Updates! Stability AI Releases SDXL Version 1, Automatic 1111 Improvements

[Midnight-Phantom]

I think I should let people know about some important updates in case they missed the recent news. Firstly, Stability AI has released their latest SDXL version 1, and oh boy, it's good! If you use Stable Diffusion, I highly recommend trying it out; it's significantly better even in the base version. Additionally, the community will create their own versions based on that, so check Civit AI regularly for updates.

For those who are unfamiliar with SDXL, it comes in two packs, both with 6GB+ files. One is the base version, and the other is the refiner. To generate an image, use the base version in the 'Text to Image' tab and then refine it using the refiner version in the 'Image to Image' tab. If you are using Automatic 1111, note that and remember that the default resolution of the image is 1080x1080. If you are using Comfy UI, create a workflow based on that. Also, SDXL includes a VAE file, so don't forget to add it to your VAE folder along with the other files mentioned earlier.

Before proceeding, please make sure to update everything to the latest version. If you are unsure how to do that, go to your SD folder where all the SD files are located, click on your Windows path, type 'cmd,' and then hit enter. A CMD GUI will pop up, and then just type 'git pull' and press enter to update Automatic 1111 or even Comfy UI if you do the same thing in the folder for it.

(Note: This will only work if you have previously installed Git, Python, and added them to your path during the initial installation. My advice is to download Python version 10 from the Windows App Store for smooth performance.)

Running a large neural network file can be challenging for desktops, so I recommend editing the 'webui-user.bat' file with an editor like Notepad++ and make changes to the '[set COMMANDLINE_ARGS]' line. Add '[--xformers --opt-sdp-no-mem-attention --no-half-vae]' to it and then save the changes. This will prevent errors like CUDA out of memory and blank images.

Lastly, don't forget that you don't need the Lycoris extension anymore. In the new update of Automatic 1111, Lora and Lycoris can both be kept in the Lora folder, reducing issues caused by misplacing files in the future.

Finally, make sure to fully update your system, especially your graphics drivers, to the latest version for a smoother experience.


That's all in this thread I just posted the important bits I thought you would appreciate, if you want detailed things like what are the little improvements in UI or other things go to the GitHub or Hugging Face page of the publishers.

 

[ForestDweller]

I just learned about Tensor.Art and holy shit it's a lot better than NovelAi. Too bad you can't subscribe for more credits though.

Still got a lot of hand or leg errors and weapons are still often fucked. You have to use img2img and pray the pose is similar to the reference image. And even then, hand or leg errors still occur if you weaken the img2img influence since you want your character's appearance, not the reference.

Spoiler: Image

 

[Bartun]

This sounds interesting

 

[HellerFeed]

Not really a tech guy, but is there a video tutorial for this?
I am scared that I might mess up with cmd and just format my PC instead of updating my automatic1111 UI.

I also noticed my automatic1111 is an outdated UI. Although I have a 4 GB graphic card I can generate images no more than 700px x 700px.
If I were to upgrade to the newer model, would it function with my 4gb graphic card?

Should I just stick to the current automatic111 or upgrade?

 

[Midnight-Phantom]

I just learned about Tensor.Art and holy shit it's a lot better than NovelAi. Too bad you can't subscribe for more credits though.

Still got a lot of hand or leg errors and weapons are still often fucked. You have to use img2img and pray the pose is similar to the reference image. And even then, hand or leg errors still occur if you weaken the img2img influence since you want your character's appearance, not the reference.

Spoiler: Image

To be honest i heard it for the first time and checked it out quite well for a free online service .. .. if any of you wanna check it out go for it ..

This sounds interesting

yep yep, really bro I thought the new base version will be good and then the community will create new models like Dreamshaper based on v 1.5 on top of that, but oh it's awesome out of the bag already.

Not really a tech guy, but is there a video tutorial for this?
I am scared that I might mess up with cmd and just format my PC instead of updating my automatic1111 UI.

I also noticed my automatic1111 is an outdated UI. Although I have a 4 GB graphic card I can generate images no more than 700px x 700px.
If I were to upgrade to the newer model, would it function with my 4gb graphic card?

Should I just stick to the current automatic111 or upgrade?

Yep go to youtube and i am pretty sure you would find them .. or you even can find in-depth written instructions in google, check them out,

tho if you are already running it in your device I think you should have all the required apps for it already, so it should go smoothly ..

Go there I have recorded how to update automatic 1111 via the git-pull command in there ..

< https://discord.com/channels/1125463561104265288/1125464517938262016/1134550312120758363 >

and yep it will work quite well as in the previous update they optimized the UI, if you want you can read the change log in their Git Hub..

I would recommend to edit the WebUI-User.bat file and put in:

--xformers --opt-sdp-no-mem-attention --no-half-vae

and save it before so it will run more smoothly.

 

[Comiak]

I tried to update with gitpull but it tells me this:
Please commit your changes or stash them before you merge.
Aborting
Updating baf6946e..68f336bd

it doesn't update anything. is there a way around this?

 

[Midnight-Phantom]

I tried to update with gitpull but it tells me this:
Please commit your changes or stash them before you merge.
Aborting
Updating baf6946e..68f336bd

it doesn't update anything. is there a way around this?

it looks like you might have accidentally tweaked the main repository, which is probably why you're having trouble pulling the latest updates from the cloud. To fix it, simply open CMD or Powershell in the SD folder and type

git reset --hard

to reset the whole repository. After that, run

git pull

to fetch the updates

=============================================================================================
Or you can copy the models, Lora, and other files that you have to a separate folder. Then, delete the whole SD repository. Afterwards, check out the Python and git modules, as well as your graphic drivers for updates.

Alternatively, you could download the latest 64-bit version of Git from - GIT (select 64bit) and remember to add Git to the system's path during installation. Also, get Python 10 from the Windows Store; it includes all Python 10 components, so you won't need to download Python 10.6.0 separately, as it is outdated. By downloading from the Windows Store, it will automatically be added to the system's path.

You will find Nvidia graphic automatic driver installing software in: Drivers for Nvidia GPUs

Once you've completed these steps and ensured your device is ready, go to the folder where SD was before. Open CMD or Powershell, copy-paste the following command, and press Enter:

git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git

.

It will start downloading Automatic 1111 fresh. Wait for the download to finish. During the download, you might encounter a yellow-coloured error related to the PyTorch version, indicating that a new version is available. The error message will also provide instructions on how to install the new version. To copy the command, simply press the Ctrl key and click the mouse and select the command part then use Ctrl+C to copy it.

If the installation is not finished, you can wait or open a new CMD or Powershell window in the same folder, paste the command using Ctrl+V, and press Enter to update PyTorch quickly. After the Automatic 1111 download is complete, the CMD window will inform you. You can then close the window.

Check that all the required files are downloaded in the folder. Scroll down to the bottom and locate the file named "webui-user.bat." Edit it using Notepad or any other text editor, modifying only the "set COMMANDLINE_ARGS=" to

set COMMANDLINE_ARGS=--xformers --opt-sdp-no-mem-attention --no-half-vae

and save the file.

Create a link of the "webui-user.bat" file by pressing Shift+Ctrl and pasting it to your main Windows homepage. Then move back all the models, Lora, embeddings, and related files where they should be.

Close all windows, restart your device, and run the disk cleanup tool to improve system performance. To find the disk cleanup tool, search for "disk clean" in the Windows search bar, and follow the instructions it gives you, select all the boxes then click on clean up system files, and repeat the process, when all boxes are ticked click ok and it will clean all the temp or cache files.

After the cleanup, run the link you created. The UI will run and first It will download the xformers files and other necessary components. The UI will be up and running once the process is complete.

Finally, use the web UI as you did before. And don't forget to download the extensions, download them again to enhance your experience with the UI."


These are the once I use personally and you could also check them out.

- https://github.com/LEv145/--sd-webui-ar-plus

- https://github.com/butaixianran/Stable-Diffusion-Webui-Civitai-Helper

- https://github.com/Bing-su/adetailer

- https://github.com/adieyal/sd-dynamic-prompts

- https://github.com/continue-revolution/sd-webui-animatediff

- https://github.com/Mikubill/sd-webui-controlnet

- https://github.com/zanllp/sd-webui-infinite-image-browsing

- https://github.com/canisminor1990/sd-webui-lobe-theme

- https://github.com/Physton/sd-webui-prompt-all-in-one

- https://github.com/AUTOMATIC1111/stable-diffusion-webui-wildcards

- https://github.com/viyiviyi/stable-diffusion-webui-zoomimage

- https://github.com/ddPn08/maji-merger

 

[Comiak]

it looks like you might have accidentally tweaked the main repository, which is probably why you're having trouble pulling the latest updates from the cloud. To fix it, simply open CMD or Powershell in the SD folder and type

git reset --hard

to reset the whole repository. After that, run

git pull

to fetch the updates

=============================================================================================
Or you can copy the models, Lora, and other files that you have to a separate folder. Then, delete the whole SD repository. Afterwards, check out the Python and git modules, as well as your graphic drivers for updates.

Alternatively, you could download the latest 64-bit version of Git from - GIT (select 64bit) and remember to add Git to the system's path during installation. Also, get Python 10 from the Windows Store; it includes all Python 10 components, so you won't need to download Python 10.6.0 separately, as it is outdated. By downloading from the Windows Store, it will automatically be added to the system's path.

You will find Nvidia graphic automatic driver installing software in: Drivers for Nvidia GPUs

Once you've completed these steps and ensured your device is ready, go to the folder where SD was before. Open CMD or Powershell, copy-paste the following command, and press Enter:

git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git

.

It will start downloading Automatic 1111 fresh. Wait for the download to finish. During the download, you might encounter a yellow-coloured error related to the PyTorch version, indicating that a new version is available. The error message will also provide instructions on how to install the new version. To copy the command, simply press the Ctrl key and click the mouse and select the command part then use Ctrl+C to copy it.

If the installation is not finished, you can wait or open a new CMD or Powershell window in the same folder, paste the command using Ctrl+V, and press Enter to update PyTorch quickly. After the Automatic 1111 download is complete, the CMD window will inform you. You can then close the window.

Check that all the required files are downloaded in the folder. Scroll down to the bottom and locate the file named "webui-user.bat." Edit it using Notepad or any other text editor, modifying only the "set COMMANDLINE_ARGS=" to

set COMMANDLINE_ARGS=--xformers --opt-sdp-no-mem-attention --no-half-vae

and save the file.

Create a link of the "webui-user.bat" file by pressing Shift+Ctrl and pasting it to your main Windows homepage. Then move back all the models, Lora, embeddings, and related files where they should be.

Close all windows, restart your device, and run the disk cleanup tool to improve system performance. To find the disk cleanup tool, search for "disk clean" in the Windows search bar, and follow the instructions it gives you, select all the boxes then click on clean up system files, and repeat the process, when all boxes are ticked click ok and it will clean all the temp or cache files.

After the cleanup, run the link you created. The UI will run and first It will download the xformers files and other necessary components. The UI will be up and running once the process is complete.

Finally, use the web UI as you did before. And don't forget to download the extensions, download them again to enhance your experience with the UI."


These are the once I use personally and you could also check them out.

- https://github.com/LEv145/--sd-webui-ar-plus

- https://github.com/butaixianran/Stable-Diffusion-Webui-Civitai-Helper

- https://github.com/Bing-su/adetailer

- https://github.com/adieyal/sd-dynamic-prompts

- https://github.com/continue-revolution/sd-webui-animatediff

- https://github.com/Mikubill/sd-webui-controlnet

- https://github.com/zanllp/sd-webui-infinite-image-browsing

- https://github.com/canisminor1990/sd-webui-lobe-theme

- https://github.com/Physton/sd-webui-prompt-all-in-one

- https://github.com/AUTOMATIC1111/stable-diffusion-webui-wildcards

- https://github.com/viyiviyi/stable-diffusion-webui-zoomimage

- https://github.com/ddPn08/maji-merger

Thanks for the help, was able to update with the simpler method. I gave sdxl a try, and it's just not working right. I'm not sure what's wrong but all my other models work just fine.

 

[Midnight-Phantom]

Thanks for the help, was able to update with the simpler method. I gave sdxl a try, and it's just not working right. I'm not sure what's wrong but all my other models work just fine.


View attachment 20785

SDXL is a whole new neural architecture, so it operates with entirely different parameters. The user interface needs significant upgrading and optimization before it can perform like version 1.5 for all the people. If you encounter any issues, try generating images without any additional elements like lora, ensuring they are at the full 1080 resolution. Choose the SDXL VAE option and avoid upscaling altogether. First, get acquainted with the model's basic usage. Also, avoid overcomplicating the prompt, instead of using (girl:0.1), simply use (girl).