🤝｜tech-support

[mystic lava]

should I

[ornate elk]

you can try

[mystic lava]

yeaaa

ok its slightly faster

still not as fast ad dream

[ornate elk]

it wont be faster than that

so better stay with sdxl models

[mystic lava]

but flux looks so good i swear

whats the best sdxl model

in ur opinion

and what abt midjourney

[ornate elk]

midjourney is an online service

[mystic lava]

true

costs moneys

sadaly

thats why i like thsi

local running

[ornate elk]

depends on the subject

there are a lot of good models

some can do better realism, some can do better anime stuff

[mystic lava]

best realism

and best anime one

[mystic lava]

any issues u can see

that might be causing performance drops

😭

yea that looks sick

[grizzled cypress]

On forge ui it says it’s using computation dtype as torch.float32, is there a way to change that to a lower number? Like 16 or 8? And if so would it be faster?

[ornate elk]

for realism try Juggernaut XL

for Anime try Perfect Pony XL or AAM Anime Mix

[mystic lava]

tried using a different model

it crashed

wtf

[ornate elk]

that workflow only works with flux

[mystic lava]

ssaame with the dream model

ah

[ornate elk]

here is an sdxl workflow

[mystic lava]

ah

i see

i need a vae

[ornate elk]

here an sdxl vae:
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main

the sdxl.vae.safetensors

[mystic lava]

its faster

but not as good as flux

by far

[mystic lava]

oh wrd

ty

[honest perch]

I am a noob and first time trying to install this, I used the AUTOMATIC1111 from github, stable-diffusion-ui and did the nvidia option and got the zip, ran the update.bat and then the run.bat but it gives me this

and this is the website

[ornate elk]

yea, the webui doenst ship the default model anymore

you need to download a model from civitai.com

like dreamshaper for example

https://civitai.com/models/4384/dreamshaper

[loud pond]

hey quick question, if i wanted to make sprites using stable diffusion. Is there any lora's which could help me with making something like that.

[ornate elk]

yea there are pixel art and pixel sprite loras

[loud pond]

oh cool ill check em out any diffusion models you reccomend?

[ornate elk]

nope sry

[loud pond]

bro whats going wrong with this

it keeps coming out blury

this is how its meant to look like

do you know if its the promt or can it just not do it?

[ornate elk]

your two loras that are not the pixel lora are to strong

[loud pond]

oh i thought 1.4 was low doesnt it go up to 10?

[ornate elk]

-2 to 2

[loud pond]

oh alright thanks a bunch!

[ornate elk]

np

[wet flame]

what is flux how do i use it ?

[stable dagger]

Where can I find Image_Crop_Location_Exact node in Cofyui Manager?

[loud pond]

flux is an image generator like stable diffusion. Apparently it is better than SD3.

[cosmic furnace]

Be quiet or Corsair psu ? hmgm

rmx shift vs straight power 12 hm

[grizzled cypress]

Does using multiple loras in the same prompt work well?

[ornate elk]

best would be to use it with zluda

[loud pond]

how do i scale down an image post processing?

https://github.com/Astropulse/sd-palettize apparently i have to use this but im not sure how

[loud pond]

also found out the issue when the width and height are at higher values like 1024 then the pixels become less distorted

[balmy lily]

im getting a "No backends available!" error on swarmui, how fix?

[elder mauve]

getting this

[loud pond]

found out how to do it i needed to go to scripts and activate it there

but one question is there anyway to decrease the size of an image from 740by1024 to 48by 64?

[tepid quest]

hi masters!!!!

im using webui forge and im really new in this world... Im triying to transform an old green real statue from city into a real person

so should have same detail but with real skin.... tried lots of methods but all my outputs are green persons hahaha

any tip? 😄

[loud pond]

hi im new myself, are you using image to image?

[tepid quest]

yep

[loud pond]

cool so i found this video quite helpful

[tepid quest]

let me check hahaha

[loud pond]

https://www.youtube.com/watch?v=PUwLT9JwCs8&list=PLp8fvhoFxLAhjQMHUuX7MdaQLetuskUgI&index=6

[tepid quest]

thx brother!

let me see 🙂

[loud pond]

also maybe try using some lora's

maybe someone's made a lora which would allow you to do this

[tepid quest]

im using some dude but still skin green hahaha

if you want to try 😛

[loud pond]

maybe just change the colour from green to skin colour

like in the text box write 'green skin colour changed to (whatever skin colour you want)'

[loud pond]

i would try but my pc is pretty low end so it takes me about 5 minutes to load one picture lol

[tepid quest]

ohh ok ok bro thanks anyway

If there are an a good soul let me explain, im running out of money, but im my city this 31 will be a festival for short histories with 1700€ rewards

i changed my mind about what histories use this last days and try to replicate a real history about a lady who dies waiting her husband on front of the sea...

so im running out of time and still cant get how to make her face looks real...

tried a copple of ai programs and nothing 😦

[snow lance]

try this one

https://huggingface.co/spaces/DamarJati/FLUX.1-DEV-Canny

[grizzled cypress]

Is there a way to merge the pony model with animagine for instance in a way so I can get the huge amount of characters and style from animagine while still maintaining the creativity of the pony model?

They’re different kinds of models, because one is sdxl and the other is pony so i’m unsure

[mystic lava]

My images keep coming out looking like this

with this weird effect

work space

[odd marten]

I have recently downloaded Stable Diffusion Webui, I have configured everything, but as soon as I try to generate an image I get this error: ‘RuntimeError: tensor.device(). type() == at::DeviceType::PrivateUse1 INTERNAL ASSERT FAILED at ‘C:\C:__w\1\s\pytorch-directml-plugin\torch_directml\rc\cs\dml\DMLTensor.cpp’:31, please report a bug to PyTorch. unbox expects Dml at::Tensor as inputs’

What can I do?

[ornate elk]

Can you try an other model?

[ornate elk]

Hey, what's your GPU ?

[odd marten]

6800 XT

[ornate elk]

Ah then you need to follow my AMD Automatic1111 Webui with ZLUDA install guide

[odd marten]

I had followed an installation guide for AMD precisely so as not to run into this error. So I have to delete everything?

[ornate elk]

The error you got is from directml which isnt recommended if you card is supported by zluda

[ornate elk]

Yep you need to follow the
AMD Automatic1111 with ZLUDA guide from here:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides

[odd marten]

ZLUDA, right?

Okay ty so much

[ornate elk]

Yep

[odd marten]

I'll try

How can i delete all my Webui?

[ornate elk]

Just delete the stable-diffusion-webui-directml folder

Thats enough

[odd marten]

Oh okay, ty

[ornate elk]

And make sure you have python 3.10.11 64bit installed

[odd marten]

Should Python and Git not be reinstalled?

[odd marten]

Yup

[ornate elk]

Git should be already installed

No need for reinstall

[odd marten]

Nice, ty

[ornate elk]

If you have any questions about the steps.
Feel free to ask me

[odd marten]

I hope this will work

[odd marten]

Ty so much again, dude

[ornate elk]

Should work 🙂
Others with 6800 got it working

[odd marten]

Do you think the 6800 is a good video card for generating images? 1 year ago I was using Nvidia's 2060 ti ahahah

[ornate elk]

Yea I think its pretty good

16gb vram is a big advantage

[odd marten]

I see, ty againa hahaa

I am still new to this field, do you recommend models?

[ornate elk]

No problem, let me know how it goes

[odd marten]

Checkpoints

[ornate elk]

For the first installation you will need 1 checkpoint

[mystic lava]

got it fixeed

dw

im a genious

ty tho

[odd marten]

I see

[ornate elk]

I would recommend Dreamshaper v8 because its lightweight and okay.
Later you can use better ones

Here is a link:
https://civitai.com/models/4384/dreamshaper

[odd marten]

ty

[fierce hollow]

There are circumstances that force me with Forge to go back to AUTOMATIC1111, I installed SD.
What values do I need to enter here? Help me please (file: webui-user)
Now I have this
I want to achieve the fastest possible image generation. I have a 3070ti

@echo off

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

call webui.bat

[ornate elk]

With a 3070ti (8gb vram)
You need --xformers --medvram-sdxl --no-half-vae

[fierce hollow]

set COMMANDLINE_ARGS= --xformers --medvram-sdxl --no-half-vae ?

[radiant fog]

hello i have an issue im not sure how to fix it:

C:\Program Files\Comfy UI\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
Total VRAM 12282 MB, total RAM 65310 MB
pytorch version: 2.3.1+cu121
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 4070 Ti : cudaMallocAsync
Using pytorch cross attention
Traceback (most recent call last):
File "C:\Program Files\Comfy UI\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\main.py", line 220, in <module>
server = server.PromptServer(loop)
^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Program Files\Comfy UI\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\server.py", line 90, in init
self.user_manager = UserManager()
^^^^^^^^^^^^^
File "C:\Program Files\Comfy UI\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\app\user_manager.py", line 22, in init
os.mkdir(user_directory)
PermissionError: [WinError 5] Access is denied: 'C:\Program Files\Comfy UI\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\user'

C:\Program Files\Comfy UI\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable>pause
Press any key to continue . . .

[mystic lava]

@ornate elk

Hes my friend

I thought u might have an idea wsp

4070 ti

comfy ui

[ornate elk]

Yes

[ornate elk]

Make a new folder on C drive and call it Ai, then move the comfyui in that folder.
Then relaunch.
You problem is a permission error

[hallow latch]

Hey .. How do I add models, lora's etc .. that are on another harddrive?

[radiant fog]

ah ok

thank you

[ornate elk]

Np

[ornate elk]

For which webui?

Auto, comfy, forge ?

[odd marten]

@ornate elk It's me again, I'm following the whole installation process and have arrived at the paragraph to get ZLUDA working.

I'm at the first step which says to install a program from AMD and it also says not to put a tick on ‘The Pro Driver’ but in the installation process of the program I don't find any such wording. Shall I send screenshots?

[ornate elk]

Idea for what?

[ornate elk]

Hey, the pro driver (at the bottom) is disabled by default so leave everything as it is and install

[hallow latch]

Auto

[odd marten]

I understand, but if I install this programme, will it overwrite the version I currently have of ‘Amd software: Adrenalin Edition’?

[ornate elk]

Then you need --ckpt-dir "D:\pathToModels\

[ornate elk]

Nope. Its a seperate Programm just for the rocm hip sdk

[odd marten]

Ty

[hallow latch]

Cool got 🙂 Just a new issue ..

I just put in a new GPU and runnin ESXI and a Windows Machine. Got drivers installed and the ui and all it running but when I tried to do my first test with a model I got
```RuntimeError: CUDA error: unknown error CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

I am looking up on google and something about *PyTorch CUDA *?

[ornate elk]

You have GPU passtrough on the esxi ?

[hallow latch]

Yes

[ornate elk]

Make sure the latest nvidia driver is installed

[hallow latch]

That is installed too

Im gonna try a reboot first of all before I say anything more.

[mystic lava]

like how to rfix it

also should fluix ai take 10 - 15 mins to generate one image on 20gb vram rx 7900 xt

that feels rlly slow

[odd marten]

@ornate elk Sorry again, which ZLUDA .Zip i have to install?

[mystic lava]

rocm6

amd

[odd marten]

6, okay ty

[mystic lava]

wait

whats ur gpu

[hallow latch]

Reboot solved it ❤️

[odd marten]

6800 xt

I ignored the third step, which was for gpu's stronger or weaker than that

[mystic lava]

yea

[odd marten]

I'm having trouble with the fourth step, but slowly I can understand ahahah (The System Variables one)

[mystic lava]

ah ok

the paths right

[mystic lava]

,

[ornate elk]

Send a screenshot so I can check if its wrong or right

[odd marten]

I'm editing ‘Path’ and I put the folder ‘ZLUDA’ as it says in the fourth bullet point, but it also says I have to put ‘%HIP_PATH%bin’ and I didn't understand this

[mystic lava]

%HIP_PATH%bin

as a path

path

new

%HIP_PATH%bin

ok

ok

ok

run it

[odd marten]

OH, OK

ty

[mystic lava]

ngl

think im using hte wrong version of Flkux

it should NOT be this slo9w

on such a strong gpu

ima try the 16 version

not 8

[hallow latch]

What recommendation would you say to run the set COMMANDLINE_ARGS= when you run an 3060 with 12 GB ?

I start with set COMMANDLINE_ARGS=--no-half-vae --theme=dark --xformers --listen

But should I have --medvram-sdxl added or not?

[mystic lava]

@ornate elk this can NOT be that good

for speeed

on my system

that was deadass 4 steps

[tepid quest]

I will test now! Thanks my friend!!!

[tepid quest]

Ou mama!! I will!! Thank 🙂

[ornate elk]

Looks good

[ornate elk]

Which model did you used?

[ornate elk]

Same as you did with the zluda folder path

Shark won't work right now

[mystic lava]

one you recommend

fp8 I think

[mystic lava]

so its as fast as it should be goiung

sakura is using the same model on simiilar hardware but nvidia and getting 40 second waiting times

[ornate elk]

I'm getting 60 seconds with my 7900xtx

[mystic lava]

wtflip

[ornate elk]

For a 1024x1024

[mystic lava]

is tat wit 20 steps

cause when i do 20 steps

it takes fucking half an hour

[tepid quest]

tried inwebui forge with the canny mode, depth and some others with the safetensor for flux but looks like dosnt work atm

[mystic lava]

that cnat be right man

[tepid quest]

what values u used my friend? hah

[mystic lava]

U look ai generated

iddk how to say htat

😭

[ornate elk]

You could try also sdxl with canny

[mystic lava]

am i just cooked

[tepid quest]

yeah bro maybe i have to change from flux to do this hehe

[hallow latch]

Hey CS1o .. Am I forced to have PCI/PCIe passthrough enabled to run auto 1111?

[ornate elk]

Idk, but could be needed

[tepid quest]

Anyone knows whats values can i use on hugginface flux canny to get this?

all my outputs looks soo different

and dont respect the pose

[odd marten]

@ornate elk Thanks again for all the help, it works fine, there are just a few small problems that don't let me do some things:

• When I install a new Checkpoint, it takes a very long time to load it and I have to restart webui
• When I generate an image, it shows me that it has already finished, but before showing it to me it takes about 15 seconds
• Sometimes for one of these problems listed above, I get a ‘connection errored out’ error and have to restart webui

[ornate elk]

Hey np, glad it worked.
what's the checkpoint you tried when this happens.

Also how much normal RAM do you have?

[odd marten]

Where can I see it? But do you mean the Ram in the PC?

[ornate elk]

Yes

In taskmanager under performance tab

[knotty turret]

@ornate elk trying to inpaint but doenst matter anything always the same result this gray color

trying to change the color hair

idk why not working right

[ornate elk]

Hey can you show your inpaint img2img settings?

[knotty turret]

yeah

i already done alot inpanting but not working now for some reason

[odd marten]

16Gb

[wind chasm]

Long shot but does anyone know how to fix this error on kohya _SS ? I might be being dumb but I couldn't find a discord https://github.com/kohya-ss/sd-scripts/issues/1411

[knotty turret]

my mistake didnt change the sdxl VAE

working normal now lol

😅

[ruby yoke]

@limber pier might know

[limber pier]

that's kinda weird, it's not really an error message

but i'm going to guess you don't have enough vram

6gb is very little

[wind chasm]

oh no thats not my post

im getting that error on a 3090 , 5900x 64gb ram pc

[limber pier]

oh, wtf

[wind chasm]

just better than posting a wall of text

I can't for the life of me figure it out

[limber pier]

it would be good to post your exact error here

you can paste screenshots

[wind chasm]

File "C:\Users\Owner\AppData\Local\Programs\Python\Python310\lib\runpy.py", line 196, in _run_module_as_main
return _run_code(code, main_globals, None,
File "C:\Users\Owner\AppData\Local\Programs\Python\Python310\lib\runpy.py", line 86, in run_code
exec(code, run_globals)
File "G:\Kohya_GUI_Flux_Installer_v17\kohya_ss\venv\Scripts\accelerate.EXE_main.py", line 7, in <module>
File "G:\Kohya_GUI_Flux_Installer_v17\kohya_ss\venv\lib\site-packages\accelerate\commands\accelerate_cli.py", line 48, in main
args.func(args)
File "G:\Kohya_GUI_Flux_Installer_v17\kohya_ss\venv\lib\site-packages\accelerate\commands\launch.py", line 1106, in launch_command
simple_launcher(args)
File "G:\Kohya_GUI_Flux_Installer_v17\kohya_ss\venv\lib\site-packages\accelerate\commands\launch.py", line 704, in simple_launcher
raise subprocess.CalledProcessError(returncode=process.returncode, cmd=cmd)
subprocess.CalledProcessError: Command '['G:\Kohya_GUI_Flux_Installer_v17\kohya_ss\venv\Scripts\python.exe', 'G:/Kohya_GUI_Flux_Installer_v17/kohya_ss/sd-scripts/flux_train_network.py', '--config_file', 'G:/models trained/config_lora-20240829-144248.toml']' returned non-zero exit status 3221225477.

[limber pier]

has kohya ever worked for you? or has this always been an issue

[wind chasm]

so kohya works normally

unless I try the Flux version

and then its like Nah bro

I've tried "he who shall paywalll everythings" installer

and I tried their instructions on the git

EXACT same error

re-installed every requirement as a bach,on their own

[limber pier]

my guess is the config is the problem

or your launch command

it's a pretty new version of kohya so there's some weirdness with some of the options

what's your toml look like

[wind chasm]

toml_file ?

[limber pier]

subprocess.CalledProcessError: Command '['G:\Kohya_GUI_Flux_Installer_v17\kohya_ss\venv\Scripts\python.exe', 'G:/Kohya_GUI_Flux_Installer_v17/kohya_ss/sd-scripts/flux_train_network.py', '--config_file', 'G:/models trained/config_lora-20240829-144248.toml']' returned non-zero exit status 3221225477.

yeah, the config for the job

[wind chasm]

Ahh see I thought it was accelerate as that threw a bunch of errors

[verbal fractal]

Hey guys, i am new to A111 with a few experience of Fooocus. Can you help me to get the better results of it ? I am lost with all options. I have a 1060 nvidia with 6 gb memory. Thanks a lot

I need some informations : what checkpoint to use for photo realism, best quality, should i need a refiner and / or VAE ?, wich parameters to use with my old GPU

[wind chasm]

Its not something dumb like becuase the Config_lora bla bla is in a different folder its throwing a hissy fit ?

if it is ima throw my pc out a window

[limber pier]

cache_latents_to_disk = true
caption_extension = ".txt"
dynamo_backend = "no"
logging_dir = "logs"
max_train_epochs = 100
max_train_steps = 8292
mem_eff_attn = false
min_snr_gamma = 5
mixed_precision = "bf16"
resolution = "1024,1024"
save_every_n_epochs = 9
save_model_as = "safetensors"
save_precision = "float"
sdpa = true
seed = 42
train_batch_size = 1
train_data_dir = "/home/clownsharkbatwing/training_set/"
unet_lr = 1

that's a working toml

training_set needs to have a folder named "1_blah" or something like that

1 = number of repeats

and in that is your data

so training_set/1_blah/1.png 1.txt etc

[wind chasm]

yeah I get that, I usually set it through the gui

but i guess its not with SD scripts ?

[limber pier]

i don't bother wit hthe gui

i don't think it makes it any easier

[wind chasm]

im just getting pissed off with ai-toolkit XD

[limber pier]

accelerate launch --num_cpu_threads_per_process 1 flux_train_network.py --pretrained_model_name_or_path /home/clownsharkbatwing/ai/ComfyUI/models/unet/flux-dev.safetensors --clip_l /home/clownsharkbatwing/ai/ComfyUI/models/clip/clip_l_flux.safetensors --t5xxl /home/clownsharkbatwing/ai/ComfyUI/models/clip/t5xxl_fp16.safetensors --ae /home/clownsharkbatwing/ai/ComfyUI/models/vae/ae.sft --cache_latents_to_disk --save_model_as safetensors --sdpa --persistent_data_loader_workers --max_data_loader_n_workers 2 --seed 42 --gradient_checkpointing --mixed_precision bf16 --save_precision bf16 --network_module networks.lora_flux --network_train_unet_only --cache_text_encoder_outputs --cache_text_encoder_outputs_to_disk --fp8_base --highvram --save_every_n_epochs 1 --output_dir /media/clownsharkbatwing/SeagateHub1/flux_lora --timestep_sampling sigmoid --model_prediction_type raw --guidance_scale 1.0 --loss_type l2 --optimizer_type prodigy --learning_rate 1 --network_dim 64 --optimizer_args "weight_decay=0.01" "d_coef=0.8" "d0=5e-5" "use_bias_correction=True" "safeguard_warmup=True" "betas=0.9,0.99" --config_file /home/clownsharkbatwing/ai/test/prodigy_flux.toml --output_name flux_lora_test --max_train_epochs 100

[verbal fractal]

thanks

[limber pier]

there's my launch command

[wind chasm]

thats the only toml I got in the directory where I shave my models

[limber pier]

irc buckets don't work yet with flux

try what i just gave you with the paths updated

you'll have to be in the venv for kohya first

so be sure to do

source venv/bin/activate

or whatever it is on winderz, i forget

[wind chasm]

so cmd into the venv and throw that in

[limber pier]

gotta activate it, not cd into it

[wind chasm]

🙃

am I being really dumb ?

the coffee hasn't kicked in yet

i do not

[ornate elk]

And what's the checkpoints size?
If its an sdxl model and 6gb you have to increase the Windows Pagefile

[odd marten]

5/6 gb

[elder mauve]

i can use "upscale models" in auto1111?

[ornate elk]

Yep than you need to increase the pagefile

[elder mauve]

https://civitai.com/models/141491/4x-nmkd-superscale

like this

[wind chasm]

-_-.... its in scripts not bin

[ornate elk]

Yep

[elder mauve]

how?

[ornate elk]

Put them into models/esrgan

[elder mauve]

where do i place it

there is no folder

do i make one?

thank you

ooooo

[wind chasm]

returned not zero exit status 2

WAIT

ok yeah I might just give up here XD

C:\Users\Owner\AppData\Local\Programs\Python\Python310\python.exe: can't open file 'G:\Kohya_GUI_Flux_Installer_v17\kohya_ss\flux_train_network.py': [Errno 2] No such file or directory

THEN WHAT IS THIS

[verbal fractal]

@ornate elk please

[wind chasm]

@limber pier thanks for you help im giving up I fix one error I get another xD il jsut wait for it to be updated or something else

[odd marten]

How can i do that?

[vocal burrow]

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

[odd marten]

ty, But by how much should I increase it?

The Checkpoint is 5,18 gb

Oh, okay, sorry

It makes sense, of course. Thanks, I'll check now

@ruby yoke My Swap Space file size is "11776 MB", and the local disk where Stable Diffusion is installed has ‘1.61 TB’ available

Should I do the same thing to avoid running into those errors?

And i have to do "11776" x 2?

The problem is that I am afraid of messing up, so I don't know by how much I should increase it

[ornate elk]

16gb min and 24max

For C

And make sure its only enabled for C and no other drive
Then restart the PC

[hallow latch]

If I host an GPT on a secondary PC in my network.. and not using listen..

I should be able to access it right?

[limber pier]

something must be wrong with the path

or maybe a permissions issue

[loud pond]

hi @ornate elk is it possible for stable diffusion to give a side view image of a character which i rendered already?

would i have to use image to image?

[ornate elk]

its possible, best would be to use Controlnet IP-Adapter for that
or you use the same prompt but with side view or controlnet openpose

[loud pond]

oh ok, where can i find this controlnet IP0Adapter?

and is it for stable diffusion?

[ornate elk]

ye

you have to install the controlnet extension first

[loud pond]

oh ok and whats the IP-adapter thing?

[ornate elk]

thats a model for controlnet

it can analyse an image, like colors or composition

and then process is into the genearted image

[loud pond]

WOAH that sounds so cool is there a video or a link on how i can use this?

[loud pond]

wait so can i find it on cavit ai thing?

since its a model

where would i go to find it then?

alright ill try and find one

[ornate elk]

https://huggingface.co/h94/IP-Adapter/tree/main

[loud pond]

cool thanks! ill install controlnet and then ill try and download this

[ornate elk]

no problem

[loud pond]

oh wait really?

so i just need to install sdxl and paste it into my sd thing right?

cuz rn i ahve sd 1.5

[ornate elk]

no you can use it for 1.5 too

the link has 2 folders

one is for sdxl

the other for 1.5

controlnet works with both

[fierce hollow]

Question 1) Why I don't have openpose displayed
Question 2) Why do I uncheck controlnet on my own?

https://imgur.com/a/nKbZG5i

[elder mauve]

any way to generate images of same face?

[ornate elk]

your using forge?

[ornate elk]

with controlnet extension and IP-Adapter

[elder mauve]

which extension?

roop?

[loud pond]

alright thank you both

[ornate elk]

roop is outdated

[elder mauve]

ah i see , will use reactor

[ornate elk]

so for Faceswap Reactor is okay or using a seperate tool called Roop-unleashed

[fierce hollow]

Yes

[ornate elk]

controlnet is already included in forge so you cant install it

[fierce hollow]

What about openpose?

[ornate elk]

here i tried with sdxl and canny:

[ornate elk]

is supported by controlnet

you still need the model for it

[coarse pumice]

is pony realism a model or a lora?

[ornate elk]

model

if a file is larger than 2gb its always a model

[fierce hollow]

I don't see a separate menu with openpose where you can edit the position

[ornate elk]

you have to click on the preview button

then on edit on the right

[fierce hollow]

After I put the models in - will it show up?

[ornate elk]

the openpose model in the dropdown

if you mean that

[fierce hollow]

No - it's a model, but there should be an openpose button where you can edit the skeleton itself, I don't have it....https://imgur.com/a/eC5kYPC

[ornate elk]

openpose extension is broken, you would need to fix it first

check the cmd

you need to do pip install basicsr into the venv

[fierce hollow]

I just installed it, as well as Forge itself....

[ornate elk]

yea

[fierce hollow]

Should I enter this in cmd in the folders with Forge?

[ornate elk]

first run this command in the forge folder
venv\Scripts\activate.bat

and then: pip install basicsr

[fierce hollow]

I'm probably too stupid... But I can't find the “Venv” folder in the “Scripts” folder, there's no file you specified.
https://imgur.com/a/u3JURgF

[loud pond]

which version should i get are they all the same or is one better than another?

i have no idea what the descriptions mean 😅 , im very very new to stable diffusion

what would you suggest?

im just going to use this to get the same image but from a different angle

ok! cool, also if you dont mind where should i put the installed files 😅

[viscid oxide]

What's a good discord channel or reference to help me figure out issues with training a Lora?

[ornate elk]

its not inside scripts

its in the main folder

[ornate elk]

models/controlnet

[loud pond]

thanks

[tidal thunder]

Hello guys, i need help about stable diff forge, i'm missing svd and z123 tabs (and extensions)

[ornate elk]

In the latest version they got removed as forge is experimental.
You would need to download the old forge or the forge fork

[tidal thunder]

i've check it out, people texting about this thig, but i don't know how to install this https://github.com/lllyasviel/forge-legacy-extensions

[ornate elk]

Here is the old version:
https://github.com/lllyasviel/stable-diffusion-webui-forge/releases/download/previous/webui_forge_cu121_torch21_f0017.7z

As direct download

[tidal thunder]

thank you so much

[wheat flare]

Been having some trouble with the upscaler for SDXL1.0 in AUTOMATIC1111, running on an Nvidia GTX 1660 Ti. I've tried both of these configurations (first is default, second was recommended to me previously):

With the first config, the upscaler gets stuck on 98% and just sits there

[prisma shadow]

HELP: Stable diffusion says i need a check point how do ii get one i asked gpt and looked it up but cant find anything

[wheat flare]

With the second config, it doesn't even get started

What should I do?

[hallow latch]

Not a problem so.. but when I do Upscale on Img2Img .. I get 2 files of the same image?? Why?

[wheat flare]

Are the resolutions the same?

[hallow latch]

Yeah

[wheat flare]

What do you have the upscaler set to?

[wheat flare]

Try fiddling with the scale factor to see what happens

I don't recognise this UI so I can't help more than that

[hallow latch]

Used this all the time :3 But it's a new setup with the new GPU

[ornate elk]

You always need to set hires steps to 10

[ornate elk]

Download one from Civitai.com

[ornate elk]

Because in the settings there is an option that does that. Its a .jpg and a .png right?

[hallow latch]

Yes it was

Turned this off.

Thanks

[white stirrup]

i get this error when trying to use flux

yep

straight from the official site

[candid moss]

hey i need some help: im using forgeUi with a flux model but also while using the adetailer extension running with a sd model.
The issue i have is that no matter what inpainting denoising strength i use, it always only uses very low denoising strength.
In my most recent generation i had it set to 0.92 but the original image was still influencing the result too much.
(If i set it to 1 i get weird results.)
Any help is appreciated cause im probably just stupid and i hope its not just a forge/adetailer bug :/

[grizzled cypress]

I keep getting weird blue-ish images with comfyui for some reason

[white stirrup]

do you by any chance know where i can get the flux fp8 clip

[woeful knot]

Why the hell do my tile resample steps go from 10 to 30 just by changing the denoise level? The only thing I did there is set it from 0.3 to 1. It did nothing in both cases by the way. It's very hit and miss for some reason, sometimes it works sometimes it doesn't, I have no idea what I'm doing wrong

[white stirrup]

I do not know i was told to change the weight type and clip to the fp8 versions

I have the weight type but not the clip

Nevermind someone sent it to me

I like how both of you responded to that question at the same time in seperate servers

[woeful knot]

Before and after. Notice how it also made the image 8 pixel taller for no apparent reason, just to fuck my life up in photoshop a bit more

I'm having the same exact result with no controlnet model selected, I don't know what's breaking but something is breaking. It says it's loaded in the console, I have no idea what's going on

yeah that's what I just did, I deselected the model so it just ran with the upscaler and maybe the preprocessor, it kept saying the model wasn't selected so the 10 steps were the preprocessor I'm guessing. And the result was the exact same in terms of quality

Is that the sdxl tile model yall use? [4d6257d3]?

[ornate elk]

Its normal that the steps get more, when you increase the denois

[ornate elk]

Sdxl tile resample can't improve the image like the 1.5 tile resample does

[ashen loom]

@ornate elk hey brother im having a new issue today? idk what caused it to get on the fritz, {RuntimeError: mixed dtype (CPU): expect parameter to have scalar type of Float} im runnin the amd version it was working perfectly

[ornate elk]

Hey, when do you get that error?

[ashen loom]

when i try to generate anything. it just locks up for about 1 or two mins.

[woeful knot]

Yeah I'm gonna stick with upscaling with just the base model and ultimate sd upscale. Thanks tho

[ornate elk]

Can you share a full CMD log

[ashen loom]

ofc

[ornate elk]

Or upscale with the tile resample of 1.5

[ashen loom]

hopefully thats what you need

[ornate elk]

Yep

Can you show me the txt2img settings when you get that?

[ashen loom]

its with any generations. i typed cute girl

[hallow latch]

Hey @ornate elk , can I ask a totally random thing that is not to do with tech support? Would love to hear what u think

[ornate elk]

Okay very strange.
Whats your GPU?

What you can try first is to delete the venv folder and relaunch the webui-user.bat

[ashen loom]

okay i will give it a shot

[ornate elk]

Im off now. If it doesn't work, we can fix it tomorrow

[ashen loom]

coolio

[ashen loom]

amd rx 6800

[ashen loom]

i deleted the venv and it worked @ornate elk

[sterile coyote]

I keep getting this error, anyone can help me with this?

nvrtc64_112_0.dll cannot be found under: C:\SD-ZLUDA\stable-diffusion-webui-amdgpu\ .zluda

But the file seems to be where it should be

[zenith bronze]

hi, my webui bat file is stuck at this when i launch it, can anyone help?

[errant dust]

Having trouble upscaling images, where it keeps finishing itself at 50% total progress for no apparent reason and I've changed every setting imaginable. Anyone know what could be the cause?

[sterile coyote]

I followed the guide of CS1o step by step but the ZLUDA I downloaded didn't come with config files I think

https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides#amd-automatic1111-with-zluda

That are my env var settings under "Path"

Yes

I'm running a 6700 XT

Downloaded the respective files, did everything what the guide said

Would it be an issue if I accidentally checked the pro drivers for the AMD HIP SDK 5.7.1 from here: https://www.amd.com/en/developer/resources/rocm-hub/hip-sdk.html ?

Ok, ty

Any idea how I can see if amd pro 23.q4 is installed and how to uninstall it?

[worthy stratus]

hello, may i use the free credits in dreamstudio to test stable diffusion for an article review I am writing about Stable Diffusion? I am planning to publish the image results together with the article review, but I'm not sure if Im allowed to do that

[worthy stratus]

okay, thank you, i sent them an email

[fair oxide]

anyone can suggest how to fix this error i get when trying to install Searge LLM for comyfui
## Execute install/(de)activation script for 'D:\ComfyUI\ComfyUI\custom_nodes\ComfyUI_Searge_LLM' Ignoring llama-cpp-python: markers 'platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"' don't match your environment

im on windows using comfyui

the python version i have installed is 3.10

[thorn pendant]

i running sd and it show this, then run normally, do it effect much on sd

[runic moth]

Does anyone know where I could get help with Flux?

[tall grove]

gm

[thorn pendant]

why sir?

understandable sir

[runic moth]

is it possible to use nf4 flux model in comfyui?

does it require a different text encoder? I can use flux1-dev-fp8 unet and text encoder is t5xxl_fp8_e4m3fn, this works but if I change to flux1-dev-bnb-nf4 then it gives crazy errors.

a whole bunch like this

size mismatch for img_in.weight: copying a param with shape torch.Size([98304, 1]) from checkpoint, the shape in current model is torch.Size([3072, 64]).

looks like that requires a custom node according to reddit, time to install the comfy manager 💀

[elder mauve]

anyone getting this last line in comfyui?

[solemn moon]

sup
What have i missed? Is flux on Auto yet?

[elder mauve]

nope

[runic moth]

oh my god this is driving me absolutely crazy

https://github.com/comfyanonymous/ComfyUI_bitsandbytes_NF4

it says use the comfui-gguf node because its better

install that node, it cant load the flux nf4 model because its not a gguf model

so I install this node instead, even though its deprecated

this node only wants to load stable diffusion models and the flux models dont appear on the list

I seriously want to tear my hair out

[runic moth]

finally got it to work x_x

[ornate elk]

Always run the webui-user.bat

Make sure --xformers is in there too

[thorn pendant]

somehow it lost

[ornate elk]

Hey, check the AMD Adrenalin Version.
It should show 24.8.1

[ornate elk]

What's your GPU again?

[fair oxide]

I am not sure if I could ask the question about the prompt itself here, but I would like to generate realistic character for a game. However, I need to generate the body parts separately. The reason is so that I could move the body parts independently. Also I need the dismembered version of the limbs. Let say I want to display the arm is chopped off and the character is still shooting with the other arm. I have tried with SD1.5 and SDXL but can't really generate what I want. Is that even possible? Do I need to train myself?

[thorn pendant]

6gb sirr

[ornate elk]

Then you need --xformers --medvram --no-half-vae

[thorn pendant]

i'm just changed to opera, idk if my bowser effect sd's effective

[thorn pendant]

where i need to put in on edit tab

[ornate elk]

In the line commandline_args=

[thorn pendant]

ty sir

[ornate elk]

Let me know if you still need help

[thorn pendant]

my img have about 3xx x 4xx size, i i want upscale it withour break its img, how can i do that?

[ornate elk]

Without breaking ?

You mean without crashing?

[thorn pendant]

this is before and after up scale, idk how to keep it quality sir

[ornate elk]

How do you upscale?

With which method

[thorn pendant]

by extra tab sir, 4x

[ornate elk]

Ah thats not the best way to upscale

[thorn pendant]

can you teach me how sir?

[ornate elk]

Sure. Do you want to upscale ai generated images or normal images ?

[thorn pendant]

both sir

[ornate elk]

Okay then you should use the SD upscale script in img2img

[thorn pendant]

just doing that?

is it wwork for both ai and normal img sir?

[ornate elk]

Yes but its better for ai generated images

For normal images you can use the free tool Upscayl for example

Or the extras tab of auto1111

[thorn pendant]

understood and ty sir

[thorn pendant]

somehow sd work pretty fast at first then start slower when i keep using it
idk if that's just my though or smt i cant imagine

[odd marten]

Excuse me for replying today, but I was busy yesterday. I downloaded Stable Diffusion Webui to my local disk ‘D:’ which is an external hard disk that has 2 TB of free space. Should I move it, or can I increase the file size spacing there?

[last flax]

yo does this seem right?

[odd marten]

It says I currently have ‘11176MB’ allocated on my local C: disk, in what sense are you asking me to put a minimum of 16 and a maximum of 24 if they are Megabytes?

Anyway, as mentioned I installed Stable Diffusion on my local disk D:, should I move it to local disk C: ?

[tidal thunder]

for some reasons it didn't work and not convenient to use, so i want to ask, is that possible to install this things and where to put it in?
https://github.com/lllyasviel/forge-legacy-extensions

these files

[last flax]

where do i place my lora's i dont have a folder for it

[odd marten]

go to "Models" and then "Lora"

[last flax]

ooh aight

[odd marten]

Create a folder "Lora"

If that doesn't work, I really don't know what to tell you. You'll have to ask some admin or someone who knows about it

[last flax]

alrightt thanks a lot

found lora in built in extensions

[feral otter]

I keep getting "onnxruntime" errors on ForgeUI, it usually shows up when it runs the reactor process: "[E:onnxruntime:Default, provider_bridge_ort.cc:1992 onnxruntime::TryGetProviderInfo_CUDA] D:\a_work\1\s\onnxruntime\core\session\provider_bridge_ort.cc:1637 onnxruntime::ProviderLibrary::Get [ONNXRuntimeError] : 1 : FAIL : LoadLibrary failed with error 126 "" when trying to load "D:\Stable Diffusion\Forge-Stable\system\python\lib\site-packages\onnxruntime\capi\onnxruntime_providers_cuda.dll"

Any tips of diagnosing the issue?

[ornate elk]

You still need to increase and set the pagefile only for C

As its mostly the fastest SSD of the system

No need to move the webui

[odd marten]

I see, but how much? I don't quite understand what you mean by minimum 16gb and maximum 24gb, because I get MB

[ornate elk]

16000 min

And 24000 max
So yes 16gb and 24gb

[odd marten]

Oh, okay, let's see

[odd marten]

Right?

[ornate elk]

Perfect

Now restart the PC

[ornate elk]

Models/lora

[last flax]

found it

thanks a lot

[odd marten]

Yup

ty again, i hope this work

[ornate elk]

Hey you can use comfyui with hypertile and stable video diffusion nodes.
But you can't download these 3 folders and move it into the new forge. That won't work

[viscid badge]

I have an issue, error code 3, it can't launch Python. I'll show you the error here:

I have python 3.12.5

[thorn pendant]

seiga have those thing on her head, but smh i can make it, what should i do

[vocal burrow]

not compatible, install any python from 3.10.6 to 3.10.11

[odd marten]

@ornate elk I restarted the PC and tried Stable Diffusion, no luck.... It took 100 seconds to load a model and then gave me ‘Error’ written in red above. And to make matters worse, Discord wouldn't let me open either, I had to put the file size spacing back from before

[viscid badge]

same error

not same now

[ornate elk]

Is the external drive a SSD or HDD ?

[odd marten]

Oh God, how can I see it? All I know is that it's a hard disk

I see on the amazon purchase history

[ornate elk]

Delete the venv folder and relaunch the webui-user.bat

[odd marten]

https://www.amazon.it/dp/B0BMGTK46D?ref=ppx_yo2ov_dt_b_fed_asin_title
@ornate elk

[ornate elk]

Thats a HDD

Its good for storing files but not good for installing games or Stable diffusion on it

[odd marten]

I have a local system disk D: with 1tb free. Do I move it there Stable Diffusion?

[ornate elk]

Is the system disk an SSD or HDD ? xD

[odd marten]

EHM, how can i see it?

Oh, right, amazon

https://www.amazon.it/dp/B07YD579WM?ref=ppx_yo2ov_dt_b_fed_asin_title
@ornate elk

SSD, yes

[ornate elk]

Thats good

Move the whole SD-Zluda folder to the D drive

You then need some commands to fix the update path for the webui.
I can show them in a few minutes

[odd marten]

Ty so much

I hope this will fix the problems

[ornate elk]

Also how much free space do you have on C ?

[viscid badge]

i try

[odd marten]

Every time I reboot the PC it changes, but it usually says I have ‘51’ gb free

[ornate elk]

After you moved the SD-Zluda to D: do the following:
To fix the webui update process go into the stable-diffusion-webui-amdgpu folder and click in the File Explorer bar (not searchbar) and type cmd and hit enter.

Then use the following commands one by one:
git init
git remote add origin https://github.com/lshqqytiger/stable-diffusion-webui-amdgpu.git
git fetch --all
git reset --hard origin/master
git branch --set-upstream-to=origin/master master
git pull
Then launch the webui-user.bat

[odd marten]

ty, i'll try now

[odd marten]

By mistake instead of cutting, I copied the folder to the local disk D:, is that a problem? Or do I simply delete the folder on the previous local disk?

[viscid badge]

dont work

[ornate elk]

you can delete the other one

[ornate elk]

can you show the full cmd log

[viscid badge]

its all i have

[ornate elk]

okay thats fine

[ornate elk]

can you uninstall all python versions you find and then install python 3.10.11 64bit at its default path ?

[odd marten]

I did everything and started the webui, it again did the long load for the model and in the terminal it gives me this error:

[ornate elk]

whats the model you try to load?

[sinful frost]

is anyone using the new forge? The negative prompt box is greyed out

[ornate elk]

Its greyed out when you select flux as model

Because flux can't use negatives or dont need them

[sinful frost]

Oh! you're right! I thought I was on SDXL!
Thank you

[feral otter]

After I run image generations for a while on my computer it just starts making blobs instead of images. If I turn ForgeUI off and then restart it they seem to generate well again, same exact prompt.

Is there a build of of files or something that causes this? Can I fix it?

[viscid badge]

already done

[ornate elk]

Can you relaunch the python installer, select modify, click next and then check "add python to system environment"

[tidal thunder]

okay, thanks

[thorn pendant]

i have soem struggle on makeing this 1st img have those thing on hair like 2nd img
i tried fix thew negligee and char lora but smh it fighting each others tho
i want to make the things like 3rd img
hope someone can fix it
ty

[novel temple]

hello people, i seem to have an issue with stable diffusion
i generate images with pony and SD crashes on me
i can generate 1 image but the 2nd makes it crash, i get connection errored out and no explanation why

[prisma shadow]

I have 3 models in side "models/Stable-diffusion" and when I try to generate a image it is always a blank grey screen any fixes?

[novel temple]

ok it got weirder
i changed my gen settings to be more GPU efficient
now i get this after 1st gen

[last flax]

any thing that could help me fix this?

[vocal burrow]

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

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

[vocal burrow]

what command line args ?

[last flax]

thanks a lot!!!

[ornate elk]

flux is not supported in auto1111 right now

[ornate elk]

whats your gpu?

[loud pond]

hi @ornate elk, do you have instructions on how to get an ai which will give videos locally?

[novel temple]

4070 12GB

[ornate elk]

you need to have --xformers in the webui-user.bat

[novel temple]

i didnt, then i added it and now i removed medvram and medvram-sdxl

seems to run better

and actually run

[prisma shadow]

Thank you so much ive been trying to figure this out for hours all i had to do was change the model

[ornate elk]

no problem 🙂

[loud pond]

how would i get a video of a character walking? do i use comfyui with vid diffusion and then do i use control net on top of that?

[ornate elk]

hmm controlnet only helps if you do video 2 video

[loud pond]

alright

thanks

[ornate elk]

you can use AnimateDIff to generate short videos

[loud pond]

yeah cuz i wanted to use stable diffusion to make a walk cycle

this was the best i got with text to image

but its not using complex pictures so i was thinking maybe something could animate the character walking instead

and from their i could use that

[ornate elk]

someone already has done that on this discord

[loud pond]

wait fr where?

do you know where i could find it

[ornate elk]

he used controlnet openpose and a walking animation from a webseite

[ornate elk]

its burried under millions of messages in this discord

[loud pond]

oh

[loud pond]

do you know which website?

[ornate elk]

found it:
#🏞｜general-with-images message

[ornate elk]

i have bookmarked it but cant remember the name xD

[loud pond]

ah ok

[ornate elk]

found it

https://posemy.art/

[loud pond]

bro thank you so much ive been trying for days to get a walk cycle xd

[ornate elk]

No problem 🙂

[jolly turret]

Why're there no preprocessors ?💀

[sterile coyote]

Edit: Wait, I need to double check, sorry

[loud pond]

GOT IT

THANKS SO SO MUCH MAN

[rich garden]

where can i download Stable Diffusion 1.5? i only found this link with a 404 error. ping me please~ https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/main

[sterile coyote]

So I followed the guide for Automatic1111 with ZLUDA 2 times, exactly as described here, step by step:
https://github.com/CS1o/Stable-Diffusion-Info/wiki/Installation-Guides#amd-automatic1111-with-zluda

But I always get this error:

Commit Hash: 595ce8dc2c661282d4a874a859e56cb5d3a4b59ab
Failed to load ZLUDA: Could not find module 'C:\SD-ZLUDA\stable-diffusion-webui-amdgpu.zluda\nvrtc64_112_0.dll' (or one of its dependencies). Try using the full path with constructor syntax.
Using CPU-only torch.

@ornate elk I checked the AMD Adrenalin Version, it's 24.8.1

[calm cairn]

is SDXL supposed to be bad at inpainting? I keep getting error messages that 1.5 models dont give me

[onyx solar]

Are you trying to download the base 1.5 model or are you actually looking for the software that works with 1.5 models? You do not need to have Automatic1111 version 1.5 to create Stable Diffusion 1.5 model-based images. These are 2 different things.

[onyx solar]

SDXL works fine with inpainting. People would likely be able to assist more if you post the details like the errors you're getting, the settings you're using, etc.

[calm cairn]

ah sorry, i can do that

[hasty orbit]

--xformers --no-half-vae

^For Linix is there anything special I have to do with this?

Or is it as simple as putting into the webui-user.sh file

[worn topaz]

Hey there! I'm trying to get SF3D up and running, and followed all the instructions in the installation, but couldn't figure out how to get it to run

I even followed the comfy UI extension instructions (minus the part where it said to go to sf3d_code_release cause that doesn't exist) but I couldn't figure out how to actually use the software

Is there something I'm missing? I'm using Windows if that helps; all other requirements (like python 3.8+ and VS2022) are good to go

[arctic zinc]

Is there a laptop recommended for stable diffusion? I have a Mac and it’s way too slow generating images.

[rich garden]

tbh idek.. i was just told to download SD 1.5 cuz i'm still new to the whole generating AI art thing

i currently have 1.10.1

[ornate elk]

Can you show me the full cmd log ?

Also try to delete the .zluda and venv folder and relaunch

[ornate elk]

You don't need the base 1.5 model

Its over 2 years old and doesn't generate good images

Better use community made models based on 1.5 or SDXL

[sharp valley]

is it possible to use flux on stable-diffusion-webui

when I try to use it, it seems it uses either a random model or a model used previously

even though the software still reports back as the correct model for some reason

[ornate elk]

Nope, flux is currently not supported in Auto1111

[sterile coyote]

Thank you, CS1o, the deletion helped, I'm one step further now, trying to generate my first image, currently waiting if it works 🙂

[ornate elk]

Perfect np!

[sterile coyote]

...picture generation with the default model and a downloaded model worked as well 🙂

[tidal thunder]

may i get the help about animatediff in stable diff forge? there's missing module

[onyx solar]

What I'm saying is that they're 2 different things. Automatic1111 is not Stable Diffusion 1.5 itself, just one of many applications that can use 1.5 models. Having the version of the application you currently have is perfectly fine and it can use 1.5 models.

[dusk shuttle]

I have a question related to running on automatic1111 on AWS. Is this the right channel to ask?

[rich garden]

[hollow ledge]

https://www.youtube.com/watch?v=zo6w-Md7DUo

[turbid carbon]

So I got this massive error code. Do I need to reinstall or update stuffs?
And yes, Forge is the only option I can use on this potato.

[ruby yoke]

you could use a standard method of animating a sprite sheet

[dusk shuttle]

Are you also using the same seed as the prompt you are trying to replicate? Maybe your seed is set to random.

I have a what might be a noob question.
I've been trying to follow this guide to run auto1111 on AWS EC2: https://github.com/blekmus/aws-stable-diffusion?tab=readme-ov-file
I got the auto1111 webui installed and when I run "bash webui.sh", its gives me a local url http://127.0.0.1:7860
The last step in the guide is "Connecting", To establish a secure connection to the WebUI using ssh tunnel. I used putty, made the connection.... and now what? I tried the local url in my browser but it just says connection time out.
I am new to AWS so i dont understand the issue. Added my ip address to security policy of my instance.
Can someone help please? How do I access the webui of auto1111?

[spare wigeon]

Yes, fortunately most of the pictures comes with a seed so im also entering that.

[loud pond]

what? how do i do that?

also @ornate elk do you know if its possible to make 3d art like for blender using ai?

[ornate elk]

not with stable diffusion webui, but there are other tools

[loud pond]

oh really like what?

[ornate elk]

dont know the name rn sry

[loud pond]

oh alright, btw i found something like this

https://civitai.com/models/87432/3d-room-blender

would this be possible to just make and import to blender?

[ruby yoke]

usually, you're calling them in a game. here's a tutorial for gamemaker https://youtu.be/vdM4HkqJ020?si=fQe-fQTGanxGQDny

[loud pond]

cheers, i recently made all the characters moving thanks to CS1o, all i needed to go was to animate them

[sinful frost]

This is random, but does anyone have a problem making mediapipe_face work in Adetailer on Auto1111? I keep getting this error message: nothing detected on image 1 with 2nd settings.

[thin cave]

im having trouble with installing stable diffusion 1.9.4 on arch linux

[lime imp]

Guys, who knows, this message appears when start generation
TypeError: '>' not supported between instances of 'int' and 'str'
generation doesn't work

This error appeared after restarting SD forge

oh I got it! I accidentally hit this half-knock

and the value was set to 2

[molten rampart]

Can I ask here a problem with Forge and Flux? I'm going crazy 😦

I don't understand: Is my 4080 16gb not powerful enough to run dev fp16 or fp8? D:

[azure magnet]

Hi, how is call the extension for using controlnet on comfyui?

[ornate elk]

It should be

Use nf4 model

[steady thistle]

I have Swarm I copied over from my previous PC. I have a brand new PC now and I installed Git and Python 3.10.11 (and added to Path). Can I get it Swarm to work on this PC without redownloading it?

I was about to delete SwarmUI.sln and running the install-windows.bat

[ornate elk]

swarm ui also needs .Net 8

but you may need to reinstall it

[lunar perch]

when u remix images in civitai it auto-applies the loras without showing their names in the prompt window.

forge webui requires you to type the lora name as <lora:name> but the actual lora names are often not the same as they appear in civitai, and it's a pain to go track them down in the local folder

how do I easily grab the actual lora name I need to enter in <lora:> format into hte prompt window, from civitai?

[quaint idol]

Hello can someone help me with an error "error: RPC failed; curl 92 HTTP/2 stream 0 was not closed cleanly: CANCEL (err 8)" I think its my mobile phone internet speed so I tried hotspot then usb cable I get at the maximum 28mbps for a speed test, git has cloned stable diffusion and I still have the same error with webui user.bat

before the error where it says "remote: Enumerating objects: 580, done" it gets to about 19% then there is more text with an error 128?

[fair oxide]

how can I point automatic11 to use my second gpu for installing? i have an amd gpu in my first slot and its trying to use that one and an nvidia cardc in my second slot. I want to use the second slot gpu not amd gpu

[steady thistle]

Ah yes, thank you, that seems to have done it.

Yeah, that was it. 16 vram on my Nvidia RTX 4700 Ti Super with 32 gigs ram and I'm finally cooking with gas.

[quaint idol]

....is there a way to enable http version 1.1 for git when using webui.bat

and to enable a larger buffer

[sudden hawk]

Hello, I have this grey image when I try to generate with the checkpoint ariesFp16NoEmaNoVae_v32. can someone help me pls ? It was working yersterday so I don"t understand

[ruby yoke]

did you change your latent size?

[sudden hawk]

no

[ruby yoke]

did you change anything at all?

[sudden hawk]

absolutly not, thats why i dont understand

maybe i need to update my SD? i don't remember how to do it

[ruby yoke]

and you've rebooted, yes? just to make sure nothing's hung in cache?

[sudden hawk]

i can try

[ruby yoke]

please do a completely shutdown, power off, leave off for 10 seconds, then power on and boot up. to clear everything and load everything fresh

[sudden hawk]

it still don't work..

[ruby yoke]

okay, i'm lost on why it's having a problem. i doubt it's because you need to update anything. i don't recall seeing a new version of webUI and stable diffusion itself doesn't get updated, we just get new advanced models.

if you can wait till @ornate elk gets back online, he might have some solutions

[sudden hawk]

okei, thanks for your time 🙂

[cerulean furnace]

new to Stable diffusion and i have AMD 6700xt 12 gb VRAM do i install it on my Windows 11 or to create a dual boot with linux mint and to install Stable Diffusion there ?

[quaint idol]

what is the origin repository for stable diffusion? I changed it to [e-mail] and cant use git pull

[quaint idol]

You can definately install on Windows 11 Im in the same process Ive been trying for about 8 hours

You need a good internet speed or git can show errors

[cerulean furnace]

but i read that it's better to install on linux if you have an AMD gpu no sure if that's true

[quaint idol]

Im new too I guess linux could be faster but I guess it would depend how much detail you want to push out of your pc with quallity of pictures would be, I bet it would be fine on windows 11

Im having download errors using a mobile Ive read if you have better internet theres not much trouble

im using Nvidia so I cant really help

try on 11 first it should still be fast gee I'm using an RTX 2080Ti and quite old CPU but ive read a picture generates in about 7 to 10 seconds

while I wait what GPU is it

Ill have a look online

[quaint idol]

There are different results when comparing different versions, the benchmarks say 11 is ok for one version and linux is ok for another version

https://www.pugetsystems.com/labs/articles/stable-diffusion-linux-vs-windows/

[cerulean furnace]

will my Sapphire rx 6700xt nitro works with rocm on linux ?

[ruby yoke]

there are AMD guides you need to go through. look for the push pin icon at the top of this channel, they're pinned

[south vigil]

yo guys after using the latent upscale my picture gets a bit weird, does anyone know why?

[ruby yoke]

wow, that's some ugly aliasing

[south vigil]

ye, what could be the reason?

[ruby yoke]

i'm going to blame your upscaler - which one are you using?

[south vigil]

omg im not sure why i didnt thought about changing the upscale model, sometimes its a bit overwhelming with all those settings lmao xP

ah no wait, the latent upscaler has no model right?

it looks like this, after the first ksampler the picture is aight and after the second the problem hits

not always but often

after the second ksampler the actual upscale begins where it works actually fine

[ruby yoke]

okay, well, that narrows down where the issue is coming from at least so start playing around with the values there

or just leave it if the final result is good

[south vigil]

yeah thats true, I thought maybe someone would know what setting it could be either denois eor cfg i thought

[south vigil]

the final result is sadly not "good" it looks fine but the ultimate upscaler doesnt correct the issue it just enhances the quality

[ruby yoke]

okay well, it's fairly late at night for most of the members that hang out here. you might need to wait a few hours till others log back on

[thin cave]

im having trouble with installing stable diffusion 1.9.4 on arch linux

[ornate elk]

You then need to add --device-id 1 in the webui-user.bat

[ornate elk]

You can use my AMD Zluda guide if you want to use it on windows, or you can dualboot or use WSL2 to install it on Linux

[ornate elk]

Denois 0.4 is to high when not using Controlnet tile model.
So try 0.3 or 0.2

[south vigil]

thank you this helped me

one more question, I have a lot of problems with hands and watched some videos but all of them didnt work, is there anything you can recommend doing for getting better hands?
or maybe even a photoshop tutorial for fixing hands? i relaly try my best but cant find anything useful

[ornate elk]

Fixing hands is really hard and mostly people only do it with Photoshop.
To not get bad hands you would need to use sdxl or pony models.
And Controlnet openpose for 1.5 models

[south vigil]

I see so there are actualy lora models for sdxl and pony i will research bout that, do you maybe have a good photoshop tutorial? the last time i tried following some tutorials i got hedache even though i use it regularly for thumbnails

[ornate elk]

Nope sry, and yes there are loras for hands or embeddings but they can't fix them reliable too

[south vigil]

alright thanks for the information, so to get consistent good works i gotta be able to use photoshop properly

[ornate elk]

Depends on what you want to do and how important the hands are in the image

[south vigil]

in my case the hands are heavy weighted sadly

[ornate elk]

Or use sdxl or pony models to not get bad hands in the first place

[south vigil]

yeah i will actually experiment a bit out with those first, if i have embeddings installed i only need to use the key word to activate them right?

[ornate elk]

Yes

[south vigil]

thanks so much that enhanced my understanding a lot

[ornate elk]

no problem

[molten rampart]

Can someone please help me with Forge and Flux dev? There is no way to make it work, but I've done everything right

[ornate elk]

whats the problem?

[open wharf]

i see that i need to python version 3.10 to install the Automatic 111 UI. i have 3.12 is it possible to downgrade

[ornate elk]

You need to uninstall 3.12 then install 3.10.11 64bit
Then delete the venv folder and relaunch the webui-user.bat

[scenic plaza]

hello, can someone help me why the stable diffusion a1111 ui takes up 15gb of vram just by starting it without any generation?

[shut hazel]

Hi, I want to train an inpainting model where I go from multiple input images to an output with stable diffusion. Is this possible? If yes, can you guide me through the process?

[open wharf]

by webui.bat am i to assume you are reffering to the Stable diffusion installation?

[ornate elk]

If you installed the Automatic1111 Webui you always need to launch it with the webui-user.bat

Not webui.bat

[ornate elk]

Whats your GPU ?

[scenic plaza]

rtx 3090

[ornate elk]

hmm, do you use sdlx models?

try switch to an 1.5 based 2gb model and relaunch

[shut hazel]

Any idea about this guys?

[unkempt root]

how to add dreambot?

[scenic plaza]

Changed to ForgeUI seems to be fixed but now im getting "Refiner is currently under maintenance and unavailable. Sorry for the inconvenience." ?

[ornate elk]

yea refiner is not needed

[scenic plaza]

okay thanks

[errant thistle]

im seeing clip skip 2 under a lot of png info.
Does clip skip reduce or increase image quality at all?

...better question actually. Is clip skip something thats better to use on some models rather than others? And does the reccommended amount change?

[jolly mango]

sorry how to use this? i only joined recently, and did not find instructions in the server guide

[ornate elk]

Hey, you can install stable diffusion localy on your PC.
Check the link of the pinned messages for install guides.

[lament jasper]

Hello I need to make a stable difusion model
should be an image blender 2 images, and fine tune it with lora or dreambooth, than deploy it on replicate

if you are interested on this job and want to help me
DM ME

[oblique meadow]

Depends on the model, for just normal sdxl models, clip skip usually increases creativity at the cost of some prompt following. PonyXL models which are popular as well, require clip skip of 2.

[fair oxide]

this didnt work, i do wanna mention that my gpu isnt recognized in task manager since its an old mining gpu

[ornate elk]

What GPUs do you have ?

[fair oxide]

6600 and p102-100, basically a 1080 ti

[ornate elk]

Ah okay

[fair oxide]

got it working with LLMs just not sd

[ornate elk]

Don't think that will work then

[fair oxide]

i think i may know away

ill try later

[open wharf]

just wanted to be clear about it

[solemn moon]

Argh, thinking of moving onto forge I guess.. is there something similar to lobe where forge UI can be customized?

I read that it doesn’t use negatives

[rare mountain]

guys how can i set up an sxdl workflow so i can use sxdl and loras like this? https://civitai.com/images/12873849 I've found sxdl is a huge pain in the ass to use with comfy ui for some reason compared to sd 1.5 or flux and am struggling big time trying to replicate the image in the link

[solemn moon]

Read it here 😅

facts, i pop in to help if i can and try keep up with everything, comfy too hard, i click install missing nodes from a workflow and nothing happens so gonna move onto forge seeing as though Auto has stopped Auto'ing

[spare wigeon]

Greetings y'all! Im a complete beginner with AI, but recently i've wanted to dip my hand trying AI art. I've learned a few things here and there, but there something im not understanding at all.
On the website CivitAi, you can get LoRAs and Models to get your machine to draw specific artstyle and/or characters, but even if I use the exact same settings, seed and prompts of a posted picture, im not getting anywhere close to the artstyle depicted. Is there a reason why?
Also, i'm wondering how can our machine learn what's "good" and what's "bad", since we don't have an option to tell it "that's a good image" or "that's a bad image".

[ornate elk]

To get the same image like on civitai is not alwaye possible but make sure you do it like this if you use Automatic1111:
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.

[fair oxide]

trying to use controlnet but this keeps happening, i could only find one thread on it and it didnt help worth a shit, send help

[ornate elk]

Can you share the whole CMD log and the txt2img + Controlnet settings

[fair oxide]

cmd log?

heres my settings

[ornate elk]

The cmd

The black window

Also do you use sdxl or 1.5 ?

As model

[fair oxide]

pony

[ornate elk]

Ah okay

You may need an other Controlnet openpose model

[fair oxide]

wdym

[ornate elk]

I mean that you may need to use an other Controlnet openpose model

Like the bottom one from here:
https://huggingface.co/lllyasviel/sd_control_collection/tree/main

[fair oxide]

which one

[fair oxide]

this??

[loud pond]

for some reason control net keeps giving me front pictures and i want a view of my character walking away from the player

whats wrong?

[ornate elk]

The bottom one

[fair oxide]

bet

lemme download it and try

[ornate elk]

Hmm did you downloaded controlnet manually instead of using the extensions tab for install?

[fair oxide]

i dont remember tbh

i downloaded it forever ago

[ornate elk]

The webui too ?
Maybe both is outdated then

[fair oxide]

😔 man ig i gotta install the newer version?

[ornate elk]

Looking at the cmd and the folder name I would recommend a clean reinstall

[vocal sundial]

hi, can anybody help me with this? I just updated forgeui and they started showing up

[fair oxide]

alright ill try that

anything i might need to know before i reinstall

ill be doing it manually since my internet is ass

[ornate elk]

Do it with my guide from the pinned tabs

[fair oxide]

bet

[ornate elk]

And then you can move the models over

And install controlnet via the extensions tab in the webui

[vocal sundial]

also getting these errors:

[ornate elk]

Try updating the extensions

[steady thistle]

I can't get ReActor to install. Sigh. I opened the cmd from my AI folder and python_embeded\python.exe -V gave an error, but python.exe -V showed me 3.10.11

[vocal sundial]

ControlNet is integrated and it is deactivated

Regional prompter I already downloaded again from the github and does not show up anymore in the UI

[steady thistle]

I might be moving forward on it. Crossing fingers

[ornate elk]

Okay

[steady thistle]

Could you help me troubleshoot my ReActor install? I keep getting "fail" on FaceSwap also

[proper swallow]

Is there a reason why some prompts I see shows:<lora:Detail Tweaker LoRA add_detail:0.7> but in my prompt it shows lora:add_detail:0.7 when clicking on the lora add detail.

[steady thistle]

I can't seem to get Insightface to install even tho I went through the git with python.exe -m pip install insightface-0.7.3-cp310-cp310-win_amd64.whl

[vocal sundial]

so, should i just ignore them?

i want to keep using regional prompter

[steady thistle]

Ah, with Swarm, I have to go into the dlbackend folder to do comfy stuff

mesh_core_cython.cpp
insightface/thirdparty/face3d/mesh/cython/mesh_core_cython.cpp(36): fatal error C1083: Cannot open include file: 'Python.h': No such file or directory
error: command 'C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.41.34120\bin\HostX86\x64\cl.exe' failed with exit code 2
[end of output]

note: This error originates from a subprocess, and is likely not a problem with pip.
ERROR: Failed building wheel for insightface
Failed to build insightface
ERROR: ERROR: Failed to build installable wheels for some pyproject.toml based projects (insightface)
Ok
Done

I think installing Swarm vs Comfy was a bad idea. I should just do ComfyUI and just dl the 6 or so gigs

[ornate elk]

Forge is an experimental webui. The problem is that the extension can break at any time

[ornate elk]

You need to download the insightface WHL file before you can install it

[steady thistle]

I think I'm fighting with two locations with SwarmUI, but I don't think I need the SwarmUI frontend anyway now that I'm getting into comfy more

[ornate elk]

Ah okay

[vocal sundial]

I was just trying to get SD3 to work

guess I'll revert

[ruby yoke]

did forge ever put sd3 support in?

[vocal sundial]

Idk, i just assumed since it has flux options

[steady thistle]

Okay, I haven't opened Manager yet, should I put insightface WHL into a certain folder? And which?

[ornate elk]

In the folder where comfyui is

[steady thistle]

Okay, cool. Thanks.

Okay, got it and done. Now I'll try and start installing stuff as I think I have all the prerequisits

[ruby yoke]

they're not the same, and i'd be surprised if forge added sd3

[vocal sundial]

I was just going to give it a try, but updating forge broke everything

[steady thistle]

I dunno, it failed again on a fresh ComfyUI install, carefully as I could following the instructions.

[ruby yoke]

anything in your update logs?

[steady thistle]

Ooh, I think it's a pip update issue

nope

Okay, I'm getting this error even though I have insightface:

I tried updating pip but it's reading the wrong version. I have 24.2 but it says 24.1.2

[vocal sundial]

where's it?

[ruby yoke]

it would likely have been in your console window

[vocal sundial]

I posted them earlier

[steady thistle]

Can anyone explain this and how to fix it?
WARNING: The scripts pip.exe, pip3.11.exe and pip3.exe are installed in 'D:\Ai\ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI_windows_portable\python_embeded\Scripts' which is not on PATH.

[ornate elk]

you have to install python 3.10.11 and check "add python to path"

[steady thistle]

If I didn't update my drivers I'm going to do something terrible to myself.

[steady thistle]

I was very careful to do that... Let me see exactly what I did... Do I have to uninstall and reinstall?

I kept going back to it but I can't find the PATH option

[ornate elk]

you can also run the python installer again, click modify,, next, and then select "add python to environment variables"

[steady thistle]

Okay, I can confirm that it was already checked

I opened up this PC box yesterday and didn't update the drivers yet. Maybe that will help. No idea.

[vocal sundial]

SD3 support?

[ruby yoke]

those are pull requests on github

[vocal sundial]

=/

dunno then, sorry

[steady thistle]

I was able to check Python versions in multiple folders and it's available. I don't know why that pip error came up.

I keep getting an error with 'building wheel' and I have no idea what wheel it's talking about

[hardy hare]

can I use amd npu to run stable diffusion or anything at all, cuz I cant seem to find anything online. Only ryzen ai and shit of amd

[steady thistle]

I may not have installed all components required for Visual Studio Community

Nope, that didn't do it. Still no wheel whatever building

Is this maybe an onnx error?

[ornate elk]

What npu is it?
A 780M and a 680M is supported by Zluda for AMD

[steady thistle]

omg, I mismatched my Python version. It's fixed. Thanks for the assistance.

[steady thistle]

All this was in red and didn't work properly. Ideas?

It starts when I use the Face Restore Model, I think

@ornate elk Do you have any thoughts on what's happening there? It's when I turn on GFPGANv1.3.pth to improve the face quality.

[ornate elk]

Use Codeformer with reactor

Gfpgan is not good for realistic faces