[SUPPORT] - stable-diffusion Advanced

[Holaf]

Nope, sorry.
But you can look at /entry.sh
It's this script that does everything (It can be a bit messy, I'm not used to do this kind of thing)

For AMD GPU I didn't do anything mainly because I don't have one on my servers, so it will be hard for me to test.

Edited September 23, 2023 by Holaf

[hans-o]

Have been using automatic1111 just fine and wanted to give fooocus a shot after it's v2 update, it's booting up with the following error:

 

Quote

RuntimeError: Detected that PyTorch and torchvision were compiled with different CUDA versions.
PyTorch has CUDA Version=11.7 and torchvision has CUDA Version=11.8.
Please reinstall the torchvision that matches your PyTorch install.

 

Is there an easy way to manually set the versions? thanks

[Holaf]

Sorry I was busy lately, I will fix this asap

edit :
simple fix : )
You stop the container, go in the Fooocus folder and remove the venv directory. When you'll lrestart the container it will reinstall all the correct dependencies 👍

Edited October 7, 2023 by Holaf

[ShadowVlican]

On 9/12/2023 at 10:20 AM, Avsynthe said:

Hey all,

 

I'm having an issue where stable diffusion never releases RAM on AUTOMATIC1111. The more I generate, the higher it goes.

The server went down today and I couldn't figure out why the last snapshot of the system showed 99% memory used of 64GB. I realised SD is just compounding away. This happens no matter what model I use with VAE models increasing it quicker for obvious reasons. Switching models makes no difference, it just continues on.

 

I've had to limit SD to 20GB RAM and so it'll eventually crash when it hits. Is anyone else experiencing this?

 

i turned OFF -medvram and now system RAM doesn't leak anymore!

i read this thread to figure it out: https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/6850

[JapanFreak]

Hi noob here

i tried to install this added the gpuid to the container but i still failed and i have no idea why and what to do

maybe i have the wrong nvidia driver should i install the open source one or other version ?

Edited October 19, 2023 by JapanFreak

[Holaf]

I have the same driver, so it should work 🤔
Can you try to uninstall/reinstall the plugin ?
can you also check if the driver is running with the command "nvidia-smi" ?

[Jeryd]

Try taking out the specific GPU ID and use use "all" (no quotes).

[JapanFreak]

51 minutes ago, ubermetroid said:

Try taking out the specific GPU ID and use use "all" (no quotes).

Error response from daemon: Unknown runtime specified nvidia.
See 'docker run --help'.

The command failed.

thats what i get with all

 

2 hours ago, Holaf said:

I have the same driver, so it should work 🤔
Can you try to uninstall/reinstall the plugin ?
can you also check if the driver is running with the command "nvidia-smi" ?

thats what i get when i runn the command it says off so i am not sure it works

as for reisntalling i tried like 6 times every time i get Error response from daemon: Unknown runtime specified nvidia.
See 'docker run --help'.

thanks for your replays

Edited October 19, 2023 by JapanFreak

[Holaf]

I believe that the runtime comes with the nvidia driver, so for me it's a bug with the nvidia plugin or your server, but not with my container 🤷‍♂️

[ados]

Installed and was working ok with Easy Diffusion but really wanted to get Lama Cleaner for constant image fixing I do.
It seemed to install but will not stay running, the docker just stops before the web UI will load.

The last line in log before closing is: ModuleNotFoundError: No module named 'torch._utils'

[Holaf]

Thanks for the info, I'll take a look at this tomorrow.

[Holaf]

@ados
I tried a fresh installation of lama cleaner alone, and a fresh installation of lama cleaner with easy-diffusion already installed, and it worked in both cases 🤔

Can you stop your container, delete both folders "cache" and "50-lama-cleaner" and then relaunch your container ?
I think it's your best chance to get it working 🤞

[danbru1989]

Any ideas why ControlNet is disabled when using Invoke? I noticed that my models/sd-1/controlnet folder is empty.

[Holaf]

I tried to do a fresh install and invokeAI downloaded by itself controlnet models (it took some time after I could connect to the webUI)
Perhaps you could try to remove or rename the actual invokeAI folder and try to install it again

[danbru1989]

I renamed the 03-invakeai folder and started up the container. It re-downloaded Invoke and ControlNet is accessible now. Thank you!

 

Maybe there was a better way to do it that would have retained all my assets and settings, but this worked and I didn't have anything important there.

[Holaf]

Honestly, the way InvokeAI handles models/settings is still a bit of a mystery for me 😅

[blanketred]

How do you install it as CPU only? I don't have a GPU in my unraid server. Thank you!

[ados]

On 10/31/2023 at 8:33 PM, Holaf said:

@ados
I tried a fresh installation of lama cleaner alone, and a fresh installation of lama cleaner with easy-diffusion already installed, and it worked in both cases 🤔

Can you stop your container, delete both folders "cache" and "50-lama-cleaner" and then relaunch your container ?
I think it's your best chance to get it working 🤞

Before seeing this I tried switching to SD.Next and also had issues with it not loading an docker crashing.

This was with a fresh docker container too.

I applied the suggested fix of deleting the cache and SD.Next folders in container appdata and has not got it running.

The only model I have it working with is easy diffusion. 😢

 

Should mention the SD.Next crash is caused by: 

-bash: line 1: 73 Illegal instruction bash webui.sh --listen --port=9000 --insecure --medvram --allow-code

Edited November 5, 2023 by ados

[Holaf]

strange 🤔

I'm working on a new version, I hope to push it on dockerhub this week-end.
It's a bit long to do because I'm modifying all scripts and I have to test multiple installations from scratch 😣
I hope it will work for you with this next update

[FoxxMD]

Hi @Holaf and thanks for the amazing container.

 

I'm using SD.Next with almost everything working correctly. However using Models -> CivitAI -> download for embeddings logs an error like this:

 

2023-11-14T17:42:13.998575208Z 17:42:13-994976 ERROR    CivitAI download: name=easynegative.pt               
2023-11-14T17:42:13.998614237Z                          url=https://civitai.com/api/download/models/9536path= 
2023-11-14T17:42:13.998627355Z                          temp=models/Stable-diffusion/22863553.tmp size=0Mb   
2023-11-14T17:42:13.998638689Z                          removed invalid download: bytes=25323                 
2023-11-14T17:42:14.001201822Z 
2023-11-14T17:42:14.002561798Z Exception in thread Thread-68 (download_civit_model_thread):
2023-11-14T17:42:14.002596562Z Traceback (most recent call last):
2023-11-14T17:42:14.002618000Z   File "/usr/lib/python3.10/threading.py", line 1016, in _bootstrap_inner
2023-11-14T17:42:14.002935406Z     self.run()
2023-11-14T17:42:14.002967860Z   File "/usr/lib/python3.10/threading.py", line 953, in run
2023-11-14T17:42:14.003280728Z     self._target(*self._args, **self._kwargs)
2023-11-14T17:42:14.003312062Z   File "/opt/stable-diffusion/04-SD-Next/webui/modules/modelloader.py", line 174, in download_civit_model_thread
2023-11-14T17:42:14.004375606Z     os.rename(temp_file, model_file)
2023-11-14T17:42:14.004402584Z FileNotFoundError: [Errno 2] No such file or directory: 'models/Stable-diffusion/22863553.tmp' -> 'models/Stable-diffusion/easynegative.pt'

 

I have no issues downloading from huggingface or downloading regular sd 1.5/lora models from civitai. Additionally, after manually moving files into appdata/stable-diffusion/models/embeddings and restarting the container it still does not move them into the correct folder in 04-SD-Next.

 

2023-11-14T17:52:43.096595854Z removing folder /opt/stable-diffusion/04-SD-Next/webui/models/VAE and create symlink
2023-11-14T17:52:43.109880467Z moving folder /opt/stable-diffusion/04-SD-Next/webui/embeddings to /opt/stable-diffusion/models/embeddings
2023-11-14T17:52:43.118850832Z sending incremental file list
2023-11-14T17:52:43.163582694Z 
2023-11-14T17:52:43.163651074Z sent 155 bytes  received 12 bytes  334.00 bytes/sec
2023-11-14T17:52:43.163666876Z total size is 100,167  speedup is 599.80

 

I have to manually move files

 

appdata/stable-diffusion/models/embeddings -> appdata/04-SD-Next/webui/models/embeddings

 

after which point they are recognized by SD.Next

[FoxxMD]

On 10/10/2023 at 8:23 PM, ShadowVlican said:

 

i turned OFF -medvram and now system RAM doesn't leak anymore!

i read this thread to figure it out: https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/6850

 

This didn't work for me but the comments in that issue were the clue I needed. For anyone else running into memory leaks when using SD.Next:

 

1. Exec into the container and install updated malloc library

 

apt update
apt -y install libgoogle-perftools-dev

 

2. Create the file /mnt/user/appdata/stable-diffusion/04-SD-Next/webui/webui-user.sh and add this contents:

 

export LD_PRELOAD=libtcmalloc.so
echo "libtcmalloc loaded"

 

3. Make the file executable

Open unraid web terminal and run

chmod +x /mnt/user/appdata/stable-diffusion/04-SD-Next/webui/webui-user.sh

 

EDIT: @Holaf can you please add libgoogle-perftools-dev as a dependency in the image?

 

OR add a hook into entry.sh so end-users can run a script as root on container start/creation so these kinds of modifications can be made without needing to modify the docker image.

Edited November 16, 2023 by FoxxMD

[echofire]

On 11/15/2023 at 1:29 PM, FoxxMD said:

 

This didn't work for me but the comments in that issue were the clue I needed. For anyone else running into memory leaks when using SD.Next:

 

1. Exec into the container and install updated malloc library

 

apt update
apt -y install libgoogle-perftools-dev

 

2. Create the file /mnt/user/appdata/stable-diffusion/04-SD-Next/webui/webui-user.sh and add this contents:

 

export LD_PRELOAD=libtcmalloc.so
echo "libtcmalloc loaded"

 

3. Make the file executable

Open unraid web terminal and run

chmod +x /mnt/user/appdata/stable-diffusion/04-SD-Next/webui/webui-user.sh

 

EDIT: @Holaf can you please add libgoogle-perftools-dev as a dependency in the image?

 

OR add a hook into entry.sh so end-users can run a script as root on container start/creation so these kinds of modifications can be made without needing to modify the docker image.

THANK YOU!  This resolved my issue as well with my 02-sd-webui installation.

[Holaf]

Just pushed a big update, I tested as much as I could, but I'm sure I let some bugs go through ...
Don't hesitate to report here

[Joly0]

Hey @Holaf would it be possible, to get the code for the dockerfile and scripts that you used to build this container on github, gitlab or anywhere so people can look at it and maybe improve it? Also, is it possible to get a change log. You said you pushed a big update, but i would like to know what changed? Having a changelog and having the source code would help to better understand what changed

[Jeryd]

What is the big update?