🤝｜tech-support

[ornate elk]

i can show you an example, give me 2 mins

[nimble stump]

also, why some lora models wont show in my stable diff?

i put the lora models to lora folder

but some dont show up

[ornate elk]

because Loras are either made for 1.5 or sdxl models.
so they only show up when a compatible model is selected

[nimble stump]

ohh damn

[ornate elk]

on civitai you see the for which model the lora is made

always check the info box

there it says "Base Model 1.5"

[nimble stump]

oh yea

[upbeat kettle]

hi

[nimble stump]

hi

[upbeat kettle]

i am Mr capybara

[ornate elk]

hi

[upbeat kettle]

hi

[fair oxide]

its good to see some ppl on youtube are leveraging AI and uploading remastered 4k videos of the classic rock music

[ornate elk]

oh an important thing is that you should activate FP8 mode for SDXL, in the settings.
that lowers the vram usage of these big 6gb models

[fair oxide]

isn't fp8 redundant compared to fp16?

[ornate elk]

in speed it doesnt change but it needs less vram, so for 4gb vram cards it can be helpfull

[fair oxide]

was looking up ..

[dark viper]

oh damn

I forgor the existence of fp8

[viral knot]

fp4 when

[tiny garden]

hmm... when I do regular human characters, I keep hitting the problem that the pictures are showing some not well rendered pixel artifacts or are generally not well rendered... (see also the example image) what can I do to get rid of them?

[fierce hollow]

What is the difference between Batch count and Batch size

[fierce hollow]

Is there a way to take the same photo from multiple angles? Example: A car, front and side view, relatively speaking, so that 1 image is divided into two parts in the left car in front with the right car on the side

[nimble stump]

how can i do that?

also what is fp8?

[nimble stump]

anyone knows how can i change fp?

anyone knows what could be problem?

yesterday it worked, now it says error everywhere and i cant use it

could be becouse i added too much lora?

[oblique parrot]

I asked an API question in the SD3 channel because it’s an SD3 creative upscaler question. But should I also ask it here?

Not sure which channel is more appropriate for the question.

[vocal burrow]

batch count, how many batch should be completed
bach size, how many images should be created in a single batch

[vocal burrow]

add fp8_storage to your quicksettings list

should show up at the top of the page after that

[nimble stump]

well currently i cant use webui cuz it says connection errored

[vocal burrow]

what does the log says ?

[nimble stump]

any idea how can i fix this?

im noob, where is log?

[vocal burrow]

everything in the console

the black window thingy with white text

you closed it didn't you.

[nimble stump]

yes but dont have anything in log

theres only images folder

and only 1 image i generated

[vocal burrow]

1. if you closed the console then you closed the program
2. there can be no "image folder" as you picture it in a console

that's the console

you have to keep it running when using sd

[nimble stump]

sd model failed to load

is this the problem?

[vocal burrow]

what's your gpu ?

[nimble stump]

i have 2 gpus, 1 is 4gb and 2nd is 2gb

[vocal burrow]

ok, what are your gpus then

[nimble stump]

nvidia geforce gtx 1050 ti and amd radeon tm vega 8 graphics

[vocal burrow]

ok so gtx 1050ti

amd radeon tm vega 8 graphics is a chipset iitc

[nimble stump]

but yesterday everything worked

today i added few more models and it says error

when i take all models out and run it, it still says error

even without all the models

[vocal burrow]

open your webui-user.bat with notepad find the
set COMMANDLINE_ARGS=
and turns it into
set COMMANDLINE_ARGS=--xformers --medvram

[nimble stump]

i do have it

i followed @ornate elk s tutorial

[vocal burrow]

ok then copy paste the full content of your log into a .txt and dump said .txt in there.

[vocal burrow]

I'll look at it after doing my errands

[fallen quest]

Successfully running WebUi with SDXL and loras on Web UI with 4gb of vram must come close to a miracle.

Does it work if you switch to a 1.5 model?

[vocal burrow]

full log with an error

[nimble stump]

this is the full log

dont have more there

[vocal burrow]

trigger the error

[nimble stump]

i did

this what it shows

even if i trigger the error

and if i press any key, it closes

what should i do?

can someone help pls

[fallen quest]

Sorry, can't help you with your error. But I have some advice: comfyui and stable swarm are way less vram hungry and much more optimized. That might be a better option for a machine with 4gb vram.

[ornate elk]

How much RAM do you have?

It seems like you tried to load a sdxl model

Remove the sdxl models from the models/stable-diffusion folder
And then relaunch the webui-user.bat

Then activate the FP8 sdxl mode in settings

[oblique parrot]

The API documentation says "(Caution: may contain cats)" for fetching a result.

Is that a joke about peoplegeneraitng cat pictures, or is it saying the results may need to be fetched multiple times and concatenated?

[maiden coral]

Hello. Is there a way to make DWPose to extract the pose from an anime image? I can't make it work, but for other ones it works.

[split comet]

Anybody have any info on what exactly brackets and numbers do? Or could provide me a link? thnx

[maiden coral]

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

[maiden coral]

the link above

[split comet]

thanks

[maiden coral]

just find "{}" and you'll automatically get your answer

[ornate elk]

You need to click on the explosion icon in Controlnet

Then you get the pose of the input image

[maiden coral]

I don't see any explosion icon. I only see load picture -> DWPose Estimator -> Preview picture

[ornate elk]

Do you use auto1111 webui?

[maiden coral]

but when I load an anime picture it doesn't load the pose

no

COmfyUI

[ornate elk]

Okay then I can't help :/

[maiden coral]

it works on some picture and on some doesn't

nevermind, I'll try to figure it out, youtube videos didn't help, it always works for them

[vocal burrow]

try asking in #🧣｜comfy-ui (and provide your workflow)

[maiden coral]

finally got it to work

it works only for certain images, it mostly doesn't work with anime images

[slender coral]

Installed posex plugin and it doesn't have window to change the pose

how it have to look like

[maiden coral]

Which one is it? I can't find it in the ComfyUI manager, but I want to try to implement Openpose, let me check it. Try that one.

[slender coral]

Wanna use it instead of getting pose from image because of control pos return allways black image

[maiden coral]

it needs to be an image that's real like, it doesn't work too much on anime images

sometimes it works, sometimes doesn't

[fierce hollow]

Is there a way to take the same photo from multiple angles? Example: A car, front and side view, relatively speaking, so that 1 image is divided into two parts in the left car in front with the right car on the side

[maiden coral]

https://www.youtube.com/watch?v=iHt-dQER_Po&ab_channel=MemorySlashVisionstudios check this video, it might help

oh wait, car, that's different, maybe try to make multiple clips

[fierce hollow]

Alas... I don't speak English and I'm new to SD

[nimble stump]

hey guuys how can i fix errored connection?

please help someone

[maiden coral]

I don't know, but I had it in A111. I transferred to ComfyUI. What are you PC specs? Such as GPU (+VRAM), CPU, RAM

[nimble stump]

gtx 1050 ti

[maiden coral]

tell me all the specs I'd written out

[nimble stump]

ram is 8gb

[maiden coral]

that's too low

[nimble stump]

but yesterday everything worked perfectly

today saays errored connection

[maiden coral]

you need at least GTX 1650 4 GB VRAM (but better have something like 12 GB VRAM and GTX 3080 or 3070 or a higher model)

[nimble stump]

im tellin u, yesterday it worked

[maiden coral]

yeah, I had that too, I tried to figure it out and even consulted here but nothing helped, maybe if someone knows the solution

but I couldnt' find anything on the internet

I know

I had A111 too

[nimble stump]

yes its annoying

im tryin to fix this all day

[maiden coral]

ComfyUI doesn't have these problems, or maybe Forge if you want something like A111 but with a better memory management

but your RAM is too low, 16 GB is minimum

[nimble stump]

too low for what tho

[slender coral]

haven't knew it! Thanks!

[maiden coral]

for image generation, look up the specs for AI img generation with stable diffusion, SD 1.5 has generally lower requirements

[maiden coral]

yeah, figured it out today when I saw only a black picture preview

[nimble stump]

im tellin u yesterday i generated images without any problem

[maiden coral]

I wrote you before that I had the same error but couldn't figure it out. Maybe someone who knwo A111 better will know.

[nimble stump]

hmm

ive seen ppl on youtube fixing this problem by deleting venv folder

but that didnt work for me

[ornate elk]

I already answered you how to fix it

[maiden coral]

it's different, did you change anything before?

[ornate elk]

Remove the sdxl models (6gb) from models/stable-diffusion folder.

Then restart

[nimble stump]

alr ill try

[ornate elk]

You can move them to downloads folder, no need to delete them

[nimble stump]

yea yea ik

[ornate elk]

If it works again you need to go into the settings of the webui and activate FP8 mode for sdxl

[past birch]

I'm trying to make a deform video and I cant see the video after I generate it
can someone help me?

[fallen quest]

Have you checked the filepath that it told you (bottom right in your image)?

[past birch]

there is just this file

[nimble stump]

did i do it right?

@ornate elk

now can i use 6gb models?

[ornate elk]

Yes thats ok

Also if you only have 8gb of RAM you need to increase your windows page file

[nimble stump]

how

can i do that

will it strenghten my generator?

[naive dragon]

just a quick question i use both okada voice changer and the stable diffusion not at the same time but one after the other and it seems that both have a heard time starting up now bcs the voices are loading slower and the images are getting generated way slower in the beginning is this true or am i trippin?

[nimble stump]

how can i increase my windows page file?

[ornate elk]

Yes

I can give you a link

[nimble stump]

after i do that, then can i use 6gb models?

[ornate elk]

Yes it should work.
https://www.thewindowsclub.com/increase-page-file-size-virtual-memory-windows

[nimble stump]

alr tysm!

there it shows i have 8gb ram

but there it shows i have 109802mb

[vocal burrow]

cause that's not your ram. that's the size of your pagefile

[nimble stump]

how much should i give to pagefile?

[vocal burrow]

16gb

[nimble stump]

dude i have 8gb ram

[vocal burrow]

dude that's not your ram

it's a file on your disk that will act as ram whenever you ran out of it

[ornate elk]

You need to set it to 16000 min and 24000max

[nimble stump]

alr thxxxx

[vocal burrow]

it will be slow af (especially when you run out of ram and start using that file) but that's the only way for you to run stuff like so

[nimble stump]

thx guyssss!

[ornate elk]

Make sure you have at least 15gb free space on C drive
Also restart the PC after adjusting the pagefile setting

[nimble stump]

i have 90gb free

ok ill restart my pc now

thx again

[ornate elk]

Also make sure you clicked Apply after enabling FP8 mode in the webui

[nimble stump]

i didd

[ornate elk]

Hey, did you get the webui with zluda to work.
I tried okada too and it works normal

[nimble stump]

yoooooooooo 6gb models work noww

but in the lora still it says error

and everywhere else except text2image and image2image

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker

[nimble stump]

do i just copy the url from the webui?

theres just ip address

few numbers

[naive dragon]

i did yes but sometimes it gives an error that it lost connection or something i dont have it now but i can try to make a screenshot when it happens

[nimble stump]

omg it workssssss

[nimble stump]

you gotta be angel

god bless you

[ornate elk]

No problem 😁

[unique bluff]

Hello, I tried opening forge sd and the following error appeared:

    main()
  File "/Users/duda.obladen/stable-diffusion-webui-forge/launch.py", line 47, in main
    start()
  File "/Users/duda.obladen/stable-diffusion-webui-forge/modules/launch_utils.py", line 541, in start
    import webui
  File "/Users/duda.obladen/stable-diffusion-webui-forge/webui.py", line 17, in <module>
    initialize_forge()
  File "/Users/duda.obladen/stable-diffusion-webui-forge/modules_forge/initialization.py", line 50, in initialize_forge
    import ldm_patched.modules.model_management as model_management
  File "/Users/duda.obladen/stable-diffusion-webui-forge/ldm_patched/modules/model_management.py", line 122, in <module>
    total_vram = get_total_memory(get_torch_device()) / (1024 * 1024)
  File "/Users/duda.obladen/stable-diffusion-webui-forge/ldm_patched/modules/model_management.py", line 91, in get_torch_device
    return torch.device(torch.cuda.current_device())
  File "/Users/duda.obladen/stable-diffusion-webui-forge/venv/lib/python3.10/site-packages/torch/cuda/__init__.py", line 803, in current_device
    _lazy_init()
  File "/Users/duda.obladen/stable-diffusion-webui-forge/venv/lib/python3.10/site-packages/torch/cuda/__init__.py", line 309, in _lazy_init
    raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled```

I have an Intel Mac with a intel iris pro graphic 6200 1536 MB as my gpu
Is there any way to fix that error and open forge sd?

[heavy perch]

I still need help with mov2mov tab not appearing

[ornate elk]

Here is the fix for mov2mov:
https://github.com/Scholar01/sd-webui-mov2mov/issues/146#issuecomment-2054157728

[nimble stump]

anyone knows how to make video out of images?

[hexed kestrel]

hello , i have a question, when i use SD 1.5 checkpoint i generate the img very fast but when i use SDXL checkpoint ,( i m on 1111) i can take very very long time to generate the img ; my PC spects : Processeur 12th Gen Intel(R) Core(TM) i9-12900H 2.50 GHz
Mémoire RAM installée 32,0 Go (31,7 Go utilisable)
Type du système Système d’exploitation 64 bits, processeur x64
Nvidia Gforce RTX 3070 Ti

[timid notch]

i got a problem~~can someone help?

i wanna use Temporal-Kit extention,and i am trying to download it,but then i find out that i can not open webui~~

[heavy perch]

already tried that but it didn't work
can i maybe show you in vc

[timid notch]

than i delete Temporal-Kit in extention folder, but the webui still can not open

why~~?can i fix this problem?

[ornate elk]

Hey, you need to edit your webui-user.bat
At the line commandline_args=
You need to add:
--xformers --medvram-sdxl --no-half-vae
Then save and relaunch

[ornate elk]

You need to delete the venv folder and then relaunch the webui-user.bat

[timid notch]

ok i am going to try you way~~@ornate elk

which venv folder?can you tell me the name of the folder?

not working

i think it is need some module?but i didnt delete anything else~

[ornate elk]

Do you have a update.bat in your webui folder?

[timid notch]

yes

[ornate elk]

Run that and then delete the venv folder again

[timid notch]

this update bat？

yes,it is working~~real magic,@ornate elkyour guys really help a lot~thanks

[nimble stump]

any idea how to fix this?

i know i should lower the resolution but 1024x1024 isnt much

and it worked for me till now

idk what happend

is there any way how to clear cuda memory? or make it stronger?

[maiden coral]

do it in lower resolution and then upscale, also don't use hires

[nimble stump]

i tried it now and it destroys the quality

[maiden coral]

also depends on what model you want, SD1.5 was trained for 512x512, SDXL 1024x1024 but can go lower if needed

[nimble stump]

its more blurry and noisy

[maiden coral]

upscaling adds detail, it only depends on what sampler etc. you use

[nimble stump]

dpm++2m

[maiden coral]

Have you tried tiled sampling? Is that option there?

it'll break the img into tiles of lower resolution and then stitch them again together when the image is processed lowering the amount of VRAM you need

[nimble stump]

its like, it makes whole pic loads to 99% and then after it finishes the image disappears and shows that cuda is out of memory

[maiden coral]

try if there's that option, I only know that it works in ComfyUI but I don't have forge/A111 anymore

[nimble stump]

hmm idk but this prolly makes copy of the same pic doesnt it?

[maiden coral]

what parameters you use in user webui bat?

[maiden coral]

no, the process will be slower but you can do higher resolution thanks to that

[nimble stump]

hahahahaha now it rendered again in 1024x1024 with no problem

sometimes it renders, sometimes not

[maiden coral]

Do you use forge or A111?

[nimble stump]

a1111

[maiden coral]

better use ComfyUI or Forge

Forge is better A111

although I haven't tried it it has better optimization

[nimble stump]

first i used forge but some ppl here said a1111 is easier and better

[fading crypt]

What's the trick to getting SD Forge to refer to the A1111 directories? Trying various paths but it won't work REM set A1111_HOME=./stable-diffusion-webui

[maiden coral]

isn't it the other way around 🤔

[nimble stump]

also many things didnt work for me in forge

[fading crypt]

I've tried using an absolute path, too, but it didn't work either

[nimble stump]

mixed opinions here, but more ppl said i should download a1111

[maiden coral]

you can slowly learn with ComfyUI, you just match nodes such as upscaler, sampler etc.

[nimble stump]

how big is comfyui?

gb wise

[maiden coral]

I think 3 GB? at most, even lower

well, my folder is way too big because of models

[nimble stump]

do i have to delete my a1111?

[maiden coral]

you don't have to, but idk how to link models form A111 into ComfyUI

[nimble stump]

ill think about it, now im having fun with a1111

the only problem is that sometimes it renders 1024x1024 sometimes not

[ornate elk]

you need to install the tiled diffusion extension, it can help to generate at higher resolution

[nimble stump]

from the civitai?

[ornate elk]

nope from the extensions tab

[ornate elk]

i can show you how it should look like, give me a sec

[nimble stump]

i dont have anything in the available section

[ornate elk]

click on load from

then you get a list

[nimble stump]

ok got it

[ornate elk]

@fading crypt
here is how the forge webui-user.bat should look like:
you have to change only the first path to yours:
`@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=

@REM Uncomment following code to reference an existing A1111 checkout.
set A1111_HOME=H:/AI-Generator/stable-diffusion-webui
@REM
@REM set VENV_DIR=%A1111_HOME%/venv
set COMMANDLINE_ARGS=%COMMANDLINE_ARGS% ^
--ckpt-dir %A1111_HOME%/models/Stable-diffusion ^
--hypernetwork-dir %A1111_HOME%/models/hypernetworks ^
--embeddings-dir %A1111_HOME%/embeddings ^
--lora-dir %A1111_HOME%/models/Lora ^
--vae-dir %A1111_HOME%/models/vae ^
--controlnet-dir %A1111_HOME%/models/ControlNet
git pull
call webui.bat`

[ornate elk]

then restart the webui, and then activate only the tiled vae option

[nimble stump]

are there more extensions that help rendering?

do i leave it at default when rendering?

[ornate elk]

yep, just activate it should be enough

there are no other extensions that can help lower the vram usage

[nimble stump]

got it, thxx!

[heavy perch]

@ornate elk

[pure kayak]

Hello all, two questions, does anyone have a good installer for sdxl? Also, where would be a good place to ask for a custom Lora or TI? Thanks a lot!

[ornate elk]

hey, in the pinned messages of this channel you find my install guides for the Automatic1111 webui, with that you can load 1.5 and sdxl models, loras, embeddings etc

After the installation you can go on Civitai.com to download custom models and Loras

[ornate elk]

hey, for me it worked

[heavy perch]

did I put it in the right place?

[naive dragon]

this messages pops up now and then and it stops the process and then i normally relaunch it and its fine but if i tried to do anything after the error message came i wont be able to do anything at all

[ornate elk]

yes the m2m_ui script has to be replaced, then you need to restart the webui

[ornate elk]

do you have the full cmd log?

[naive dragon]

😬 ill do that next time sry i pressed enter in the cmd so it closed

[ornate elk]

okay np, the full log can help to figure out whats going on

also do you use sdxl models maybe?

[naive dragon]

ill keep that in mind for sure

[naive dragon]

what is that and how do i find out i am using that?

[nimble stump]

anyone knows what models does liberxx0 use?

would kill to know that

[ornate elk]

there are 1.5 based models (2gb) and sdxl based (6gb) models

[naive dragon]

it is a sdxl then

[ornate elk]

okay, and do you also use hires fix with sdxl models?

[light geyser]

is there any reddit post or something that explains those samplers?

DPM++ 2M SDE SGMUniform
DPM++ 2M SGMUniform
Euler A SGMUniform
Euler SGMUniform
LCM Karras
DPM++ 2M SDE Turbo
DPM++ 2M Turbo
Euler A Turbo

[naive dragon]

ehm i dont think so i didnt download any hires fix specifically

[ornate elk]

Here:
https://stable-diffusion-art.com/samplers/#Samplers_overview

[naive dragon]

i only downloaded 1 model thats it

[fading crypt]

Hmm, I tried that, didn't work. But I did find a fix, I put the arguments on set Commandline_args= line and that did it

[ornate elk]

hires fix is a setting of the webui

[twilit kraken]

eyy yoo I need helpp

[naive dragon]

lol ye i do only have one model i just began recently lol

[twilit kraken]

it saying that its watiing for file to be created and its been spamming that for like days

[ornate elk]

alright ^^ hopefully its not the base sdxl model

[naive dragon]

its not

[ornate elk]

can you show us what you mean? also which webui you are using

[naive dragon]

i followed your guide in the pinned

[twilit kraken]

I downloaded webui 1.7.0

[ornate elk]

oh you used the auto install tool, thats not recommended. Also 1.7.0 is outdated. We are on 1.9.3 now.
better follow my install guide. its in the pinned message of this channel

[naive dragon]

no i dont think i have hires on atleast if its not in your guide i didnt rly change anything from default just the model

[twilit kraken]

which comment? is the pinnd comment lmao

[ornate elk]

okay alright, yea then we have to wait for the error to occur again to get more information

[naive dragon]

🫡 will do

[ornate elk]

are you using nvidia or amd gpu?

then i link you the correct one

[twilit kraken]

nvidia 3060

[light geyser]

it doesn't speak of the turbo samplers or SGMuniform

[ornate elk]

Here:
#🤝｜tech-support message

[ornate elk]

oh true, i didnt saw these samplers anywhere are they from an extension?

[light geyser]

they come preinstalled with webui forge

[ornate elk]

ahhh okay

[fading crypt]

If I train a LORA on SDXL 1.0, it should work with PonyXL, correct? Or do I need to train the LORA with Pony?

[light geyser]

with pony

[heavy perch]

so I need to delete m2m_ui or?

[ornate elk]

in the link i sent you, there someone made a m2m_ui file you need to download and replace the existing one

[fading crypt]

Another LORA question for A1111: is it enough to use lora:Whatever:1 in the prompt or does the LORA trigger word also need to be used?

[twilit kraken]

CS1o I need your help again

[heavy perch]

just did that, still nothing

[ornate elk]

is your webui updated?

and your extensions

[heavy perch]

umm I think so? can i screen share in vc

[ornate elk]

ok

at the webui bottom you see the versions:

[ornate elk]

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

you will need xformers if your on nvidia

[heavy perch]

this?

[ornate elk]

yes

[ornate elk]

right click and edit the webui-user.bat
at the line COMMANDLINE_ARGS=
you need to add --xformers
then save.
Then delete the mov2mov extension again.
Then launch the webui-user.bat
install the mov2mov extension again via the extension tab, then replace the one file again and restart the webui

[heavy perch]

like that?

[ornate elk]

correct

[heavy perch]

Let's gooo it worked tysm, sorry for a lot of questions 😅

[ornate elk]

good! no problem ^^

[nimble stump]

how can i make video out of image

?

is it possible to do in stable diff a1111?

[twilit kraken]

why my python not workin

[vocal burrow]

cause either you didn't install one or didn't check the "add to path" checkbox during installation

[naive dragon]

i still have cmds open now 😅

[nimble stump]

probably not supported version of python

[uncut heron]

i was told no

[nimble stump]

what version do you have?

i think the latest 2 doesnt work

[vocal burrow]

Yes it's possible, there's plenty of tutorials out there to do videos. And there's many workflows to do so.

[uncut heron]

what?????

[uncut heron]

@ornate elk told me no and i consider him a guz who knows everything

[vocal burrow]

Maybe you asked the question in a weird way that oriented his response.

It's possible but not out of the box. you have to install heavy extensions and it will be very taxing on your hardware

[uncut heron]

i got 4060ti 16 gb and 4x16 RAM that are runing at 2k mhz will that do?

[vocal burrow]

sure

[ornate elk]

hmmm we still cant see to to the top

can you copy it as text file?
Also when does the error appear?

[ornate elk]

wait, me? i normaly would say what options you have

not that its not possible

but as aryetis said, its possible but not as easy like txt2image

[nimble stump]

any idea why it wont change model when i select other one?

im on dreamshaperxl and cant change it no matter what

always puts me back to dreamshaper

[vocal burrow]

what does the log say?

[vocal burrow]

full log please, it's cut off

[nimble stump]

nvm fixed it somehow, restarted it few times and it works

[naive dragon]

i made a recording of the whole thing but whats above that screenshot is jsut what gets loaded in the beginning it when i start the generation

[ornate elk]

hmm okay, is your AMD driver updated?
at which resolution do you generate?
check the vram usage in taskmanager when generating.
and make sure background programms are close.
For example Wallpaper Engine has to be terminated.

[naive dragon]

AMD is up to date
resolution i generate 512 512
okay about the background programs i do have discord on and the google tab with sb

[naive dragon]

it crashed btw perfect example

[ornate elk]

okay, you have to check your Windows Pagefile settings

seeing a hdd on 85% doesnt look right xD

[naive dragon]

😅 ill do that

[ornate elk]

you know how?
if not i can send you a guide

https://www.thewindowsclub.com/increase-page-file-size-virtual-memory-windows

there you need to make sure the Pagefile is only enabled for the C:/ drive and NOT for any other drive

[naive dragon]

how do i do that

[naive dragon]

lol you cant read that 😅 hopefully you dont have to 🙏

[ornate elk]

uncheck the first box

then select the D: drive

and then check "Geen wisselbestand"

and then click on the C: drive

and there select "Grootte wordt door het system behered"

click OK and then you have to reboot the PC

and then make sure you have 15gb or more free space on C:

[plush stirrup]

Hello, I'm kadir, can I get your opinion on something? The quality of the video drops a lot while doing faceswap in my artificial intelligence videos. can you give me an idea about this?

[heady pine]

Could someone tell me what they believe to be the best way to create images with the same face? I have watched a few videos, but most of them, apart from two, were all different, and the two that were the same had faces that were noticeably different from the image they started with.

[ornate elk]

hey if you want realistic faces then you can use the Reactor Extensions for faceswap.
if you want to implement a face in different styles like anime or comic, then you would need the controlnet IP-Adapter

[nimble stump]

guys how can i fix these?

those bumps and lines

someone pls help

any idea?

[heavy perch]

@ornate elk any idea why this doesn't work?

like nothing comes up

[ornate elk]

Have you pressed "Load from" ?

Then you should get a list

[ornate elk]

With the Turbo LCM model you need to use less steps and lower cfg

8 steps and cfg on 2

[nimble stump]

did it on one of these but still theres bumps everywhere

[maiden coral]

I think cfq 1 is more appropriate

1 to 2

[ornate elk]

Maybe reduce the lora strength

[nimble stump]

alr

[ornate elk]

Set them to 0.5 to test

[nimble stump]

just did

its rendering

[heady pine]

Alright, thank you.

[ornate elk]

How long does it take for an image?

[nimble stump]

2 mins max

but still theres bumps on the ground

i have 2 loras in the prompt

set them to 0,5

[ornate elk]

Try remove them to test

[nimble stump]

cfg to 2 and samppling steps to 8

but still the same result

[nimble stump]

ok

[ornate elk]

And if it still looks like that then you should test a completely different prompt like
Cute Cat and in the negative prompt
Blurry, deformed

[nimble stump]

ok

have you had this problem too?

or anyone there?

[ornate elk]

Yes, but such artefacts can come from many things

Prompt, model, lora,

Turbo LCM is also a special version of sdxl

[nimble stump]

its really annoying, the bumps are usually on the front ground and never on the character

[ornate elk]

You could also try with an other model

[nimble stump]

ok

thiss is without the loras

but if you look closely she has those bumps on her face

and body

u see them?

[ornate elk]

Yes

Try with an other prompt or model to test further

[nimble stump]

alr

dreamshaper

no loras

prompt: dog

cfg 2 sampling steps 8

would like to hear your opinion

pls

[ornate elk]

Okay try an other model

For example juggernautXL v9

[nimble stump]

dont have that one

[ornate elk]

Then any other sdxl you have

[nimble stump]

i got realvisxlv40

sampling and cfg at 2 and 8?

[ornate elk]

Nope, cfg 7 and 20 steps

[nimble stump]

ok

its rendering but i didnt add loras to the prompt

next one ill do with loras

0.5

[slender coral]

How can I make stable Diffusion to cover Person in some places like on this image:

[nimble stump]

different model same prompt without loras

still bumps on the ground

could it be that extension tiled vae?

or idk

still bumps, different model with 2 loras both at weight 0.5

[gentle sundial]

Hi Guys! if any one can help mw with my issue , I'd really appreciate it!
I have auto 1111 installed locally and it updated automatically after that if I use any lora , the image generation time just get 10 times slower , any idea what to do? thanks!

[ornate elk]

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

[ornate elk]

Can you try the dog again?

[nimble stump]

with loras or without?

[ornate elk]

Without

[nimble stump]

alr its cookin

cfg and samples still same?

[ornate elk]

7 and 20

[nimble stump]

yea yea

[ornate elk]

Cover?
Like not generating the face?

[nimble stump]

but on the first dog there wasnt any bumps

or were there any?

[ornate elk]

There were bumps

[nimble stump]

hmm

then what could it be

[gentle sundial]

Thanks for the quick reply !rtx 3050 ti laptop , display memo 4gb

[ornate elk]

Okay, make sure you have
--xformers --medvram --no-half-vae
In your webui-user.bat

[nimble stump]

couldnt it be that tile vae extension?

[gentle sundial]

@echo off

set PYTHON="C:\Users\user\AppData\Local\Programs\Python\Python310\python.exe"
set VENV_DIR=
set COMMANDLINE_ARGS=--medvram --xformers
set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.9,max_split_size_mb:512
set SAFETENSORS_FAST_GPU=1
git pull

call webui.bat

[nimble stump]

XDDDDD

bumps everywhere !

[ornate elk]

I don't think so

[gentle sundial]

it's very fast but only whan I add lora it becomes 10 x slower

[wet flame]

well my gpu old one got toasted

what is recommended but best value gpu i can get

is 3060 good?

[gentle sundial]

yes

[nimble stump]

any idea

[wet flame]

yeah randomnly getting black screen on my monitor

[ornate elk]

Remove the py torch cuda allocated conflict garbage allocation line

[wet flame]

any cheap 12 gigs?

4070 is so expensive

i wonder if 8 gigs 4070 is same as 3060 12

btw is ali express cheaper than amazon?

[nimble stump]

what should i do now @ornate elk

[ornate elk]

The last thing that could help is the sdxl VAE

[nimble stump]

this one?

i downloaded it yesterday but idk how to use it

[ornate elk]

You have to go into Settings, user interface, Quicksettings, there add
sd_vae
And then apply and reload ui.

Then you get a VAE dropdown

[gentle sundial]

I just did that and still the same, rendering 10 images takes 3 hours and without loras 5 minutes

[ornate elk]

What's the lora and model?

[nimble stump]

ok got it

[ornate elk]

And is your webui updated?

[nimble stump]

yes

[gentle sundial]

all loras and models I tried everything, it was working fine before update 🙁

[nimble stump]

i got another bar beside the models bar

called sd vae

is that it?

[gentle sundial]

yes I put git pull so it update automatically

[ornate elk]

Okay and fore the updated It was faster?

[nimble stump]

now can i add the loras too?

[gentle sundial]

yes it was very fast like 20 image in 20 minutes

[ornate elk]

Then you may need to reenable a setting that was disabled after the update. But its to late now, I can't check which one

[slender coral]

Stable diffusion won't generate the astronaut suit covered with flowers

This is the best result I get

[nimble stump]

it can

[gentle sundial]

yeah exactly I think this is the reason but I don't know what setting

[ornate elk]

Ah okay, that's something for #📝｜prompting-help

[gentle sundial]

is there an xformer option in the settings?

[ornate elk]

Found it: its under compatibility

I'm now off gn

[nimble stump]

they are less but still there

if you look closely

but much better i think

[ornate elk]

Yea much better

[nimble stump]

any idea how to get rid of them completly?

[wet flame]

can you use amd gpus yet?

man amd gpu are performing better and cheaper pricewise

[gentle sundial]

it did not fix it, but thanks very much and good night, I'll try to find a solution for this and I'll get back to you if I wont

[maiden coral]

You may want to check this video if you're using ComfyUI. https://www.youtube.com/watch?v=_C7kR2TFIX0 it's about conditioning using conditioning nodes

[slender coral]

Thank you! Saved it, Gonna check this video tomorrow after school

[viral knot]

you discovering new comfy vids 🙂

[maiden coral]

yeah, watched a few today, after tests I'll try all that stuff

[viral knot]

nice

[maiden coral]

conditioning seems as the most practical

[naive dragon]

Thank you for helping me with that problem.
i just downloaded the three extensions that are on the screenshot and i am getting this error message: RuntimeError: mat1 and mat2 shapes cannot be multiplied (154x2048 and 768x64)
what did i do wrong 😬

[vocal burrow]

most likely missmatching sd15 / sd2 / sdxl stuff

eg : using sd15 loras with sdxl model, using sd15 controlnet models with sdxl model

[naive dragon]

so the lora's are not compatible with the version of my model? and there is no way to use them or atleast not easily?

[vocal burrow]

can't know for sure, I don't know what you're using.

But no you cannot (easily) missmatch stuff

[naive dragon]

alright thank you for that

[vocal burrow]

what are you using ?

[naive dragon]

ehm what do you mean specifically?

[vocal burrow]

what settings are you using, like what model ? what prompt ? what does the full log says ?

[naive dragon]

im using a nsfw model names: mfcgWashedMix_v10 with 3 lora's of different anime characters but two of them are sd1.5 and one is other because its a lycoris

😅

[vocal burrow]

pretty sure mfcgWashedMix is sdxl (pony) so yup that's not gonna work with sd15 loras

[naive dragon]

ye i was told that before with something else

[dapper kraken]

aye guys, how do you use a gpu with 4.0 dedicated memory ?

some people managed to get it working

[viral knot]

comfyui

[dapper kraken]

aight

[ornate elk]

Hey, checkout the pinned messages of this channel.
There you find a install guide for Auto1111.
It works with 4gb vram

[ornate elk]

Yes its possible, but for ai stuff nvidia performance still better

[undone badge]

Dumb question but does anyone happen to know what the goto method for installing SD on NixOS is?

[dense tapir]

Any way to fix that? Literally just reinstalled Python, kind of clueless here!

[ornate elk]

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

[dense tapir]

2060S

[ornate elk]

Okay looks fine, but why the conda install stuff ?
SD should already install the right versions

[dense tapir]

Honestly don't remember, any reason to keep it?

[ornate elk]

No remove the conda line and the pip lines

Then delete the venv folder

And relaunch the webui-user.bat

[dense tapir]

Thank you my good sir you always come to my aid

[ornate elk]

No problem 🙂

[dense tapir]

Is the venv folder like SD's cache or something like that?

[ornate elk]

Not exactly, in the venv are the core files of the functions of SD stored.
But extensions sometimes also install their stuff in it.
That could mess something up over time.
Deleting the venv will recreate it at the next launch reinstalling xformers and the rest

[dense tapir]

Thanks!!!

I got this in cmd

File "C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\fastapi\params.py", line 4, in <module> from pydantic.fields import FieldInfo, Undefined ImportError: cannot import name 'Undefined' from 'pydantic.fields' (C:\Stable Diffusion\stable-diffusion-webui\venv\lib\site-packages\pydantic\fields.py)

[ornate elk]

Oh wait

[ornate elk]

This is a known bug caused by the controlnet extensions

In the webui-user.bat
Add: --update-all-extensions
Then save and delete the venv folder again and relaunch the webui-user.bat

[dense tapir]

Same result

[ornate elk]

Okay then in the stable-diffusion-webui folder
Click in the explorer bar (not searchbar)
There type cmd and hit enter.
In the cmd type and run these 3 commands after each other.

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

Then relaunch the webui-user.bat

[dense tapir]

That fixed it thank you again

You're great

[ornate elk]

Perfect np

Now your updated in every way, hopefully it consumes less RAM now.
But it also depends on the Model your using.
SDXL models use more RAM than 1.5 ones

[dense tapir]

Yeah now it only spikes up after the last iteration which is fine with me!

[ornate elk]

Thats the VAE step.
You can reduce that with the extension called Tiled Diffusion. It has an option called Tiled Vae where the VAE step gets split so it doesn't consume so much vram/ram

It can help when upscaling to larger resolutions

[dense tapir]

Oh that's what that's for lol, I have it

What are some recommended options for it?

[ornate elk]

Nice, only enabling tiled vae is enough and leave its settings on default.

[dense tapir]

Hmm, speaking of upscaling, any way to speed it up in txt2img?

[ornate elk]

There isnt much other then using not more than 10 hires steps. (Also don't use 0 hires steps when hires fix is active)

[dense tapir]

I'll remember that

[fair oxide]

      const requestData = {
      text_prompts: [
        {
          text: prompt,
          weight: 0.5
        }
      ],
      width: "1024",
      height: "1024"
    };

    const response = await fetch("https://api.stability.ai/v1/generation/stable-diffusion-v1-6/text-to-image", {
      method: "POST",
      body: JSON.stringify(requestData),
      headers: {
        "Authorization": `Bearer ${process.env.NEXT_PUBLIC_STABLE_DIFFUSION_API_KEY}`,
        "Content-Type": "application/json"
      },
    });

      if (response.ok) {
        const imageData = await response.arrayBuffer();
        setImagines([imageData]);

error:

Help!!

[unique bluff]

Is there any way to fix the following error? AssertionError("Torch not compiled with CUDA enabled") AssertionError: Torch not compiled with CUDA enabled

It’s not allowing me to open forge

[nimble stump]

is there any way to get rid of these bumps completely?

[ornate elk]

What's your GPU?

[unique bluff]

Intel iris pro graphics 6200

I do understand support for intel gpu is limited, but I want the forge install to open and run at least simpler images (the a1111 didn’t work due to a lowvram error- there’s probably a fix but idk what it is either, hence why I’m trying forge)

[ornate elk]

You should try vladimanics sdnext, it has the best Intel support

Your GPU seems not supported by Forge

[foggy pewter]

Hi where can I find the best noobies guide for installing/using SD? This pinned message shows that there is a channel for that, which I don't have access to. "Find the most suitable method for yourself by visiting our ⁠No Access channel!"

[ornate elk]

Hey, in the pinned messages of this channel you find the best install guide

[unique bluff]

Oooh

Totally makes sense now

I’m tryin to install it but my Payton version in incorrect, guess I’ll need to downgrade it

[foggy pewter]

I wasn't sure if the guides were complete or just for the GUI, but it seems Automatic loads the models for you so I guess that covers it all. Thanks.

[unique bluff]

Is there any way to open using the correct python version? I have both 3.12.3 and 3.9 (the one I need to open dd with)

[ornate elk]

You need Python 3.10.11

That works with any webui

[unique bluff]

Yeah, but it keeps pulling the 3.12.3

[ornate elk]

Automatic is the GUI.
You get the models from Civitai.com

[unique bluff]

Even after uninstalling 3.12.3

[ornate elk]

Are you on windows or Mac?

Or linux

[unique bluff]

Mac

None of the commands I’ve tried seem to work so the version that’s used is not the latest but 3.9 I have installed

[open fox]

On Mac I use comfyUI without any trouble. Speed is slow compared to the Nvidia cards out there. But at least it works.

Just created a virtual environment. Used git clone to get the comfyui from GitHub and pip installed the requirements. Uses the mps backend as torch does since a while.

[nimble stump]

is it good to render with euler a + karras?

[unique bluff]

Update: I had to uninstall the automatic in order for the correct python to work

[ornate elk]

Euler a or DPM++ 2m Karas are good fast samplers

[nimble stump]

whats the best one?

[ornate elk]

There is no best one

But these two are the most used

Then there is also DPM++ SDE karras

But its slower

[nimble stump]

is euler a + karras good?

still bumps

no matter what, cant get rid of them

can someone help pls

some guy told me to upscale it but it rendered completly different image

wtf is this

[knotty cosmos]

AI Art Enthusiast Seeking experts for Generating 4MP Images Similar to MidJourney

Hello Everyone!

I'm an AI art enthusiast currently facing issues with accessing my MidJourney account and thus, the MidJourney community.
Consequently, I am looking for a talented AI enthusiasts who can generate high-quality images that are similar to those produced by MidJourney.
The images should be at least a resolution of 4MP.

Project Details:
-Generate high-quality AI-generated images.
-Each image should be at least 4MP in resolution.
-The style and quality should closely mimic those seen in MidJourney outputs.

Requirements:
-Proven experience in AI image generation.
-Ability to produce high-resolution images.

Compensation:
Hourly rate: $5 to $10, depending on experience and portfolio quality.

This is a AI image generation project and we will discuss further details, including specific budget and timelines.
If you are interested, please reach out with to me to generate AI art.

Looking forward to collaborate with you!

[fair oxide]

Someone help

[empty schooner]

hi! I'm having some issues with a style LoRA I made. It works well as long as I don't have a negative prompt - but as soon as I do I might as well have not used it. Anyone else encounter the same thing and have any tips?

this is for SDXL!

[wheat helm]

getting the error message "bad sampler name: DPM++ SDE Karras. Please, check your schedules/ init values."? I checked my scheduling of course but not sure what exaclt im supposed to be looking for. Any dealt with this before or have any ideas?

[open fox]

This is the example from the website of openAI but it is the actual v2 API not the one you use.
A quick Check wether your JSON Payload is well-formated delivers "no".

[meager ginkgo]

is it possible to give bodies to headshots ive already generated?

[wheat helm]

for some reason its working now? I changed my sampler to euler a and its working.

[open fox]

And if you want to use SDXL you would need the Engine "stable-diffusion-xl-1024-v1-0" and needs to use the correct image height and width

[fair oxide]

I've read the doc and the code is accordingly

Letme show you again

[open fox]

The code above shows the 1.6 Engine not the sdxl

[fair oxide]

Okay I'm changing the engine again

Will it fix this issue?

[open fox]

No it won't fix it

[fair oxide]

      const requestData = {
      text_prompts: [
        {
          text: prompt,
          weight: 0.5
        }
      ],
      width: "1024",
      height: "1024"
    };

    const response = await fetch("https://api.stability.ai/v1/generation/stable-diffusion-xl-1024-v1-0/text-to-image", {
      method: "POST",
      body: JSON.stringify(requestData),
      headers: {
        "Authorization": `Bearer ${process.env.NEXT_PUBLIC_STABLE_DIFFUSION_API_KEY}`,
        "Content-Type": "application/json"
      },
    });

      if (response.ok) {
        const imageData = await response.arrayBuffer();
        setImagines([imageData]);

The error remain same

I'm following this https://platform.stability.ai/docs/api-reference#tag/Text-to-Image/operation/textToImage

[open fox]

Yes but you use the V1 Version not the actual v2 Version of the api.

First step i would try to create the body with in a separate step just to make sure the requestData is able to be JSON.stringify

so varBody = JSON.stringify(requestData)
and in the response body: varBody

Just to make sure theere is no problem with the data

[fair oxide]

Would you like to contribute to my code file?

@open fox

[wheat helm]

omfg accidentally just overwrote my deforum settings

is there a way to restore the file

im in file properties and its saying theres no previous versions to restore to bruhh

[open fox]

Try this please:
This is a simple call. you just need to replace the API Path and fill in the SDXL engine in the first row and your Api Key

[nimble stump]

how to properly upscale image ?

[fair oxide]

 organization does not have enough balance to request this action (need $0.002, have $0 in active grants, $-0.27 in balance).","name":"insufficient_balance"}
``` is sdxl paid? I just tried a fresh key

@open fox

Anyways, i fixed the parameters issue

[nimble stump]

what?! is it paid?

[empty vector]

Hello guys, I really hope someone can figure this our…

I moved from Seaart to A1111 and now I‘m getting different images with the exact same settings and same seed! I don’t know what it is but on A1111 the colors and shadows are a little bit darker for photorealistic images.

Does someone have a clue why? Is there any setting in Seaart’s backend that might cause this difference? Is someone able to get the exact same picture on Seaart as on A1111? If so, how??

[limpid oyster]

Having weird issue! When I launch Auto1111 the UI is partially unresponsive until I refresh the browser and then it is fine. Any ideas?

[limpid oyster]

NM, it was an extension issue 😛

[ornate elk]

It's not possible. You only can try set the RNG to CPU in the settings

But we don't know what seart uses

[gentle sundial]

@ornate elk Hi again , I'm still facing the same issue we talked about yesterday , when I add lora render time increase 10 x and when I add more loras it even get's longer to render an image , 1 image 10 minutes , do you know what setting is that got reset when I updated A1111?

[fair oxide]

Kill me bro

Before i die in my own thoughts

[ornate elk]

hey did you checked the cumprod_fp16 thing ?

[gentle sundial]

I dont know where to find this?

[ornate elk]

this was changed from 1.7.0 to 1.8.0
so enable, then hit apply and test

[fair oxide]

11:13 pm for you too?

[ornate elk]

so enable, then hit apply

[fair oxide]

Damn bro

Where u from?

Let's go sleep

[gentle sundial]

same🙁 , to give you more info my version before update was installed in 8/2023 basically old , and the updated to most recent

[fair oxide]

Yeah we don't support old versions

Basically anything before update won't work

[light geyser]

how the hell do i prune a sdxl model to fp8 without the results being a mess?

[fair oxide]

you have to compress the fp8 from the model to see the results

[ornate elk]

hmm okay, can you show me the webui versions at the bottom?

[light geyser]

???

[fair oxide]

i don't know what i just said honestly

[light geyser]

ight

[light geyser]

do you know how i can prune sdxl to fp8?

[fair oxide]

i don't bro i'm sorr

[light geyser]

alr

np

[gentle sundial]

version: v1.9.3  •  python: 3.10.6  •  torch: 2.1.2+cu121  •  xformers: 0.0.23.post1  •  gradio: 3.41.2  •  checkpoint: a3d7311d85

[gentle sundial]

also I get these errors when I open webui but it works anyway

[LyCORIS]-WARNING: LyCORIS legacy extension is now loaded, if you don't expext to see this message, please disable this extension.

fatal: unable to access 'https://github.com/AUTOMATIC1111/stable-diffusion-webui.git/': Could not resolve host: github.com
venv "C:\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]

[ornate elk]

the lycoris extension can be deletet, but before that move the lycoris models to the lora folder

[gentle sundial]

copy I'll do that now

[gentle sundial]

I removed the folder and deleted extension restarted webui, it still give me the lycoris error but the fatal error is gone , and rendering still the same

takes very long

[ornate elk]

Can you delete the venv folder and relaunch the webui-user.bat?

Then check for any other errors

[gentle sundial]

yes I did that before

shou;ld I do it again?

wait I'll send you all of the script

[ornate elk]

Can you update your extensions in the extensions tab then restart the webui?

[gentle sundial]

yeah I did 4 times

[ornate elk]

Okay, and what was your GPU again?

And the slowdown only starts when adding a lora in 1.5 ?

Can you show me your extensions too?

[gentle sundial]

3050 ti rtx laptop

[gentle sundial]

no it happened after I uptaded A1111 to the recent version

[gentle sundial]

yes 1 sec

[ornate elk]

Okay I think the issue is that you have 3 lora extensions where 2 are abandoned and one is 9 months old.

You need to go into the extensions folder.
There delete all lycoris extensions.
Delete the loractl extension.

If you don't use animatediff then delete it to.

Openpose editor is broken so delete or fix it by installing basicsrs

Then delete the venv folder. And relaunch the webui-user.bat

[unique bluff]

Hello, I’m generating images but they come with a very bad quality to the point where it’s unusable. Is there any fix?

[ornate elk]

Can you show an example?

[unique bluff]

(A picture from mobile device because i don’t want my computer to overload lol)

[ornate elk]

Okay and the settings?

[gentle sundial]

Thank you very much , 'll do that and get back to you

[unique bluff]

Base model is cetusmix_whalefall2
344X512 of size

Would you like the prompts as well?

[ornate elk]

Nope but sampling steps and cfg

And is a VAE used?

[unique bluff]

Nop

Let me see how I can check those, I’m new to the next interface

[ornate elk]

Also try a 512x768 image

[unique bluff]

Now the quality is a bit better (I guess there’s was steps missing yet?) but still lower than what I wanted i guess

[unique bluff]

20 sampling steps on Karras and low order

[ornate elk]

20-30 steps is good

[unique bluff]

Also it cropped the image when i said on negative cropped

[ornate elk]

Cfg on 7

What's your positiv prompt?

[unique bluff]

Originally I meant to use another prompt but it didn’t generate anything so i had to change it

masterpiece, top quality, best quality, official art, beautiful and aesthetic, extremely detailed, high quality, highres, smile, upper body, looking at viewer, girl, solo focus, mature female, black sweater, intricate design, purple eyes, very detailed eye, long hair, went, purple hair, hair between eyes, shiny skin, white background

[ornate elk]

Okay seems normal

Have you tried the 512x768 ?

[unique bluff]

Is there any way to make this work?
lora:scmore_v5:0.5 (masterpiece, top quality, best quality, official art, beautiful and aesthetic:1.2), extremely detailed, high quality, highres, smile, upper body, looking at viewer, (1girl:1.4), solo focus, (mature female:1.4), (black sweater:1.2), intricate, design, purple eyes, (very detailed eye:1.5), long hair, bangs, purple hair, hair between eyes, (shiny skin), white background

[unique bluff]

Will do now

[ornate elk]

I think it should work

But I haven't used sdnext

[unique bluff]

I’m trying it on the bigger size and with anime style in hope of making the quality better + achieving the look I’m going for

[unique bluff]

Now the quality is much better

Still not the style I’m hoping for, but it’s nice to have usable pictures

Thank you!

[ornate elk]

No problem, 1.5 models are trained on 512x512 resolution so going lower than that can result in bad quality

[gentle sundial]

They should call you the God of AI ! it worked , thank you so much!

[deft lynx]

Hi

None of my extensions, although installed

they dont show up on the ui

does anybody know a fix

i already restarteed

[unique bluff]

Okay, I tried generating using a VAE and got a ton of errors (plus it took ages longer to get to 100% and used way more ram, but I guess that’s expected)

This was the last error

[indigo tree]

Why are my XL models generating weird images in Automatic 1111? I have updated to the most recent version of A111 and updated pytorch as I was getting an error to update but the problem persist.

[unique bluff]

Is there any workaround?

[onyx solar]

You're using the wrong VAE.

[unique bluff]

The errors are still popping despite not using VAE. How do I fix that and could it be related to the model used?

[radiant stratus]

Someone leaked an image from a game I play but it looks like they ran it through an edge detection algorithm to destroy any data that could have been used to identify them, how could I go about using stable to convert it back into what looks like more normal art?

[viral knot]

can you post the pic? im so confused as to what you even talking about lol

[solemn moon]

Salutations
Does anyone know where to find reActor
Im either bad at finding what im looking for or its dissa..

says its installed, its in my extentions folder, but cant find it in the webui 4 1111

[deft lynx]

same

[brisk cedar]

that wont help you identify them

[dense tapir]

Hello guys, is there any way to upscale an already transparent image and keep the transparency? The Extras tab only lets me upscale my transparent image, put a solid over the transparent part, and then mask it again which results in a transparent image with flaws.

[dense kernel]

Hello Everyone!
I have a Dream Studio account that I have been using for about a year, I regularly buy credits. However, I have come along an issue after last recharge, that my account has been disabled though I bought credits of $100 but now there are 0 credits and account has been disabled. I am not able to use any models, nor have they mailed me any warnings or alerts. This is the error that keeps on appearing while I try to buy more credits:

"Something went wrong! Your account has been disabled. Contact mailto:[e-mail] to appeal if you believe this was done in error."

I have also mailed the customer support, but there has been no response from their side yet.

[oblique parrot]

can anyone explain the creative upscale creativity range of 0 to 0.35?

Is this logarithmic or something?

[ornate elk]

I don't think you can, but there is an extension called Rembg to remove the background of an image

[ornate elk]

The errors says something about IP-adapter and hypernetwork

[radiant stratus]

https://i.imgur.com/h0ws1DA.jpeg

[radiant stratus]

Oh I know what a lot of the stuff is actually

[solemn moon]

check for updates, instal and restart, close and rerun webui.bat

in extensions tab

[ornate elk]

@deft lynx Hey, reactor only appears if installed correctly after its install guide:
https://github.com/Gourieff/sd-webui-reactor?tab=readme-ov-file#installation

[ornate elk]

Use Controlnet extension with Aninelineart model

But you can't get the original image back, you can just let the ai guess and prompt for it

[radiant stratus]

Oh yeah I'm aware I know the limits of this

[oblique parrot]

I’ve been experimenting and yeah, I think it works to take the creativity percent and do this to get the value to use:

cr_range = 0.35
creativity = cr_range * cr_pct ** cr_range

(seriously i can’t be the only coder here)

[terse lichen]

so i am getting this error and have installed the packages properly can i pls get some help

[oblique parrot]

I'm suddenly getting ripped off. I'm getting 403 errors in the creative upscaler, but being charged 25 credits each anyway. And when it seems "successful" I'm suddenly getting 404 errors and cannot retrieve anything.

[timid notch]

i have a very silly problem~~

i wanna download this model,but i think i need to download base model first~

and i serch sd 1.5 in civiai,

there are lots of choice~~which one should i use?can someone tell me ?thx

they all sd 1.5so confoused

so confused

[fierce hollow]

Is there a way to use SD to create a video that is difficult to distinguish from reality? I tried the "Deforum" plugin, it is clear that this is a neural network and each frame is slightly different, are there other tools or neural networks with local access?

[open fox]

First you are downloading a Lora. These Lora are small “addons” to regular models and adds concepts or styles to the base models. The Lora you showed here works with all stable diffusion models that are based on 1.5 or 1.6 models. (512x512px base size).

[thin quarry]

just pick something you like \ style you want, they all trained on different things (make sure it's model tho)

[open fox]

You can download any model. Realistic ones or dreamshaper etc.

[terse lichen]

can i pls get some guidance on this @open fox

everything is installed properly

[open fox]

Well you got anymore error messages. So far I am missing these reason while the import does fail.

[ornate elk]

AnimateDiff

[open fox]

Or you can try stable diffusion video.

[fierce hollow]

https://imgur.com/a/oTzC2mB ?

[terse lichen]

no its only that error msg i am getting when trying to install the inference core node which is missing

[fierce hollow]

Can I go into more detail? I'm new to this topic;(

[open fox]

Well it total there a some free models that covers movement. Deforum is more or less rendering image each frame.
Animatediff got a frame context and with enough memory you can get 80+ frames of cihensive animation.
Stable diffusion video model (svd xt1.1 for example) can generate simple movement and animation from an image.

[fierce hollow]

I have 3070ti, 32GB RAM

[open fox]

Without the error message it will be hard to find the source of the Error. Maybe you try to install the custom node manually by cloning the repository.

[open fox]

Which gui are you using.

[terse lichen]

thats what it said in the installtion process on github

[fierce hollow]

Automatic111

[open fox]

Then there are probably other people around who can help you. With comfyui I know some workflows

[fierce hollow]

I looked at the option with AnimateDiff - the problem there is that there are "Motion module" submodels In civitai, alas, they are not available only on SD 1.5 on SDXL...

[ornate elk]

There is one for sdxl

[fierce hollow]

https://civitai.com/models/331700/odinson-sdxl-animatediff ?
As for me, it is extremely bad ; ( Ideally, I would do with the help of all this - the movement of people

[ornate elk]

Oh you mean motion module

I thought you mean animatediff model

[fierce hollow]

Yes, yes

[fierce hollow]

The task is this: To make a video where a man of 30 years old drinks coffee, after which he starts reading something on his phone, then gets up and goes out, say, somewhere, is this real?

[ornate elk]

Only if you make multiple short videos and cut them together

[fierce hollow]

I don't mind, but how and through what can I achieve this? You told me there are 3 options in total

• Deforum
• Animatediff got
• Stable diffusion video
  Which one is better?

[open fox]

For that I would go with animatediff and prompt traveling

[fierce hollow]

how do I find prompt traveling? https://imgur.com/a/bpcrqJ8

[open fox]

Prompt traveling is used to describe the different frames.
https://civitai.com/articles/2943/how-to-use-prompt-travel-with-animatediff-tutorial

Beware the example of the tutorial is full NSFW

[fierce hollow]

Okay, thanks, I have to leave now, tomorrow I will ask for your method and write it off) Thank you for your time and I'm sorry if I asked stupid questions

[rocky zephyr]

hi, is stable video down?

the api and the website stablevideo are not working for me

[long sage]

I'm rewriting my discord chatbot in python, and am wondering, what would be the fastest way to go from no model loaded in vram to an image. The setup i used before was ok, but id like it to go faster.

Having the model always loaded is also not possible, because there is an llm loaded in vram when the image generator is not used

[long sage]

Here's my old code:

global imagen_pipe
await unload_llm() # unloads the text model to free up vram
imagen_pipe = AutoPipelineForText2Image.from_pretrained("stabilityai/sdxl-turbo", torch_dtype=torch.float16, variant="fp16")
#imagen_pipe = AutoPipelineForText2Image.from_pretrained("runwayml/stable-diffusion-v1-5", filename="*emaonly.safetensors",torch_dtype=torch.float16, variant="fp16")
imagen_pipe.to("cuda")

imagen_pipe(prompt=prompt, num_inference_steps=1, guidance_scale=0.0).images[0].save("image.png")
print("image generated")
await message.channel.send(file=discord.File("image.png"))
del imagen_pipe
torch.cuda.empty_cache()
await load_llm()

[ancient wadi]

Hey so i just downloaded Forge UI on a computer and when i generate an image my computer will chrash afterr an indeterminate period of time

[balmy oyster]

Guys, what if ControlNet doesn't show up after installation? Or do I need to press something?

[ornate elk]

Check the cmd, maybe there is an error

[ornate elk]

What's your GPU?

[ancient wadi]

im running with an RTX 3080

10 gigs of vram

[ornate elk]

Make sure your power supply can support your GPU under full load.
Also make sure your GPU is connected with every power cable so that there is not free slot on the GPU.

[elder mauve]

why GPU is recommended for running SDXL models?

and how do u figure out how much vram u need?

[ancient wadi]

ok so before i had mistakenly ran forge through the run.bat, when i restarted and ran it through the webui bat it started downloading pytorch

imabye again?

[ornate elk]

It works on 6gb vram and more.
Recommend would be at least 8gb and best 12gb or more

[ancient wadi]

and then it returns error code 1

[elder mauve]

so i need more VRAM than the model size?

so entire model is loaded into vram?

[ornate elk]

Yes

[elder mauve]

i have 16GB vram

but amd

[ornate elk]

Ohh

[elder mauve]

:/

[ornate elk]

Then your lucky ^^ since February AMD works great with SD now

[elder mauve]

i am running fooocus cuz i was unable to run SD using automatic1111's web gui

or what method is recommended?

i only know of these 2 web gui

[ornate elk]

The old webui was based on directml, thats what Fooocus uses too. But thats really slow and not needed anymore.

The new way is to run auto1111 with zluda.

You find a full guide for that in the pinned messages of this channel.
And if you have any questions feel free to ask me. I'm on AMD too.

[elder mauve]

i am running fooocus with zluda

[ornate elk]

Ah okay

Then you should try auto1111 with zluda

[elder mauve]

but i noticed same speed , as with directml?

[ornate elk]

Nope

[elder mauve]

thank you! which card do u have?

[elder mauve]

yes?

[ornate elk]

Zluda is always 2-4 times faster

[elder mauve]

it is not? or i am doing smth wrong

[ornate elk]

Your doing something wrong then

[ornate elk]

I'm on a 7900xtx

[elder mauve]

i am running like this
"D:\zluda\zluda.exe D:\foocus\run.bat "

[ornate elk]

Yea that's not how it works

If you got the same speed its still using direcml

[elder mauve]

.\python_embeded\python.exe -s Fooocus\entry_with_update.py --directml pause

and this is contents of run.bat

[ancient wadi]

hey just making sure, i can use python 3.10.11 with forge right?

or does it have to be 3.10.6

[ornate elk]

Yep 👍

[ancient wadi]

lets go

[elder mauve]

how can i use it then?

[ornate elk]

There you see it uses directml

[elder mauve]

oh

[ornate elk]

I didn't tried to get Fooocus working with zluda. But auto1111 and comfyui can work with it

[elder mauve]

i will try auto1111 once again

[ornate elk]

Would recommend using auto1111 with zluda. Its the best experience

[elder mauve]

let me check the guide

what speed do u get ?

how many interations per second for sdxl?

[ornate elk]

Need to start it real quick

On a 512x512 1.5 image I get 16 it/s compared to 6 it/s on directml

[elder mauve]

also where can i get feedback about the pics i generated?

[elder mauve]

on 896x1152 i am getting 11 seconds per interation , i am on 7800xt

[ornate elk]

For image questions we have #📝｜prompting-help

[ornate elk]

Let me try this resolution, I guess without upscaling?

[elder mauve]

yes , native

[ornate elk]

Getting like 2.6-2.9 it/s

[elder mauve]

wow

[ornate elk]

It took 8 seconds

[elder mauve]

for me it takes almost a minute :/

[ornate elk]

Yea directml is bad

[elder mauve]

how many iterations did u run

[ornate elk]

You can't even upscale to high

[ornate elk]

20 steps

[elder mauve]

wow thats really good

excited to get mine working now

[ornate elk]

this took me 50 seconds, 720x1280 native upscaled by 2 with hiresfix on sdxl with multiple loras:

[elder mauve]

what seed and prompt?

[ornate elk]

now its wqhd desktop wallpaper

[elder mauve]

i tried to make a comparison table in excel haha

same prompt

its loong

[ancient wadi]

ok so i think i solved my issue

i uh

didnt have git installed

lol

[ornate elk]

wow ^^
i used the aamXLAnimeMix_v10, i think its one of the best anime models for sdxl

[ornate elk]

did you found the guide?
let me know if you have any questions, also download the latest zluda files and replace yours

brb

[elder mauve]

😂 its very rough but if u know a way to improve the process , i would be happy to hear , let me know what u think

[elder mauve]

yes , very detailed, following it rn

[elder mauve]

i get this on step 5

[ornate elk]

Show me your path settings

[elder mauve]

this?

[ornate elk]

Yes looks okay

[elder mauve]

also there is that

[ornate elk]

Then open up a cmd and type
Pip cache purge
Then delete the venv folder and relaunch the webui-user.bat

[ancient wadi]

so my computer no longer crashed but i get the error "connection errored out" after about a minute or two of the ui running,

[ornate elk]

How much RAM do you have?

[ancient wadi]

32 gb of ddr4, and 10 gigs of vram

[ornate elk]

Okay make sure your gpu driver is updated

[ancient wadi]

yep done that

[ornate elk]

Hmm okay, then I would recommend installing automatic1111 webui with my guide.

[ancient wadi]

yea if i cant get forge to work il just cut my loss and go to automatic 11111

[ornate elk]

Forge can be a bit buggy, so I would recommend start with the most used webui

[ancient wadi]

its been working for an extended peirod now, no idea why

[elder mauve]

i think its stuck here

now i got this

[ancient wadi]

also why does image width and hight have such a drastic effect on the image generation, the difference between 512x512 and 1024x1024 is huge

[elder mauve]

because SDXL was trained on 512x512

1024x1024 can introduce artifacts iirc

better to upscale from 512

[ornate elk]

Sdxl was trained on 1014x1024

[elder mauve]

😮

[ancient wadi]

but face generation is way bettwe on 1024

[ornate elk]

1.5 on 512x512

[ancient wadi]

ah mabye this

[ornate elk]

Is your GPU driver updated?

[ornate elk]

Make a complete PC restart and then relaunch the webui-user.bat again

[elder mauve]

should i update?

[ornate elk]

Not needed now

Im on 24.3.1 too

[elder mauve]

gonna restart real quick now

WARNING: ZLUDA works best with SD.Next. Please consider migrating to SD.Next.
Using ZLUDA in H:\SD-Zluda\stable-diffusion-webui-directml\.zluda
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug  1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.9.3-amd-5-g674c5cac
Commit hash: 674c5cacdac3392279653be6b3da0c8e62898693
Skipping onnxruntime installation.
You are up to date with the most recent release.
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
H:\SD-Zluda\stable-diffusion-webui-directml\venv\lib\site-packages\pytorch_lightning\utilities\distributed.py:258: LightningDeprecationWarning: `pytorch_lightning.utilities.distributed.rank_zero_only` has been deprecated in v1.8.1 and will be removed in v2.0.0. You can import it from `pytorch_lightning.utilities` instead.
  rank_zero_deprecation(
Launching Web UI with arguments: --use-zluda --medvram-sdxl --update-check --skip-ort
Press any key to continue . . .```

same

[ornate elk]

Hmm strange. It also doesn't use the zluda files in C

[elder mauve]

what does that mean?

it couldnt find zluda?

[ornate elk]

Yea but it tried to use its own

[elder mauve]

i delete that one?

[ornate elk]

Make sure your not running any other Programms in the background. Also are you on windows 11 or 10 ?

[elder mauve]

then it will fallback to the zluda in C?

[elder mauve]

11

[ornate elk]

Did you downloaded the latest stable zluda version?

[elder mauve]

i copied the existing one from my downloads

let me check with latest version

[ornate elk]

So you didn't downloaded the one from my guide?

[elder mauve]

nope

downloading now

[ornate elk]

Lol

[elder mauve]

i thought all are same , guess not

oh they are very different

damn

[ornate elk]

Nope that's why I said to replace your existing zluda with the guide ones

[elder mauve]

done

lets try again

[ornate elk]

If it doesnt work you may need to delete the venv folder again

[ornate elk]

Did it launched?