🤝｜tech-support

[fallen nymph]

can i use --xformers --autolaunch in the commandline args

[karmic crown]

Edit the webui-user.bat (right click), At the line COMMANDLINE_ARGS= You add: --skip-torch-cuda-test --no-half --use-cpu all --opt-sub-quad-attention

[karmic crown]

xformers is for nvidia

[fallen nymph]

oh i see

what does it do though

[karmic crown]

use those from guide

[fallen nymph]

should i use medvram?

i got 12 gigs

[karmic crown]

i think so, but you can check later when some of windows guys came

they will help you better, i am mac guy

🙂

[fallen nymph]

okay i'll try with the args from the tutorial above first

les see

welp brb in 2-5 mins got to install venv files

[karmic crown]

i am not sure if you can use --medvram-sdxl but i guess not, since it is not mentioned in guide

[fallen nymph]

i can scale images up to 1920x1080 right? i'll need them as wallpapers for my website (just local project that'll be hosted)

[karmic crown]

ok CS1o is here 🙂

[ornate elk]

you need --medvram --opt-sub-quad-attention --opt-split-attention-v1 --upcast-sampling --no-half-vae

[lilac cobalt]

I have run it with all the commands you have told me but it still does not work.

[fallen nymph]

wow 12 gigs of vram to use medium settings 😄

[ornate elk]

yes but for that you need the tiled diffusion extension with tiled vae

[ornate elk]

on Linux you wouldnt need it ^^

[fallen nymph]

eh i am on windows soo

[ornate elk]

but windows isnt good supported rn for SD with amd

[karmic crown]

Edit the webui-user.bat (right click), At the line COMMANDLINE_ARGS= You add: --skip-torch-cuda-test --no-half --use-cpu all --opt-sub-quad-attention

and then just run webui-user.bat

[fallen nymph]

sadly

though i prefer amd's software anyday in the week over nvidia's

[ornate elk]

even 24gb amd vram needs --medvram 🥲

[ornate elk]

so true

[karmic crown]

and i can bet that it works faster than my macs 🙂

using a1111 on macs and 1660 should teach you to be patient 🤣

trust me, you are lucky

[fallen nymph]

u got 24 gigs?

oh okay

it ran

automatically

even without --autolaunch

[karmic crown]

forget autolaunch

where people found those outdated information? 🙂

[ornate elk]

yes, RX 7900XTX and in the old PC a GTX1080

[fallen nymph]

sheesh

amd > nvidia sadly nvidia is supported in wider range

[wind sun]

I installed everything like the videos are showing and yet when i try to generate something, nothing happens, can someone help

[fallen nymph]

bro wow

quite the load on the gpu 💀

[karmic crown]

that is the point

[fallen nymph]

@ornate elk when i wanna upscale image from 512x512 to 1920x1080 which resize mode should i use? resize and fill?

[ornate elk]

you should use 540x960 and then hires fix upscale by 2, denois near 0.5, hires steps 10

that will get you a FullHD image

[fallen nymph]

hmm

okay lemme try

[ornate elk]

but like i said you first need an extensions to prevent out of vram errors if your going for that resolution

512x768 upscaled by 2 should work without extensions

[fallen nymph]

which extentions

do i get

😄

[ornate elk]

Tiled Diffusion, click the extensions tab then on Available then on Load from

[fallen nymph]

yeh ik

this?

any others ?

[karmic crown]

for that purpose - no, but you can install many others

[ornate elk]

if your into anime, get the boorutag autocomplete extension

can also help for non anime

[fallen nymph]

are there any styles extentions

[ornate elk]

yes i think so

[wind sun]

@ornate elk ay can i steal your attention in your DMs for a bit, please

[karmic crown]

i suggest to check those extensions

[ornate elk]

if you think it cant be discussed here, sure

[wind sun]

can be

[karmic crown]

i said CHECK! not install, install what you want 🙂

[lilac cobalt]

still does not work

[karmic crown]

show me your webui-user.bat

[fallen nymph]

@ornate elk would the img2img work to upscale the image?

[karmic crown]

@lilac cobalt

[lilac cobalt]

wait

[ornate elk]

only with the sd upscale script

[fallen nymph]

because it'd be faster to generate 4 images 540x980

[ornate elk]

you can rerun images, put them into png-info tab and click send to txt2img, then enable hires fix

[fallen nymph]

right, thanks

[lilac cobalt]

Is it ok?

[karmic crown]

you need to add those above after = in line 7

and to remove that on the bottom

[wind sun]

so the problem is i did every step from the video tutorials and when i launch the stable diffusion webui it send me yk in the web with everything looking perfect but then when i type something in the promp and click the generate button, noyhing happens like im not clicking anything

[ornate elk]

what does the cmd shows?

[lilac cobalt]

now shoud work

[karmic crown]

try go generate simple image like 1girl with 512x512px

[fallen nymph]

can't see hires fix in img2img

[wind sun]

simple than a cat with a hat ?

[karmic crown]

it was for @lilac cobalt

[wind sun]

mb

[karmic crown]

sorry for confusion

[ornate elk]

Its in txt2img, in img2img you would need the SD upscale script

[karmic crown]

resize is not enough?

[fallen nymph]

lol

[karmic crown]

i havent try to be honest

[ornate elk]

How did you installed it?

[fallen nymph]

are these settings fine

[ornate elk]

Won't work for amd GPUs cause to much vram usage

[ornate elk]

Yes but you also need to set the resolution to 512x512 and the denois to 0.2

Also choose an upscaler

At the bottom

[wind sun]

what do you mean ?

[fallen nymph]

that good?

[ornate elk]

Looks good

[fallen nymph]

is the resize mode correct?

[ornate elk]

You can use that or the just resize

[fallen nymph]

alright, i'l test them both

also peak pfp

[ornate elk]

What install guide did you followed?

[fallen nymph]

wasn't it supposed to be 1920x1080 😄

[lime elbow]

I had to redownload SD and I don't seem to have the start up UI I had before where I could set command line arguments, auto update etc anyone know where it is?

[fallen nymph]

webui-user.bat

open empty notepad and drag webui-user.bat there

[lime elbow]

interesting

do i save the text file when im done?

[fallen nymph]

yes

[lime elbow]

and then I just run it with the webui shell script?

[fallen nymph]

mhm

[ornate elk]

If your initial one is 540x960 then yes

[lime elbow]

fanks

[fallen nymph]

it was but it didn't upscale

[ornate elk]

Pls make a new folder and follow my guide:

Here is a quick Guide to install Automatic1111 Webui for Nvidia GPUs (Stable Diffusion)

You need to install Git 64bit: https://git-scm.com/download/win
and Python 3.10.11, 64bit: (any python above or 3.10.6) No 3.11, 3.12 https://www.python.org/downloads/release/python-31011/
And check "add python to path" when installing Python.

1. Make a new Folder on your drive (not on Desktop, downloads, documents, Programms, Onedrive) and name it Ai for example:
   C:\Ai\
2. You go into the folder you created in this case Ai, then click in the File Explorer bar (not searchbar) and type cmd then press enter.
3. Then you copy and paste this command:
   git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
4. Press enter and after its done you can close the cmd and launch the webui-user.bat inside the stable-diffusion-webui folder.
   After the Installation that can take a while. Youll get an URL http://127.0.0.1:7860
   Thats the webui you open in Browser.

[karmic crown]

540x960 vertical
960x540 horizontal

🙂

easy, right? 🙂

[ornate elk]

Then something is wrong

Try Resrgan4x as upscaler

[fallen nymph]

yeah you flipped the res

lmao

is there a way to disable hiresfix in txt2img?

it's upscaling every image i generate

[ornate elk]

Uncollapse it

So its hidden

[fallen nymph]

i see

[wind sun]

so no 3.12, i have to download only from 10th series ?

[karmic crown]

that is most confusing option from a ui perspective

[fallen nymph]

it takes quite long to generate them pics

[fallen nymph]

the hiresfix?

[ornate elk]

Yes. 3.10.11 is the Windows latest

[karmic crown]

ux desiner from my team just said "no comment" 🙂

[fallen nymph]

yeah i think tick box would be better for ux

[wind sun]

oh i have 3.12

[ornate elk]

Then you need to uninstall it if you dont need it

[fallen nymph]

@ornate elk how long does it take u to generate 4 images on 20 steps

u prolly got x3d cpu aswell

[ornate elk]

On what settings?

[lilac cobalt]

the image is being generated but it is taking a long time

[karmic crown]

you are using cpu onlu

what is your intel again?

how much vram?

[fallen nymph]

idk 512x512

960x540*

[lilac cobalt]

GPU: Intel(R) UHD Graphics

[wind sun]

right

[karmic crown]

you need to use CPU only, with options i sent, sorry

Intel Arc GPUs can work, but not UHD

[lilac cobalt]

RAM de adaptador 1,00 GB (1.073.741.824 bytes)

vram= 1,00 GB (1.073.741.824 bytes)

but are there ways to use stable diffusion online?

[fallen nymph]

yeah but most website limit you

[lilac cobalt]

I want to generate images ia. but either I have a few uses a day or it's paid for.

[wind sun]

what about that ?

[ornate elk]

I can test later

[karmic crown]

Unfortunately you only have a few choices - free but limited online services, paid online services, slow generation on your PC. sorry 🤷‍♂️

good graphic card is the key for fast SD image genearation

[fallen nymph]

i cannot generate 4 images tho, runs out of memory for some reason ;d

[karmic crown]

try to crate 4 baches instead of 1 with 4 images

it should help

[fallen nymph]

i'll give it a try, thanks

[karmic crown]

so 4, 1, not 1, 4

[fallen nymph]

yep, understood 😄

but my images are like cartoonish with low quality, in the prompt i've written 4k blah blah

[karmic crown]

what model are you using

do not use default one

go to civitai and find a decent one

[fallen nymph]

sdxlTurboUnstable

[karmic crown]

turbo is for fast generation, not for quality

wellcome @twin jackal

i already know why you are here 🙂

[twin jackal]

Hi, if anyone is running Mac with Intel plus eGPU successfully, then I'd be super happy to hear about your A1111 setup! I'm on Intel MacMini i7 and MacBook Pro i9 with RX 6900 XT (eGPU)

[karmic crown]

@twin jackal please open terminal, go to sd folder and type those command and send the result

venv/bin/python --version
venv/bin/pip --version
git --version

then do
git pull

[twin jackal]

Python 3.10.13

pip 23.3.1 from /Users/torquee/AI/stable-diffusion-webui/venv/lib/python3.10/site-packages/pip (python 3.10)

[karmic crown]

also fine

[twin jackal]

git version 2.43.0

[karmic crown]

what git pull gives you?

❯ git pull
Already up to date.

or something else?

[twin jackal]

Yes: Already up to date.

[karmic crown]

great

show me your webui-user.sh

[twin jackal]

#!/bin/bash
#########################################################

Uncomment and change the variables below to your need:#

#########################################################

Install directory without trailing slash

#install_dir="/home/$(whoami)"

Name of the subdirectory

#clone_dir="stable-diffusion-webui"

Commandline arguments for webui.py, for example: export COMMANDLINE_ARGS="--medvram --opt-split-attention"

#export COMMANDLINE_ARGS=""
export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half-vae --medvram-sdxl"

python3 executable

#python_cmd="python3"

git executable

#export GIT="git"

python3 venv without trailing slash (defaults to ${install_dir}/${clone_dir}/venv)

#venv_dir="venv"

script to launch to start the app

#export LAUNCH_SCRIPT="launch.py"

install command for torch

#export TORCH_COMMAND="pip install torch==1.12.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113"
export TORCH_COMMAND="pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cpu"

Requirements file to use for stable-diffusion-webui

#export REQS_FILE="requirements_versions.txt"

Fixed git repos

#export K_DIFFUSION_PACKAGE=""
#export GFPGAN_PACKAGE=""

Fixed git commits

#export STABLE_DIFFUSION_COMMIT_HASH=""
#export CODEFORMER_COMMIT_HASH=""
#export BLIP_COMMIT_HASH=""

Uncomment to enable accelerated launch

#export ACCELERATE="True"

Uncomment to disable TCMalloc

#export NO_TCMALLOC="True"

###########################################

[karmic crown]

looks fine

ok, run webui.sh and show me your terminal when you get ui in browser

[twin jackal]

################################################################
Install script for stable-diffusion + Web UI
Tested on Debian 11 (Bullseye)
################################################################

################################################################
Running on torkelmini1 user
################################################################

################################################################
Repo already cloned, using it as install directory
################################################################

################################################################
Create and activate python venv
################################################################

################################################################
Launching launch.py...
################################################################

Python 3.10.13 (main, Nov 1 2023, 17:29:04) [Clang 14.0.3 (clang-1403.0.22.14.1)]
Version: v1.6.1
Commit hash: 4afaaf8a020c1df457bcf7250cb1c7f609699fa7
Launching Web UI with arguments: --skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half-vae --medvram-sdxl
no module 'xformers'. Processing without...
no module 'xformers'. Processing without...
No module 'xformers'. Proceeding without it.
Warning: caught exception 'Torch not compiled with CUDA enabled', memory monitor disabled
2023-12-11 18:04:08,719 - ControlNet - INFO - ControlNet v1.1.422

ControlNet preprocessor location: /Users/torquee/AI/stable-diffusion-webui/extensions/sd-webui-controlnet/annotator/downloads
2023-12-11 18:04:08,815 - ControlNet - INFO - ControlNet v1.1.422
Loading weights [84d76a0328] from /Users/torquee/AI/stable-diffusion-webui/models/Stable-diffusion/epicrealism_naturalSinRC1VAE.safetensors
Creating model from config: /Users/torquee/AI/stable-diffusion-webui/configs/v1-inference.yaml
Running on local URL: http://127.0.0.1:7860

To create a public link, set share=True in launch().
Startup time: 7.7s (prepare environment: 0.3s, import torch: 3.0s, import gradio: 0.8s, setup paths: 0.7s, initialize shared: 0.1s, other imports: 0.8s, load scripts: 0.7s, create ui: 0.9s, gradio launch: 0.3s).
OMP: Warning #191: Forking a process while a parallel region is active is potentially unsafe.
OMP: Warning #191: Forking a process while a parallel region is active is potentially unsafe.
Applying attention optimization: sub-quadratic... done.
Model loaded in 7.1s (load weights from disk: 0.5s, create model: 0.8s, apply weights to model: 4.6s, apply dtype to VAE: 0.1s, move model to device: 0.1s, calculate empty prompt: 0.9s).

...sorry had to split it up due to too much text

[rough locust]

guys im using foocus for generating images and variations.
does webui has something like faceswap feature like foocus

[karmic crown]

cool, so no errors

[rough locust]

i want to try pose editor inside webui but generated images are not like foocus. foocus has different generation i guess

[karmic crown]

go to settings, find sysinfo

click on this link

[twin jackal]

Yep got loads of info...

[karmic crown]

search for GPU, and let me know if you can find anything, i am not sure if your eGPU is visible there

have you tried on another mac (without eGPU)?

[twin jackal]

"nvidia_gpu_models": null,

"randn_source": "GPU",

"prepare environment/torch GPU test": 0.000392913818359375,

[rough locust]

since i have been here i observed that u have so much knowledge about webui and extensions. do you have cool faceswap extension for generated images.

for example i will upload face and i will say chopping down trees and it will generate that face chopping down

or is there built in option for that

[twin jackal]

so far only way I can get it to work is using --no-half when starting: ./webui.sh --no-half

[karmic crown]

yeh, i just saw that i do not see my radeon in my iMac

anyway

[twin jackal]

Nope not tried on other macs... Without eGPU it does not start

[karmic crown]

give me a second to find a script

[rough locust]

are u nvidia gpu

❯ ./webui.sh --medvram --xformers --no-half-vae

try this

[twin jackal]

no I'm on RX 6900 XT eGPU

[karmic crown]

@twin jackal

save as test.py

put it inside your sd folder

and run

venv/bin/python test.py

you should get this

❯ venv/bin/python test.py
tensor([1.], device='mps:0')

which indicates that metal rendering is working properly

[twin jackal]

Yes tensor([1.], device='mps:0') 👍

[ornate elk]

yes, did it worked?

[twin jackal]

I also see in GPU history that the 6900 is running (fine) when generating (noise)

[karmic crown]

ok

everything is fine so far, so lets see whats next

[rough locust]

why my open pose editor doesn't look like this

[karmic crown]

which model are you using?

and which settings

[ornate elk]

1:23 for 4 images, Steps: 20, Sampler: DPM++ 2M SDE Karras, CFG scale: 7, Size: 960x540,
model was an 1.5 2gb model

[rough locust]

juggernaut xl and default

its like this

[twin jackal]

...wonder why the --no-half works and makes it generate images (slowly), anyways we let it ripe a bit and if it hits you what it could be or any theories then do not hestitate. Super thanks again 🙏🏻💚

[karmic crown]

shoe me model and settings in your ui

which model you are using?

do you have some simple 1.5 model like photon?

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

but any 1.5 model will do

[fallen nymph]

lool

[twin jackal]

v1-5-pruned.ckpt, epicrealism mainly and some more

[karmic crown]

https://civitai.com/models/25694/epicrealism

this one?

click generate

and check the terminal

let me know if you have any errors

it should look like this

[twin jackal]

Yes that one!

[karmic crown]

oops, i forgot to change the model, cfg is too high for lcm 🙂

5/20 [00:08<00:28, 1.92s/it] - m1 mac

no errors, no warnings

what you have?

[twin jackal]

Model loaded in 7.5s (load weights from disk: 0.7s, create model: 0.7s, apply weights to model: 4.8s, apply dtype to VAE: 0.1s, move model to device: 0.1s, calculate empty prompt: 0.9s).
100%|███████████████████████████████████████████| 20/20 [00:11<00:00, 1.68it/s]
100%|███████████████████████████████████████████| 20/20 [00:09<00:00, 2.03it/s]
Total progress: 100%|███████████████████████████| 40/40 [00:22<00:00, 1.74it/s]
100%|███████████████████████████████████████████| 20/20 [00:09<00:00, 2.03it/s]
100%|███████████████████████████████████████████| 20/20 [00:09<00:00, 2.07it/s]
Total progress: 100%|███████████████████████████| 40/40 [00:20<00:00, 1.97it/s]
Total progress: 100%|███████████████████████████| 40/40 [00:20<00:00, 2.07it/s]

[karmic crown]

?!

[karmic crown]

and settings?

under genrated image, you should have something like this

1girl
Negative prompt: nsfw
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2145502047, Size: 512x512, Model hash: ec41bd2a82, Model: photon_v1, Clip skip: 2, Version: v1.6.1

Time taken: 45.0 sec.

what you have?

[twin jackal]

Yes 1girl
Negative prompt: nsfw
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3516725563, Size: 512x512, Model hash: 84d76a0328, Model: epicrealism_naturalSinRC1VAE, Denoising strength: 0.7, Clip skip: 2, Hires upscale: 1, Hires upscaler: Latent, Version: v1.6.1
Time taken: 20.3 sec.

[karmic crown]

colapse upsaceler

and any other of those integrated ones if you have them

both of those should be colapsed

and not used for now

also, epiCRealism sugest using of clip skip 1, you are using 2

[twin jackal]

Ok they are not active (upscaler to 1 and no refiner)

[karmic crown]

1 is active

you need to colapse it

"Hires upscale: 1, Hires upscaler: Latent,"

you are using hires

also, go to settings, find User interface

add those here

[twin jackal]

Ok thanbks took that away...

[karmic crown]

that way you can easily control vae and clipskip

[karmic crown]

this is very strange, can you try with other model?

Prompt: simple explanation of the image (try first without extra keywords)
Negative: "cartoon, painting, illustration, (worst quality, low quality, normal quality:2)"
Steps: >20 (if image has errors or artefacts use higher Steps)
CFG Scale: 5 (higher config scale can lose realism, depends on prompt, sampler and Steps)
Sampler: Any Sampler (SDE, DPM-Sampler will result in more realism)
Size: 512x768 or 768x512

see this

try with 30 steps

and set cfg to 5

and use 512x768

not 512x512

[fallen nymph]

@ornate elk rip i cannot generate 4 pics at once

[ornate elk]

With batch size?
Dont use that with amd

[karmic crown]

@twin jackal and if you do not mind, try this model, plese https://civitai.com/models/84728/photon

[stuck knot]

Does anyone know what ComfyUI custom node this is from?

[fallen nymph]

batch count aswell

😄

[ornate elk]

Batch count should work

[fallen nymph]

depends

if i got enough vram

[ornate elk]

Also everytime you run out of memory you need to restart SD

[fallen nymph]

oh

why didn't u say that earlier 😄

why though?

[ornate elk]

The vram memory doesn't get cleared after an failed attempt so its stuck and full

Until a restart

[fallen nymph]

i can

generate 2 images tho

but it takes quite a while, around 3-5 minutes

on 35 steps*

[ornate elk]

Yes ik

[fallen nymph]

with dpm++ 2s a karras

and dreamshaper model

[ornate elk]

It still is able to generate stuff but slower and smaller

So restart and your good to go

[fallen nymph]

i ramped up

my fans

on 100%

just to make sure hotspot temp doesn't go above 90 lol

[twin jackal]

Ok tried photon with those settings

[fallen nymph]

@ornate elk does it always save the images automatically?

[ornate elk]

Yea its recommended to set a custom fan curve for amd cards in adrenalin

[ornate elk]

Yes

In the outputs folder

[fallen nymph]

ya but sometimes if the drivers fail and you reinstall

u have to do it all over again ;d

[ornate elk]

Yes also after a windows update

[fallen nymph]

yeah just found it, thats why i asked

[karmic crown]

@twin jackal and if you put --no-half instead of --no-half-vae?

leave everything the same

[twin jackal]

Ok will have to restart fr terminal with --no-half. One sec...

[karmic crown]

export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half --medvram-sdxl"

like this

in webui-user.sh

[fallen nymph]

@ornate elk what was the res

960x540?

[karmic crown]

if that not working, try
export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half --medvram"

let see what works for you

[fallen nymph]

also some youtube had an image about all the sampling methods

like images per steps etc

is that pic around here

[karmic crown]

@twin jackal btw. how much VRAM your eGPU has?

[fallen nymph]

@ornate elk wtf is this

[karmic crown]

hahaha

[ornate elk]

booru auto tag

[fallen nymph]

lol

[ornate elk]

its an autocorrect for words

type blon and youll get blonde hair for example

[twin jackal]

16BG Vram

[karmic crown]

@fallen nymph https://github.com/DominikDoom/a1111-sd-webui-tagcomplete

[fallen nymph]

okay let's see if after a restart i'll be able to generate 4 images

[fallen nymph]

yep thanks

[karmic crown]

try with export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half --medvram"
first

[ornate elk]

propably not with batch size but with batch count

[fallen nymph]

yeah trying with count

[twin jackal]

export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half --medvram-sdxl"

[fallen nymph]

lmao @ornate elk

[ornate elk]

woah xD

[fallen nymph]

curve looking cool ngl

[karmic crown]

so, it works with --no-half, but not with --no-half-vae?

wait, wait, wait... we just talked about similar problem on radeon on windows

[twin jackal]

Yes!

[karmic crown]

it looks like your radeon will not work with --no-half-vae

show me the speed of 512x512 20 steps

if it is less than 30 seconds you are faster than my m1

actually, lest time it

[fallen nymph]

eyy i did it @ornate elk

took like 2-3 mins but still

[ornate elk]

nice

[karmic crown]

that looks like batch size, not a bachcount 🙂

[fallen nymph]

but it's count

🙂

[ornate elk]

nah its just the grid you get as first image

[karmic crown]

@twin jackal Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3475476718, Size: 512x512, Model hash: ec41bd2a82, Model: photon_v1, Clip skip: 2, Version: v1.6.1

Time taken: 36.5 sec.

[fallen nymph]

@ornate elk

about that

what settings do i do

[twin jackal]

export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half --medvram" a few seconds longer

[fallen nymph]

this too

[ornate elk]

if you upscale images you need to enable Tiled Vae, not Tiled diffusion.
set the encoder to 1024 and decoder to 128 and your fine

[fallen nymph]

so like this and enable it only when upscaling, gotcha

what about the diffusion

[twin jackal]

A photo of a robot walking in snow
Negative prompt: cartoon, painting, illustration, (worst quality, low quality, normal quality:2)
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 5, Seed: 2145502047, Size: 512x512, Model hash: ec41bd2a82, Model: photon_v1, Version: v1.6.1
Time taken: 16.3 sec.

[karmic crown]

you can experimetn with --no-half --no-half-vae --medvram --medvram-sdxl but do not change the other parameters, please let me know what works best for you

[formal trout]

Hey, to upscale in latent in img2img without SDUpscaler, with tiles, it is possible with SDXL?

[lime spindle]

what the best hires fix for sdxl

[twin jackal]

A photo of a robot walking in snow
Negative prompt: cartoon, painting, illustration, (worst quality, low quality, normal quality:2)
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3475476718, Size: 512x512, Model hash: ec41bd2a82, Model: photon_v1, Version: v1.6.1
Time taken: 14.2 sec.

[ornate elk]

with the diffusion you can make different prompts for different parts of an image

[ornate elk]

didnt tested it

[fallen nymph]

u said i'd need it if i generated 4 images? or am i mistaking

[karmic crown]

@twin jackal try to remove --no-half and --medvram too, but if you get this error Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check. restore --no-half

pleas let me know what works the best for you

and please post a comment on github, so i can add your comment to next version

i didnt experienced your problem, but i guess it is realted to eGPU

[ornate elk]

nope, you dont need tiled diffusion, you only need tiled vae, for larger images (upscaled)

[fallen nymph]

okay

well

i am not able to generate 4 images now

it stops automatically lol

[twin jackal]

Ok many thanks, could be eGPU as you do not have these probs. I'll play around and will let you know what works best. Normal generation speed is acceptable, but upscaling and img2img has been so slow until now. But we'll see with your tests. Thanks again!

[karmic crown]

i have imac with radeon and m1, both works, but they are both slow

your egpu is much faster

even with no-half, which is known to be very slow

[twin jackal]

hmmm that's interesting. Wonder how it would rock without no-half...

[karmic crown]

--no-half-vae works faster, but obviously you can't use it 😦

and i am almost sure that without --no-half, you will get an error i mentioned

i have similar situation with 1660 on my sons pc

it requires --no-half, and cant work with --no-half-vae 🤷🏻‍♂️

[twin jackal]

Hmm, similar problem it seems. Strange. Will familiarize also with the settings, perhaps something is messed up there, hopefully the thing you mentioned above... Let's see..

[ornate elk]

It also depends how much GPU heavy programms are open at the same time

[fallen nymph]

just youtube 😄

and spotify

what do i do for fitness pictures, which model is the best

[ornate elk]

Okay, for example close Wallpaper Engine If you have it

[fallen nymph]

right

;d

cs

is there any way to save the settings as i leave?

because i have to redo all everytime

[karmic crown]

@twin jackal learn a few tricks and you can do this #🏞｜general-with-images message

[formal zodiac]

Anyone know how to make Stable Diffusion images actually take the negative prompts into consideration? Like for example, extra limbs is in the negative prompt but the image still generates extra limbs

(I'm using the VTAI UI if that helps any)

[twin jackal]

Wow with 7 steps... Thanks yes will dig in deeeply into it. The Photon model seems to rock. 👍🏻

[formal zodiac]

This is what my screen is showing atm

The reason why I am using img2img rn is because I am attempting to get the two anime characters to stand into the pose like the image I submitted

[karmic crown]

it is a combination of a few tings LCM lora, LCM sampler SDXL renderer lora

[ornate elk]

you can set it maybe in the ui_config.json

[karmic crown]

those two

https://civitai.com/models/195519
https://civitai.com/models/171159

• LCM sampler

[ornate elk]

for poses you need controlnet

[formal zodiac]

What's that?

Sorry I'm a beginner lol

[ornate elk]

its an extension thats really powerful in how images get generated. you can set an image and it will use/output the same pose or the same silhouette as the person in the image

[formal zodiac]

Ooooo, how do I get it?

Is it on huggingface?

[ornate elk]

under Extensions tab, click on available, then on Load from, then search for
sd-webui-controlnet and click install

then restart
then you need the controlnet models from here:
https://civitai.com/models/38784/controlnet-11-models
put them into the models/controlnet folder

[karmic crown]

One Piece fun was faster than me, as usual 🤣

[formal zodiac]

ValueError: unknown url type: 'sd-webui-controlnet'

That's what pops up when I search sd-webui-controlnet

[twin jackal]

🙏🏻 will check and try. 🔥 😎

[ornate elk]

ok then search for controlnet

and install the one from the bottom thats named sd-webui-controlnet

[formal zodiac]

It happened again

[ornate elk]

oh you didnt clicked on load from

[formal zodiac]

I did

I think

[ornate elk]

you typed in the wrong field i guess

[formal zodiac]

Ohhhhh yeah I did

There's two

SD-CN-Animation and sd-webui-controlnet

[ornate elk]

the sd-webui-controlnet

[formal zodiac]

Okay

It's downloading I think

[rough locust]

how can i create images based on image of person

[placid edge]

does anyone use Fooocus? I was wondering how the controlnet works for that ui? I know in A1111 you can use depth, softedges etc, but I can't see that in fooocus's controlnet tab

[rough locust]

does foocus has controlnet?

[placid edge]

I do see an option for controtnet under the advanced option and in the directories

[rough locust]

i don't see

[fallen nymph]

@ornate elk broski do u got the samplers comparison pic

[ornate elk]

nope, but euler a, dpm++ sde karras and 2m karras are good ones

[fallen nymph]

what about the 3m

[ornate elk]

didnt used them that often

i think they require more steps

[fallen nymph]

bro im waiting for 3-4 minutes

per picture

it's taking so long lol

been generating pics for hours lmao

[karmic crown]

I like UniPC, and LCM (but that one is separate beast)

[fallen nymph]

but how do i prevent the ai of changing the hairstyle

lol

[ornate elk]

with hires fix?

[fallen nymph]

no

just this

[placid edge]

There is an "advanced" tab

[karmic crown]

i am testing different command line options with SDXL on M1 and i want to kill myself

[ornate elk]

ah okay, well txt2img would be faster

[placid edge]

There is also this too

[ornate elk]

even with hires enabled

[fallen nymph]

ye but im changing a picture of my friend

lmao

using ai to give him better shape and stuff 😄

[rough locust]

hmm

[ornate elk]

ah okay

[karmic crown]

this takes forever 😦

[placid edge]

I doesn't seem like it's exact like a1111's controlnet

[fallen nymph]

how do i prevent the ai changing the hair @ornate elk

[rough locust]

sometimes i hate foocus

[fallen nymph]

it uses the shadow and transforms it as a hair

lol

[rough locust]

but sometimes i generate images better than webui with fooocus

[placid edge]

I will say that it's a lot faster than A1111 from my experience

and a lot easier to use

[rough locust]

and styles are very useful

[ornate elk]

lowering the denois

[fallen nymph]

what is

[rough locust]

fooocus

[placid edge]

yeah exactly, the preset styles are so nice

[fallen nymph]

what's fooocus even

;d

[rough locust]

alternative to webui A1111

[fallen nymph]

oh

is it good

[rough locust]

depends

[placid edge]

it's easier to use and faster from my experience

[rough locust]

when i type exactly same prompt in 2 editors, i get very good results in fooocus rather than a1111

[fallen nymph]

can u resize images, use styles, extentions etc

[rough locust]

i cannot generate something like that in a1111. idk maybe its skill issue

[placid edge]

Stopping jobs in automatic 1111 is very scary

[rough locust]

thats the difficult part

fooocus has not extensions

[karmic crown]

https://tenor.com/view/head-bang-gif-4680729

[placid edge]

I just started using it and it's a pretty big letdown

[karmic crown]

Steps: 10
Sampler: Euler a
CFG scale: 7
Size: 1024x1024
Batch count: 4
Time taken: 11 min. 57.2 sec.

[placid edge]

but for like regular stuff, it's really good imo

[rough locust]

and you only can use SDXL models

[rough locust]

rip gpu?

[karmic crown]

M1 mac

[rough locust]

ah

and also you cannot select width, height in fooocus

[placid edge]

I didn't even know that part, yeah that's a bit of a let down

[rough locust]

not just sdxl

checkpoints trained with sdxl counts too

[rough locust]

juggernaut xl is good model for fooocus

but i cannot use hypernetworks idk why

[placid edge]

I usually use juggernaut and idk what hypernetworks are lol

[rough locust]

for me, its camera angle

[placid edge]

I think it's decent once you figure out what that image prompt tab does

[rough locust]

its mostly manipulatin image

faceswap, etc.

if i could use A1111, i would like to use it

it has no fooocus stylex

[placid edge]

Yeah but A1111 seems to run so slow on my pc, like I just did 60 steps in 30 seconds and in A1111 it would be like a min

[placid edge]

yeeeeaaaahh I don't like that lol

[fallen nymph]

@ornate elk sorry for the lots of mentions n stuff, does sd take space in C 😄 suddenly 10 gigs is gone

[rough locust]

i found a extension for that 😄

https://docs.google.com/spreadsheets/d/1AF5bd-fALxlu0lguZQiQVn1yZwxUiBJGyh2eyJJWl74/edit#gid=0

[placid edge]

This is exactly what I was searching for hahaha

[rough locust]

https://github.com/thundercat71/Automatic1111-Fooocus-Styles

[placid edge]

I might switch back to A1111

[ornate elk]

thse are temp files in cache. open up a cmd and type
pip cache purge
youll get free 10gb back

[karmic crown]

@ornate elk and what should i type to get faster SDXL on mac 🤣

[rough locust]

btw check dm's i will explain something

[placid edge]

ok

[fair oxide]

Is there a channel for like...feedback on images?

[karmic crown]

promting-help?

[fair oxide]

Thanks

[fallen nymph]

that in sd folder or doesn't matter?

[ornate elk]

doesnt matter

[fallen nymph]

kk

[rough locust]

bro why even my prompts are same, but results are totally different

not even talking about seed

[karmic crown]

with same seed?

and same sampler?

[rough locust]

which sampler is the best for you do you have advice for me

[karmic crown]

i use LCM or UniPC most of the time, but I am using LCM models or lcm lora

[candid quarry]

Hello, so for some my SD broke after updating my OS

Now I get RuntimeError: No HIP GPUs are available

[karmic crown]

please note that if you are trying to replicate some image from civitai, you need to have same GPU type (for example nvidia)

i used completely same settings models and everything and my result was completely different, then i tried on my sons PC (with nvidia) and i instantly got almost identical image

some small details were different

[karmic crown]

which os, wchic GPU

[candid quarry]

openSuse Tumbleweed, Vega 64

[rough locust]

and sometimes this is happening

but in fooocus i can get brown eyes with one try

[karmic crown]

❯ python3 --version
❯ python3.10 --version

what you are getting?

[candid quarry]

Python 3.11.6

You think it might be the OS updating python?

[karmic crown]

you need to have 3.10

install 3.10, and change this

[candid quarry]

hmm, I could swear I had 3.11 before but idk

[karmic crown]

# python3 executable
python_cmd="python3.10"

❯ python3.10 --version
Python 3.10.13

i have that on my linux server

[candid quarry]

where?

[karmic crown]

webui-user.sh

[candid quarry]

I used webui until now, but my friend told me to switch to comfy

[karmic crown]

i cant help you much with comfy, sorry

[candid quarry]

thats fine for now

Im sure I can figure it out

[karmic crown]

remove venv, and let webui create a new one, than it will work

[candid quarry]

I just find it weird that python version affects rocm detecting my gpu?

[karmic crown]

looks like comfy works fine with 3.11

for a1111 it should be 3.10

[candid quarry]

well I tried starting comfy with 3.11

[karmic crown]

will this agony even finish 🤣

fortunately this is just a test with parameters i know it should not be using, but i want to prove my point

[fallen nymph]

it doesn't recognize pip wow

[candid quarry]

I cant even manage to get an old python version 😭

[karmic crown]

are you using cmd? 🙂

[candid quarry]

I treid pyenv and it just throws a million errors

[fallen nymph]

cmd/windows terminal doesn't matter lol ;d

[karmic crown]

after your message that && doesnt work, i do not beleive you anymore 🙂

use cmd

[fallen nymph]

still doesn't ;p

[karmic crown]

i am using ubuntu, i didn't use suse for at least 10 years 😦

does this help? https://software.opensuse.org/package/python310

[candid quarry]

Im a bit scared the versions will conflict with each other

if they are not specifically managed with something like pyenv

ill just try, worst case I load a snapshot

[karmic crown]

on ubuntu I added a custom repository and installed 3.10 without problem, along with system version

and there is no conflict at all

you can try with 3.11, but i am not sure will it work

just remove venv first

and let it regerate that

i am talking about a1111 now

for comfy see the instalation notes on github

i am sorry but i cant help too much 🤷🏻‍♂️

[candid quarry]

If one works, all work

at least in my experience

I started out with Fooocus

and I always ran webui with Fooocus' venv

[ornate elk]

Did you installed python correctly?

[fallen nymph]

i had it installed

for years 😄

also

also resize doesn't work for some reason

[candid quarry]

It still doesn’t work

I give up for now

[dusk oar]

Couldn't install torch. Help please.

[ornate elk]

Then its maybe not set in the windows Path

[fallen nymph]

how do i check lul

[karmic crown]

https://realpython.com/add-python-to-path/

[ornate elk]

You can run the python installer again then click modify and then next, and then check "add python to system variables"

[acoustic dust]

i need help with this problem

[karmic crown]

what are your OS and your GPU?

[acoustic dust]

nvidia geforce rtx 3050

[fallen nymph]

@ornate elk problem was that since i'm not in an environment

[karmic crown]

your command line parameters are wrong

[fallen nymph]

i have to use python -m pip instead of pip

😄

[acoustic dust]

how i can fix?

[fallen nymph]

told u terminal is the same as cmd @karmic crown

[karmic crown]

it should be, but you clearly told that && wasn't working

[fallen nymph]

ye idk why it didn't

perhaps windows terminal doesn't support multiple commands at once

i am not sure

[karmic crown]

how much vram? 4 or 8GB?

[acoustic dust]

16 ram

[karmic crown]

i guess you need "--xformers --medvram-sdxl --no-half-vae", but i am not sure if that is the best for your card, at least i am sure you need --xformers

you can try with those, and then someone else will correct me there are some better combination for your card

[acoustic dust]

ok. i'm gonna try. tysm

[karmic crown]

find webui-user.bat

[acoustic dust]

yea

[ornate elk]

He needs --medvram

Not medvram-sdlxl

3050 has only 4gb vram

[karmic crown]

CS1o knows better

he has 16 (at least he said that)

[ornate elk]

Yes because he dont know vram is not ram
In the screenshot you can see the total vram amount

[acoustic dust]

how i need to write

[karmic crown]

CS1o is PC guy, i am Mac guy, for PC, always listen to him

[ornate elk]

--xformers --medvram --no-half-vae

At the line COMMANDLINE_ARGS=

[acoustic dust]

ok tysm

[ornate elk]

Np

[acoustic dust]

@ornate elk i have the same error

[ornate elk]

Show the webui-user.bat

[karmic crown]

run webui-user.bat, not webui.bat

[acoustic dust]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS= --xformers --medvram --no-half-vae
git pull
call webui.bat

[karmic crown]

@ornate elk check out this

SDXL test 1:
export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half-vae --medvram-sdxl"
Time taken: 11 min. 55.2 sec.

SDXL test 2:
export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling"
TypeError: BFloat16 is not supported on MPS

SDXL test 3:
export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half"
TypeError: BFloat16 is not supported on MPS

SDXL test 4:
export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half-vae"
TypeError: BFloat16 is not supported on MPS

SDXL test 5:
export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half-vae --medvram"
Time taken: 11 min. 57.2 sec.

SDXL test 6:
export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half --medvram"
Time taken: 48 min. 51.6 sec.

SDXL test 7:
export COMMANDLINE_ARGS="--skip-torch-cuda-test --upcast-sampling --no-half-vae --medvram-sdxl"
Time taken: 11 min. 39.9 sec.

SDXL test 8:
export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-split-attention-v1 --upcast-sampling --no-half-vae --medvram-sdxl"
Time taken: 11 min. 40.4 sec.

Steps: 10
Sampler: Euler a
CFG scale: 7
Size: 1024x1024
Batch count: 4

on M1 Mac 🙈

[ornate elk]

Oh xD I think you won't get it faster than that

[karmic crown]

check test 6!

[ornate elk]

Run the webui-user.bat

[ornate elk]

Thats insane xD

[acoustic dust]

I still can't generate images

idk

[karmic crown]

when I see "use --no-half" 🙈

but today i leaned something new and strange, see this #🤝｜tech-support message

he has eGPU, and that Radeon requires --no-half, with --no-half-vae it generates only noise

#🤝｜tech-support message

😱

[ornate elk]

He runs torch 2.2

Dont know but that could be issue

[karmic crown]

i am using 2.2

[ornate elk]

Can you show a screenshot of the cmd after launching the webui-user.bat?

[karmic crown]

on mac --no-half-vae requires 2.2

[ornate elk]

Oh okay

[acoustic dust]

sure

[karmic crown]

i am forgot but it is possible that --medvram-sdxl also requires 2.2 on mac

i will test when 1.7 is released

[acoustic dust]

this is good?

[karmic crown]

remove this space here, and try again

[ornate elk]

What's size is the model your using?

[acoustic dust]

36,000 KB ;-;

36 mb

[karmic crown]

he has space between = and --xformers, is that important on windows?

[ornate elk]

Nope, should work

[ornate elk]

Thats the issue. 36mb is not a model file

[acoustic dust]

oh

[fallen nymph]

is there any way

to use the AI on anything but the face of a person?

[ornate elk]

Its mostly a lora so move It out if the models/stable-diffusion folder

In models/lora

[ornate elk]

Yes thats called inpainting

[fallen nymph]

how does it work 😄

[acoustic dust]

I found another model. it has 2 GB

this one is good?

[ornate elk]

Yes, every file thats larger than 1.98gb is fine as model (checkpoint)

[acoustic dust]

okk. I will let you know if I still have problems with something xD

tysm for help !

[fallen nymph]

i got 6 gigs ones 😄

[ornate elk]

No problem 🙂

[ornate elk]

Rip vram

xD

[fallen nymph]

ye don't ask 💀

so how does

inpainting

work

[karmic crown]

🙂

[ornate elk]

To get inpainting to work for AMD you need to add --no-half to your webui-user.bat
But that will use a lot more vram for generating so remove --no-half for normal txt2img stuff.

[acoustic dust]

is working lmao. ty again <3

[ornate elk]

Perfect np!

[fallen nymph]

ye i have --no-half already

so how does it work

[karmic crown]

https://stable-diffusion-art.com/inpainting_basics/

like that

[ornate elk]

In img2img click on inpaint. Set an image. Then Mask the Face.
Then select not masked only.
Then prompt for something

[fallen nymph]

but it makes weird stuff around the face

how can i fix those

it's like blurry pixels

like these

are those settings correct?

[ornate elk]

That happens mostly when you inpaint non ai generated images. You could try to change the denois. Or use an other model like the 1.5-inpainting model

[fallen nymph]

aaah

okay

could you link the model

[ornate elk]

https://huggingface.co/runwayml/stable-diffusion-inpainting/tree/main

[fallen nymph]

just this right

stable-diffusion models right

as the others

[unique prairie]

Excuse me, how would I turn off node shadows and straighten my node connections in comfy ui?

[formal trout]

Any config file or guide for training lora on SD XL with something like RX 6700 12 GB?

[ornate elk]

I know of only one guide for an 7900xtx and thats pretty complicated and I think its Linux only

[formal trout]

I use Linux only so good 🤙

I didn't want to walk that path alone

I had trained some SD 1.5 with relative success, they were a mess but worked if tweaked a bit

[karmic crown]

https://aituts.com/stable-diffusion-lora/

[formal trout]

but I wanted to train on SD XL, setting seems challenging

[karmic crown]

check training part

or this https://stable-diffusion-art.com/train-lora/

i just found those, so do not blame me if that is not helping 🙂

and this one https://techtactician.com/how-to-train-stable-diffusion-lora-models/

[formal trout]

hey great, seems to be very complete, thanks! I'll check it out when I finally got into it

[karmic crown]

https://aituts.com/sdxl-lora/

and this one

https://colab.research.google.com/github/Linaqruf/kohya-trainer/blob/main/kohya-LoRA-trainer-XL.ipynb

something should help 🙂

[formal trout]

great thanks! I mostly know all the process (with some doubts nonetheless) but the settings for my card and not taking ages. I'll check that out.

[karmic crown]

google colab is fastes way i guess 🙂

[acoustic maple]

I keep getting upstream connect error or disconnect/reset before headers. reset reason: connection failure after the same calls working fine before. Is there an outage on the text-image api?

[fallen nymph]

@ornate elk using the module

still does these blurry pixels

[short sail]

wait nvm i fixed it 🔥

[odd fractal]

yall does he mean environmental values??

did everything according to code and it still didnt work?

[odd fractal]

nvm i got it up and running 🙏

[unique anchor]

Is there any extension that ensures a111 doesn't generate NSFW images without replacing it with a black image?

[terse mulch]

The image on left was generated with controlnet txt2img as part of a 360 frame sequence. I used EBSynth to make a sequence without flickering but the crossfading segments in that animation are kind of ugly.

I figured I could get a less flickery output directly from SD by using the same frames i used for controlnet, with the EBSynth output as the img2img prompt.

But the regenerated image also has the blobby oversharpened look that happens with really high CFG values, even if I use extremely low CFG values. How do I make it not do that? Shouldn't using the image generated by controlnet text2img, used as the img2img prompt give the exact same result?

the only way I can get it to not have the oversharpened look is turn the denoise way down, but then it's just the same image, rather than a re-generation of the source image with some variation (trying to hide the double exposure look of the EBSynth crossfaded frames)

[fair oxide]

hello, I am looking at the api options, and I wonder if you can teach with images to have better results, sorry I'm a noob and I have no idea, any information is appreciated.

[fallen nymph]

@ornate elk

if i wanted phone wallpaper what res should i use then upscale by 2

[dusty bronze]

How can i make old blurry and grainy images sharp and clear? I am using Fooocus with Juggernaut XL but when i do imageprompt it changes all the faces

[ornate ruin]

Hi 😃 'torch is not able to use GPU' yesterday my Automatic1111 was working just fine. I tried to bulk create a few images (5 batches the size of 2 nothing too huge, 28 steps 2,4x upscale) 8GB VRam RTX4060

But now it won't load up anymore smh just showing this error

It also says I should put --skip-torch-cuda-test in the arg command lines. Is that the only way to make it work again? Seems kinda weird imo.

[faint badge]

How come, whenever I upscale with the exact same prompt. It looses all it's detail in the face plus it distorts it?

Here happens the same

Generating vs result

[ornate elk]

Upscaling always decrease quality.

You can try Latent upscaler that can make Additional details

[faint badge]

I mean img2img

my mistake

[ornate elk]

Img2img shouldn't change the quality.

[faint badge]

It shouldn't

Yet it still does

[ornate elk]

But if you upscale with img2img it will

Just resize is upscaling

Changing the resolution with denois 0.2 is also upscaling

[faint badge]

But do you see what I mean?

[ornate elk]

Yes its normal

[faint badge]

Difference between faces in these pictures

[faint badge]

How come this problem only started showing up 2 weeks ago?

[ornate elk]

I dont know what your trying to do and also what settings your using in img2img

[lost crescent]

I want to read the generative ai data in my PNG files - but in an app indpendent from ComfyUI or AUTOMATIC1111?!

[ornate elk]

So your upscaling your imagw with Denois 0.5

That means it will change the output image slightly

If you would use denois 0.1 you would get nearly the same image

[faint badge]

Yes, I know. But no matter the denoise. With img2img, why does it remove all the detail in the face, even distort the eyes, mouth, whatever? It shouldn't be doing that right?

[faint badge]

This is the great example of what preview shows, and what the result gives me

Here it goes from 80% to 100%

[ornate elk]

That depends also on the tags. Use some negative prompts and good quality prompts.
Also lower the denois and the mouth won't be disorted

Upscaling always reduces quality as it isn't the native resolution

Normaly you should upscale with the SD upscale script in img2img

There you can define the upscaler that gets used.

[karmic crown]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/5461#discussioncomment-7830815

https://tenor.com/view/crazy-alert-crazy-alert-alerta-alerta-loca-gif-166249322369313028

--no-half --medvram on ARM macs = ☠️☠️☠️

[next tapir]

which of these is the correct one?

i downloaded juggernaut v7

[karmic crown]

i forgot when i download anything manually since i installed civitai browser plus

[next tapir]

i used wget to download that

and it created these files

[karmic crown]

I haven't used wget for at least 10 years. if you really wanna play with a command line, then use curl

curl -L https://civitai.com/api/download/models/240840\?type\=Model\&format\=SafeTensor\&size\=full\&fp\=fp16 -o "Juggernaut XL.safetensors"

but my recommendation is to install civitai browser plus if you are using a1111

ooops, it should be lowercase o, not the capital one

[rough locust]

me too

[summer tinsel]

Hi, is there an option to fill in a company VAT number when making a payment through Stripe to get rid od the additional VAT? Stripe allows that function, but I am unable to do that in case of Stability payments. Thank you.

[chilly sandal]

why does it do thaaaat? face looking decent while rendering, but the final product looks like a sewer... hgow the hell can I get nice face/eyes? (Ik I can inpaoint after, but is there no way to make it look decent from the start?
(also, does anyone know good eyes lora? like sparkly, 3d eyes style)

[karmic crown]

better ask in #📝｜prompting-help

[ornate elk]

Higher resolution would make it better. But for 1.5 models, upscaling is the key

Hires fix can help

Reducing steps can help sometimes too

[karmic crown]

i am generating 768x1024 with 1.5 in 8 steps without upscaler with lcm and sdxl renderer lora 🙂

[chilly sandal]

lower steps is better?.... I need to look into how steps work xd

[karmic crown]

i am using special loras and sampler that helps reduce number of steps and gives more clarity

[ornate elk]

Not always, but your mostly fine with 20-30

Also it depends on loras and sampler too like Viking said

[chilly sandal]

ty for the advice, I'll look up the loras u sugested and try to tweak some stuff

[karmic crown]

btw, read this https://invoke-ai.github.io/InvokeAI/help/SAMPLER_CONVERGENCE/

depending on sampler images between steps can be similar or totally different

usually samplers with A at the end are those with most drastic changes between steps

[chilly sandal]

I hate inpainting and love it too... I dont change anything in the promt, but focusing on the face "fixes" it... most of the times

[chilly sandal]

uuu, gotcha, I'm using Euler A most of the times, so that makes sense

what happens if I increase the "resize to" parameters in inpaint... is is possible to increase the pixels of just one portion of a picture? (faces, people, etc?) or does it upscale the whole picture?

[karmic crown]

https://stable-diffusion-art.com/samplers/

also check this one

[chilly sandal]

yeap, went througha bit of this one already
guess it's time for a lecture tonight xd

[karmic crown]

many things will just start to make sense when you read that

like, why my image looks totally different at step 5 and step 15

and how can you get almost the same image with 20 and 15 steps, but i can not

[plucky marsh]

does anyone know if it's possible to install models inside stable diffusion via google colab on Fooocus?

Fooocus does not create google drive SD folder nor it does use it

i can install the models locally just fine but i can't seem to find a way to install them online 😢

[azure magnet]

Help please

https://tenor.com/view/tense1983-rage-csgo-troll-afk-gif-14645059

[dusky wing]

Hi everyone Is it weird that if I generate an image (768 x 768) the first part of the rendering is really quick (2.60 it/s) and then suddenly it goes realllly slow, it looks like it is even freezing. The waiting time also starts going up instead of going. I am using a Nvidia RTX3060 Ti

[karmic crown]

if you are using upscaling, that takes much more time than image generation, check the console, and show what you have there when slowdown occuers

[dusky wing]

Aah okay so upscaling takes more time in general?

[karmic crown]

absolutely, it needs more time to generate bigger image

[dusky wing]

Aah thanks a lot! So it makes more sense to generate a few smaller images first, and upscale the ones I like later?

[karmic crown]

collapse the hires part of ui (that will stop it) and just generate image with like 20 steps, and check how it goes, you should not notice any slowdown

[dusky wing]

It's a lot faster indeed, and no slow down 🙂

You're a legend, thanks!!

[ornate elk]

btw are you using xformers ?

[dusky wing]

Yes I am!

[ornate elk]

okay good 🙂 also --medvram-sdxl is needed if your using sdxl models with your 8gb vram

[dusky wing]

That's new to me, thanks, Ill enable that! I just write that behind "set COMMANDLINE_ARGS=" right?

[ornate elk]

yes

[dusky wing]

Awesome, thanks!

[ornate elk]

no problem, it should increase the generation speed a lot for sdxl models

[karmic crown]

@ornate elk have you ever seen this error?

Traceback (most recent call last):
  File "/Users/francesco/stable-diffusion-webui/launch.py", line 48, in <module>
    main()
  File "/Users/francesco/stable-diffusion-webui/launch.py", line 44, in main
    start()
  File "/Users/francesco/stable-diffusion-webui/modules/launch_utils.py", line 432, in start
    import webui
  File "/Users/francesco/stable-diffusion-webui/webui.py", line 13, in <module>
    initialize.imports()
  File "/Users/francesco/stable-diffusion-webui/modules/initialize.py", line 14, in imports
    import torch  # noqa: F401
  File "/Users/francesco/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/__init__.py", line 1219, in <module>
    from .storage import _StorageBase, TypedStorage, _LegacyStorage, UntypedStorage, _warn_typed_storage_removal
  File "/Users/francesco/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/storage.py", line 307, in <module>
    class UntypedStorage(torch._C.StorageBase, _StorageBase):
TypeError: 'torch._C._TensorMeta' object is not iterable

[ornate elk]

nope but i would update python and delete the venv folder after

[karmic crown]

he tried, more than once

but the same error keeps on popping up

[ornate elk]

did he also has the latest nvidia drivers?

[karmic crown]

mac

[ornate elk]

ohh

[karmic crown]

i am trying to figure out how to help him

[ornate elk]

did he also used
pip cache purge before deleting the venv?

that could help if the inital torch version is corrupt

[karmic crown]

yes

btw, i am currently making a1111 installer skript for mac

[ornate elk]

oh thats awesome!

[karmic crown]

new install, just clone
existing purge cache, remove venv, hard reset

🙂

ok, I done some cleanup

❯ ./a1111-setup.sh
Checking for Homebrew...
Already up-to-date.
Detecting processor...
ARM processor detected.
Installing cmake...
cmake is already installed
Installing protobuf...
protobuf is already installed
Installing rust...
rust is already installed
Installing python@3.10...
python@3.10 is already installed
Installing git...
git is already installed
Installing wget...
wget is already installed
Existing installation detected...
Forcing use of latest file versions...
HEAD is now at 4afaaf8a add changelog entry

@coral galleon hello

[coral galleon]

where do I find it?

[karmic crown]

my script is not finished yet

i am working on it

so, you have m1 mac and you created completely new folder and you are getting the same error?

right?

[coral galleon]

yes, exactly

[karmic crown]

have you copied anything from old installation?

or you just run it?

[coral galleon]

in my home folder (without any stable-diffusion-webui folder): ➜ git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui; ➜ cd stable-diffusion-webui; ➜ vim webui-user.sh; ➜ rm -rf venv; ➜ ./webui.sh; output: "TypeError: 'torch._C._TensorMeta' object is not iterable"

[karmic crown]

lets see what i will get

i am getting the error i mentioned in the guide

ModuleNotFoundError: No module named 'torchvision.transforms.functional_tensor'```

can you try to run it with this command line?
export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling"

are you getting the same error?

wait a second

Yours
Python 3.10.0b1 (v3.10.0b1:ba4217537c, May 3 2021, 17:17:02) [Clang 12.0.5 (clang-1205.0.22.9)]

Mine
Python 3.10.13 (main, Aug 24 2023, 12:59:26) [Clang 15.0.0 (clang-1500.0.40.1)]

also, waht is your macos version? sonoma or some older?

[vestal sigil]

this thing happens to some pictures when I try to upscale them with R-ESRGAN 4x+ Anime6B

[coral galleon]

Sonoma 14.0 (23A344)

[karmic crown]

can you please unistall that beta versionof python 3.10 and install it with brew, like this

brew install python@3.10

btw have you installed all of those

brew install cmake protobuf rust python@3.10 git wget

[karmic crown]

done! 🤣

https://github.com/viking1304/a1111-setup

[ornate elk]

Nice! saved for future mac questions ^^

[plucky marsh]

Does anyone know why my google colab stable diffusion via fooocus times out all the time after i install the model Realistic Vision V6.0 B1?

As soon as I add image prompt it times out, unfortunately locally i cant run it cause its toasting my gtx 1660 ti 6gb gpu to bits lol

[plucky marsh]

Works just fine if I run it without any image prompt. As soon as I add Image prompt of any kind it won't run and returns ^C (Connection Timed out)

[plucky marsh]

I saw some solutions online like to place export COMMANDLINE_ARGS="--no-gradio-queue"

but i have no clue where to place that since im using fooocus_colab

[vocal burrow]

it's an automatic1111 command line arguments not a fooocus one (although something similar can probably be implemtented by the dev)

[vocal burrow]

a 1660 should be plenty enough for that

[polar cove]

Any reason i dont have the ip adapter option in my control net?

[vocal perch]

Hi Everyone
Is it possible with stable diffusion to add items in empty room without changing walls floor etc.

[coral galleon]

i tried to uninstall the python beta and reinstalling it an it worked

thank you

[karmic crown]

please add info to github, in case someone else experience the same issue

[karmic crown]

outdated version? which version do you have?

[lone helm]

Good afternoon everyone. I have a problem with generation, at first I saw a normal picture in the preview, but after generation I got this:

What's wrong and how to fix it?

[karmic crown]

are you using LCM?

that can happen with LCM with too large cfg

what are your setting used for generation?

[lone helm]

What is that 'LCM'?

[karmic crown]

just forget it for now, if you do not know what it is, we can get back to that later

what are your settings used for generation?

copy the info below the picture after you generate it

[lone helm]

I change settings to many time

Steps, models, samplers

[karmic crown]

this

[lone helm]

Ok, i will make new generation

[karmic crown]

create an image and send a screeenshot

[lone helm]

...

[karmic crown]

make a screenshot of left part while you are waiting

like this

[karmic crown]

wait wait

you are trying to use 400x400 with SDXL

collapse those two (arrows in the right top corner) - that will deactivate them, until you open that again

use 768x768

10 steps

euler a

what you got?

[lone helm]

Still in progress

[karmic crown]

you can stop it

and please try what i just sugested

[lone helm]

Ok, thank you for your time

[karmic crown]

which model do you use btw?

[lone helm]

ALbedo XL

[karmic crown]

ok, thatn those setting shuld be fine

maybe image will not be great, but at least it should look normal

[lone helm]

omg

[karmic crown]

what is your GPU?

that is too slow

[lone helm]

1060 - 6 gb

[karmic crown]

ouch, that will be probably painfully slow for SDXL

but you should get relatively decent speed with 1.5

do you have some parameters in webui-user.bat?

[lone helm]

I have 1080 ti, but it burn a little

[karmic crown]

screenshot this part

is xformers N/A or have a version?

[lone helm]

Ye

[karmic crown]

thats look fine

show me your webui-user.bat

and tell me do you launch webui-user.bat or webui.bat?