🤝｜tech-support

[final mason]

idk i only downloaded the cetus one and have the base one

[ornate elk]

We can fix it but I dont have the time right now sry :/

[final mason]

all good no worries i can wait

[ornate elk]

I'm on in 1 hour, then I can help

[final mason]

and i tryd it with the base model and its still doing the inf reload

[ornate elk]

You can delete the config.json the ui-config.json and the venv to fully reset everything

[final mason]

kk ty

[willow kettle]

OHhhhh I see. So I could try to hack ReActor's installation requirements if I wanted it to be cleaner...

[willow kettle]

Actually ReActor's install.py seems to have some logic that tries to get onnx right

[final mason]

its fixed now i think so i cant use a differnet vae with cetus mix? to make non hires look better?

[ruby yoke]

don't switch the VAE that comes with any model. switch your samplers and schedulers

[ornate elk]

Exactly

[ornate elk]

How does the images look ?

Are the colors normal or washed out?

[final mason]

blurry kinda

its naked so i cant really show unles i just show the face but i have blurry in negative

not hires i mean i do have the vae extesion thing idk if i should try it

this

[karmic crown]

are you using 1.5 or SDXL?

[final mason]

how do i see that

[karmic crown]

which model do you use?

[final mason]

https://civitai.com/images/1322390

[karmic crown]

you have info here

set Clip skip to 2 (same as in that image)

[final mason]

hires or no

[karmic crown]

try without first

to make a clean image

[final mason]

wheres clip skip

[karmic crown]

you try this lora https://civitai.com/models/171159/sdxlrender

it can help

i like i t a lot

[final mason]

ill try it do i need to enable clip thing in the ui

[karmic crown]

i have webui only in virtual machine at this memont, give me a second to open it up

[final mason]

cuz ik theres a clip thing for ui but idk

[karmic crown]

i will show you. please wait a second

[final mason]

k

[karmic crown]

i was testing my installation script, and i obviously removed venv, so it installing packages now

but it will be finished fast

[final mason]

k np

im just eating while trying to figure this out

[ornate elk]

its called "clip stop at last layers"

[karmic crown]

downloading default model 🙈

[ornate elk]

can you generate a non nsfw image to test?

[final mason]

where at im alittle confused and ya sorry i was making sure my prompts i use worked

[ornate elk]

go into settings, User interface / user interface, quick settings

there add: sd_vae, and clip stop at last layer

[final mason]

ya im there i dont see the clip stop

[final mason]

i have sd vae

[karmic crown]

type stop, click on clip_stop and add it

[final mason]

ah ok

thats weird lol

[karmic crown]

now you will have that dropdown

and you can change it whenever you want

[final mason]

still have this issue not a huge problem tho

[karmic crown]

after reload you should heave clip skip

[final mason]

ya i do ty

[karmic crown]

it took about 5 seconds to reload in a virtual machine 🤷🏻‍♂️

[final mason]

wellp

i might take a break from this unless you know a quick fixc

[karmic crown]

show what you have in cmd/terminal

[final mason]

maybe its this?

idk

cuz i think its saying it dont like the decimal

dont know tho

[ornate elk]

only use clip skip 1 OR 2

not 1.5

[final mason]

rip ok

its working now ill try the lora thing you said

[karmic crown]

use a slider, or type 1 or 2

it must be whole number

i got some great results with 1.5 models and that sdxl renderer lora

CS1o might remember some of those, since i showed them to him

[final mason]

will 2 work on mine?

[karmic crown]

use the setting recommend by the model authonr

and check what they used on example images

[final mason]

oh it says 2 ok i see now

[final mason]

one more thing sorry how do i use the lora thing

i put it in the folder

[karmic crown]

you have examples there

put it in loras folder

you have Lora folder in models

lora:SDXLrender_v2.0:0.51

[final mason]

ya i put it there

[karmic crown]

you are adding this in prompt

lora:lora_file_name:strength

[final mason]

both of those?

[karmic crown]

?

lora:lora_file_name:strength <- this is how you use loras

lora:SDXLrender_v2.0:0.51 <- this in example for this lora

but if your file name is for example SDXLrender, than you need to change the prompt

lora:SDXLrender:0.51

part between : and : must be the same as filename

[final mason]

so this?

or am i stupid

[karmic crown]

what is the filename in lora folder

[karmic crown]

in that case use the original from civitai

[final mason]

think its this

[karmic crown]

lora:SDXLrender_v2.0:0.6

[final mason]

oh

[karmic crown]

do not push it too hard

0.6-0.9

[final mason]

k trying it now

[karmic crown]

always read the instructions left by author

[final mason]

im illiterate lol but ill try to im trying to master this sorry for asking alot of questions

[karmic crown]

i love the idea behind that lora

he took a bunch of sdxl images and trained 1.5 lora on them, in order to improve images generated by 1.5

[final mason]

cool

[karmic crown]

right is the exactly the same image, just with this lora

[final mason]

im trying to get it not blurry but i dont think hires is going to work nope it does and idk really how to use this

[karmic crown]

that looks fine, i haven't use that one in a while

[final mason]

forgot to put these in

i tired it without hires i cant really tell a differernce

see how its blurry

[ornate elk]

mssing vae

[final mason]

?

[karmic crown]

show me what you have here

[karmic crown]

most likely

[karmic crown]

under that

[ornate elk]

download this vae and put it into models/vae
https://civitai.com/models/22354?modelVersionId=88156
then selecte it in the dropdown

[karmic crown]

text after genration

[final mason]

i still dont under stand vae was told it has to be compat with my model?

cuz that broke last time

[karmic crown]

this thing, where i can see your prompt and used resources

[final mason]

this?

[final mason]

sorry i have it in list format

[karmic crown]

this

[ornate elk]

it will work

[final mason]

ok

[ornate elk]

it works with every 1.5 based model

[karmic crown]

that models does not have "baked" vae, so it need external one

use the one that CS1o posted

[final mason]

k trying it

[karmic crown]

@ornate elk wtf is 1.5 Hyper?!!!!!

lcm on steroids i guess, i guess (based on the name)

[ornate elk]

its new and fast ^^ i heard its better than lightning and lcm

[final mason]

might be better i cant really tell

[karmic crown]

ByteDance as expected

[ornate elk]

do you have it selectet?

[final mason]

what selceted?

[karmic crown]

VAE

[ornate elk]

the vae

[karmic crown]

you need to change automatic, to that one

yes

that

[final mason]

think it is slightly better

than the other one

ty both for the help btw

[karmic crown]

this one is gone ?!

try this vae

https://civitai.com/models/276082/vae-ft-mse-840000-ema-pruned-or-840000-or-840k-sd15-vae

[final mason]

ok

[karmic crown]

looks like it is here
https://civitai.com/models/23906/kl-f8-anime2-vae

[final mason]

meh idk about this one

it would prob be better with hires but not idk

[ornate elk]

you also need the adetailer extension

oh wait nope, that would require --no-half for directml

[final mason]

k ty

ah

kk

[ornate elk]

i can show you upscaling in img2img

[final mason]

ok

should i try it with the extesion you said but without hiress ? or is it really only meant for hires?

[ornate elk]

tiled vae is for hires

[final mason]

ok dont think it really works i can try again

[ornate elk]

try upscale by 1.2
set hires steps to 10
and enable tiled vae
then set the vae encoder to 1024 and decode to 128

[final mason]

ok im almost pos it will error me but ill try

what upscaler do i use

[ornate elk]

set that to esrgan4x+ anime6b

[final mason]

the one with the R - in front right

ok

hmm does this look hires compared to this

[ornate elk]

check the resolutions of the images

i also found the good vae for colors and anime:
download the blessed2.vae.pt
and put that into models/vae and select it
https://huggingface.co/NoCrypt/blessed_vae/tree/main

[final mason]

what res do you recommend

and ok

[karmic crown]

768x512 for 1.5 models

or 512x768

[ornate elk]

try if you can upscale by 1.5

[karmic crown]

than you can upscale if you want

[final mason]

doubting it

also what am i looking at with the huggin fface thing sorry wheres the download

[ornate elk]

the little download arrow

on the right

[final mason]

sorry my vission sucks lol

and i cant do 1.5

sadge

[ornate elk]

okay then you need to use img2img for upscaling larger

[final mason]

k so rn i need to focus on getting better prompts for poses then ill do that cuz im getting ok poses but somtimes the legs are bent weird or other limbs like hands are bent'

every thing else is working flawlesly

spoke to soon

idk if that vae is working

with hires 1.2

ok now im not getting it to work with the other one tf did i do

[ornate elk]

everytime you get an out of memory error you have to restart SD fully to clean the stuck vram

it wont work otherwise

[final mason]

ah

i see

[karmic crown]

@final mason i found one of my images i made with 1.5 model
#🏞｜general-with-images message

[final mason]

nice looks good

[karmic crown]

i used this model

https://civitai.com/models/84728/photon

you can try and see fi you are getting blurry images with that model too

[final mason]

well now why am i getting errors now

[karmic crown]

but i am sure i tried the model you are using and that it worked fine

[final mason]

ya its bettter hires ofcours but for somereason its crash ing i got it to work b4

i restarted

[karmic crown]

which one of those you have downloaded?

[final mason]

the first

[karmic crown]

try coda2

https://civitai.com/models/6755?modelVersionId=48569

[final mason]

not sure if i fully restarted one sec

and ok

[karmic crown]

maybe this model?

https://civitai.com/models/23521/anime-pastel-dream

[final mason]

ok im lost ill try that then but i got it to work once now its not

[ornate elk]

would recommend any anime model with 2gb size

also make sure to close every other programms and games and Wallpaper Engine

[final mason]

like steam>?

[ornate elk]

everything that could use gpu

multiple browser tabs

etc

[final mason]

why is it crashing me

it worked b4

[ornate elk]

have you restarted the cmd? (webui-user.bat)

also have you enabled tiled vae ?

that settings is off by default

[final mason]

yes and the 1024 128

typo

why it keep breaking on me

[ornate elk]

can you show the settings?

[final mason]

this is how i had it b4 im pretty sure

[ornate elk]

your webui is broken

[final mason]

😭

what i do

[ornate elk]

make sure its whitelisted in any browser adblocker

also make sure you close the browser and reopen it

[final mason]

ya error again oof ok i might take a break anyways and come back to this

[ornate elk]

alright

[final mason]

been at this all day ill come back later

tyty for the help

[ornate elk]

yea no problem 🙂

[young sierra]

whenever I upscale my pictures too much on stabble diffusion they start looking ugly!

anybody knows how to fix this..?

it looks great but its like ehh idk

the original one is this

[ruby yoke]

use something different to upscale with other than the AI you're generating with

[young sierra]

like what?

[ornate elk]

show us the settings you used

seems like a to high denois

[young sierra]

sure brother

[ornate elk]

thats the original, and now show us your settings when upscaling

[young sierra]

i just upscaled again

i click on this button to upscale

[fading heron]

Thank you for your time making these guides. I ran into problems with step 5. I can't find the cublas.dll and the cusparse.dll files in the ZLUDA directory. I found them in the SD-Zluda directory. I tried using those, but ComfyUI won't launch. A terminal window opens when I use the .bat file, but it closes with a second or two.

[ornate elk]

oh dont do that. it uses the default settings and these are bad

[young sierra]

Oh.. Right

[ornate elk]

you first have to set them in hires fix

denois to 0.5 or lower
hires steps to 10

then you can use that button

if you close hires fix again

[young sierra]

let me try it

OPH

OH SHIIIIIIIIIIIIT

HOLY CRAP

THIS LOOKS AMAZING

let me try to upscale

it still looks weird

but not as bad

by the way i have a powerful pc

if that means anything (rtx 3080 32 gb ram+i9)

[ornate elk]

make sure you have --xformers --medvram-sdxl
in your webui-user.bat for the best performance

[ornate elk]

try set the upscaler t oesrgan4x+ or latent bicubic

[ornate elk]

the button shouldnt do anything because you already generate it with hires fix enabled

[young sierra]

oh makes sense

[young sierra]

alright a second

ill do it fast

dw

this one?

so do i add it like this

this is the one I've opened

[ornate elk]

you already have xformers so you only need to add --medvram-sdxl behind it

between autolaunch and xformers

[young sierra]

like this?

[ornate elk]

yep

[young sierra]

Awesome bro thank you

lemme try this

[ornate elk]

now you can easily use sdxl models with good performance

[young sierra]

Cool by the way

i'd like to make the resolution higher

[young sierra]

quality is great

but its not 2160x1920

like i would want

[ornate elk]

change the resolution to half of it then enable hires fix

upscale by 2

then you have that resolution

it also depends on the model. 1.5 models dont like a to hig hnative res

sdxl models work better with highres

[young sierra]

ok bet lemme try that

lmao i have no idea how to do prompt stuff

i typed something so random

[ornate elk]

okay yea thats an 1.5 model

with that you have to stay nearer to 512x512
you can try 540x960 to get a Full-HD iomage when upscaling by 2

[young sierra]

bet

[young sierra]

it looks a lot less crazy

so id say thats an improvement at least

[ornate elk]

try switching it to widescreen format ^^

but yea its better

the rest is prompting

[young sierra]

how do i do that?

[ornate elk]

click on the double arrow

[young sierra]

ig here?

[young sierra]

Oh ok

when i clicked on it, it did this

if you mean this button

[ornate elk]

yes thats what i meant

[young sierra]

Yup widescreen now, thank you

I'm gonna check on poe.chat

if i can find anything that can translate my words into proper prompts

since my prompts skills are very bad

wow so fucking clean

[ornate elk]

awesome!

[young sierra]

ty

[stark zodiac]

Hi there, I am new here, and I am trying to utilize StreamDiffusion. I am running into an error when making an engine: NameError: name 'StreamDiffusionWrapper' is not defined

Any recommendations on how to proceed? Thanks!

[stark zodiac]

I fixed it

🙂

[ornate elk]

let us know how ^^

[oblique carbon]

Maybe someone can help me. I'm running SD through ComfyUI and I keep getting these error message: "Exception in callback _ProactorBasePipeTransport._call_connection_lost(None)
handle: <Handle _ProactorBasePipeTransport._call_connection_lost(None)>
Traceback (most recent call last):
File "asyncio\events.py", line 80, in _run
File "asyncio\proactor_events.py", line 165, in _call_connection_lost
ConnectionResetError: [WinError 10054] An existing connection was forcibly closed by the remote host" I think it is due to conflicting dependencies but I'm not sure which ones. Anyone experienced this or have any ideas?

[stark zodiac]

I ran the update dependencies from the Webui a second time and it fixed the issue. I ran into something else so now troubleshooting further.

[ruby yoke]

what version of python are you using?

[oblique carbon]

Python version 3.10. Would there be any benefit in my upgrading to 3.11?

[karmic crown]

it should be the same, but stay on 3.10, since that is a recommended version

[ruby yoke]

this is wha i'm running

[oblique carbon]

Thanks for that, because I was just about to upgrade to 3.11, which co-pilot said in preferable. That'll teach me for listening to co-pilot.

[ruby yoke]

and it's workign just fine

[karmic crown]

a1111 itself should work fine with 3.11, but it is posible that some extension will not work properly

[ruby yoke]

he's on comfy, not a1111

[karmic crown]

i am sorry

in that case go with 3.11

[ruby yoke]

@oblique carbon what version is pip and what version is cuda?

[karmic crown]

just do not use 3.12

pytorch does not work on 3.12 fully

[ruby yoke]

no, i tried that, had to downlevel

[karmic crown]

"PyTorch support for Python 3.12 in general is considered experimental. Please use Python version between 3.8 and 3.11 instead. This is an existing issue since PyTorch 2.2."

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

so 3.11 and 3.10 should work fine

for a1111 3.10 is preffered version, even i had no problems with 3.11 (on Mac)

[oblique carbon]

I will upgrade to 3.11 I think. As for my pip version, it's 24.1.1 I can't find my Cuda version: I was pretty sure it was installed, but I'm now not sure. I'm going to go get the NVIDIA CUDA Toolkit.

[ornate elk]

make sure that not any third party antivirus is blocking the connection or vpn etc

[oblique carbon]

Yes, that was something I considered. I shut down my Norton antivirus, but it kept on happening. And my Cyberghost VPN doesn't seem to be the cause either.

[ruby yoke]

it'll be in your start up log for comfy unless your start up is crashing before it even gets there

[ruby yoke]

get off the vpn.

[ruby yoke]

your entire issue is ConnectionResetError: [WinError 10054] An existing connection was forcibly closed by the remote host"

[oblique carbon]

I am using PyTorch 2.1.1 + cpu. Is this something I should upgrade?

[ruby yoke]

this is mine

[oblique carbon]

That's the same version of pytorch as I'm using. However, I just noticed this: "no module 'xformers'. Processing without..." That is a problem isn't it?

[ruby yoke]

that is a problem

[oblique carbon]

I'll go pip install it

So I installed xformers. There was an very old version running, which it uninstalled. In the process, it upgraded my PyTorch. But then it warned: ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
torchvision 0.16.1 requires torch==2.1.1, but you have torch 2.3.0 which is incompatible." Perhaps I should revert to torch 2.1.1.

[ruby yoke]

downlevel torch

[ornate elk]

what webui are you using?

[ruby yoke]

he's using comfy

[ornate elk]

ah then it already should use the correct torch version by default, but also he should use python 3.10.11

[oblique carbon]

So I reverted to torch 2.1.1, and now it says: "pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
xformers 0.0.26.post1 requires torch==2.3.0, but you have torch 2.1.1 which is incompatible." Should I be running comfyui in a virtual environment?

[ruby yoke]

we already went through that. 3.11 works fine, but he had an old version of xformers and when he upgraded that, pip upgraded torch

[ornate elk]

ah okay

[ruby yoke]

are you running it in a virtual enviroment?

[oblique carbon]

No, I'm not. It's just that both ChatGTP and co-pilot seem to think I should be to prevent conflicting dependencies. I was wondering if this was something people do?

[fair oxide]

Where are outputs from animatediff stored? I can't seem to find them

[ruby yoke]

i think first of all you should close the sessions to copilot and chatgpt, and then you might want to switch over to the #🧣｜comfy-ui channel and see if they can figure out what is going on with your install

[fair oxide]

anyone?

[ruby yoke]

https://education.civitai.com/beginners-guide-to-animatediff/

it says "Outputs will appear in the animatediff/samples/modelname/sample directory."

[fair oxide]

Where is "animatediff" though? I have it installed for automatic1111 in the extensions folder but have no samples folder

[fading heron]

I am having trouble with step 5 of the guide you wrote. Would you be able to give me a hand when your free?

[ruby yoke]

all i know is what's on that page. did you read through it? you might need to create the directory

[fair oxide]

Yeah, been reading through the docs for a while

I'll watch a tutorial I guess

[ruby yoke]

are you on Purz' discord? there's a lot of animatediff people on there

[fair oxide]

nope

[ruby yoke]

i can DM you an invite if you'd like

[fair oxide]

that'd be great, if you don't mind 🙂

[ruby yoke]

i don't mind at all

[oblique carbon]

I think I'm going to format my SSD and have a clean install of Windows, because no matter which version of the various dependencies I install, it seems to conflict with something else. Thanks for the help.

[ornate elk]

sure whats the issue?

[fading heron]

I can't find the cublas.dll and the cusparse.dll files in the ZLUDA directory. I found them in the SD-Zluda directory. I tried using those, but ComfyUI won't launch. A terminal window opens when I use the .bat file, but it closes with a second or two.

I'm using the Install ComfyUI on AMD Windows with Zluda instructions

[ruby yoke]

okay, but first i'd really suggest you talk to the guys in the #🧣｜comfy-ui comfy channel

[ornate elk]

you have to download the ZLUDA folder first, check step 2:

[fair oxide]

where is this from?

[ornate elk]

from my guides:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

[fair oxide]

oh derp I'm looking at the wrong chat thread lmao

[fading heron]

I already have a directory called Zluda on my drive. It holds a folder called Zluda-master on it. Do I delete one and replace it with the one I downloaded.

I had already installed Automatic1111

[ornate elk]

you can do that, but also make sure the folder is added to your environment paths

[ornate elk]

best would be to download the latest zluda files

if your master folder is in path you can also just replace those files inside

so check your environment path first

[fading heron]

Is this correct?

It's my PATH variable, sorry

I followed the steps and it crashes when I launch the bat file.

[ornate elk]

if its in the section for environment variables and not in User variables then its fine

cant see that on this screenshot

[ornate elk]

did it worked?

[crude halo]

Any idea why SDforge says I have CUDA 12, when my system says 11.8? Apparently it breaks reactor 🤔

I tried to delete and rebuild forge's VENV but it still says CUDA 12

[ruby yoke]

any possiblity you have more than one version of cuda installed?

[crude halo]

It's possible, I messed up a lot. Do you know how I can check/uninstall them all?

[final mason]

its still doing the runtime error @ornate elk from last time idk but im too tired i might have to fix it tomorrw

[crude halo]

hmmm I dont see cuda 12 in the windows app list, only 11.8 🧐

[final mason]

wondering if this is usfull

[crude halo]

now it says it can't find cuda even though it's installed... For some reason it's missing from the venv directory created by forge 🤔 I tried adding my cuda install to the path but no effect

[karmic crown]

you messed up your python

1. what CUDA version have you installed?
2. what SD environments do you have?
3. which Python versions have you installed?

i have a feeling that your answers for 2 and 3 will contain more than 1 answer

[crude halo]

oh so it can come from python... I was afraid of that

I installed many different visual studio programs to try different AIs

Im supposed to have cuda 11.8 but nvidia-smi returns 12.5, even though I just uninstalled cuda from the control panel... Im currently trying to uninstall all nvidia drivers

I should have python 3.10.7

[karmic crown]

what is your os? win?

your torch version should match your installed cuda version (i am almost sure, since i am a mac user)

but i am usually right 🙂

so you have only two choices 11.8 or 12.1 as far as i can tell

if your system have 12.5 than you have newer version that does not work with torch, which will explain your problems

try to downgrade cuda to apropriate version

https://developer.nvidia.com/cuda-toolkit-archive

[tired mortar]

Hello, my text2img with control net is incredibly slow. Usually it only takes me a few seconds to create a regular image, but with control net open-pose, the generation takes up to 25 minutes which is incredibly long. Is there anything i could do?

[ornate elk]

Hey, what's your GPU and what's inside your webui-user.bat?

[tired mortar]

im running a 1070 6gb

set COMMANDLINE_ARGS=--autolaunch --xformers --listen --opt-split-attention --no-half

[ornate elk]

Remove --opt-split-attention and --no-half
And add: --medvram-sdxl

A 1070 has 8gb vram

[solemn moon]

i saw my git pull did alot of new things

wondering if sd3 is available now on auto?

[ornate elk]

With the next stable release.
Yesterday was dev release of 1.10.0

[solemn moon]

awesome, cant wait, will be slower for me im guessing but looking forward to it

[ornate elk]

Why should it be slower?

[solemn moon]

i mean xl is slower than 1.5, so jsut following the trend 😅

will still be using it though regardless

im getting this message pop up, not sure what to do

stable-diffusion-webui-master\venv\lib\site-packages\diffusers\models\transformers\transformer_2d.py:34: FutureWarning: Transformer2DModelOutput is deprecated and will be removed in version 1.0.0. Importing Transformer2DModelOutput from diffusers.models.transformer_2d is deprecated and this will be removed in a future version. Please use from diffusers.models.modeling_outputs import Transformer2DModelOutput, instead.

[ornate elk]

That can be ignored

[solemn moon]

okei thanks 🫡

[hazy salmon]

yoo, i need some help. Im using Automatic 1111, and trying to run a Lora (https://civitai.com/models/159333?modelVersionId=443092), but I cant seem to get it to work.

At first it wasnt even showing up so I went into settings, and toggled hidden directories in Extra Networks.

This made the Lora show up, but it had no effect whatsoever on generation when prompted

Then I went into all settings and toggled "use old lora" setting, and this just straight up gave me an error when trying to generate an image (RuntimeError: mat1 and mat2 shapes cannot be multiplied (154x1024 and 768x320)

I dont know what to do next, can anybody help or point me in the right direction?

[ember bough]

hey guys

thinking of getting lenovo yoga 7 pro

rtx 4050 6gb

you think it's enough for sd?

[fading heron]

No, after launching the batch file, a terminal window opens for a second then closes. After that, nothing.

[ornate elk]

Hey, disable everything you turned on.
Loras only will show up and work if a compatible model is selected.
In this case the models needs to be based on sdxl

[ornate elk]

Can you show me the path settings as a full screenshot?

[hazy salmon]

So if I understand correctly. What I need is a different base model to get the LoRA working?

[ornate elk]

Exactly

[hazy salmon]

Oh my god thanks

Lmao that’s very doable

[ornate elk]

Np, always check the info box at the lora page. It mentiones "based on" 1.5 or sdxl

[hazy salmon]

Ohhhhh

I see

That’s weird tho cus the description of the LoRA mentioned a model called Dreamshapers which I was using, but it gave me the error. Dunno, but it don’t matter I’ll just find out that works

[ornate elk]

Dreamshaper v8 is 1.5 based
Dreamshaper XL is sdxl based

[hazy salmon]

ah this makes much sense

[hazy salmon]

Thanks 🙏 I get easily confused with the specifics and tiny details

[ornate elk]

Np ^^

[errant tundra]

ComfyUI: Any input swithes that only load the selected inputs instead of all inputs? Was trying to use ComfyRoll Suites input switches to switch between models and clips and such. But it's loading everything instead of just what's selected which is taking up a lot of memory. I don't want to have to redo a bunch of sections of my workflow everytime I want to switch some things around.

[ember bough]

can you help me with this one please?

[ornate elk]

That card will work with SD, but its not great

[azure magnet]

Hi, how could i use controlnet for SDXL?

[hazy salmon]

it works now!! thank so muchhhhh

[azure magnet]

How much vram should i have to use controlnet XL full models?

[fiery island]

I have an issue, at the end of deforum running I get an error

Stitching video...
Video stitching done in 22.56 seconds!
Deleting raw images, as requested:
[mov,mp4,m4a,3gp,3g2,mj2 @ 00000213cbf25d00] moov atom not found
[ WARN:2@13997.832] global cap.cpp:166 cv::VideoCapture::open VIDEOIO(CV_IMAGES): raised OpenCV exception:

OpenCV(4.10.0) D:\a\opencv-python\opencv-python\opencv\modules\videoio\src\cap_images.cpp:274: error: (-215:Assertion failed) number < max_number in function 'cv::icvExtractPattern'

[karmic crown]

go to deforum's server and ask them there, you will get a much better answer there

[ornate elk]

12gb vram

[azure magnet]

Thanks, the models are on illyasviel's hugginface page right?

[ornate elk]

Yes

[azure magnet]

Ok then last questions: Do you know if regional prompter work with controlnet?

[ornate elk]

Didnt tested it so idk

[azure magnet]

Sorry i meant if you know if regional prompter work with SDXL?

[uncut heron]

how do you use inpainting and controllnet with ConfiUI? do i need to instal them with ComfyUI-Manager?

[ornate elk]

Yes that works

[ornate elk]

Yes

[azure magnet]

Do you know if composable lora works with SDXL?

[ornate elk]

Dont know

[azure magnet]

Do you know if LLuL works with SDXL?

[uncut heron]

are there any good ones ?

[fading heron]

hope this helps.

[ornate elk]

Dont know sry, but checkout the "Auto1111 alternatives" in the comfyui manager

[ornate elk]

I dont think so

[ornate elk]

Looks good

Is your driver updated?

[azure magnet]

Ok last question i promise. Do you know if animatediff works with SDXL, and works well or should i avoid it?

[fading heron]

is this the proper ZLUDA file for windows amd?

the zip file?

[ornate elk]

It works. I used it. But it requires a lot of vram

[ornate elk]

Yes

[azure magnet]

Is it consistency?

[fading heron]

I did the "AMD HIP SDK for Windows" update

I just updated my card's drivers and tried the batch file again without any luck.

[umbral nest]

Friends, I have RTX 3050 VRAM 4 laptop, does it work, does it work Stabel Diffusion

[ornate elk]

can you delete the venv folder again and relaunch?

[ornate elk]

hey, yes it does

follow my install guide from the pinned messages

[fading heron]

which one? I have one in the "ComfyUI" folder and one in the "stable-diffusion-webui-amdgpu" folder in the SD-Zluda directory

[umbral nest]

Thank you

[ornate elk]

the comfyui folder

ill guess auto1111 still works for you?

[fading heron]

it does, yes. I've been thinking of wiping it all out and doing the comfyui install, but I wanted to be able to use both to compare.

[fading heron]

it does, yes. I've been thinking of wiping it all out and doing the comfyui install, but I wanted to be able to use both to compare.

[ornate elk]

yea im running both too

with an 7900xtx

[fading heron]

I"ve been very happy with this card.

[ornate elk]

yep me too, its a beast!

pared it with an 7800X3D

[fading heron]

damn, nice

[ornate elk]

try deleting the venv folder. Then go into the comfyui folder, click in the explorer bar
type cmd and hit enter
then type
git pull
then relaunch the Start-Comfyui.bat

[placid torrent]

Yo

When I use SD3 and A1111 which model do I need?

Do I need to activate this?

[ornate elk]

nope you dont need to activate this.
and you need the SD3 medium 5gb model

[placid torrent]

oh okay thx you but why I dont need to activate

[fading heron]

Git pull returned "Already up to date". Batch file closed after running. Nothing else opened.

[ornate elk]

because the 5gb model doesnt need it.
if you want to use the 4gb model then you need to enable it + downloading the 5t file which is 5gb

[placid torrent]

oh okay but in comfy ui I needed to download the clips manually but the t5 not

[ornate elk]

ok

i had to download all of them in comfy

[placid torrent]

mhh okay so I only need the medium model without activating t5 inside a1111

the first or second one?

[ornate elk]

can you edit the Start-Comfyui.bat and add the "pause" like in my image here:

then save and relaunch

[ornate elk]

the incl_clips

[placid torrent]

ah thx but why I dont need the text encoder activated

[ornate elk]

because this model version has them already included, thats why its named incl_clips = inclusive clips

[placid torrent]

oh thx you

but its strange that this model is smaller than

[fading heron]

Done

[ornate elk]

np, but if you have a gpu with 12gb or more vram you can also enable it and download the t5 clip if you want

[placid torrent]

I have 24gb

[ornate elk]

and you deletet the venv before?

[placid torrent]

So I can use the same model with t5 then?

where do I get the clip?

[fading heron]

I did earlier, I'll delete it and try again.

[ornate elk]

https://huggingface.co/stabilityai/stable-diffusion-3-medium/tree/main

[placid torrent]

I already have the f16

lal

any recommended settings for sd3

to make widescreen

[fading heron]

After venv had been deleted.

[placid torrent]

why is my hedgehog fat

[ornate elk]

okay open up a cmd and type
pip cache purge
then delete the venv folder again and rerun the Start-Comfyui.bat

[fading heron]

I can't run the command, nothing recognizes the pip command

[ornate elk]

then you need to install python 3.10.11 and check "add python to path"

[placid torrent]

mh

[fading heron]

I have it installed and had to manually add the path to the PATH variable.

[ornate elk]

the nru nthe installer again, click "modify" click "next" and then check "add python to environment variables"

[fading heron]

Done. I was able to run the pip command in the terminal. Had to again add the python path. Still same crash after deleting venv and running the batch file.

[ornate elk]

crash?

[placid torrent]

Any reason why my images are unsharpen

[ornate elk]

is your AMD Driver updated? are the zluda files inside the ZLUDA folder on C ?

[fading heron]

I updated my drivers earlier. I am wondering about the zluda file. I made it from the zip file, but it is mostly .dll files with a single .exe file. Looks different to me from when I installed for Stable Diffusion

[ornate elk]

the inside of C:\ZLUDA\ folder should look like this:

[fading heron]

ok, that's what mine has.

[ornate elk]

@fading heron can you show me the inside of this folder?

[ornate elk]

like i thought these are the wrong torch files

thats why it doesnt work

[fading heron]

great, what do I need to do?

[ornate elk]

delete the venv folder.
then edit the comfyui start bat
replace this line
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
with this:
pip install torch==2.3.0 torchvision==0.18.0 torchaudio==2.3.0 --index-url https://download.pytorch.org/whl/cu118

then save and relaunch

[fading heron]

It Launched!!!

Thank you, your amazing

does it usually take a long time for the first job?

[ornate elk]

yes

like with anything that runs with zluda, it will take a long time for the first run

but after that its normal speed

[fading heron]

understood. Thank you again.

[ornate elk]

i can also send you a workflow for fast SD3 gen if you need one

[karmic crown]

you do not need torchaudio for a1111, but ok 🙂

[ornate elk]

its comfyui

[karmic crown]

i do not know that they have something that use audio either

blah... torchaudio is in requirements for comfy

[ornate elk]

comfyui supports stable audio

[karmic crown]

really?

that explains a lot

[ornate elk]

there is a node for it yes

https://github.com/lks-ai/ComfyUI-StableAudioSampler

[karmic crown]

i need to try that, but suno spoiled me

[ornate elk]

you also need to manually replace the files in the torch/lib folder

the 2 zluda files (cublas and cuparse) like stated in the guide

[ornate elk]

@fading heron let me know if it worked ^^

[fading heron]

Sorry, Afk for a bit. It crashed the first time, then I copied the two files you said. I am now trying again.

[ornate elk]

ok make sure you renamed them before copy

[fading heron]

I did. Looks like it processing right now.

It worked!!! Thank you

[ornate elk]

perfect 🙂 np

[final mason]

idk why this is happening i restarted my pc i swear it worked once with the settings you said yesterday @ornate elk

should i give up on hires

[ornate elk]

yea maybe your better off using img2img sd upscale script

[final mason]

ok but id rather do that when i have better prmpts cuz i keep getting like weird bent hands and im trying to get cat girls but idk the tail is being weird

also can you explain how to do the image thng

like do i need to re put the prompts in the image2 thing

[ruby yoke]

you can't prompt that out. you use a fine tuned model or you use a lora

[next frost]

Having a bit of an issue that I haven’t ran into before if anyone could give me a hand?

[errant tundra]

Hi someone told me I could use rgthree's stuff to accomplish something but I'm having some issues

[errant tundra]

please use SHIFT+WIN+S to use snipping tool to grab screenshots of specific content on your screen

much better than a phone pic

[vocal burrow]

welp it's telling you everything

you're using an ooooooold torch

=> --reinstall-torch to update it

[next frost]

I’ll give it a try again, typed the prompt last time on cdm but it didn’t recognize it properly screwed up on my end there

[vocal burrow]

that's an argument not a command line

you put it in your webui-user.bat, on the set COMMANDLINE_ARGS= line

run it once with it and it will reinstall / update torch.

remove it afterwards or you'll be stuck reinstalling (and thus wasting time) at every launch

[next frost]

no luck

[vocal burrow]

you're launching the webui-user.bat aren't you. Not the webui.bat

[ornate elk]

put the image you want to upscale in img2img
then select Just resize or (latent)
then set the resolution to 512x512 (wont be the output res)
set the denois below 0.3
at the bottom select SD upscale script.
then select an upscaler like Resrgan4x+ Anime6b
then generate

[final mason]

ok ty

[final mason]

is it possible to use the same pic but change the pose of the girl?

[next frost]

the red one is the one I've always used to launch sd and the yellow one was the one where I pout the reinstall prompt. not sure if that helps to see where im messing up at, my bad if its a bit frustrating helping me figure out this problem

[vocal burrow]

you're supposed to run webui-user.bat (the yellow one)

This one will call webui.bat (with the flags you've set in it)

if you launch webui.bat directly well.... it will just launch without the stuff you've set in webui-user.bat

(also do yourself a favor and turn on file extensions in explorer)

[ornate elk]

nope thats not so easy

[final mason]

darn

[errant tundra]

omfg losing my mind over a stupidly dumb bug

AHHHHHHH

[next frost]

definitely helped, its updating rn 🫡

thanks fam

[tacit brook]

Could you guys please take a peek at my stability matrix A1111 startup log, and hopefully explain an easy fix?

python related

[azure magnet]

Hi there is someone that could help me with controlnet for SDXL?

[onyx solar]

You probably should be more specific than that with your ask.

[azure magnet]

How can i use the pose in the pose pack with SDXL controlnet automatic1111

[ruby yoke]

upload a refrence image that's in the pose you want your generated image to be in

[azure magnet]

I wanna use the pose on civitai in controlnet for SDXL. What model i need?

[ruby yoke]

do you have a link to the item on civitai?

[azure magnet]

https[:]//civitai.com/models/147743/openpose-hands-on-own-hips

[ruby yoke]

okay so you use this instead of a refrence image. it's for SD1.5 so use any SD1.5model you prefer. and prompt for something like a cute girl wearing a snowsuit - don't prompt for a cute bunny wearing clothes

it says, right there, that the base model is SD 1.5 not SDXL

[azure magnet]

So now i try to search the pose for SDXL

fuck

[ruby yoke]

you could just use controlnet open pose instead of looking for something off civitai

[ruby yoke]

here https://huggingface.co/thibaud/controlnet-openpose-sdxl-1.0/tree/main thats for SDXL

[azure magnet]

I wanna use pose made from something.

[azure magnet]

Need I only the 5GB file?

[ruby yoke]

i'm not sure what you're trying to accomplish but all it is, is a pose. it just makes it eaiser for you to get your prompted character to be in that pose.

[azure magnet]

Just wanna try a standing pose for better result

[ruby yoke]

okay, well - if you use open pose that i gave you the link for, then you can upload an image of someone in the pose you want, or just move the poser into the pose you want, and that'll influance how the subject in the image you generate is standing. you should watch this tutorial https://youtu.be/-bsGo3bSx0M?si=mhuML7tYCE1Adaxl

[scenic seal]

I'm trying to launch Stable Diffusion, but it keeps giving me code 9009 witch is it can't find python, yes I'm checking the bottom box when installing python, and I've tested both install directories and neither of them work. idk what to do

[golden ruin]

hey guys i have a wierd problem right now. I generated images that were 1080x1080 with 40 sampling steps (using auto1111 webui) and it only took me 5 min, i have a RTX 2060. But for some reason the generation takes way longer now. from 5min to 20min. My gpu is at 100% and my drivers are up to date. I even restarted my pc. Can you guys help me?

[ruby yoke]

put python in your system path and reboot your box

[ruby yoke]

what's in your system logs? what else do you have running in the background? when did you reboot last?

[golden ruin]

where can i see my logs. I had nothing in the backroud running. i rebooted after the problem pooped up

[ruby yoke]

system logs - are you on windows or mac?

[golden ruin]

windows

[ruby yoke]

10 or 11?

[golden ruin]

11

[ruby yoke]

you might need to clear the gpu cache. from the solar winds site: https://www.solarwinds.com/resources/it-glossary/windows-event-log

[golden ruin]

i got this error

under this thing

[ruby yoke]

okay, search google for nvlddmkm and read through the hits

[golden ruin]

it says that a file is corrupted

or not installed

[ruby yoke]

search google for "how to clear GPU cache" and try that before reinstalling device drivers

[golden ruin]

ok

[ruby yoke]

make sure you are searching for information on how to clear the GPU cache of the GPU you have

[golden ruin]

i cleared the cache but still have the same problem

[ruby yoke]

i think you might need to contact the tech support of the company that makes your GPU

[golden ruin]

thank you

[golden ruin]

i figured out the problem. If you have a nvidia gpu you have to download the studio driver on nvidias website and restart your pc. it is even faster now.

[little bough]

Hi, I'm trying to use a custom SD3 checkpoint in ComfyUI in a workflow that works just fine with the stock model but when I try to generate I get the error shown in the picture.. when i switch the model back to the stock checkpoint, it generatesjust fine and when i switch to this i get the error, any ideas why this may be? Thanks

[ornate fern]

#🤝

[ornate elk]

Uninstall python and conda if installed.
Then download and install python 3.10.11 64bit
Then delete the venv folder
Then relaunch the webui-user.bat

[errant tundra]

Hi so I am using PonyDiffusionXL V6 and I'm trying to make an image of a female anthropromorphic dragon character that has black scales with a red underglow, but the black scales hardly ever show up and my main issue rn is that I am getting a full dragon head instead of a more anthro/human dragon head and face and I can't seem to correct this. Here is part of my prompt:

(human, anthro:1.5), (Anthropomorphic female dragon:2) with a detailed human like face on a dragon like head, (black scales glowing red from beneath:1.25), and (dragon-like body features:2), laying on back in the water of a cherry blossom circled lake

I've also tried to use the PXL Human-Anthro slider lora but that doesn't seem to help me. I either get full bipedal dragon or a human body with a dragons head or a full on human.
Please ping me in replies

[ornate elk]

Hey, thats something for #📝｜prompting-help
And isnt an technical issue or question

[errant tundra]

I asked there like 9 hours ago and got no help. Figured I might get a quicker response here

for a server with 26K online, figured I would get help faster but nope

but my bad

[ornate elk]

People will answer if they know. I can help you too if you show me your score and source tags in #📝｜prompting-help

[errant tundra]

mind moving over to #📝｜prompting-help then

[fair oxide]

im on amd gpu, do you recommend using zluda or directml?

[ornate elk]

What's your GPU?

[fair oxide]

uhm

let me check

[fair oxide]

amd rx 580 series

[tender eagle]

hello gentlemen

[fair oxide]

i think its not supported anyways

[tender eagle]

i came here seeking help

I have this error that I dont know how to get rid of and its driving me crazy

DLL load failed while importing onnxruntime_pybind11_state

[ornate elk]

Thats currently not supported with zluda.

[ornate elk]

Hey, can you show the full cmd log?

[fair oxide]

but zluda is better than directml right?

[tender eagle]

ok

[ornate elk]

Yes, its faster and uses less vram.
And the 580x can use it but currently in SD there is an error that needs to be fixed.
On github someone with an 580x opened an issue already

[fair oxide]

thanks ill keep my eye on the changelog

[ornate elk]

What's your GPU?

[tender eagle]

I'm trying to use stable with the GPU as the same dude you were talking to

rx 580

8gb

[ornate elk]

Lol

[ornate elk]

This is the issue currently:
https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu/issues/479

[fair oxide]

thanks

[tender eagle]

i know amd its not currently supported by stable

[ornate elk]

So you both should follow my AMD Automatic1111 with Directml guide

To get it working

The guide is in the pinned messages of this channel. Or on the first link of the pinned messages.

[tender eagle]

ok I'll try thanks

[ornate elk]

Np, let me know if you have any questions

[tender eagle]

but just to be sure so that i odnt waste time

can it work on amd cpu?

[ornate elk]

Yes it can also work on CPU

[tender eagle]

gpu i mean

[ornate elk]

Yes it will work on your card

[tender eagle]

ok

is my gpu not supported by zluda?

so I should use directml

[ornate elk]

Yes

[ornate elk]

Yes currently there is an bug with it

Linked above

[tender eagle]

when i install python should i disable path lenght limit?

[ornate elk]

Yes

[tender eagle]

ok

RX 5xxx cards additional need: --no-half

should i add that?

since mine doesnt have 4 digits

[ornate elk]

You can try first without it. And if you get an error (when generating an image) you can add it later

[tender eagle]

the same thing happens

what the FUCK is onnx?

I went online thinking okay maybe i have to install something

but there is nothing

i donwload some weird file that i dont know what to do with

[ornate elk]

Is your GPU driver updated?

[tender eagle]

yeah

[ornate elk]

OK, then open up a cmd
And type
Pip cache purge
Then delete the venv folder again

Then relaunch the webui-user.bat

[fair oxide]

why is my model loading that long

[tender eagle]

ok

[tender eagle]

that can happen

[ornate elk]

What's the model?

[fair oxide]

i got it from civitai

[ornate elk]

Is its a 6gb file?

[fair oxide]

yes

6.5 i think

[ornate elk]

Okay, thats an sdxl model

It requires a lot of resources and most likely won't work properly with directml and 8gb vram

You also need 16gb RAM to load it

[fair oxide]

oh okay

but almost every model is sdxl

what filter do i have to use?

[ornate elk]

SD 1.5

[fair oxide]

is that the maximum?

[ornate elk]

Thats the best for your gpu

[fair oxide]

okay

[ornate elk]

There are a lot of good 1.5 models

[fair oxide]

what are these ones can i even use them

[ornate elk]

Nope these need to be used in an other webui or tool

And pony is sdxl based

Also dont use 1.4, 2.0, 2.1 and SD3

You can use 2.1 but there are way less and not so good models

[fair oxide]

okay

this model is sd 1.5 but its 7gb

[tender eagle]

the same problem

I think its something to do with GPU

because if i insert skip torch cuda test

in the webui

i can use stable but its really slow

but i havent been able to generate anything

i either get a grey screen or connection error

[ornate elk]

Don't use that xD

Make sure they are 2-4gb max

[ornate elk]

Edit the webui-user.bat
Don't use --skip-torch-cuda-test
Instead add --skip-ort
Then delete the venv folder and relaunch

[tender eagle]

looks like this

should i delete everything and add just that line?

or on top of whats already there

[ornate elk]

On top

[fair oxide]

why are my results so ass i used the exact same generation data

[tender eagle]

oh my god

i cant believe it

it opened

let me see it it works

[ornate elk]

To get the same image like on civitai is not alwaye possible but make sure you do it like this:
On the image in civitai click on the blue copy prompt Settings button.
Then paste that into the positive prompt.
After that Click on the blue/white arrow at the right.
It should now apply everything automaticly.
Then generate.

[stiff obsidian]

there are more: load loras and (hidden on civiatai) embed "Civitai Safe Helper (Minor)", now you may come closer or not... ENSD may also play a roll

[mellow spire]

Hey guys, is there an extension that can face swap a video? mov2mov is not working anymore for later versions of stable diffusion. is there an alternative?

many thx

[ornate elk]

Mov2mov can be fixed with that:
https://github.com/Scholar01/sd-webui-mov2mov/issues/146#issuecomment-2054157728

[mellow spire]

thx, now the tab is there, but it's not working

uploaded a video with an image I want the video to be replaced by and it's just like this

[hallow latch]

Hey, I wanna setup a local stable diffusion at home.. but I need a graphic card then. Would a 3060 with 12 GB be okay?

I can find a new one for like 400$ but maybe a used one for cheaper

Or would you recommend more ?

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker

[ornate elk]

A 3060 would be a good card as its price/vram is good

More vram is always better but nvidia pricing is ...

[hallow latch]

Yeah I looked at like a 40xx and the price was like 900$ instead for same ram xD

I'm just gonna use this one for chat gpt, stable diff and decoding for Plex. Maybe something else later but that's my aim for now 🙂

[coarse ridge]

comfy ui manager

when installed just this line below shows up

any ideas on how to fix?

https://github.com/ai-dock/comfyui

using this docker image

[fading heron]

I am having the same problem with the manager. I installed it using the guide, but it does not show up on the menu like the image shown above

I'm on their tech support site regarding this same problem, with no answers yet.

[coarse ridge]

yeah idk im probably switching to auto1111 for now

i haven't encountered this many bugs ever lol

comfyui was faster for me but what use is it without upscaling, inpainting, controlnet etc

[fading heron]

I am just learning to use this. Started with Automatic1111, I just got Comfyui working. except the manager issue.

[hallow latch]

Is automatic or comfy like better or worse?

Or is it just the theme that is different?

If its like worth switching between them?

[obtuse knoll]

Oh gods python is so damn asinine... I know I had fixed this issue previously, but I don't remember how I did it and an update undid my changes. Obviously I'm not renaming the folder... can someone help?

[coarse ridge]

Comfy is better

[hallow latch]

Do you know in what ways?

[obtuse knoll]

It's not better, it's different, it depends on your personal use case and needs.

I suggest running both and deciding for yourself.

You can use symlinks to not have to double up all your checkpoints and lora.

[hallow latch]

Oh okay, I might have to test it out then. But how do you use it and what is your needs that automatic 111 can't do for you? 🙂

[obtuse knoll]

I use a1111 (forge) because I need to run SDXL with decent speed on only 12GB of vram

[hallow latch]

I mostly just run SD 1.5 models with an 2080ti

[low tree]

Hello !
I generated a chair using stable diffusion, but now I would like to change the color of the chair to black, do you know how I can go about it? THANKS

[hallow latch]

Head over to #📝｜prompting-help 🙂

You can get more help there about prompt

[low tree]

thanks

[obtuse knoll]

A1111 is kind of like an SUV, it'll get you everywhere comfortably and has a lot of space to stuff it with extensions. Comfy is an amphibious F1 car with offroad tires and tank controls. It'll get you to the same places and maybe more, but it'll take you longer to get there and you'lL be bending over backwards to understand how to master the controls.

A1111 forge is an SUV with nitrous and supercharger.

It's faster but you'll crash a lot 😂

[ornate elk]

have you done the steps from my guide to install it ?

[coarse ridge]

real

although im pretty sure there's kinda a lot of stuff you cant really do

in a1111

that's in comfy

altho i dont wish debugging anything in comfy on my worst enemy

[obtuse knoll]

Yea text2gif in A1111 is an unmitiagated nightmare for example

Almost all good SD animations are done in Comfy

[fading heron]

followed them exactly. rebooted and nothing

[fallen escarp]

I'm moving from diffusionbee (v1.7.4) to automatic1111 and for the life of me cannot get the same results using the same generation settings (model, samplers, etc.). Is the application itself such a key factor for the output? I really liked the more vibrant output from diffusionbee and am trying to get similar results in A1111, but haven't come close. Anyone with the same experience?

[ornate elk]

Does the cmd shows anything red?

Make sure the webui is whitelisted in any browser adblocker

[ornate elk]

hey, can you show a comparisson ?
washed out colors are mostly a vae issue

[fading heron]

Nothing in the terminal is red. everything is running great, except I can't see the manager.

[ornate elk]

hmm delete the manager again and git clone it again

[fading heron]

Just did, same issue. I posted the terminal return.

[ornate elk]

can you show the full cmd log?

[fading heron]

That should be it

[ornate elk]

no way, i updated my comfyui stuff an now the manager button is gone too xD

[fading heron]

lol, so now who do you ask for tech support?

[ornate elk]

good question xD

[fallen escarp]

Absolutely. In the image the left column is diffusionbee output (brighter hair), the right A1111 (darker hair). All settings the same, A1111 uses a VAE. It might be me, but although A1111's images might me more realistic, diffusionbee's are more to my liking. And I've been trying to replicate, to no avail...

[ornate elk]

ah seems both okay

if the model is the same i see no issue here

auto1111 uses a vae so the colors are mor accurate and the eyes better

[ornate elk]

i guess its an issue from the comfyui manager

we have to wait for an update

[fading heron]

ok, np. Still have a ton to learn.

[ornate elk]

you can still install extensions by git cloning them into the custom nodes folder

will keep you updated if i get it fixed

[fading heron]

do you know where i can find info on how to use LORAs?

ok, thank you

[fallen escarp]

But do you agree that the left column looks a bit brighter than the right? I can't figure out why it does that with the same models/seeds/settings. Is it the app used having an influence?

[ornate elk]

in comfyui? create a lora node and select them.

[ornate elk]

maybe because it doesnt use a vae, that let it look more pale/brighter

[fading heron]

I put two into the loras directory in comfyui. I added a node, but it is not finding the loras.

[ornate elk]

hmm maybe try linking your stuff from auto1111 to comfyui, in my guide its explained

loras show up for me

[fading heron]

Restarting helps alot. thank you for your help.

[ornate elk]

no problem

[dry ivy]

Question:
Would it be a good idea to upgrade Python from 3.10.6 to 3.10.14?

[dreamy current]

Is it normal that when i put the pony model on my forge sd it needs 237s to load it and that it uses almost a 100% of my memory?It also says that i have no embedding which is false .I have a NVIDIA GEFORCE GTX 1660 SUPER (6 VRAM)

[sonic vortex]

hello guys, i could need some expertise on setting some stuff up, is someone willing to help me?

[ornate elk]

better to upgrade it to python 3.10.11

[ornate elk]

yes thats normal

[ornate elk]

if you provide more information, sure

[dreamy current]

Even the embedding part?

[ornate elk]

yes, because an embeding will only show when a compatible model is selectet. so you have 1.5 embeddings, and they wont show up when sdxl or pony is selectet

[dreamy current]

But i see people using 1.5 embeddings on pony, how is that possible?

[sonic vortex]

i want to set up slxl for img2img editing, already installed the automatic 1111 webui, but the thing is, that i am not getting good results on img2img generations, i cant really find a good setup tutorial 🤷‍♂️

[ornate elk]

thats not possible, i mean they can add it to their prompt, but it wont do anything

[sonic vortex]

addtional information, for img2img, for now only the sd2.1 model works: v2-1_768-ema-pruned.safetensors

[ornate elk]

sdxl and 1.5 should work normally in img2img

whats the exact issue?

maybe you can show an example

[dreamy current]

CS1o should i enable fp8 for sdxl?

[ornate elk]

yes

makes sense for your gpu

[sonic vortex]

i basically put a photo of a person in, as promt something like futuristic city + futuristic clothing and i got this, used DPM++ 2M KARRAa, cfg on 7, DS 0.45

[ornate elk]

what settings are you using?

[sonic vortex]

addition to this, trying the xl model gives me the: NansException: 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.

[ornate elk]

whats your gpu? and whats inside your webui-user.bat?

[sonic vortex]

GPU is a 1070ti 8gb

[ornate elk]

i mean the img2img settings, like resolution, steps, etc

[sonic vortex]

something close to this

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--medvram-sdxl --xformers --no-half-vae --no-half --precision full --disable-nan-check --api --opt-split-attention

call webui.bat

[ornate elk]

then you need to edit the webui-user.bat
at the line COMMANDLINE_ARGS=
you need to add: --xformers --medvram-sdxl --no-half-vae
then save and relaunch the webui-user.bat

[ornate elk]

please remove --no-half --precision full --disable-nan-check --api --opt-split-attention

some of them slow down and wont help if any errors occur

[sonic vortex]

so only this ones?

[ornate elk]

yes

my old GTX1080 runs with just these

[sonic vortex]

while starting, i get the message, that theres no module named xformers, should it be like that?

[ornate elk]

no

make sure you run the webui-user.bat

not the webui.bat

[ornate elk]

if your using sdxl models the required resolution is 1024x1024 because thats what sdxl is trained on

[sonic vortex]

ok i see, the xformer got installed, can i start the webui from now on?

[ornate elk]

only and ever start the webui-user.bat

not webui.bat

[sonic vortex]

is there a specific reason why, in all solutions i got told to use that upon usage

[ornate elk]

yes the reason is simple,
if you start the webui.bat (which hasnt --xformers --medvram-sdxl --no-half-vae) in it, all these optimisations wont get load and used for image generation, making it slow or even crash

they only work when launched with the webui-user.bat which is the file every webui user should use

[dreamy current]

Is it normal for my forge to generate a 832x1216 image using pony to take 6.8 s/it?

[ornate elk]

yea pony and sdxl are resource hungry and your gtx1660 doesnt support half type precision (fp16).
using fp32 for generation makes everything really slow and using even more vram that this card not has

[dreamy current]

would you recommend me to stop using pony and use another model instead?

[ornate elk]

yea with that crd i would stick to 1.5 models and then use hires fix to upscale them. faster generation and also good possible quality

[dreamy current]

And what kind of card would suppport that with pony?

[dreamy current]

But i want to use some loras from pony

i guess ill just stick with sd 1.5

[ornate elk]

every newer gpu than a GTX1660

[dreamy current]

I see, thanks

[frigid notch]

Is there any way of adding in new ui taps on A1111? VAE hypernetwork ClipSkip?

This is mine btw.

Cause I'm seeing others having those around like this. is there anyway to acquire these? Or which is recommend to use?

[ornate elk]

@sonic vortex let me know if you tried it again. make sure to set the resolution to 1024x1024

[sonic vortex]

working on it, getting some results, working with the prompt and settings rn

is there something youd recommend me more rathen then these? @ornate elk

[frigid notch]

oh yeah, just getting started with my new 3090 btw.

[ornate elk]

Looks okay, but you can also do 30 steps

[ornate elk]

Also how much RAM do you have to store that much models in it?

[sonic vortex]

vram or normal ram?

[ornate elk]

Yes go into Settings / User interface / Quicksettings, there add sd_vae, clip_stop at last layers etc
Then apply and reload

[ornate elk]

Normal

[sonic vortex]

in total 16gb for my os

python uses around 9gb atm

[ornate elk]

You should decrease the model amount that stores in ram than

[ornate elk]

Set the two top sliders to 1