🤝｜tech-support

[lone helm]

I launch this

[karmic crown]

please find this part in your opened cmd window

Launching Web UI with arguments: --skip-torch-cuda-test --opt-sub-quad-attention --opt-split-attention-v1 --upcast-sampling --no-half-vae --medvram-sdxl

what you have there?

[lone helm]

Launching Web UI with arguments: --autolaunch --medvram --xformers
Loading weights [a40d817f46] from C:\AI\stable-diffusion-webui\models\Stable-diffusion\albedobaseXL_v13.safetensors
Running on local URL: http://127.0.0.1:7860

[karmic crown]

so you have --medvram --xformers
please wait for some of guys with nvidia to tell you what you should also add

i am mac guy

[ornate elk]

There isn't more he can add :/

[karmic crown]

and CS1o is here - I left him in your hands 🙂

[lone helm]

Thank you

[karmic crown]

@ornate elk he got estimate of 25:30 for 768x768 10 steps euler a

should he just give up on SDXL?

my m1 mac generate that in 2:30

and my mac is extreamly slow 🙂

[lone helm]

idk, mb you have NASA mac

[ornate elk]

if its that slow, i would advice staying with 1.5 models

what do you mean with burnt? is it broken?

[karmic crown]

unfortunately I saw how a1111 works on a100 with 80GB. I still have nightmares 🤣🤣🤣

[ornate elk]

oh wait he used a refiner + sdxl

[lone helm]

exactly

[karmic crown]

i told him to collapse those

[ornate elk]

okay rip :/
did you tried 768x768, 20 steps only with an sdxl model and not refiner opened?

[karmic crown]

i told him to try with euler a and 10 steps, but he show estimate of 25:30

[karmic crown]

can you screenshot the whole a1111 screen?

[ornate elk]

collapse hires fix too

then it should be faster

[karmic crown]

interupt the genration

do what he told you and try again

you should get image in a few minutes (i have no idea how slow it is)

[ornate elk]

ill say 3 to 5 mins

but didnt saw actual tests

[ornate elk]

if your using sdxl models (6gb models)
make sure these two are not opened:

[lone helm]

i need restart my lag-machine

[karmic crown]

i do not know if you have seen his message, but you were right - he used some beta version of Python 3.10 ☠️
It works now

[ornate elk]

ohhh that would explain the errors ^^

[lone helm]

So, im here

And i will try again

Better

[ornate elk]

yea better but you wont get faster, you may can use sdxl turbo models

but i would use 1.5 models and then upscale them for better quality

[lone helm]

almost without artifacts

[ornate elk]

oh whats your vae?

[ornate elk]

you need no or an sdxl vae

because that one is for 1.5

[lone helm]

You mean this problem "artifacts" because i chose wrong VAE?

[karmic crown]

yes

you cannot mix sd and 1.5 resources

that will make your life miserable

[white thorn]

Hello, im trying to use ultimate SD upscale (linux+A1111+AMD card btw) and I got : *NansExeption : a Tension with all NaNs was produces in Unet. this could be either because theres not enough precision to represent the picture or because your video card does not support half type. Try setting the upcast cross attention layer to float32 option in settings>stable diffusion or using the --no-half argument to fix this. use disable nan check to disable this check. *

So I tried the option , didnt changed anything, disable nan check make my image all black and no half do this :

OutOfMemoryError: HIP out of memory. Tried to allocate 2.85 GiB (GPU 0; 15.98 GiB total capacity; 13.82 GiB already allocated; 1.64 GiB free; 14.28 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_HIP_ALLOC_CONF

Do I understand it right if I say it seems pytorch reserve 14GiB ? is there a way to fix this ? if someone could help

[polar cove]

That's weird. I did a 'check for updates' and no updates avail although it was from august, I just uninstalled the extension and manually installed from the gitub link.

all good now

[ornate elk]

Hey, don't use --no-half and --disable-nan-check
What are your settings when you get that error?

[summer tinsel]

Sorry about the late reply on this. We had to investigate the issue a bit more. Many prompts result in blurry images even though the images can be hardly considered as nudes. For example front view of a fat guy's face and shoulders, a woman in a swimming suit on a surfboard (I can even see that the woman has clothes on in the blurred image), and many more. This is causing quite some confusion among our users. Is there anything that can be done about it?

Another question I posted here and wasn't answered: We are VAT registered Europe company. When making a purchase of credits, we should be able to fill in the VAT number in Stripe which should result in getting rid of the AVT added to the price. This option is however not available in the case of payments to Stability. Am I missing something? Should we include our VAT number somewhere else?

Thank you very much.

[dim ermine]

Hi there, after lots of hours trying to figure out why is my brand new MBPRo M3 Pro ( 12 coreCPU - 18 GPU ) is slow. I realise that is a prolem only from MAC. Is there a solution or any solution ( Coding or other ) on the way?

My solution so far :

1. Selling & buy a M3 Pro plus ( 16 core CPU = 36 or 48 GPU )
2. Work with a GPU server ( Runpod - Vas.ai or other... )
3. Try Core ML ( But not sure if it's complete like the Original Auto1111? The interface seems too simple... )

Thanks for you answers, it is driving me crazy that slow process 🤦‍♂️✌️

[placid edge]

Maybe try some other went say ComfyUI which uses less vram or fooocus, this is based on comfy and runs a lot better on my system

[dim ermine]

I tried... Still slow

[dim ermine]

What system you have?

[placid edge]

I’ve got a 2070 super with a Ryzen 2600x and fooocus or comfy works so much better

[dim ermine]

Got a MAC = No bueno...

[lone wolf]

@ornate elk Thank you again for helping me install AMD SDXL ❤️ But I ping you now as I RMAd my 7900 XTX and got a 4090 xD Do I need to reinstall SD or just change the commands in webuser?

Sorry about that xD

[hardy canopy]

Do you guys know how to change the models directory path for Fooocus?

[ornate elk]

hey np, you need to reinstall SD with my SD Nvidia Guide:
#🤝｜tech-support message

[placid edge]

I have been seeing people running sd on macs though.

[lone wolf]

No need to edit the webuser ? 🙂

[ornate elk]

Hey, install auto1111 with @karmic crown install script for Mac.
It also makes performance adjustments:
#🤝｜tech-support message

[ornate elk]

You only need to add: --xformers --no-half-vae

[ornate elk]

nope, but you can use Symlinks. You can do that with cmd commands. or Best tool for that is the LinkShell Extension, can link it if wanted.

[dim ermine]

Give me that :

File "<string>", line 23, in <module>
ModuleNotFoundError: No module named 'torch'
[end of output]

note: This error originates from a subprocess, and is likely not a problem with pip.
error: subprocess-exited-with-error

× Getting requirements to build wheel did not run successfully.
│ exit code: 1
╰─> See above for output.

note: This error originates from a subprocess, and is likely not a problem with pip.

[ornate elk]

You need uninstall the amd drivers install nvidia drivers.
Then open up a cmd and type
Pip cache purge
Then delete the venv folder

Then rerun the webui-user.bat

[fierce raptor]

I remember the main page of focs has many intro how to set models directory path

[dim ermine]

Geeting to deep for me thanks. But now, my PNG info make system bug ( connection errored out ) since à type those ( --xformers --no-half-vae ) when launching webui. It did something?

[clever cargo]

I am about to go to bed.. If anyone can reply directly to this message I will try to respond when I next have time.

I am having so much trouble understanding how to run StableDiffusion on my own PC. I have spent several hours downloading and uninstalling and redownloading things. I got it to work for awhile the very first time but once I noticed that it was NOT utilizing my GPU then I tried to uninstalling and attempt a different method of install. Unfortunately now I am exhausted and I have work in like 5 hours or so

I Have:
NVIDIA GeForce RTX 2060 (GPU)

AMD Ryzen 5 1600 Six-Core Processor (CPU)

and

idk how much ram but at least 8 GB(Most likely more..maybe when I get more time I can get the exact info for my RAM......)

Can anyone please help me take the first steps :'c

I am lost and desperately excited to start

Thanks for your time!😭

[ornate elk]

Can you post the full cmd log?

[ornate elk]

Hey, you can follow my install guide. And if you get problems post a screenshot of the full cmd log.
Here is the guide:
#🤝｜tech-support message

[turbid elk]

does someone have a colab to train an own sd model? kohya ss suddenly stopped working for me, always gives me some error

[wide frigate]

guys, someone can help me with this error?

Cloning into 'C:\Users\Fenix_Kayami\Documents\GitHub\stable-diffusion-webui'...
remote: Enumerating objects: 29021, done.        
remote: Counting objects: 100% (69/69), done.        
remote: Compressing objects: 100% (44/44), done.        
error: RPC failed; curl 18 transfer closed with outstanding read data remaining
error: 7948 bytes of body are still expected
fetch-pack: unexpected disconnect while reading sideband packet
fatal: early EOF
fatal: fetch-pack: invalid index-pack output

[worthy bluff]

hey guys, im following a guide to install Automatic 1111 with Stable diffusion 1.5 and i keep getting an error
ERROR: Could not find a version that satisfies the requirement torch (from versions: none)
ERROR: No matching distribution found for torch

I have an AMD GPU/intel CPU, maybe i need the CPU only version or something but idk how

[dusty meteor]

hello - i'm looking for some help from you kind folks with getting the Web UI working successfully on an Intel Mac Pro 2019. I have two GPUs installed, an AMD Radeon Pro W5700X (MPX) 16GB and an AMD Radeon RX 6900 XT (Reference) 16GB, so should have plenty of power to render.

I did the "vanilla" install for Mac following the instructions at https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Installation-on-Apple-Silicon
I could load up the Web UI, but could only get it to generate grey squares or garbled images. This is just with the regular "v1-5-pruned-emaonly.safetensors" model.

Today I came across @karmic crown's "fix"(?) to get Automatic1111 installed correctly on Macs at https://github.com/viking1304/a1111-setup and ran it.... I think successfully. However, even after doing this, the Web UI is still not generating proper images. I'm using the prompt "man on the moon" with the SD 1.5 model and getting a grey square.

A log of my Terminal output after running sh a1111-setup.sh is available at https://drive.google.com/drive/folders/1R-rnbz-S1yo4Jzc061AYO12GY5_33xtv?usp=drive_link along with a screenshot of what I'm getting from the Web UI.

I'm sure it's something to do with the arguments the Web UI is being loaded with, but not sure....

Launching Web UI with arguments: --skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half-vae --medvram-sdxl

(I'm very new to using Discord, so apologies in advance if I'm breaking any rules or demonstrating poor etiquette! I'm open to feedback 🙂 )

[ornate elk]

hey, you installed the wrong webui. AMD needs the directml one.
Before following my install guide, delete the wrong webui folder.
then open up a cmd and type
pip cache purge
that will delete the wrong files for nvidia sd stuff.

Here is the guide:
#🤝｜tech-support message

[ornate elk]

it seems like you got an connection error while cloning the git repo. make a new folder and try again. also make sure to have a stable connection

[wide frigate]

200kb-1 mega connection for now

[ornate elk]

kb isnt good :/

[young stirrup]

A tensor with all NaNs was produced in VAE. This could be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

anybody know why I keep getting this error time after time when trying to upscale to 1024x1024?

[feral mango]

hi my automatic 1111 is losing conection very often. I am running it locally and everytime i do a big high rez or download too many models...i need to restart the bat file because it loses connection

[ornate elk]

You need to add --no-half-vae to your COMMANDLINE_ARGS

In the webui-user.bat

[ornate elk]

Make sure the webui is whitelisted in any browser adblocker. Also what browser are you using?

[feral mango]

firefox

how i do that step =O?

this are my settigns atm
set COMMANDLINE_ARGS=--no-half-vae --opt-channelslast --opt-sdp-no-mem-attention

[ornate elk]

Remove --opt-sdp-no-mem-attention

And use --xformers instead

The error comes from the no-mem-attention as it ignores the vram limit

[dusty meteor]

@ornate elk is this the best place to ask about this?

[worthy bluff]

your setup worked like a dream, its all running and i can even choose the different versions, now its just figuring out how to get realistic images and use this thing for my art project haha

[lean fog]

Is there anyone available to ask about using FOOOCUS?

Im running it on AWS

[ornate elk]

Hey, thanks for the kind response! For questions about prompting you can also ask in #📝｜prompting-help
But I would suggest you checkout civitai.com and look at models like realistic vision

[ornate elk]

Hey, can you try to add --no-half to the other args and then try again?

Also here is the best place to ask

But for Mac stuff, Viking knows more

[dusty meteor]

Thanks for the reply - will try that and see

[dusty meteor]

Sorry if this is a stupid quesiton, but how do I run it once off with different args? or should i edit the webui-user.sh file?

like should I include all the args i saw it using on the previous run ( --skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half-vae --medvram-sdxl ) and just add the one you mentioned ( --no-half ) ?

or are the ones previously included already included if I just run ./webui.sh and I only need to append "--no-half" ?

[dusty meteor]

ah.... i just added it to the long list of args previously included and now i'm actually getting images! yay! thanks!

[dusty meteor]

on my 3rd image generation, i'm running into this error:

RuntimeError: MPS backend out of memory (MPS allocated: 11.67 GB, other allocations: 15.42 GB, max allowed: 27.17 GB). Tried to allocate 160.00 MB on private pool. Use PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.0 to disable upper limit for memory allocations (may cause system failure).

[lean fog]

anyone know how to use AWS

[karmic crown]

@dusty meteor Macs with eGPU works a little different than "normal" Macs, command parameters I used in scripts should be changed fro eGPU

[dusty meteor]

i'm not using eGPUs, both of my GPUs are PCI cards in the Mac Pro

or are you saying that your script https://github.com/viking1304/a1111-setup is specifically for Macs with eGPUs?

[karmic crown]

my script is for ARM and Intel Macs with Radeon

I only have M1 MacBook and Intel iMac with Radeon

https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/5461#discussioncomment-7822007

this guy had a problem with eGPU, maybe it is related to your issue

[dusty meteor]

right, my 2 cards are AMD Radeons so I'm good 🙂

[karmic crown]

he also has radeon as eGPU

😦

we need to find paramteres that works for you

[dusty meteor]

so CS1o's suggestion of adding --no-half to my args got the UI generating images.... i just started to run into the error then

RuntimeError: MPS backend out of memory (MPS allocated: 11.67 GB, other allocations: 15.42 GB, max allowed: 27.17 GB). Tried to allocate 160.00 MB on private pool. Use PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.0 to disable upper limit for memory allocations (may cause system failure).

[karmic crown]

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

try those 5

let me know if any of those works for you

[dusty meteor]

regarding the "MPS backend out of memory" error, i found this https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/9133 which led me to Apple's Dev Docs about Metal-Acclerated PyTorch training https://developer.apple.com/metal/pytorch/

so i installed the latest PyTorch

i'm now wondering.... can i just run ./webui.sh again? Or do I need to purge something first to ensure it uses the latest PyTorch that's been installed?

[karmic crown]

my guide and script both use metal, and development version of torch, so thats fine

for some reason i cannot copy the link to this exact message

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

find that message from Torqueeee and read that part

we are trying to figure out what will work for him (and you)

PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.7 does not work for him, nor me

he said that PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.0 works, but than generation is slow

[dusty meteor]

@karmic crown - just clarifying that the Web UI is now generating images for me - I'm no longer getting grey squares 🙂 Just checking you saw that, right?

[karmic crown]

with --no-half-vae --medvram-sdxl = no image
with --no-half --medvram-sdxl = image + error about memory limit

correct?

is this your case too?

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

Again got:
RuntimeError: MPS backend out of memory (MPS allocated: 13.04 GB, other allocations: 14.03 GB, max allowed: 27.17 GB). Tried to allocate 160.00 MB on private pool. Use PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.0 to disable upper limit for memory allocations (may cause system failure).

[dusty meteor]

correct.... but the error is only after i do like 2 generations

[karmic crown]

have you tried this one?

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

it works super slow on my system, but it might work for you

[worthy bluff]

i just noticed im getting an error no module 'xformers'. Processing without...

Do i need to do something special with an AMD GPU to install xformers? i read its good for the blocks in processing images

[dusty meteor]

i'm currently running the Web UI with this

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

[karmic crown]

use one of those --no-half --no-half-vae not both

[dusty meteor]

and it actually hasn't thrown that MPS error yet..... after 5 generations.... using different models

[dusty meteor]

ah got it! will take out "--no-half-vae"

on a side note..... is this forcing it to use full precision instead of half precision? and thus making generation slower for the GPU?

[dusty meteor]

I believe xformers is only for Nvidia GPUs - https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Xformers

[karmic crown]

generally yes, but it is needed if --no-half-vae does not produce image

i have that situation with nvidia 1660 in my sons PC, i need to use --no-half, or i cant get image at all

[dusty meteor]

yeah i understand. but it seems like a fairly significant advantage of the GPU to be without!

[karmic crown]

#🤝｜tech-support message

eGPU on mac

[dusty meteor]

so i've been running with "--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half --medvram-sdxl" now and i've actually not been getting that MPS error

it's like installing the latest PyTorch actually fixed things with that?!?

[karmic crown]

show me this part of ui

[worthy bluff]

i see

[dusty meteor]

that's the only thing i changed..... i think...... unless my system was actually out of memory previously..... which would be weird.... because i have like 576 GB of RAM - or does MPS only use VRAM, not regular RAM?

[karmic crown]

that's good, you are using latest development version of torch

[karmic crown]

this should work for you

how long does it take for 1.5 to generate?

and how long it take for sdxl?

[dusty meteor]

is there a specific prompt i should try?

and for 1.5, i'm using "v1-5-pruned-emaonly.safetensors" model, is that right?

and for SDXL i should use ^^^ ?

[karmic crown]

just use 512x512, 20 steps, DPM++ 2M Karas, no refiner, no hiresfix, with any prompt and any 1.5 model

for sdxl, use 768x768 or 1024x1024 with same other params

i like this 1.5 model https://civitai.com/models/84728

but you can use any you like

[wet osprey]

this looks like 2.1, not sdxl

sdxl should be 1.0

[karmic crown]

i would not use 2.0 nor 2.1

you can try this one https://civitai.com/models/200061/hephaistosnextgenxl-lcmbf16hybrid

you can use it with LCM to get an image with less steps

actually, it has LCM lora embeded

9 steps only

[dusty meteor]

got it thanks - i'm downloading the "vanilla" version first from https://huggingface.co/stabilityai/sdxl-turbo/tree/main

[karmic crown]

it will save you a lot of time

not turbo

turbo is for 1 step image generation

[dusty meteor]

thanks - downloading now

[dusty meteor]

ok so for 1.5, it took 12.6 secs

[karmic crown]

https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0_0.9vae.safetensors?download=true

this is base xl model

[karmic crown]

nice, it takes more than 30 seconds on my m1

[dusty meteor]

confused.... on the page https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/tree/main - wouldn't the "sd_xl_base_1.0.safetensors" be the base model.... isn't the one you linked "sd_xl_base_1.0_0.9vae.safetensors" a VAE? (I haven't gotten to understanding what VAEs are yet!! 😕 )

[karmic crown]

vae cant be almost 7 gigs 🙂

that is 1.0 model with 0.9 vae "baked in" (embeded)

this is the recommended base model, since 1.0 version of SDXL vae had a problem

[dusty meteor]

ah i see.... OK

[karmic crown]

this should be very easy to understand 🙂

[dusty meteor]

thank you for making it .... VISUAL 🙂

[karmic crown]

vae fix belached colors

but... NEVER mix 1.5 with SDXL, or you will suffer without any need 🙂

like sdxl model + 1.5 vae = NO NO NO 🙂

when you find what works best for you, please leave a comment in https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/5461 and write what system you have and what works the best for you (and what didn't work)

let me know (here) if you have any other mac related problem

its 3am, so i am going to sleep

[dusty meteor]

ok so for SDXL Base 1.0 0.9 VAE model...... for a 1024x1024 img, render time was ..... i got that MPS backend our of memory error again! 😭

[dusty meteor]

will do! thanks so much for your help! good night!

[karmic crown]

can you try
--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half --medvram

[dusty meteor]

for SDXL Base 1.0 0.9 VAE model...... for a 768x768 img, render time was 41.9 secs

[karmic crown]

this is nice time

m1 needs more than 2 minutes

[dusty meteor]

yay! 🙂 doesn't really look like it's using a lot of GPU though

[karmic crown]

sd is not fully optimized for mac yet, but i am trying to help as much as i can 🤷🏻‍♂️

[dusty meteor]

does "--medvram" mean something different to "--medvram-sdxl" when I'm using SDXL ?

[karmic crown]

it should be no difference, but let see if that would fix the error or not

i am doing a wild guess

since noone tried what you are doing yet 🙂

[dusty meteor]

ok starting up Web UI again now

[karmic crown]

oh, yes

important thing

ALWAYS restart webui when you get any memory related error

3:27, i can see anymore 🙂

[dusty meteor]

i'm just doing Ctrl C twice in Terminal, and then running ./webui.sh again, is that the right way?

[karmic crown]

yes

[dusty meteor]

ugh still getting the MPS backend out of memory error for a 1024 image on SDXL

for a 768x768 it took 44s this time

[karmic crown]

you must have --skip-torch-cuda-test (no nvidia)
you must have --upcast-sampling or your python will crash

you can experiment with other things

how much vram your gpu has?

[dusty meteor]

and Photon V1 render (which is visually stunning!) took 36 seconds - 768 x 768

both GPUs have 16 GB

[karmic crown]

you can try to add
export PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.0
to webui-user.sh

i am sorry, but i relly need to go now

[dusty meteor]

sounds a bit risky.... since it can cause system failure (i've read).... but maybe 0.8 or 0.9

yes i understand..... go to bed! we can chat again

[karmic crown]

you can experiment

but only 0.0 worked for me

everything else gave me errors

i wish you luck

good night

[vernal turret]

Why can I not see images I'm generating on my mobile device suddenly?

[plucky marsh]

does anyone know what is the fix for stable diffusion constantly timing out on google colab?

[karmic crown]

can you try to disable those two lines in modules/sd_disable_initialization.py

then try this those param sets:
--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half --medvram-sdxl --precision full
--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half --medvram --precision full
--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half --precision full

let me know if there is any difference

but that probably would not help, since i just checked again what the actual error was and why i was added that
"TypeError: Cannot convert a MPS Tensor to float64 dtype as the MPS framework doesn't support float64. Please use float32 instead."

[gloomy magnet]

Hello, how to insert an image into a picture so that it looks realistic (in color, light and perspective). instead of this machine you need to insert cybertruck

[zenith nova]

where do i put this launch argument? i need to "add --api " according to google to fix my SillyTavernExtras issues but where does it go?

im running SD locally in web-ui and STE doesnt want to see it on the URL

do i put it in run.bat?

like this?

HA GOT IT

right here like that, if anyone has that issue

[paper viper]

Is there anyone to know about stablecode?

[ornate elk]

What's your GPU?

[unreal steppe]

before I start to train , it happened. How to solve?

[tame yoke]

is there a way to have a negative prompt by default

[hardy thorn]

guys, where do I find bsrgan?

[lean ridge]

whats the default value of the Noise multiplier for img2img?

[fair oxide]

NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

I tried turning upcast cross atention on, but still it does that

I just try to use image 2 imaage

[lean ridge]

try generating a text2img first

and then try img2img

it happened to me

[fair oxide]

Oh i think its because i use a XL model...

Does a1111 still not support xl models with image to image?

[karmic crown]

your COMMANDLINE_ARGS are probably wrong

what is your GPU, which OS, and what you have in COMMANDLINE_ARGS?

[fair oxide]

set COMMANDLINE_ARGS= --xformers --api
Windows 11
3090

I thought xl models didn't work with control net and image 2 image, at least last time i tried?

[karmic crown]

try to add --no-half-vae

how much vram do you have?

24?

[mint light]

Would anyone know why a ton of my Loras aren't showing up under my extra network tabs? I have a massive library and only a handful of them are showing up. I can still manually add them to the prompt and they work fine but browsing them is impossible. Is there a limit as to how many can be populated at once?

[karmic crown]

xl model and 1.5 loras? or the other way around? maybe

[mint light]

I only use 1.5, I just checked in SDnext and the extra network tab populates just fine. Its something with A1111 specifically.

[desert garnet]

why do my models show up as undefined?

what am i doing wrong?

[dim ermine]

Hello, I am on Jupiter and it's giving me this... 🤦‍♂️

root@d43691301191:/workspace/stable-diffusion-webui/models/Stable-diffusion# wget https://civitai.com/api/download/models/245598?type=Model&format=SafeTensor&size=pruned&fp=fp16
[1] 18544
[2] 18545
[3] 18546
root@d43691301191:/workspace/stable-diffusion-webui/models/Stable-diffusion#
Redirecting output to ‘wget-log’.

Thanks for help...

[light nebula]

hey guys, im trying to use kohya to train a lora for sdxl, imk trying to train using multiple gpu's, does anyone know how to do that? i can only get it to work on one gpu

[desert garnet]

im pretty sure those are integrated graphics which are nowhere near capable of running sd

[ornate elk]

SD needs an Nvidia or AMD GPU.
If you only have this Intel igpu you can only run SD in CPU mode. Thats pretty slow.

If you want to use that mode you have to add --skip-torch-cuda-test --use-cpu all --no-half
In your webui-user.bat at the line COMMANDLINE_ARGS=

[teal arch]

hey i am tring to do a gradio ui. i want to know if a image is selected and witch one. when i cklick on a image i am able to get the image id and show it in the text box. i do this with the .select event. but the problem is when i close the image agin then it does not deselected and the value does not chage. the only way i was able to do it was with the big deselct button on the botom. but i want that the same thing happes when i just use the normal x button on the gallery. here is also my code.

[ornate elk]

Can you show me the full cmd log?

Did you run the webui-user.bat afterwards?

Dont run the webui.bat

Lol xD

You have to run the webui-user.bat

The bat will launch everything

Only ever run the webui-user.bat

Nothing more to do

xD that could be an Low RAM issue

How much do you have?

[desert garnet]

can i copy part of a workflow to another workflow?

[abstract oracle]

Hello, does someone know if its possible to use stable diffusion locally with only 7 gpu (MacBook Air M1) or can I only use online versions?

[ornate elk]

Okay, that would explain the freezing.
You have to adjust your windows pagefile to make it work then

[ornate elk]

It should work on your Mac. How much ram do you have?

[abstract oracle]

8GB

[ornate elk]

Okay. That can work:
You can try the script from @karmic crown
https://github.com/viking1304/a1111-setup
He's the Mac expert

[abstract oracle]

Thanks !!

[ornate elk]

launch sysdm.cpl from the Start menu search or run box (Win+R) and navigate to Advanced –> Settings –> Advanced –> Modify
Set it to "20000" for C:
And disable it for all other drives.
Then save and restart the PC.

Okay then first enable it for C:
Set it to system managed.
And disable it for D
Then restart the PC.
If it still freezes you need to adjust it to 20000 for C

Yes

Click on C and enable system managed

Then click on D and click no paging file

Restart PC

Ohh

You need to have 15gb free space on C

Okay 👍

[ornate elk]

It will take time. Let it run

5-30mins

Yea the first installation is a heavy task xD

I'll go sleeping now

Its late here

NP, let me know if it worked 🙂

[lucid stirrup]

does anyone know what is happening here

it was working perfectly i was messing around wit downloading custom cuda for keras and now after few pictures it generates noise or just grey images

my args:
set COMMANDLINE_ARGS=--no-half --no-half-vae --precision full --disable-nan-check

[gaunt root]

Hello, I'm having issues with models/settings that run out of VRAM half way through the process. Can someone help me troubleshoot/optimize for my RTX6700?

sounds like a model/checkpoint issue. try a batch count of 1 first. I'm have a VRAM error problem...grrr

[carmine ibex]

Getting the error in ComfyUI when using Canny Controlnet for SDXL: 'NoneType' object has no attribute 'copy'

[proper pike]

How are these done? https://www.youtube.com/shorts/0ffjgBbo5k4

[graceful loom]

Why i am getting this error
error: 'invalid literal for int() with base 10: ' 3000.5''. before reporting, please check your schedules/ init values. full error message is in your terminal/ cli.

[high sage]

Hello I did all the steps before trying to install a1111 stable diffusion. Still getting this error when I run user.bat file as specified.

Can anyone help? I am really new here and did all the step by step process according to a youtube video..

[ornate elk]

What's your GPU?

[ornate elk]

Hey, for a 6700xt you have to edit the webui-user.bat and at the line COMMANDLINE_ARGS=
you add:
--medvram --opt-sub-quad-attention --opt-split-attention-v1 --upcast-sampling --no-half-vae
Then save and relaunch the webui-user.bat
Also note that you can do 512x768 upscaled by 2 with hires fix without errors. But if you want to get a FullHD image (540x960) upscaled by 2 you need the Tiled Diffusion Extension. And then only enable the Tiled VAE Option. That saves you from vram errors in that resolution.

[high sage]

Okay found it. Apparantly my python was installed in wrong place. reinstalling in sd folder fixed the issue. its fine now

[ornate elk]

Normaly you have to install python in its default place (appdata/local/programs)
The important part is to set the checkbox "add python to path"
Then SD would find it anywhere

[lucid stirrup]

rtx 3060 12gb

[ornate elk]

Remove --precision full --no-half --disable-nan-check
Then add: --xformers --no-half-vae

Then delete the venv folder and rerun the webui-user.bat

[lucid stirrup]

i already did that and the same thing happens

i tried multiple models snd samplers, same thing

[ornate elk]

You also deletet the venv folder ?

[lucid stirrup]

yep

[ornate elk]

Can you restart the webui-user.bat and show me the full cmd log?

[dim ermine]

Hello again, 🙂 Need help again.

I have upload My Ip extension, everything seems alright. But it is not showing up on the interface. Any idea plz again :🙏

[ornate elk]

The IP Adapter is a model file for ControlNet and need to be downloaded or added in the models/controlnet folder

[dim ermine]

Thanks. But I just want to add this control net option. It's a learning curve. Thanks a lot.

[ornate elk]

You need the additional .yaml file for the IP adapter too. Then it will appear

It goes in the same folder as the IP adapter model

[dim ermine]

I'Lll try thanks.

[fair oxide]

Hi all, can I ask a question here about preparing a dataset for training a lora? Or is it in another channel?

When writing tags for a dataset of images, should I exclude tags such as "solo" if I am doing it on one character?

[fair oxide]

The purpose of the tags is to prevent confusion about what is the underlying feature you are trying to capture. So, if you want to make a LoRA that matches a particular person's face but all the training photos have a blue sky in the background, you will inadvertently train the LoRA on both the face and the blue sky (since it is a pattern in your training data, like the person's face) unless you specify in the tags that there is a blue sky in each image.

So, I am trying to understand if "solo" is a tag that should be part of the underlying feature. I want the LORA to know it is one person, so if I tag solo images as "solo" will the LORA try to separate the solo-ness from the image?

I think in general you need to do trial and error. But, something you might find useful is to have the model describe photos for you and make sure you tag the things it notices that are not your underlying feature. So if you have it describe a solo photo versus one with multiple characters and this is reflected in the generated description, you probably want to tag it too.

Thanks, I put it through Auto1111 tagger and 'solo' appeared in 88% of images, so I think I will keep it in to differentiate between that and images with multiple characters,

[lucid stirrup]

the same goes when i put xformers (i removed old venv)

[mystic sequoia]

Hello, I am getting this error for image to image transformation within 512x512 image from stability.ai: "Engine ID stable-diffusion-v1-6 does not support image_to_image at this time"
Wh ich mage to imagemodel is available for 512x512 image from stability.ai?

[next tapir]

i dont think automatic1111 ui is running with my gpu

i am using an rx 6700xt, i did all of the above on linux, what am i missing?

[copper sundial]

Im trying to install stable diffusion on my PC. Unfortunately the person helping me has had to go to work and Im stuck here like
a monkey poking things with a stick

This is as far as I can get

[ornate elk]

You need the rocm drivers on linux

[next tapir]

i thought that they were gonna get installed by the script?

what's the point of the amd installation guide if it doesnt tell you how to use amd?

[ornate elk]

Hey, you can follow my easier install 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.

[copper sundial]

Ive got it up and running. But getting errors when trying to generate things

[ornate elk]

What's your GPU? And what did you installed ?

[ornate elk]

Yea its sadly not a good instruction

[copper sundial]

16GB AMD RADEON™ RX 6800 XT - HDMI, DP - DX® 12

a1111

[ornate elk]

Oh wow then you installed the wrong webui version

Ill give you may AMD install guide. One sec

[copper sundial]

I used an auto installer

[ornate elk]

Yea thats also not recommended

It isnt made for AMD

[copper sundial]

The guy is trying to help me do a very specific thing. So hopefully I'll still be able to follow what he did to get the results he did

[next tapir]

what should we use then??

[ornate elk]

For Linux you need the rocm drivers then it will work

[ornate elk]

That guy mostly has an nvidia card so you can't use his guide or follow it.
You should use my install guide.
After that you can ask him again.

[next tapir]

and how do you install that? I ran this pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/r ocm5.2/ --force-reinstall but it did not change anything

[ornate elk]

@copper sundial
you installed the wrong webui.
AMD needs the directml one on Windows.
Before following my install guide, delete the wrong webui folder.
then open up a cmd and type
pip cache purge
that will delete the wrong files for nvidia sd stuff.

Here is the guide:
#🤝｜tech-support message

[ornate elk]

Can't help much with Linux but maybe that helps:
https://rocm.docs.amd.com/en/latest/deploy/linux/os-native/install.html

[copper sundial]

How do I know if I already have python installed?

[ornate elk]

Check in Windows Settings under Apps if you have it installed

Or in the system control

[copper sundial]

Apps?

[ornate elk]

Apps or Programms

In the windows settings

[copper sundial]

oh add/remove programs?

[ornate elk]

Okay than you dont need to install python but you can install the newer one if you want

[copper sundial]

Ill skip it for now and try later if it fails. I thought I had it from doing something else(duckDNS maybe?)

[ornate elk]

Maybe or the other auto install script installed it

Python is always needed for automatic1111

[somber python]

uhh
hello ive just looked through the FAQ and the start hear and still have no idea

so how do i install and set up stable diffusion?
(i would like to use it to create a token for a TTRP session)

[copper sundial]

Join my conversation dude cos im trying to do the exact same thing

[somber python]

ah
ok
hmm
whelp that looks terrifying

[copper sundial]

Tell me about it. Im stuck waiting on what to do next cos I cant even install it when following the instructions

Just know that it is 100% doable to make the tokens. One of the people helping me has made a token easily looking like what I want.

[copper sundial]

As an aside, What VTT are you using?

[somber python]

roll 20

[copper sundial]

Foundry here

[somber python]

better but costs money

[copper sundial]

I was lucky to have some spare cash. Totally worth it. But if you do ever get it. It has a bit of a learning curve

[copper sundial]

What does this error mean?

[fair oxide]

I am trying to figure out lora training and I have 10 epochs that I am running through, about half of them seem to ignore the character I trained the lora on and fulfil everything else in the prompt, I get the same image from 5 epochs that has nothing to do with the lora I want. Does that mean those epochs were overtrained or is there something in the way I am testing it?

[leaden night]

I have 20gb of VRAM and its still jibbering about not enough GPU memory available

[mystic sequoia]

Hello, I am getting this error for image to image transformation within 512x512 image from stability.ai: "Engine ID stable-diffusion-v1-6 does not support image_to_image at this time"
Which image to image model is available for 512x512 image from stability.ai?

[ornate elk]

make sure you have enough space on C
then open up a cmd and type
pip cache purge
then delete the venv folder again and relaunch the webui-user.bat

[copper sundial]

Im halfway through re-installing Vladomatic(or whatever its called) but on D drive as way more space.
I'll install this one you linked me to on D as well and see how I get on

Ok I am up and running!

Thanks for all the help folks

now to just learn how to tweak it to be closer in style to the original

[copper sundial]

Are you using a radeon card or Nvidia?

[somber python]

Asus

[copper sundial]

Errrrm Oki im not sure i can help. If it was radeon I could have talked you through how I got mine up and running

[somber python]

eh no problems ime puzzleing my way through it with lots of googleing

[copper sundial]

Good luck

[ornate elk]

hey, follow my install guide to get startet with SD: (if you have an nvidia gpu)
#🤝｜tech-support message

[karmic crown]

yes you can use a1111

since i saw that @ornate elk told you to try with my script, please let me know if it work for you

[fair oxide]

can i ask for help here

i am having trouble with using SDXL

[karmic crown]

what is your problem?

[fair oxide]

i keep getting error messages abour running out of memory when using SDXL

[karmic crown]

gpu, vram, os?

[fair oxide]

hold on i ll show u

[karmic crown]

your command options are probably wrong

[fair oxide]

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 128.00 MiB. GPU 0 has a total capacty of 7.79 GiB of which 46.12 MiB is free. Including non-PyTorch memory, this process has 7.73 GiB memory in use. Of the allocated memory 7.21 GiB is allocated by PyTorch, and 336.09 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

[karmic crown]

what is your GPU, how much VRAM it has, and what is your OS?

[fair oxide]

i have a rtx 3070 (8GB), my OS is pop OS (linux btw)

[karmic crown]

do you have ROCm drivers? what are your command params in webui-user.sh?

[fair oxide]

i am not using that

i am following one of the examples

https://huggingface.co/docs/diffusers/using-diffusers/sdxl

i pasted this code in a .py file

from diffusers import AutoPipelineForText2Image
import torch

pipeline_text2image = AutoPipelineForText2Image.from_pretrained(
"stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
).to("cuda")

prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipeline_text2image(prompt=prompt).images[0]
image

and then i executed it

[karmic crown]

you cant used cuda on amd

cuda is nvidia

[fair oxide]

rtx 3070 is an nvidia gpu

[karmic crown]

do you need to use code? or you just want to run sd?

?!

[copper sundial]

Im getting this error: composition via AND is not supported for DDIM/PLMS samplers
Time: 0.64s

[karmic crown]

i misread something than

sorry

[fair oxide]

https://www.techpowerup.com/gpu-specs/geforce-rtx-3070-mobile.c3712

[karmic crown]

i was sure you mentioned some other card

who knows what i saw

i know what is rtx

🙂

sorry

[copper sundial]

That was me with the AMD card

<<< GUILTY

[karmic crown]

i guess i am too tired

sorry

i had a day from hell at work

[fair oxide]

is there something wrong i am doing?

[karmic crown]

but this question remains, do you need to use code? or you just want to run sd?

if you just want to use sd, better run a1111

if you need code, that is another story

[fair oxide]

i havent tried using a1111

[opaque spindle]

@karmic crown I am a newbie here and would greatly appreciate some pointers from you. From the post, https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/5461#discussioncomment-7698107, I understood that you succeeded running SD on a Mac without M1 or M2 chip but I didn't. I would appreciate your pointers on what steps I missed and need to address to bring up SD. Thanks again very much!

Here is my machine: 2019 Macbook with AMD Radeon Pro 5500M 4 GB
Intel UHD Graphics 630 1536 MB & macOS 14.1.1

Here are the steps been taken:

1. Followed this instruction at first with python 3.10.6 https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/5461

2. Updated my webui-user.sh to add these 2 lines:

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

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

3. Did two minor changes including:
   Updatd gradio==3.41.2 in extensions/Text2Video-Zero-sd-webui/requirements.txt to match the version in requirements_versions.txt
   Replaced rank_zero_deprecation with rank_zero_only

4. Didn't install xformers as I am not sure if it's needed and I don't know how to install it, either.

Result:
Despite http://127.0.0.1:7860/ is up, Generating an image trigger an error:
`AssertionError: We do not support vanilla attention in 1.13.1 anymore, as it is too expensive. Please install xformers via e.g. 'pip install xformers==0.0.16'

Time taken: 1 min. 1.8 sec.`

[karmic crown]

even if you want to use code, I advise starting with a1111. many things will be much easier for you to understand after some time

[fair oxide]

i ll try using it and see if i have better luck

[opaque spindle]

Should I just use a1111?

[karmic crown]

@cm just a second

[karmic crown]

when you understand the principles, what is what, it will be easier to understand code

i was stupid enough to start with code 🙂

@cm did you follow the original guide or mine?

also, does your Mac have two graphics?

try trhis

rename your stabledifusion webui folder if you need anything from there

or just remove it if you do not

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

download and run my scipt

let me know if it work

it is possible that you will need to tweak some params later

but it should install a1111 an run it without a problem

i am currtently working on adding some command line parameters

[opaque spindle]

I followed the original guide, https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/5461. I didn't know you have another I could follow. I am more than happy to try. I just need to set up SD on Mac without M1/M2. As long as there is one way to make it work, I am happy. Also, on cpu is very slow. I wish to speed up if the gpu can be of use.

[karmic crown]

do not expect too much

you only have 4GB or VRAM

even my iMac with 8GB is slow

and my m1 mac is slow

my guide is here

same page just lower 🙂

[opaque spindle]

Hmm... To generate one SD-XL image, how long am I expecting on 4GB non-M1 Mac.

[karmic crown]

i am not sure if you would be able to run it

lets fix your install first

and see how fast is 1.5

then we will try sdxl

[wet osprey]

Which GPU?

[karmic crown]

amd on mac

[wet osprey]

I have an R9 M295X 4GB. If you have the same one that would be convenient.

Because then I could just tell you how long it takes for me on that GPU.

[karmic crown]

i have intel mac with radeon with 8GB and it is slower than my m1 mac

[ornate elk]

you should use auto1111
follow my install guide:
#🤝｜tech-support message

[karmic crown]

and he has macbook with AMD Radeon Pro 5500M 4 GB

i have Raden Pro 580X 8 GB

and that is slow as hell

[opaque spindle]

Thank you very much! Let me try this new instruction first. Please allow me an hour or two.

[karmic crown]

script takes few minutes 🙂

but if you want to follow instructions just do it

[opaque spindle]

I will run just the script

[karmic crown]

it takes a few minutes on my m1, but i have 600Mbps connection

[opaque spindle]

Running a1111-setup.sh now

[opaque spindle]

Should I place a1111-setup.sh script inside my stable-diffusion-webui directory or another directory?

[karmic crown]

no no

my script creates stable-diffusion-webui folder in your home directory

for example /Users/viking/stable-diffusion-webui

[opaque spindle]

ok ok

[karmic crown]

if folder exists

it checks if you have venv inside or not

do you have .git inside or not

[opaque spindle]

The folder doesn't exist... Thanks. I am just waiting for the script to be done.

[karmic crown]

if stable-diffusion-webui exist
print "New installation. Cloning A1111..."
clone automatic1111

if not
print "Existing installation detected..."
purge pip cache
remove venv
ensure that correct files are used

[opaque spindle]

Thanks for explanation

My connection is slow and it's still doing brew upgrade... Please give me few more minutes.

Btw, I encountered a problem before on my machine,
brew_install python@3.10 install the latest 3.10.x but some part of decord asks for 3.10.6.... let's cross the finger

[karmic crown]

basic flow of my script:
Install Homebrew
Temporary fix Brew path on ARM devices
install required packages
(re)intall a1111
apply some fixes
run webui

[karmic crown]

forget that, for mac is 3.10.13

also, 3.10.6 and 3.10.13 are the same from users perspective

[opaque spindle]

I see. So the reason SD doesn't work is because my brew packages not working? So fix Brew path will make SD installation work?

[karmic crown]

your pythong packages were messed up

and maybe your command params

we will fix all that

[opaque spindle]

Hmm.. because after I followed the original instruction, I encountered some issues, and fixed the issues by other threads in the discussion.

But by running a1111, I will have a brand new python venv and command params will be restored to the origin state, freshly cloned into the home directory, so it should work, hopefully.

[karmic crown]

xformers is for nvidia, so something was obviously wrong

[opaque spindle]

I didn't install xformers but the webui.sh keeps asking for xformers. 😦 Is it in the configuration, command prompts?

Really appreciate your help!

[karmic crown]

those are "magic" files 🙂

webui-user.sh is the file with command paramteres for mac

[opaque spindle]

ok

[karmic crown]

my script will add some of those for you

[opaque spindle]

cool. What these parameters mean if you don't mind me asking?

Would it use GPU even though there is only 4GB?

[karmic crown]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings

we can easily test that

[opaque spindle]

Reading

[karmic crown]

i have a script for that too 🙂

#🤝｜tech-support message

[opaque spindle]

Hmm... it doesn't explain what these parameters for. It only lists what parameters are there and their default value.

[karmic crown]

--skip-torch-cuda-test - we need to use this since we do not have nvidia

--upcast-sampling without that one python on mac crashes when you try to run a1111

--no-half-vae - faster version of --no-half that use that option only for vae, otherwayse it is used for everything

--medvram-sdxl - use medvram only for sdxl

--opt-sub-quad-attention - memory optimization for the rest of us (who do not have nvidia)

--opt-split-attention-v1 - another version of memory optimization

clear enough?

whats happening with installation?

[opaque spindle]

the script is still doing brew upgrade....

[karmic crown]

how fast is your connection?

[opaque spindle]

slow.. but not sure how slow

[opaque spindle]

Thanks. Clear now.

Brew upgrade is done.. seemingly installing python packages now.. Thanks again.

[sterile marsh]

hello everyone, new to stable diffusion. I currently don't have a strong pc so I decided to set up stable diffusion through runpod cloud. When I finished the cloud setup and attempted to prompt, this is the error message I get in the screenshot. The cloud is feeding off of an RTX 3090

[opaque spindle]

Thank you! It works this time! It works with the default v1-5-pruned-emaonly.safetensors [6ce0161689]. Thanks again!

[karmic crown]

now find a model you like on https://civitai.com/

[opaque spindle]

Let me run this test script. What's for? I'd like to try sdxl next.

ok ok

[karmic crown]

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

this one is nice 1.5 model

for sdxl, since you have a gpu with not much vram i recommend lcm embede model

like this one

https://civitai.com/models/200061?modelVersionId=256258

you can get good image with less steps

how much time it takes to get 512x512 with 1.5 base model?

[opaque spindle]

Got it. So do you mean that sdxl is not necessary. I will have a nice image if I use a nice 1.5 model? Thanks again.

[karmic crown]

use the xl model i just sent you

Hephaistos_NextGENXL

[opaque spindle]

ok ok

Trying now. My machine is not fast.... I will update you as soon as there is a progress.

[karmic crown]

take your time

i recommend installing civitai browser plus, it will put everyhting in a correct folder for you

https://github.com/BlafKing/sd-civitai-browser-plus

[sterile marsh]

also getting this error too: "Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument index in method wrapper_CUDA__index_select)"

[opaque spindle]

By the way, your guide is for non-M1 machine only? or apply for all other types of machine?

[karmic crown]

It was tested and has been working without a problem on those Mac models:

M1 MacBook Pro (2021) with 16GB
iMac 5K (2019) with 16GB of RAM and Raden Pro 580X 8 GB

you tell me 🙂

it works for 98% of macs

some strange configuration might not work as expected

like one guy had laptop with eGPU

and other had 2 GPUs

[opaque spindle]

Thank you very much!

[ornate elk]

whats the model you use?

[sterile marsh]

lora model right?

[ornate elk]

make sure you dont use a lora as normal checkpoint

models are always bigger than 1.98gb

loras go into the models/lora folder

[sterile marsh]

oh so anything below 1.98gb isn't a model to use in civitai?

[ornate elk]

anything below 1.98gb isnt a checkpoint/model to create images.
A lora is used together with a model to create specific images

it wont work on its own

if you try to load a lora as model you will get the error you got in the screenshot

[sterile marsh]

ohhh I see, I thought they were the same, I'll check it out in a bit

[opaque spindle]

I just tried Photon but am getting an error:
Torch not compiled with CUDA enabled

[karmic crown]

show me full error

[opaque spindle]

*** Arguments: ('task(93j59l1r1dvpho5)', 'a write cat is talking to a white robot', 'cartoon, painting, illustration, (worst quality, low quality, normal quality:2)', [], 20, 'DPM++ 2M Karras', 1, 1, 6, 512, 768, True, 0.45, 2, 'R-ESRGAN 4x+', 10, 0, 0, 'Use same checkpoint', 'Use same sampler', '', '', [], <gradio.routes.Request object at 0x19bc0f880>, 0, False, '', 0.8, -1, False, -1, 0, 0, 0, False, False, 'positive', 'comma', 0, False, False, '', 1, '', [], 0, '', [], 0, '', [], True, False, False, False, 0, False) {} Traceback (most recent call last): File "/Users/cm/stable-diffusion-webui/modules/call_queue.py", line 57, in f res = list(func(*args, **kwargs)) File "/Users/cm/stable-diffusion-webui/modules/call_queue.py", line 36, in f res = func(*args, **kwargs) File "/Users/cm/stable-diffusion-webui/modules/txt2img.py", line 55, in txt2img processed = processing.process_images(p) File "/Users/cm/stable-diffusion-webui/modules/processing.py", line 732, in process_images res = process_images_inner(p) File "/Users/cm/stable-diffusion-webui/modules/processing.py", line 856, in process_images_inner p.setup_conds() File "/Users/cm/stable-diffusion-webui/modules/processing.py", line 1309, in setup_conds super().setup_conds() File "/Users/cm/stable-diffusion-webui/modules/processing.py", line 469, in setup_conds

self.uc = self.get_conds_with_caching(prompt_parser.get_learned_conditioning, negative_prompts, total_steps, [self.cached_uc], self.extra_network_data) File "/Users/cm/stable-diffusion-webui/modules/processing.py", line 455, in get_conds_with_caching cache[1] = function(shared.sd_model, required_prompts, steps, hires_steps, shared.opts.use_old_scheduling) File "/Users/cm/stable-diffusion-webui/modules/prompt_parser.py", line 189, in get_learned_conditioning conds = model.get_learned_conditioning(texts) File "/Users/cm/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/models/diffusion/ddpm.py", line 665, in get_learned_conditioning c = self.cond_stage_model.encode(c) File "/Users/cm/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/modules/encoders/modules.py", line 135, in encode return self(text) File "/Users/cm/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1511, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "/Users/cm/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1520, in _call_impl return forward_call(*args, **kwargs) File "/Users/cm/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/modules/encoders/modules.py", line 124, in forward tokens = batch_encoding["input_ids"].to(self.device) File "/Users/cm/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/cuda/__init__.py", line 293, in _lazy_init raise AssertionError("Torch not compiled with CUDA enabled") AssertionError: Torch not compiled with CUDA enabled

[karmic crown]

have you changed anything in webui-user.sh?

[opaque spindle]

Not that I know of. I simply ran a1111-setup.sh

[karmic crown]

have you restarted a1111 after adding models?

[opaque spindle]

2 lines are changed:
export COMMANDLINE_ARGS="--skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half-vae --medvram-sdxl" +export TORCH_COMMAND="pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cpu"

Yes, from the browser, I click on Apply and restart UI button.

[karmic crown]

every time when you get an error related to memory, stop a1111 from consle (ctrl+c)

go to /Users/cm/stable-diffusion-webui

and run ./webui.sh

[opaque spindle]

ok

after stop a1111 and run ./webui .sh, I can't even select photon 😦
Loading weights [12762fe2e4] from /Users/cm/stable-diffusion-webui/models/Stable-diffusion/photon_v1.safetensors changing setting sd_model_checkpoint to photon_v1.safetensors [12762fe2e4]: SafetensorError Traceback (most recent call last): File "/Users/cm/stable-diffusion-webui/modules/options.py", line 140, in set option.onchange() File "/Users/cm/stable-diffusion-webui/modules/call_queue.py", line 13, in f res = func(*args, **kwargs) File "/Users/cm/stable-diffusion-webui/modules/initialize_util.py", line 170, in <lambda> shared.opts.onchange("sd_model_checkpoint", wrap_queued_call(lambda: sd_models.reload_model_weights()), call=False) File "/Users/cm/stable-diffusion-webui/modules/sd_models.py", line 741, in reload_model_weights state_dict = get_checkpoint_state_dict(checkpoint_info, timer) File "/Users/cm/stable-diffusion-webui/modules/sd_models.py", line 315, in get_checkpoint_state_dict res = read_state_dict(checkpoint_info.filename) File "/Users/cm/stable-diffusion-webui/modules/sd_models.py", line 291, in read_state_dict pl_sd = safetensors.torch.load_file(checkpoint_file, device=device) File "/Users/cm/stable-diffusion-webui/venv/lib/python3.10/site-packages/safetensors/torch.py", line 259, in load_file with safe_open(filename, framework="pt", device=device) as f: safetensors_rust.SafetensorError: Error while deserializing header: HeaderTooLarge

[karmic crown]

[12762fe2e4] from /Users/cm/stable-diffusion-webui/models/Stable-diffusion/photon_v1.safetensors

this looks good

what is file size?

[opaque spindle]

6325

[karmic crown]

2132625918 Nov 21 00:18 photon_v1.safetensors

it must be 2132625918

[opaque spindle]

hmm. let me try again.

[karmic crown]

more then 2GB

that's why i am using civitai browser plus

it takes care that all files are downloaded and put in right folders

[azure sable]

Hi, today I tried to launch A1111 and this ocurred...does anyone know why? I definitely have python installed, and it was working perfectly fine last night.

[karmic crown]

open cmd and type python --version

also error?

[azure sable]

just tried, it says the same thing in command prompt.

[karmic crown]

reinstall python and click "add to path"

or whatever exact wording is

[azure sable]

do i need a specific version of python in order for it to work with A1111? in the past I remember something about not being able to use the most recent version. Is it ok to use the most recent one?

[karmic crown]

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.

[azure sable]

ok, will try that now

[karmic crown]

be sure that python --version works before you try to launch webui

[azure sable]

works now, thank you

[karmic crown]

np

[azure sable]

not sure what happened as I had that exact version of python installed already and it was working last night.

[opaque spindle]

Thanks. I downloaded it again but this I got a different error message.
MPS backend out of memory (MPS allocated: 4.70 GB, other allocations: 2.04 GB, max allowed: 6.77 GB). Tried to allocate 768.00 MB on private pool. Use PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.0 to disable upper limit for memory allocations (may cause system failure).

[karmic crown]

ok

open webui-user.bat

[opaque spindle]

or sh?

[karmic crown]

sh

i am half dead

i had a long day at work

replace --medvram-sdxl with --medvram (just remove -sdxl)

and try again

[opaque spindle]

ok

[karmic crown]

if that doesnt work, you might need to use --lowvram

[fair oxide]

can someone tell me how do i use SDXL with A1111

[karmic crown]

i guess you installed and it installed base 1.5 model

right?

[fair oxide]

yes

[karmic crown]

and that works

[fair oxide]

it works

[karmic crown]

cool

https://civitai.com/models

filter

find what you like

[fair oxide]

there are so many

[karmic crown]

https://civitai.com/user/Afroman4peace

any of those is good

find and read about LCM

lcm use less steps to produce the same image quality, so it "faster"

actually the speed is same, but since you need less steps like 10 instead of 20 or 30 it is faster at the end

[fair oxide]

https://civitai.com/models/140737/albedobase-xl?modelVersionId=238308

so do i just download it?

[karmic crown]

i am testing 2.1.2 to see if it is possible to use it instead of 2.2.0 for Macs with --no-half-vae --medvram-sdxl

i see many fixes for MPS (metal)

and arrch64 (silicon)

[ornate elk]

What's your GPU ?

[fair oxide]

rtx 3070, 8gb vram

[karmic crown]

@ornate elk will give you the best parameters for your card

[fair oxide]

can u explain to me what that website is for

i am clueless

[karmic crown]

don't worry, we all were when we came here for the first time

you have something called base models, and custom models

base models are default ones, and custom are trained by other people all around the world

civitai has a enorumus number of user trained models, loras...

it also has some credit sistem (with some free credits) so you can use it to generate images in cloud)

[fair oxide]

oh

i see some tags on each model

[karmic crown]

there are also some other places like liblib, but i do not recommend it unless you speak chinese, or if you are crazy enough like me and has some even crazier chinese friends ( like me) 🙂

[fair oxide]

i assume they tell u what the model is good at?

[karmic crown]

yes

stars are rating

this is important

sdxl models cannot be mixed with 1.5 loras

or you will get a mess

[opaque spindle]

Both --medvram and --lowvram doesn't work. MPS allocated is different in each but both failed. 😦
`--medvram:
MPS backend out of memory (MPS allocated: 4.38 GB, other allocations: 1.52 GB, max allowed: 6.77 GB). Tried to allocate 1.50 GB on private pool. Use PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.0 to disable upper limit for memory allocations (may cause system failure).

--lowvram:
MPS backend out of memory (MPS allocated: 3.60 GB, other allocations: 2.76 GB, max allowed: 6.77 GB). Tried to allocate 1012.50 MB on private pool. Use PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.0 to disable upper limit for memory allocations (may cause system failure).`

[karmic crown]

i guess you are now thinking "wtf is lora"

[fair oxide]

what model would u recommend if i wanted to create something that looks like this

https://wallpapers.com/images/hd/halo-reach-hd-wallpaper-and-background-image-h8a37l87a0ir5117.jpg

[karmic crown]

you can try to add
export PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.0
to your webui-user.sh

[opaque spindle]

ok

[karmic crown]

it will most probably work, but that can produce other errors 😦

i do not have a mac with less with 8GB of vram

we have like 10 iMacs at work, but none of those has 4GB

[opaque spindle]

Which LCM model is kid-friendly and works with just 4GB VRAM?

[karmic crown]

i cant properly test params for your configuration 😦

[opaque spindle]

I heard Mac can't upgrade its hardware. Is it true? Thanks.

[karmic crown]

especially if it is a laptop

[opaque spindle]

😦

[karmic crown]

they told me i need to wait for 2 months for my m1 mac, but i got it in 2 days, since someone else who oreded customized version desided to give up

my M1 has 16GB of RAM and 512 GB SSD

problem was SSD, since base version was 256 GB

[ornate elk]

You need --xformers --medvram-sdxl --no-half-vae

Add these to your webui-user.bat at the line COMMANDLINE_ARGS=

[fair oxide]

ok

[opaque spindle]

I found this https://civitai.com/models/117996/sdxl-config-comfyui-fast-generation-4gb-vram-refiner. I understood 4GB but not the rest. Is this a good model?

[karmic crown]

you can add more memory to iMac without too much trouble, and that all
replacing a ssd requires total disassembling of iMac which requires total disassembling the screen ☠️

that is for comfy

[opaque spindle]

What about macbook?

[karmic crown]

try this command line
--skip-torch-cuda-test --upcast-sampling --no-half-vae --use-cpu interrogat

[opaque spindle]

ok

[karmic crown]

memory is soldered to the board!!! you can add more memory only on purchase. you can't do anything about gpu, except to buy a new Mac. you can replace ssd after disassembling, but it is possible that some other disk causes problems

nice? right? 🙂

[summer portal]

Is Stable Audio included in the $20 monthly sub or do you have to also pay the $12 on top of that? Sorry if this is not the right place for this question.

[bright hill]

I'm new to using stable diffusion

I'm able to get it working properly, but I'm not able to get Stable Diffusion to use any Lora Models

I've got one I've downloaded, but it won't show up in Automatic1111

I've tried for two days to get this to work

I'm currently using the v1-5 checkpoint, which is what the Lora was trained on (I think)

Here's the Lora in question

https://civitai.com/models/152694/cecilia-alcottinfinite-stratos

Whenever I go the the Lora tab, it just says "Nothing here. Add some content to the following directories:"

And I tripple checked this

The Lora is there

I went into the settings, extra networks page.

[bright hill]

"Always show all networks on the Lora page" is *checked

[karmic crown]

that is 1.5 lora

if your model is sdxl, it will not show nor work

1.5 model + 1.5 lora
sdxl model + sdxl lora

do not mix

[bright hill]

Then what model am I supposed to use?

[karmic crown]

any 1.5 model

[bright hill]

Checkpoint is basically the same as model, right?

[karmic crown]

yes

what model you are using?

[bright hill]

Is v1-5 different from SD 1.5?

v1-5-pruned.ckpt

[karmic crown]

clikc there

you can see which model was used

[bright hill]

Thank you!

[karmic crown]

you do not need to use it, unless you want to get exact image

[bright hill]

I think I'll be getting somewhere soon now

[karmic crown]

also

this icon on the right, will copy ALL of the data

you can paste that in positive prompt

[bright hill]

ty

[karmic crown]

and then click the arrow

and all setings will be applied

like steps, sampler...

[bright hill]

I found the checkpoint

https://civitai.com/models/4855/anythingelse-v4

There's 4 of them?

[karmic crown]

this arrow

[bright hill]

Pruned would be smaller and faster to work with

[karmic crown]

yes

that is pruned

always pick safetensor if awailable

[bright hill]

Can checkpoints be turned into .pt files by just changing their filename extension?

[slow swift]

hello. The fast-dreambooth collaboration hasn't worked since yesterday. Is this happening to anyone else? Does anyone have a solution https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast-DreamBooth.ipynb?authuser=7#scrollTo=QyvcqeiL65Tj

[karmic crown]

i have no idea

probably not

[fair oxide]

hey... i finished downloading a model from civitai

[bright hill]

I remember reading that some models had to be renamed

Thank you, viking

[fair oxide]

how do i actually use it with auto1111

[karmic crown]

you need to use magic called Windows Explorer and move it into a proper folder 🤣

[bright hill]

Put it in the appropriate subfolder in the models folder

[karmic crown]

models foder

and then Stable-diffusion

as written 🙂

[slow swift]

hello. The fast-dreambooth collaboration hasn't worked since yesterday. Is this happening to anyone else? Does anyone have a solution https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast-DreamBooth.ipynb?authuser=7#scrollTo=QyvcqeiL65Tj

[karmic crown]

just install https://github.com/BlafKing/sd-civitai-browser-plus
trust me

it will put everything in correct place for you

[bright hill]

ty, got it bookmarked

[karmic crown]

i need to update my mac guid

2.1.2 works fine with my params

no need for nightly anymore

but i need to do more tests, i havent try sdxl yet

but params work

i am gooing to sleep, i am half dead

[fair oxide]

oh no

its back again

i am getting this error once again

OutOfMemoryError: CUDA out of memory. Tried to allocate 508.00 MiB (GPU 0; 7.79 GiB total capacity; 7.10 GiB already allocated; 196.12 MiB free; 7.38 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

i added this to the command line args

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

still didnt work

[ornate elk]

make sure to always run the webui-user.bat and not the webui.bat

also what are your settings?

[fair oxide]

run it?

i just edited the file

even 512x512 says its too big

i went all the way down to an abysmal 256x256 and it works

[wet osprey]

xl is memory hungry

[fair oxide]

please tell me there is a way for me to create stuff at higher resolutions 😭

[wet osprey]

it also gives very poor outputs at 512x512 or lower. I don't recommend using anything below 768x768

[ornate elk]

Yes you need to run the webui-user.bat to use xformers, after you edited it.

Dont run anything else

[fair oxide]

i run it in the terminal?

[ornate elk]

You just double click the webui-user.bat

To launch SD

[fair oxide]

ok

[ornate elk]

If you have edited it correctly it should use xformers and you won't run out of vram

In the webui in browser at the bottom you can check if the xformers version 0.0.20 is displayed there

[fair oxide]

uhhh

this is what my file looks like

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --medvram-sdxl --no-half-vae

call webui.bat

[ornate elk]

Looks good

[fair oxide]

also im on linux

[ornate elk]

Wow

[fair oxide]

isnt bat a windows scrpt file

[ornate elk]

Then this won't work xD

For Linux you have to edit the webui-user.sh file

[fair oxide]

bruh

[ornate elk]

And there remove the # at the line COMMANDLINE_ARGS

Then add the args there

Then save and launch the webui-user.sh

[fair oxide]

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

export COMMANDLINE_ARGS="--xformers --medvram-sdxl --no-half-vae"

[ocean kettle]

Is it possible to run something like Automatic1111 from the command line without the Gradio app? I want to pass in all of the arguments on the command line instead of using the web gui.

[fair oxide]

ok i have run it

[ocean kettle]

I'm on Mac

[fair oxide]

not sure what happened though, i didnt get any message in the terminal

[ornate elk]

Yes there is an mode for no webui. But you then need to use the API of auto1111 to communicate to it

[ornate elk]

Does it start ?

[fair oxide]

nope

maybe i need to remove other "#"

[ornate elk]

Then try to launch the webui.sh

Try first with the webui.sh

[fair oxide]

oh

its installing xformers

[ocean kettle]

So basically start it like normal but then communicate using the API over http?

[ornate elk]

I think so

[ocean kettle]

Ok. I knew about that but was hoping there was another way. I don't really want to learn the API. But I guess that's just how it is.

[ornate elk]

Yea its only API or webui

[ornate elk]

Then it should work now

I'm off

[fair oxide]

it didnt at first

but now i think it does

i had to kind of restart everything

[bright hill]

I'm trying to get lora working, but they aren't showing up in automatic1111

I'm trying to get this lora to show up

https://civitai.com/models/152694/cecilia-alcottinfinite-stratos

I downloaded anythinelseV4_v45

I have that specific checkpoint loaded, but the Lora isn't showing up in auto11

The Lora is in the models/Lora subfolder

[ocean kettle]

The API was surprisingly easy to implement. I used this script. https://gist.github.com/w-e-w/0f37c04c18e14e4ee1482df5c4eb9f53

[buoyant lotus]

how can i allocate more memory on stable diffsuion WEBUI?

[graceful loom]

@ornate elk please help with this...

[ornate elk]

It takes as much as it needs?
Do you get any issues?

[buoyant lotus]

it crash

[ornate elk]

Do you have only 8gb of Ram?

[buoyant lotus]

16 installed ram

[ornate elk]

Okay and it freezes while installing?

[buoyant lotus]

no when i am generating images

[ornate elk]

Oh what's your GPU?

[buoyant lotus]

it just says not enough memory

gtx 1050 ti

[ornate elk]

Okay then you have to edit the webui-user.bat (right click, edit)
At the line COMMANDLINE_ARGS=
you add: --medvram --xformers --no-half-vae
Then save and relaunch

[buoyant lotus]

it says i dont have xformers proceeding without it

[ornate elk]

It will be installed if you add it

[buoyant lotus]

ok

[ornate elk]

Then you won't get out of memory errors

[buoyant lotus]

thx

[ornate elk]

No problem

[vocal burrow]

Sorry for the delay too,
I would suggest you mail the official stability team to get clear answers to those questions as this discord isn't official support for stablity product. Therefore it's not my position to discuss about VAT and pricing.

[ornate elk]

No clue about that one. But remove the space between " and masterpiece

[buoyant lotus]

return t.to(device, dtype if t.is_floating_point() or t.is_complex() else None, non_blocking)
RuntimeError: [enforce fail at ..\c10\core\impl\alloc_cpu.cpp:72] data. DefaultCPUAllocator: not enough memory: you tried to allocate 29491200 bytes.

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument mat1 in method wrapper_CUDA_addmm)

[vocal burrow]

Try to use --device-id x command line argument where x is your gpu's device id (check for it in task manager)
https://cdn.discordapp.com/attachments/1002602742667280404/1152699149892984883/image.png

[buoyant lotus]

its 0 for me too

[marble temple]

I had something similar happen right now, after it automatically downloaded 1.7.0. Clicking generate generates a very similar error. Currently seeing if deleting venv works..

It's not the venv...

*** Error executing callback app_started_callback for F:\Pictures\Generation\Stable-diffusion-automatic1111\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\api.py
    Traceback (most recent call last):
      File "F:\Pictures\Generation\Stable-diffusion-automatic1111\stable-diffusion-webui\modules\script_callbacks.py", line 139, in app_started_callback
        c.callback(demo, app)
      File "F:\Pictures\Generation\Stable-diffusion-automatic1111\stable-diffusion-webui\extensions\sd-webui-controlnet\scripts\api.py", line 52, in controlnet_api
        controlnet_input_images: List[str] = Body([], title='Controlnet Input Images'),
    NameError: name 'List' is not defined. Did you mean: 'list'?

I get this on simply starting the webui (a1111). I haven't done anything since yesterday, the only change was that it seems to have pulled the newest version (1.7.0?) from GIT. Didn't see any issues on the github like this either 😦

Putting an image into img2img also generates errors... Seems this 1.7.0 update just broke my complete installation somehow

[vocal burrow]

extensions\sd-webui-controlnet

did you update your extensions ?

[marble temple]

That seems to have solved that issue. Can't believe I forgot to check extension updates -__-. Now on to the next issue 🙂

[quartz pier]

Excuse me,does anyone know what keyword I should use for the boys character change to be girls

[karmic crown]

here we go again, it's time for a new update of a1111 install instructions for macs

[ornate elk]

Finally 1.7.0

[karmic crown]

i am more excited about pytorch 2.1.2 to be honest 🙂

i can remove nightly from my macs and instructions

since nightly builds are just not for everyone

[karmic crown]

New stable release of PyTorch 2.1.2 is released, and that is now a recommended version for Macs since it has some MPS(metal) and AArch64(silicon) fixes.

[brazen sun]

hello there, I have an error that says "Could not allocate tensor with 1073741824 bytes. There is not enough GPU video memory available!" I have nothing else taking my gpu and I have 8 gb of vram in my amd gpu

this happens whenever I try to generate more than one image

[ornate elk]

Hey, whats in your webui-user.bat?

[karmic crown]

https://github.com/viking1304/a1111-setup/releases/tag/0.0.2

[brazen sun]

@echo off

set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--no-half-vae --opt-sub-quad-attention --opt-split-attention-v1 --lowvram

call webui.bat
Now it works but goes slower, I am guessing since I changed from nothing to --medvram to --lowvram

[ornate elk]

looks good, if you run a x580 or a 5700xt i would suggest --lowvram

[brazen sun]

I have RX 6600

[karmic crown]

@ornate elk what is x580? I have 2019 iMac with Radeon Pro 580X 8 GB

[ornate elk]

i mean 580x xD

[ornate elk]

okay with that you can try --medvram if you stay at 512x512 or 768x768 but use --lowvram if you get the out of memory errors

[brazen sun]

It only happen with batches but I think what i am doing has an option to avoid the error, but its only the exact thing I am doing

[karmic crown]

my 580x 8GB works with --skip-torch-cuda-test --opt-sub-quad-attention --upcast-sampling --no-half-vae --medvram-sdxl, but I guess that only applies for Macs

[ornate elk]

yes a 580x can work with auto1111 win and linux too

but not a good experience on windows xD

[karmic crown]

i guess metal works better for 580x

[brazen sun]

nvm thought it had an option to use cpu when out of gpu in the extension but no, still having the issue at medram

[karmic crown]

there is an option for that on macs, that's why my 580x works with those settings i just mentioned

[brazen sun]

I am going to guess this is a stupid question but gif generation is out of the question with just 8 gb of vram I assume

[sweet kraken]

hello i have been using stable diffusion with google collab but it has stopped working but i dont know how to fix it

[ornate elk]

yep, you would need to use linux if you want a better performance with amd 8gb gpu

[brazen sun]

oh welp, I am more of a casual user so, I can only wait for further optimizations xd but thanks

[ornate elk]

yea np, what can help is to overclock the gpu with the AMD adrenalin software. soemone with a 6600xt tested it

but it wont fix the out of vram errors

[brazen sun]

only thing it can do is speed up the proccess of generating I am assuming?

[karmic crown]

Steps: 10, Sampler: Euler a, CFG scale: 7, Seed: 715737927, Size: 1024x1024, Model hash: aeaaff8e1f, Model: hephaistosNextgenxlLCM_v20, Version: v1.7.0

Time taken: 9 min. 51.5 sec.

NICE!!!

Time taken: 11 min. 57.2 sec. with 1.6.1 and other torch version

for 4 batches of 1 image

[cold spindle]

Hey guys. I've just installed stable-diffusion webui directml because i want to use it with my AMD GPU. It works fine but there's some issues. Inpaint don't works at all (just generates some blur in place of mask and it's not black but transparency white). And I can't crop image in img2img tab. What should I do?

[proper swift]

install linux and use rocm

[pastel sequoia]

i have amd gpu rx6600 and inpaint don't works what should i do?

[vocal burrow]

you need --no-half for inpainting to work on AMD cards

[cold spindle]

what is "vae"?

[vocal burrow]

vae is the part of StableDiffusion pipeline that do the translation in and out of the latent space.

translating "prompt" into math stuff and math stuff back into "pixel"

(very very short and rough explanation)

--no-half-vae means "don't use half for the vae computation"

you want --no-half for inpaiting to works as in "don't use half for anything"

it will considerably increase your vram consumption tho.

[cold spindle]

oh thank you a lot!

[bright hill]

I've been having extreme difficulty in getting any loras to show up in auto11

For a test run, I just want this lora to show up

https://civitai.com/models/152694/cecilia-alcottinfinite-stratos

It's in the appropriate folder

I also have the exact checkpoint used to creat that first image

I don't understand why the lora isn't showing up

The lora tab just says: Nothing here. Add some content to the following directories:

[ornate elk]

make sure to select an 1.5 model then restart the webui

[bright hill]

restarting, gimme a sec

.bat was just restarted

anythingelseV4 was the default checkpoint

Nothing in the Lora

No change

Lora still isn't showing up

I've been tearing my hair out

I don't know why any Loras aren't showing up

[karmic crown]

@ornate elk https://github.com/viking1304/a1111-setup/discussions/2

to avoid further confusion i moved guide to my github

[ornate elk]

okay, then make sure to whitelist the webui in any browser adblock.
then restart.
if that dont help, delete the venv folder and restart

[bright hill]

Whitelisting didn't work, deleting venv

I don't have a venv folder?

[brazen sun]

hello I am following this guide https://community.amd.com/t5/ai/how-to-running-optimized-automatic1111-stable-diffusion-webui-on/ba-p/625585 and i have this error AssertionError: No valid accelerator specified for target system., anyone else?

[karmic crown]

#🤝｜tech-support message

[brazen sun]

I am at tech support, am I not?

oh nvm

sorry

[karmic crown]

🙂

[bright hill]

lol

[brazen sun]

didnt know it was a link

[karmic crown]

now you know 🙂

he will tell you later if you need to change some params based on your card, vram and os

so please write them now

[brazen sun]

my card is AMD Radeon RX 6600, 8176 MB and windows 11

[bright hill]

Okay, so I just got the Civitai helper extension

Got it operational

Downloading it from the webui made it work

Thank you for your help, Viking

That said

Why

[brazen sun]

3 questions, is this version I asked about better for amd? second, what exactly does the civitai helper do? and why the why?

[karmic crown]

1. i posted a link to an extremely simple but working guide for AMD cards on windows
2. civitai helper is an extension that helps download models, loras and other stuff from civitai.com, but i do not like it. civitai browser plus is 10x better in my opinion and that is why i use that one
3. do not ask me 🙂

[karmic crown]

https://github.com/BlafKing/sd-civitai-browser-plus is much better in my opinion

[brazen sun]

ohh okay thank you, and the question about the first was because I have a working sd on my computer but was curious if that version you showed me was better than average for amd cards?

[karmic crown]

cs1o made that guide i posted, so he can answer better

i am mac guy 🙂

[brazen sun]

oh okay, thank you anyways

[bright hill]

@karmic crown I got that exact extension

[karmic crown]

The only Radeon I can help with is the one inside Mac 🙂

[bright hill]

Thank you

But after all these struggles, I have to know

What folder is the Lora in now?

[karmic crown]

let me tell you a secret
open windows explorer
find your stable diffuson folder
inside, you have a models folder
and that one has many folders inside
in almost each of them is a text file
that looks like this

🙈

[faint shuttle]

Hi, I'm kinda new here, but I'm facing a problem with SD, whenever I try to use the pixelization model, I get this error: Attempting to deserialize object on a CUDA device but torch.cuda.is_available() is False. If you are running on a CPU-only machine, please use torch.load with map_location=torch.device('cpu') to map your storages to the CPU.
I'm running an amd rx580 as gpu
I'm aware that cuda tech doesnt exist on the amd gpus, I'm struggling to find the torch folder to change it from gpu to cpu load.

[vocal burrow]

what software are you using ?

[bright hill]

I know all about that

I was putting the Loras in models/Lora

But I renamed that one, after I downloaded my test model with the helper extension

I'm trying to find out where the helper extension placed the lora

[vocal burrow]

what do you mean "I renamed that one"