I'd argue it's quite important. But you don't have to do training, you can just reap the result from the labors of others.

[stark mulch]

correct im too dumb for that so thank you lol but if i need help with other stuff i contanc you guys man do you take nitro dude because i was gonna give up on this stuff and you helped me

[vocal burrow]

thanks, but I'm good man, don't worry.

[stark mulch]

nah not all heros wear capes like u dont worry when i cash my stuff in tommrow i got u first thing

@vocal burrow how do i use sdxls?

[umbral fox]

I have installed the following extensions in Stable Diffusion, but Lora is not generated.

https://github.com/liasece/sd-webui-train-tools

It is not generated as attached and [Ending job] is displayed on the command, but
On train-tools
There is no change even after an hour has passed since "A: 4.34 GB, R: 4.48 GB, Sys: 6.1/23.9883 GB (25.4%)" is displayed.

What I confirmed
・Stable Diffusion version change
(1.8.0→1.6.0→1.5.1→1.7.0)

・Can the Train base model be used to generate Lora unless it is adapted?

[vocal burrow]

sdxl models ? just like other models, you put them in stable-diffusion-webui/models/stable-diffusion, reload your webui or refresh the model list and then simply select them from the model dropdown at the top left of the webpage

[stark mulch]

how do i use 2 at once or more?

[vocal burrow]

if your gpu have less than 12gb you'll want to use --medvram-sdxl (and might also enabled fp8 mode for sdxl models in the settings) and if you have less than 8gb use --lowvram-sdxl

[vocal burrow]

you can't use more than one model at once

[stark mulch]

WHAT WHY NOT!??!!

so if i want cartoon and jug xl i cant?

[vocal burrow]

probably incorrect parameters used for training, try asking in #🔧｜finetune that's where the folks that do training usually hang out

[vocal burrow]

you can switch models but no you can't use two models at once.

just like you can't have two engines in your car (yes I know there are exceptions but you get the idea)

[stark mulch]

dang man do you have any idea how i can fix a face on some people with inpaint

[stark mulch]

yeah

[umbral fox]

Thank you!

[stark mulch]

@vocal burrow do you know whats the best negative prompts?

[vocal burrow]

inpaint + mask the face + tweak the cfg and denoiser values till you get something that please you.
The higher the denoiser the stronger the changes will be

[vocal burrow]

there's no best negative prompts just like there's no best brush in photoshop. It depends of what you're trying to do

[stark mulch]

and do symbols like (( do anything

[stark mulch]

ah ok

[vocal burrow]

cf https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#attentionemphasis

[stark mulch]

TYSM!

[split comet]

Im stuttering pretty hard playing games whilst generating, anyway to limit the ram or cpu?

[fervent shoal]

Anyone know this Error ?
Happen when i try to use Tiled Diffusion with a inpaint Controlnet.

But only at some images.
Mostly when the image is bigger then the tile sizes.

Any idea ?

[ornate elk]

What's your GPU and whats inside your webui-user.bat?

[timid notch]

does anyone know how to solve this problem? i installed sd but can not reopen it by webui-user.bat~~

did i use wrong version of it?

[vocal burrow]

it's telling you clearly that you didn't install controlnet extension

[timid notch]

oh thx,but not this reason,i just delete this folder, and it is working

[vocal burrow]

yes because it was this extension that required controlnet

[timid notch]

what?

[vocal burrow]

sd-webui-cn-in-extras-tab requires controlnet => controlnet is not installed => error
two ways to fix it :
1/ install controlnet
2/ uninstall sd-webui-cn-in-extras-tab

you chose 2/

[timid notch]

eeeee ohhhh,

so how to install this?

i seach the extension

but can only find this:sd-webui-cn-in-extras-tab

[vocal burrow]

you install it from the extensions tab

it should be sd-webui-controlnet

[timid notch]

this one?

[vocal burrow]

no

[timid notch]

this one?

[vocal burrow]

nop

[timid notch]

but there are no other choice~~

do i really need it?

can sd run without it?

[sinful needle]

Where do I get the model for open pose for forge? I tried installing an extension but it must be for base A1111 since it screwed the webui when i tried using it. Just need the model on the right

[vocal burrow]

ok I checked, should be this one

[vocal burrow]

forge comes with its own implementation of controlnet, so disable the usual Auto1111 CN extension when using forge

[sinful needle]

i did, but it doesnt work. there is no model on the right dropdown

usually im required to have something on the right

[vocal burrow]

And you can get cn models from huggingface, civitai, etc eg for sd15 controlnet models https://civitai.com/models/38784/controlnet-11-models

[sinful needle]

drop those in the integrated extensisons folder?

which one? none seem to have a folder to drop them in.

[vocal burrow]

put them in stable-diffusion-webui\models\ControlNet or stable-diffusion-webui\extensions\sd-webui-controlnet\models

[sinful needle]

this is the webui folder. no direct controlnet folder. extensions folder is empty since controlnet is builtin by default

never mind

im a dope

[vocal burrow]

I didn't say stable-diffusion-webui\ControlNet but stable-diffusion-webui\models\ControlNet

[sinful needle]

models / controlnet

[vocal burrow]

sure you can use SD without it, but if you install some extensions that requires it.... then you're required to install it

[timid notch]

hah need some brilliant programmer simplify the program

spend all day long to install this,tried~

[sinful needle]

when i use the model(s): *** Error running process_before_every_sampling: C:\Users\place\Desktop\Studio\Visual\Image\Ai\Stable Diffusion\Forge\webui\extensions-builtin\sd_forge_controlnet\scripts\controlnet.py
Traceback (most recent call last):
File "C:\Users\place\Desktop\Studio\Visual\Image\Ai\Stable Diffusion\Forge\webui\modules\scripts.py", line 835, in process_before_every_sampling
script.process_before_every_sampling(p, *script_args, **kwargs)
File "C:\Users\place\Desktop\Studio\Visual\Image\Ai\Stable Diffusion\Forge\system\python\lib\site-packages\torch\utils_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "C:\Users\place\Desktop\Studio\Visual\Image\Ai\Stable Diffusion\Forge\webui\extensions-builtin\sd_forge_controlnet\scripts\controlnet.py", line 555, in process_before_every_sampling
self.process_unit_before_every_sampling(p, unit, self.current_params[i], *args, **kwargs)
KeyError: 0

[vocal burrow]

give us the full error log please, put it in a .txt and drop the file in here.

It makes it easier for everyone

[sinful needle]

roger roger

[sinful needle]

Thak you for your patience

[timid notch]

Thak you for your patience too@vocal burrow

[sinful needle]

its pretty big so gotta download to see rest

only happens when i use a model

[vocal burrow]

I meant the full full log. Like everything from the start.
But from what I see it seems like you're using a SDXL model with Controlnet SD1.5 models

[sinful needle]

ill check

probably right boss one moment

[vocal burrow]

sdxl stable diffusion model => sdxl cn model
sd 15 stable diffusion model => sd 15 cn model
etc

[sinful needle]

are there controlnet models for sdxl

im gonna run this proper with the right 1.5s and see if it works still though

[vocal burrow]

cf https://huggingface.co/lllyasviel/sd_control_collection/tree/main there are many sdxl models out there

it's just one page among many

[sinful needle]

Well its not crashing this time, but it's not really adhering to my openpose model

[vocal burrow]

what settings are you using ? screenshot your sdwebui webpage

[sinful needle]

one sec. ill have to take 2 since it scrolls a bit

[vocal burrow]

ctrl+shift+s to screenshot webpage properly

[sinful needle]

alright one sec

[vocal burrow]

(maybe it's a firefox only feature tho, let met know)

[sinful needle]

gotta try a different model so i dont break TOS. for some reason epic realism was doing just fine SFW now its got some lady with her tatas out. damn it. one sec

ill try revanimated

[sinful needle]

not sure if that works on Brave browser. nothing happened

[vocal burrow]

what a shame

old fashion screenshot it is then

[sinful needle]

pdf

printed as pdf

[sinful needle]

that worked 🙂

full page

the person is barely contorted to the pose model each time. just barely

[vocal burrow]

you can tick the "OpenPose" ControlType box to preselect some preprocessor and model

it makes it easier for you

also there should be an "explosion" icon somewhere in the CN widget that should allow you to preview the stick figure picture thingy

Otherwise you can try another cn model, increase the control weight, set "control mode" to "ControlNet is more important", etc

[sinful needle]

preview yielded blank box on right of normal pose box

[vocal burrow]

Did you give CN an image to extract a pose from ?

[sinful needle]

the one above

oh

i cant just use an openposer? i have to have something extracted?

[vocal burrow]

I mean there

[sinful needle]

so i cant use the open pose model manually? used that from an openpose site thought it was easier to maneuver

[vocal burrow]

here you're supposed to give either a picture you want to extract a pose from with the preprocessor, or a pre extracted stick figure (for open pose model) looking image

[sinful needle]

yeah the extracted figure is there posted

[vocal burrow]

yes you can but then you don't need a preprocessor

[sinful needle]

understood. running without

[meager ginkgo]

Any know how I can turn this image into ai art?

[sinful needle]

voila

thanks fren. now ill test the SDXL models to see if that works. wanna use a pepe frog lora from XL

[ornate elk]

Hey, yes you can do that with stable diffusion and Controlnet lineart

[sinful needle]

do i need both the YAML and PTH file for each model or no

[meager ginkgo]

I tried to but it just produced the exact same thing

[vast crest]

Hi everyone, I saw that version 19.3 was released, my SD doesn't update, as always. these are the commands I have:

git pull

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --xformers --no-half-vae --autolaunch

call webui.bat

[vocal burrow]

you don't need the .yaml

[languid magnet]

can someone please share some good API service for text to video generation ?

[ornate elk]

You need to set preprocessor to None

[meager ginkgo]

Do I use a model?

[ornate elk]

Yes, You need a model for generating the image + a model for Controlnet (the line art model)

[meager ginkgo]

For some reason my controlnet model tab is empty

[ornate elk]

Have you downloaded the Controlnet models?

Or make a screenshot how it looks rn

[lean ridge]

hey, im getting this error when using sd upscaler on forge

TypeError: 'NoneType' object is not iterable

there's something i can do to fix it?

[ornate elk]

@meager ginkgo here are Controlnet models.
Download the model + its config file below and put them into models/controlnet
https://civitai.com/models/38784?modelVersionId=44877

After that refresh the Controlnet models with the reload button at the right

Then select lineart as model and for preprocessor select None

Set your output resolution the same or near the input image resolution

[meager ginkgo]

I had to put the line art in the preprocessor but I’m getting stuff the is adding up

@ornate elk any tips to remove move the wording?

[sinful needle]

not sure why the picture is coming out nowhere near what i planned. The PDF is just the webui printed with settings

[uncut jewel]

Anyone know how to solve that issue ?

[maiden coral]

Have you modified something in ComfyUI before it started showing this error? What were you trying to do?

[uncut jewel]

I think i installed SD ultimate Upscale, but then i removed it and did some stuffs based on some forum but i still have the issue

[maiden coral]

you only download the folder with the node or whatever you deleted and that's it, then reset the local server of ComfyUI

what stuff

[ornate elk]

Color them white

[limber pier]

workflow embedded

[meager ginkgo]

looks great but the vampire theme hurts my heart

some of my base models stopped working. don i need to replace them?

[karmic crown]

depends on what you exactly mean by "stopped working"

[limber pier]

how about a shark theme

or a clown theme

[solemn atlas]

I'm sorry but what does that mean

[uncut jewel]

I'm not sure i understood your response. Yeah i downloaded the custom nodes from ComfyUI manager and then restarted and had this issue coming

[maiden coral]

well, it won't happen out of blue, I have no problem with SD Upscale

[uncut jewel]

Then i started to look online for the error, so i tried to add the script folder to SD Ultimate Upscale folder but it wasnt solving it so i just deleted the whole folder from ComfyUI.
Then i found that person talking about an issue with the protobuf thing so i reinstalled the version based on those recommandations but i still got the issue

[uncut jewel]

Yeah i used to work with it and it was working, but then i had to delete it because i had some import issue. Then i reinstalled but maybe it's due to something else, but i'm not sure what.. That's why i kindly asking for help 🥹

Now i'm just starting Comfy and having that error straight away

[maiden coral]

Sorry then, I can't help you. I'm missing what files you deleted. Generally you just delete the folder with the node. I don't have much knowledge in ComfyUI, but it'd be helpful if you told me what you deleted.

or just download ComfyUI again and replace the files you deleted or modified

Do you have the portable version?

[uncut jewel]

Right, the thing is that i deleted nothing until i got the error, that's why it doesnt make sense lol

[uncut jewel]

Yup

[maiden coral]

Have you tried full reinstall even with the requirements? What python version you use?

[fair oxide]

I'm a newbie, can someone explain what \ (word\ ) does in prompting, and what adding multiple parenthesis does? like (((word)))

[uncut jewel]

No i've not tried that yet, i'll do that as my last option since it's going to be quite heavy with all the models and nodes i've uploaded overtime

[uncut jewel]

adding parenthesis add more weight to a specific word or sentence, so the prompt will impact heavier on those words than the rest of the words

[maiden coral]

increases the factor or in other words focus on the particular word during the image generation, you can find it here:https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features

[fair oxide]

that applies to comfyui too right?

[maiden coral]

yes

[fair oxide]

thanks!

[maiden coral]

not all things I think, but the brackets do

[uncut jewel]

Yeah i guess it's an error with Python

[stark mulch]

@vocal burrow can i get some help when i added another lora it deleted the other one and i cant get the other one back i even readded it how do i fix?

[meager ginkgo]

i use forge and i get an error "TypeError: 'NoneType' object is not iterable" when i try to use 2 of the check point models. one is a SDXL 1.0 and the other was Pony

[latent flare]

Only found this https://github.com/lllyasviel/stable-diffusion-webui-forge/issues/181

[limber pier]

here's another style for ya

[limber pier]

if you have comfyui installed you can just click drag these into it and the workflow will pop up and you can substitute the style 🙂

[narrow copper]

Hey, when you're usin forge, is there a way to set defaults?

(ie i want it to start w/ xy height width, z sampling method, etc.)

[latent flare]

Yes

[latent flare]

At least he tried pretty hard. That's why he has my respect ...

[viral knot]

where is the clownshark checkpoint on civitai? :3

[latent flare]

Should be in his pictures

Ohh.... checkpoint?

[viral knot]

well with all these clownshark pics, might as well make a full checkpoint 🙂

[latent flare]

I think it's a comfy workflow using IP Adapter

[viral knot]

yea i saw his workflow one day, it was a thing of beauty

[latent flare]

Beauty is an interesting word for it 😄

[viral knot]

lots of spaghet 🙂

[viral knot]

nice zoom out

[latent flare]

Not my work 🙂

[viral knot]

that's a...damn...lie :3

[latent flare]

A.I. did ... I just gave a picture

[viral knot]

ai did all the hard work

[latent flare]

But it still can't clean dishes ^^

[viral knot]

il clean your dishes 🙂

[latent flare]

Bedtime coming! Have a good time!

[viral knot]

gn

[stark mulch]

Does anyone know how to get better quality images and stuff

[lofty whale]

It may be the wrong place to ask, but since I'm using the Stability API it's worth a try. I'm trying to make a web app that generates images but I'm getting an error. Here's the front-end: ```javascript
import './style.css';

const form = document.querySelector('form');

form.addEventListener('submit', async (e) => {
e.preventDefault();
showSpinner();
const data = new FormData(form);

const response = await fetch(import.meta.env.VITE_API + '/dream', {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({ prompt: data.get('prompt') }),
});

if (response.ok) {
  const base64Data = await response.text();
  const binaryData = atob(base64Data);
  const bytes = new Uint8Array(binaryData.length);

  for (let i = 0; i < binaryData.length; i++) {
    bytes[i] = binaryData.charCodeAt(i);
  }

  const blob = new Blob([bytes], { type: 'image/webp' });
  const blobUrl = URL.createObjectURL(blob);

  const result = document.querySelector('#result');
  result.innerHTML = `<img src="${blobUrl}" width="512" />`;
} else {
  const err = await response.text();
  alert(err);
  console.error(err);
}

hideSpinner();

});

function showSpinner() {
const button = document.querySelector('button');
button.disabled = true;
button.innerHTML = 'Dreaming... <span class="spinner">🧠</span>';
}

function hideSpinner() {
const button = document.querySelector('button');
button.disabled = false;
button.innerHTML = 'Dream';
}

and here is the back-end: ```javascript
import fs from "node:fs";
import axios from "axios";
import FormData from "form-data";

import dotenv from 'dotenv';
dotenv.config();

import express from 'express';
import cors from 'cors';

const app = express();
app.use(cors());
app.use(express.json());

app.post('/dream', async (req, res) => {
    console.log("--POST")
  try {
    const formData = {
      prompt: "Lighthouse on a cliff overlooking the ocean",
      output_format: "webp"
    };
    
    const response = await axios.postForm(
      `https://api.stability.ai/v2beta/stable-image/generate/core`,
      axios.toFormData(formData, new FormData()),
      {
        validateStatus: undefined,
        responseType: "arraybuffer",
        headers: { 
          Authorization: `Bearer `+ process.env.STABILITYAI, 
          Accept: "application/json" 
        },
      },
    );
    
    if(response.status === 200) {

        // console.log(response)q
    //   const data = response.json()
    console.log("---DATA")
    console.log(response.data)
      res.send(response.data)
      //fs.writeFileSync("./lighthouse.webp", Buffer.from(response.data));
    } else {
      throw new Error(`${response.status}: ${response.data.toString()}`);
    }

  } catch (error) {
    console.error(error)
    res.status(500).send(error?.response.data.error.message || 'Something went wrong');
  }
});

app.listen(8080, () => console.log('make art on http://localhost:8080/dream'));```

the error is on line 10: const response = await fetch(import.meta.env.VITE_API + '/dream', {

[narrow copper]

Thank you kindly Yates!

[sinful needle]

I am unable to get my output to obey my poser model. I have the entire webui/settings on that PDF printout. It spits the poser in output on a separate pic so it looks like it's working, but it's not.

[sinful needle]

Could be that I have the wrong model? I'm using SDXL and it's SDXL, but there's a giant list here and I'm not sure which one to use: https://huggingface.co/lllyasviel/sd_control_collection/tree/main

[harsh token]

Does anyone know if there's a SDXL segmentation model?

[sinful needle]

i wonder if some loras just dont work with poser

[jagged jungle]

Brothas has anyone ever managed to have success in using controlnet open pose with forgeUI?
This thing just wont work
It works fine on the preview and all but the generated images just dont follow the pose

here is a pic of the controlnet with a random pose, so you can see that I do have the right model and it is drawing the lines, but the gen image doesnt follow them

[sinful needle]

thats what im sayin. try Unit1 and see if that helps. I think Unit0 wont.

im having similar issues. Sometimes it follows the pose and sometimes it doesn't; doubly difficult if I use LoRAs.

[jagged jungle]

Oh you having the same problem?

[sinful needle]

ye

[jagged jungle]

I'm considering trying swarmUI just cause of this, but dont know if it has controlnet, will search later

[sinful needle]

try Unit1 instead of zero under the controlnet tab. also if u have a pose skeleton select none under the preprocessor

[jagged jungle]

also ye I'm using loras, didnt try without

K, will try in a few mins and will report back

[sinful needle]

worked better without LoRAs. someone needs to raise a stink about this on the commit threads

because i DO NOT wanna go back to auto1111

[jagged jungle]

There hasnt been any updates in forge for some time it seems

So I dont have much hope

[jagged jungle]

yeah no way to do XL with it

[sinful needle]

in either or just forge

[jagged jungle]

I hope Swarm has controlnet cause I dont wanna waste time with comfyUI either lmao

[sinful needle]

so did it work with regular loras or just not XL

[jagged jungle]

so could use Swarm for controlnet and switch to forge when inpaiting since swarm doesnt seem to have inpainting

[sinful needle]

because it seemed to just be xl for me

[jagged jungle]

didnt try yet, but I can test

[sinful needle]

thanks yeah id like to know if its ALL loras giving u trouble or just XL

[jagged jungle]

Ya just let my GPU finish generating those 87 images

Its weird that I can barely see people talking about this tho

So Idk if its a general problem or if we are doing something wrong

[sinful needle]

was it just XL or no

also we just need to get the attention of the gurus here. they might be on break

[jagged jungle]

yeah, wanna at least have an answer from people who have been here for longer to know if it's a general problem with Forge or if it works on their machines

[sinful needle]

soooooo

was it just XL or not

or all loras for u too

[jagged jungle]

ye just tested, working fine with SD 1.5 loras

its something with XL then

[sinful needle]

udnerstood that saves me time. im gonna use 1.5s for now. was trying to make a sick pepe frog background with XLs

[timid notch]

hello,i wanna install some model and i open civit brower,but it is said i need some kind of api key?

can someone tells me how to get that key? or how to install it?

[harsh token]

API to connect to Civitai. I believe you can find it on your profile on Civitai? Was a while ago I did that. But honestly imo it's easier to just go to Civitai yourself and download from there. Especially since you probably usually want to view more info in regards to the models and also view user submitted images to get a good idea of what prompts are useful on that specific model.

[timid notch]

i serch the website

can not find anywhere about api key~@harsh token

i think this is the right website, right?

[harsh token]

Well yeah, like I said your account

[timid notch]

oh i saw that choice

thx

[sinful needle]

you ever hav issues with openpose with forge?

[harsh token]

don't use forge

[sinful needle]

y?

you saying you dont or saying i shouldnt?

goooooooooood talk

[harsh token]

I'm saying I don't

[sinful needle]

ah

[timid notch]

oh i have another question~~can you tell me which option do i need to enter the api key?is this the right place?

[sinful needle]

weird that auto1111 has problems with animatediff and forge has problems with controlnet poser

[tight dirge]

Did you ever figure out the solution here?

[narrow copper]

Alright, so I'm using Forge. I've got https://civitai.com/models/56519/negativehand-negative-embedding this in the embedding folder, but it's not showing up in options even with multiple refreshes. Is there a trick I'm missing?

[harsh token]

I'm not using Forge, but I'm going under the assumption that it works the same way there. Chances are you're using the wrong model. For instance a SDXL model. Because it probably defaults to filtering out stuff you can't use.

[stark mulch]

whats a pony image yall?

[narrow copper]

yeah, that's usually the case. i'll recheck. thanks. 😗

pony diffusion is a checkpoint method that's pretty good.

either that or a picture of a my little pony.

[stark mulch]

how does it work per say

[harsh token]

Do you know what checkpoint/models are?

[stark mulch]

kind of not really

and aye yo whats with the pf

pfp

[harsh token]

It's just what the AI has been trained on essentially. And currently, Pony is probably the best one there is for Stable Diffusion. Unless you're looking for something really specific, like realism.

[stark mulch]

so how do you use it properly?

[harsh token]

I'm not really sure I understand what you mean. I mean there are a billion different ways to use it. So what do you mean by "using it properly"?

[stark mulch]

yeah to get good images also how do i quality boost them?

[harsh token]

That's a bit too broad and subjective. But just go to Pony's Civitai page, read about the model since they give some decent information. Then you can check the user submitted images for it and see what they do to get that result.

[stark mulch]

yo is there a way to rename loras i can do it man with all the numbers?

[harsh token]

Should work by just renaming the file itself. Just know that the trigger words would be the same

[stark mulch]

does that apply with sd models too?

[harsh token]

I think so? Never tried it. Cause that seems like a recipe for disaster if you need to look up the names later. And I'm not using enough checkpoints that it would be necessary.

[stark mulch]

okay so i got another question how do i use this pony? https://civitai.com/models/257749?modelVersionId=290640

[harsh token]

I mean, just download it like you would any other model

[stark mulch]

I mean do i have to use it for checkpoint only and why do these things take 20 mins jesus

[harsh token]

What do you mean "for checkpoint only"?

[stark mulch]

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

that

[harsh token]

I think the VAE it recommends is just the default SDXL VAE.

You don't need to use it, you can use another one if you want

[stark mulch]

dude how do i use different styles with pony?

[harsh token]

It sounds like you just need to look up how to use Stable Diffusion as a whole

[ornate elk]

Why that?

[stark mulch]

im trying im hella new with it and its so slow that my main issue i thought nvida cards are good for this damn

[ornate elk]

Auto1111 has no problem with AnimateDiff.
And forge is known for buggy or broken controlnet

[stark mulch]

hey long time no see

[ornate elk]

Hey

[harsh token]

Nvidia cards are good. But you also need a good card in general. ESPECIALLY if you want to generate in an XL model.

[stark mulch]

so whats considerd good do i need to upgrade mine then?

[ornate elk]

What's your GPU?

[harsh token]

I don't even know what you have so I wouldn't know if it is or isn't good and if you need to upgrade.

[stark mulch]

rtx 3070ti

[harsh token]

Should be fine

VRAM is your biggest problem most likely

[stark mulch]

HELL NO it takes a whole min to generate 1 image lmao

[ornate elk]

Ah okay then you need to make some adjustments in the webui-user.bat for the best performance

[stark mulch]

wdym

[harsh token]

Then it sounds like you're doing something wrong. I used an RTX 2070 for the longest time. A 3070 Ti is at least 50% better.

[ornate elk]

You have to edit the webui-user.bat
By right click, edit, then at the line commandline_args=
you need to add: --xformers --medvram-sdxl --no-half-vae
Then save and launch the webui-user.bat

That will let you generate much faster

[stark mulch]

ok ima test it

[stark mulch]

DUDE A LIFE SAVER AGAIn

[viral knot]

wait until he learns about lcm, etc :3

[stark mulch]

whats lcm?

[stark mulch]

whats score up do and how do you know how to use it?

[ornate elk]

read the model description of Pony, there its explained

this model was trained differently than sdxl or 1.5 models

[harsh token]

As I told you before. Read the model's page. They give you a lot of information you need. In fact, that's what you always should do regardless of the model you use. Sometime they even give good baseline prompts to start with and recommended settings to use.

[soft fable]

Hi guys, I am looking to buy a PC (not laptop, since those GPUs are not the same like in desktop) in the future for using SD.
I have read a lot about needed specifications, but besides the VRAM, how would you weigh the influence of the other system elements when using SD?
So far, my list goes like this, rated from most influence to least influence:

1. Nvidia RTX 40-series with at least 12-16 GB
2. CPU - best maybe a new Intel 9 ?
3. RAM - at least 32GB - how much influence has r/w-speed on SD?
4. Storage - SSD - how much influence has r/w-speed on SD?
5. Mainboard and power supply - I don't know much about these components...

[ornate elk]

Hey, for SD image generation the VRAM matters the most, and the newer the gpu the faster it is too. (12-16gb is very good)
For Gaming make sure the CPU matches your GPU so you dont get a bottleneck. (Depends heavily on the monitor Resolution)
You can check this here: https://pc-builds.com/bottleneck-calculator/
32GB RAM would be perfect for any task. Make sure you buy a dualchannel kit.
For the Power supply checkout this: https://www.bequiet.com/en/psucalculator
There make sure your power supply gets used at least 75% and not over 90%. best is between them.

[naive dragon]

i get this error: AttributeError: 'NoneType' object has no attribute 'lowvram' and i am also unable to load any Stable Diffusion checkpoint how do i fix this? (i am using a rx 6700 xt and i followed (https://www.youtube.com/watch?v=eO88i8o-BoY) this video for the installation) the default checkpoint it comes with does work but trying to use others doesnt work.

[vast crest]

Hi, is there something very similar to this program: Kohya LoRA Dreambooth? Unfortunately it has been closed, and I can't create Loras with other programs, because I find them really too complex

[sinful needle]

i beg to differ a little bit i found a heck ton of complaints about animatediff while googlin the issue. like a LOT of them and its all the same exact error i have. some people dont have it but a lot do. i can go scrounge up every single thread about it and make a compilation to post here if you want.

[fallen quest]

What do you mean with "it has been closed"?

[ornate elk]

You could show me the error sure, maybe i can help

[ornate elk]

What do you mean with closed?

Kohya_ss gui works and is still active

[ornate elk]

That tutorial isnt good, you should use my install guide from the pinned messages of this channel to install the better working webui version for AMD, its with zluda

[vast crest]

this version is closed: Hollowstrawberry-Kohya-colab

[ornate elk]

ahh google colab

[vast crest]

Yes 😦

[slender coral]

Getting frustration every time when trying to generate a decent image using exampes from ai gallery websites. Could this be Lora smth?

Note: I am using the same settings and model

[ornate elk]

your not using the same settings, but i cant check your meta data

[slender coral]

could this be sampling method or lora?

[ornate elk]

lora or model

if you really use the same model, then you need the same lora too

and the image creator used hiresfix or sdxl

[slender coral]

about image creator is new to me

https://www.reddit.com/r/StableDiffusion/comments/17gysxn/sdxl_hiresfix_solutions_and_proper_sdxl_refiner/

could this be good tutorial?

[ornate elk]

nope, i wouldnt recommend using a refiner for sdxl

its not needed for custom models

what model are you using?

[slender coral]

I am using ponnyDiffusionV6XL

[ornate elk]

then you need to adjust your resolution, because pony is sdxl based, and sdxl is trained on 1024x1024

[slender coral]

Wow, haven't expected to be it very important :0

I have been using 218x218 for faster generation

[ornate elk]

oh thats pretty bad even for 1.5 based models

1.5 is trained on 512x512

resolution is very important

[slender coral]

thanks dude

waiting for the new image to be generated

[tacit crystal]

Does anyone know what the hell I'm doing wrong? I recently upgraded SD so I can use the XL and Pony models because it has some loras that I like. So I was doing tests and it happened to me that in the preview the image came out in relatively good quality, but then the final result is the same image but pixelated. Is there something I'm missing? because with the 1.5 models it doesn't happen

[ornate elk]

what vae are you using?

[tacit crystal]

orangemix

[ornate elk]

here you see an sdxl model used on 512x512 and 1024x1024

[ornate elk]

thats a 1.5 based vae, you need to use an sdxl vae

[tacit crystal]

ok, thanks, i didnt know that vaes also requires xl baesd

[ornate elk]

yep, same goes for loras and embeddings too

[tacit crystal]

okay thanks, I'll keep that in mind

[slender coral]

very huge difference

Also just wondering, is this good speed values for gtx 1080?

[ornate elk]

totally not 😮

i have a GTX1080 in my old PC and it was faster

whats in your webui-user.bat?

[slender coral]

could it be because I am using portable version?

[ornate elk]

if its on an external drive it should be on an SSD.
but to get better speed you need to remove --opt-sdp-attention
and instead add: --xformers --medvram-sdxl --no-half-vae

[slender coral]

Added arguments, won't moving it to SSD because still using 128gb

[ornate elk]

okay, with these args it should be much faster when using sdxl models like pony

[slender coral]

218x218

and 🥁

before/after

Thank you for explaining!!!!!!!!!!!!

[ornate elk]

no problem 🙂

[slender coral]

And also:

[naive dragon]

did someone make a video about it bcs im bad at reading ;/ (i guess ill try my best at following it) but thanks for stating that the tutorial isnt good :)

[ornate elk]

You can ask me if you have any problem or question. I'm on AMD too

[iron thistle]

is this the right place to get tech support or put a support ticket in for Stable Audio subsriptions?

[wooden hull]

hello would you be I can install a stable broadcast

?

[ornate elk]

What's stable broadcast?

If you want to install a Programm for Image generation (stable Diffusion) then your right here

[wooden hull]

sorry I use google translate because I'm not English Stable diffusion

yes

[ornate elk]

Ah NP, checkout the pinned messages of this channel for my Install guides.
There are different ones for Nvidia and AMD cards

[wooden hull]

I had installed it a while ago

thank

[tender gazelle]

Does anyone know of a cloud API that supports ControlNet?

[wooden hull]

good evening again excuse me do you know if it is possible to easily install lora from civitai?

[ornate elk]

Hey, download a lora and place it in the model/lora folder

Then you can select it in the webui

[wooden hull]

yes it is therefore not an extension that exists for civitai?

[ornate elk]

There is an extension called Civitai Browser Plus

With that you can easily install it

[wooden hull]

okay thank

[fair oxide]

I have no idea what's causing this. sometimes i can see faintly what i wanted to generate but it would still be made up of black and blue blobs

[ornate elk]

What's the model and VAE?

[fair oxide]

though this has heppened with two models so far

weird. issue seemed to have fixed itself after reloading the UI

[meager ginkgo]

@ornate elkhey is there a way to alter a seed? im try to lighten the skin tone on this image

[ornate elk]

hey you can alter the seed with the extras button:

[sinful needle]

I've been using Forge so I'll have to redownload the A1111 repo so I can replicate the bug.

[amber cloud]

guys, using AUTO1111 it seems i can't get any model on any sampler to draw 2 faces with opened eyes. when 1 face in the image the eyes are great, but with 2+ faces all the eyes are closed or deformed. how can i fix this please ?

[sinful needle]

Hey you told me to post the error, so here ya go. Here's also some proof (motion henry cavill) that it sometimes works as it quits on me at random after consecutive generations and no settings changed. It COMPLETELY crashes A1111 and I have to restart it as I can't even generate still images after that. I've tried to iteratively replicate it by slowly tacking on extra things like lengthier prompts, LORAs, and changing the motion module to something from Civitai and it seems to mainly shit itself when I add negative prompts to the mix.

And, to clarify, I do have a beefy GPU (3090RTX). Also thank you for your time.

[olive snow]

hey guys

but do you maybe know why this is happening?

I have tried 7 times.. always allocating less and less CUDA but still it does not allow me to train the model
is it because I need to allocate even less VRAM or because the accelerate command missing?

[dusky bone]

Stuck here

[native loom]

hi, how can I prevent Pinokio from downloading basic models when it starts?

[naive dragon]

i need some help with putting in the %HIP_PATH%bin i have no idea where this is doesnt seem to be in the bin folder for me ;/

[naive dragon]

is this what i put there

[naive dragon]

correct me if im wrong ofc but i think this is right 😬

[naive dragon]

i have a rx 6700 xt do i download this bcs well it is under the rx 6800 but i dont see specifc stuff i have to download extra 😅

[naive dragon]

sry for the message bomb 🙏 take your time ofc 😊 i appriciate the help

[feral cedar]

I am having a strange case of the webui not communicating with the server

can anyone help?

[nimble stump]

guys can someone link me a video how to download stable diffusion?

theres a lot vids on the youtube but idk which one is the right one

just so i install it correctly

im new to sd

[mortal compass]

Do you want the regular SD or SD Forge? If you don't know what forge is, Forge is like an upgraded version of SD with extra settings and features. I like it a lot, and I'm also a new user

[mortal compass]

Anyway, I like this guys videos. Pretty straight forward. https://youtu.be/zqgKj9yexMY?si=YkX8_wfTUbIwCLWR

[nimble stump]

thxxxxxxx

[mortal compass]

Anytime

[nimble stump]

does it have stable diff 3?

[ornate elk]

Checkout my install guides in the pinned messages of this channel.
There you find the latest guides

[mortal compass]

It's not out yet

Only on API

[nimble stump]

ohh ok

[mortal compass]

Yo @ornate elk, do you know anything about AttributeError: 'str' object has no attribute 'decode'?

Trying to generate in img2img and whatever I do in settings I keep getting this out of nowhere

[ornate elk]

Would need to see the full cmd screenshot

And the settings of img2img

[mortal compass]

Sure

[ornate elk]

Yes these paths are correct 👍

[ornate elk]

Yes you download the first rocm library there. Ignore the other two links and source.
Then follow the step further

[mortal compass]

That should cover it 😅

[ornate elk]

Is this img2img batch or is this in Extras tab?

[mortal compass]

Ah yes this is the extra tab batch

[nimble stump]

can stable diff generate videos too?

[ornate elk]

Okay have you tried without reactor to check it its the cause?

Or try with a single image before batch

[mortal compass]

Huh, I just did, and it seems to work when I deactivate Reactor

Might just reinstall it then and see if that fixes it

[ornate elk]

Yes or try with an face image input instead of a face model

[mortal compass]

That actually worked!

So the face model files are gone I reckon

They are not gone lol, they are where they should be

Its working now, I just had to remake the safetensors files

Not sure what happened to them tho

[ornate elk]

Did you converted them before?

I never used face model files

How big are they?

[mortal compass]

Oh its very dummy proof

you just load several images into the tool, name it, and click build and save

then it saves to the models/faces folder in reactor

[ornate elk]

Ohh sound easy, what tool is that?

[mortal compass]

ReActor, https://github.com/Gourieff/sd-webui-reactor, very decent face swap extension

[nimble stump]

@mortal compass i followed the tutorial vid but im stuck there, what should i do?

[mortal compass]

Thanks for your guidance CS:)

[mortal compass]

This looks like a help command, like "nameofsoftware" --help

What is above?

[nimble stump]

i opened run.bat like he did in the vid

for him its downloading something

[mortal compass]

can you give me the time stamp

[nimble stump]

it says couldnt install pip

timestamp is 1:30

any idea?

[mortal compass]

Sry had to go to wc xd im checking

[mortal compass]

Right, so, can you tell me what you did step by step

From the start

[nimble stump]

ok so i downloaded the web ui forge then i made folder named forge

extracted it there

deleted the rar file

updated the bat file

then ran the run.bat file

is it maybe because i had python already installed?

[mortal compass]

That might be the problem

[nimble stump]

should i delete the python?

[mortal compass]

Yes, uninstall python and git if you also have that

[nimble stump]

okk

[feral cedar]

by webui is not communicating with the server help

[mortal compass]

this version installs both automatically

[nimble stump]

ohh

[mortal compass]

Always read the fine print 😉

[nimble stump]

which version of stable diff is it?

the newest one?

[humble cobalt]

hello, my webui stopped working correctly suddenly. Like when i launch and i want to generate an image it just say "connection errored out" and i need to close the webui, there are no errors in logs and im confused

[mortal compass]

This is something called a "fork". SD is open source, free for anyone to use. So other programmers took the code and added a bunch of stuff and improved it, making it their own version, if that makes sense.

[nimble stump]

ohh yea ok

[mortal compass]

but you can use both SD 1.5 and SD XL models on it, you just need the checkpoints

@feral cedar @humble cobalt sry guys that's out of my knowledge 😦

[feral cedar]

is rocm6 not supported yet?

for a1111 webui

[nimble stump]

now it says couldnt launch python

nor git

[mortal compass]

I have no clue, I'm very novice when it comes to A1111, only been playing around with it for a month

[ornate elk]

What does the cmd shows when launching the webui-user.bat?

[ornate elk]

whats your gpu? and what OS

[mortal compass]

Okay, so, you uninstalled both right? Then you ran the run.bat file?

[humble cobalt]

`venv "C:\Users\lyzux\OneDrive\Bureau\stable-diffusion-webui-master\venv\Scripts\Python.exe"
fatal: not a git repository (or any of the parent directories): .git
fatal: not a git repository (or any of the parent directories): .git
Python 3.10.9 (tags/v3.10.9:1dd9be6, Dec 6 2022, 20:01:21) [MSC v.1934 64 bit (AMD64)]
Version: 1.9.0
Commit hash: <none>
Launching Web UI with arguments: --xformers --medvram-sdxl
Loading weights [89fa3f3b51] from C:\Users\lyzux\OneDrive\Bureau\stable-diffusion-webui-master\models\Stable-diffusion\xlsusjelomixPony_v10.safetensors
Running on local URL: http://127.0.0.1:7860

To create a public link, set share=True in launch().
Creating model from config: C:\Users\lyzux\OneDrive\Bureau\stable-diffusion-webui-master\repositories\generative-models\configs\inference\sd_xl_base.yaml
Startup time: 12.4s (prepare environment: 3.3s, import torch: 4.2s, import gradio: 1.1s, setup paths: 1.3s, initialize shared: 0.3s, other imports: 0.7s, load scripts: 0.7s, create ui: 0.4s, gradio launch: 0.4s).
Applying attention optimization: xformers... done.
Model loaded in 12.8s (load weights from disk: 0.7s, create model: 0.7s, apply weights to model: 9.7s, apply half(): 0.1s, hijack: 0.1s, calculate empty prompt: 1.2s).`

[nimble stump]

yes

should i download python and git again?

[mortal compass]

Ooh hold on I think I might know

Or wait nope I don't know lol

[ornate elk]

okay, you need to reinstall the Webui completely, because you installed it the wrong way (by downloading it as zip) and you installed it in a OneDrive Synced folder which causes issues. Also it never got updated.
Follow my Install guide from here and you can then move your models over after that.
Install guide: #🤝｜tech-support message

[mortal compass]

Maybe the video wasnt so good after all.. I'll look up another one that I liked, that guy was very noob friendly, hold on

[ornate elk]

or let him install the default webui from auto1111 with my guide

[mortal compass]

that might be easier xd

[humble cobalt]

okay thanks i'll try

[ornate elk]

@nimble stump Follow this guide if you have an nvidia gpu:
#🤝｜tech-support message
its also in the pinned messages of this channel.

[mortal compass]

I think you should follow CS1o's directions, he knows what he's talking about

[nimble stump]

but does it contain the forge version? as @mortal compass said the better one

[ornate elk]

no, it covers the Automatic1111 webui. which forge is an alternative of.
forge is only better if you have a GPU with less than 6gb vram.

[nimble stump]

but how much better is it?

[ornate elk]

its not really better, some stuff wont work on forge which works on auto1111

but its faster for low vram gpus

[nimble stump]

okay thx

[mortal compass]

You can get many of the settings and features without forge, you just need extra extensions and download extra files. You can install forge later when you have a bit more knowledge, it won't go away

[nimble stump]

thxx

[ornate elk]

exactly, get Automatic1111 to run and then you can install forge later too, and test it

Automatic1111 webui is the most used Stable Diffusion webui

[nimble stump]

@ornate elk

it says no such file or directory

what now

can i just donwload it manually?

from github?

and put the webui master into my ai folder?

cant download it through git hub with cmd

can someone help pls

[humble cobalt]

well, i tried to follow the guide, but the cmd say that when i want to launch it :

venv "C:\A1111\stable-diffusion-webui\venv\Scripts\Python.exe" Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)] Version: v1.9.3 Commit hash: 1c0a0c4c26f78c32095ebc7f8af82f5c04fca8c0 Cloning Stable Diffusion into C:\A1111\stable-diffusion-webui\repositories\stable-diffusion-stability-ai... Cloning into 'C:\A1111\stable-diffusion-webui\repositories\stable-diffusion-stability-ai'... remote: Enumerating objects: 580, done. remote: Counting objects: 100% (571/571), done. remote: Compressing objects: 100% (306/306), done. error: 4426 bytes of body are still expected | 499.00 KiB/s fetch-pack: unexpected disconnect while reading sideband packet fatal: early EOF fatal: fetch-pack: invalid index-pack output Traceback (most recent call last): File "C:\A1111\stable-diffusion-webui\launch.py", line 48, in <module> main() File "C:\A1111\stable-diffusion-webui\launch.py", line 39, in main prepare_environment() File "C:\A1111\stable-diffusion-webui\modules\launch_utils.py", line 411, in prepare_environment git_clone(stable_diffusion_repo, repo_dir('stable-diffusion-stability-ai'), "Stable Diffusion", stable_diffusion_commit_hash) File "C:\A1111\stable-diffusion-webui\modules\launch_utils.py", line 191, in git_clone run(f'"{git}" clone --config core.filemode=false "{url}" "{dir}"', f"Cloning {name} into {dir}...", f"Couldn't clone {name}", live=True) File "C:\A1111\stable-diffusion-webui\modules\launch_utils.py", line 115, in run raise RuntimeError("\n".join(error_bits)) RuntimeError: Couldn't clone Stable Diffusion. Command: "git" clone --config core.filemode=false "https://github.com/Stability-AI/stablediffusion.git" "C:\A1111\stable-diffusion-webui\repositories\stable-diffusion-stability-ai" Error code: 128

maybe its because i already downloaded webui before and so its doing shit with it ?

[ornate elk]

Seems like a network error

[ornate elk]

Please read my guide carrefully

It says don't install in Documents folder

[humble cobalt]

oh yeah its working now i switched on my 5G

[narrow sun]

i'm having a very bizarre issue where some models work and some do not

as in they refuse to load and just error

i've tried my hardest to google and research the issue but i'm left scratching my head

[ornate elk]

Hey, what does the cmd shows when switching models?

[narrow sun]

this is the most common error

i have a decent ish gpu so i don't think thats the issue

but i could be wrong

[ornate elk]

What's your GPU?

[narrow sun]

12gb i believe

[ornate elk]

In your webui-user.bat you should add --xformers
Then try at again

[narrow sun]

okie dokie

thank you

[ornate elk]

Also make sure your on the latest webui version

[narrow sun]

yep i think so, download it only a few days ago and still no luck

[narrow sun]

sadly the same error

[ornate elk]

The first error is a out of vram error

[narrow sun]

so my 12gb cpu isn't enough?

i just find it weird that i was able to get 2 models to work and generate some quite lovely images but most models i try don't work

[ornate elk]

It should be enough. But make sure you disable and close heave programms in the background like Wallpaper engine or games.

[narrow sun]

oh yes i always do that, nothing open except the cmd window and firefox which is necessary to open the ui

[ornate elk]

Okay and does the error comes from starting or only when switching to the model?
Or when generating?

[narrow sun]

well thats the baffling thing, it happens when i switch to the model

it won't even let me load it in

[ornate elk]

How much normal RAM?

[narrow sun]

how much ram do i have?

hmm

i'm not sure

is it this?

[ornate elk]

Ahh yes thats the issue

[narrow sun]

oh dear

[ornate elk]

8gb is not good for the sdxl based models

To fix the issue you could buy 8gb more RAM.
Or what's easier and cheaper: increase your windows pagefile

[narrow sun]

ooh yes please

that sound good

how big should i make it

[ornate elk]

Min 16000 Max 24000
Make sure its only enabled for the C drive.
And disabled for any other drive.

And then make sure you have at least 15gb free space on C

[narrow sun]

ah thank you, and should the initial size and the maximum size be the same?

[ornate elk]

No min 16000 and max 24000

[narrow sun]

oh i see

silly me

and do i need to restart my pc in order for it to take effect?

[ornate elk]

Yes 👍

[narrow sun]

thank you very much, i shall do that and let you know how it turns out

[ornate elk]

Yea no problem 🙂

[nimble stump]

can i update to 24.0 version?

pip

or no?

[ornate elk]

Not needed

[nimble stump]

ok

after the download, can i use it?

or do i need something more?

[ornate elk]

If its ready it will open up in browser.
Then you need to checkout Civitai.com
There you can download models needed for image generation.
Like realism, anime and more

[nimble stump]

okk tysm!

[ornate elk]

You can download these models and put them into models/stable-diffusion

There will be also files like Lora's these go into models/lora

[nimble stump]

kk thx bro

[ornate elk]

A good starter model is Dreamshaper v8
https://civitai.com/models/4384/dreamshaper

[nimble stump]

just wanna ask, do you get paid here for helping noobs?

also do you have any social media or somethin where i can support you?

[ornate elk]

Hey, nope I don't get paid, but I will setup a github page with some guides where people can donate if they want 🙂

[sinful needle]

i would definitely donate man. thanks for ur time

[narrow sun]

It works! thank you very much for you patience and help

[ornate elk]

Perfect! Np

[nimble stump]

first from the left shows my stable diff version?

1.9.3?

if so, isnt that old? can i update it somehow?

[onyx solar]

1.9.3 is the version of Automatic1111 and is the current release version.

[nimble stump]

so its the newest one?

[onyx solar]

As I mentioned, it is the current release version.

https://github.com/AUTOMATIC1111/stable-diffusion-webui

[nimble stump]

ive seen on yt ppl using stable diff v2

how is it possible

[onyx solar]

Automatic1111 is not Stable Diffusion, it is the user interface for various Stable Diffusion models. Stable Diffusion models come in a variety of versions such as 1.5, 2.x, XL, Cascade, & 3

[nimble stump]

oh okk

[nimble stump]

what if i delete this model?

im not gonna use him and he takes my space

[fallen quest]

You are deleting all the models you have.

[onyx solar]

If you're that concerned with space, then delete it. You could always download it again. But you do need some model to work with.

[fallen quest]

Is auto1111 still auto downloading 1.5?

[onyx solar]

I think so

[fallen quest]

Omg

[nimble stump]

im downloading better ones

im not gonna use this onw

im just asking if this one can be deleted

[fallen quest]

100%

[nimble stump]

ok

[fallen quest]

But first put another model in the folder

[nimble stump]

alr bro

[onyx solar]

"Better" is subjective. There are things you can make that model do that look amazing, like every other model. But in general, other models are easier, more specifically styled, etc.

[nimble stump]

guys i downloaded this model https://civitai.com/models/4499/unspeakable-horrors-negative-prompt

but its in txt file

[onyx solar]

That's not a model.

[nimble stump]

how can i add this to my stable diff?

[onyx solar]

That is only a negative prompt embedding.

[nimble stump]

is it possible to apply it to my stable diff?

[onyx solar]

You need to do some googling & reading to understand the basics.

[nimble stump]

how can i save my prompt?

is there any way?

[nimble stump]

can i make my own preset?

[ornate elk]

Yes, below the generate button is a style dropdown

But you can't safe steps or resolution etc

Only prompt

But you can load images back into SD to get the settings back

[nimble stump]

thank you g

[nimble stump]

yo i made some fire

but the problem is i can render only 1024x1024

if i go higher its not gonna render

can i upscale to 4k or atleast 1080x1920 these?

[amber cloud]

Hello, getting eyes closed on all models all samplers when there are 2+ persons in the image. while when there is only 1 person in the image, all is fine. is that a settings problem ? do i have to increase something or add something ? Thanks!

[ornate elk]

Do you get an error?

What's your GPU?

[high mesa]

anyone problem with gradio?

[nimble stump]

i have 2

first is 4gb and 2nd is 2gb

[ornate elk]

if you use the 4gb one then you need --xformers --medvram --no-half-vae in your webui-user.bat

and then you also need to install the tiled vae extension to get higher resolution

[ornate elk]

nope

[nimble stump]

idk idk cuz my shit lagging and cant even render 1080x1920

1080x1080 sometime can render but most of the time not

1024x1024 works fine

[ornate elk]

you need to know that you shouldnt render natively on these resolutions.
Stable Diffusion has different base models:
1.5 models are trained on a 512x512 resolution
SDXL models are trained on a 1024x1024 resolution

With your card you should use 1.5 based models (these are 2gb files).
SDXL wont work good with a 4gb vram card.
then generate at a lower resolution and then you can upscale images to higher resolutions

[nimble stump]

1024 works fine which is sdxl i believe

just rendered these

should i do 1.5 models only?

[ornate elk]

if you want faster generation speed, then yes.
also upscaling can enhance the image quality

[nimble stump]

thx for info bro