cog-sdxl-webui
cog-sdxl-webui copied to clipboard
Stable Diffusion XL training and inference as a cog model
Cog-SDXL-WEBUI Overview
The Cog-SDXL-WEBUI serves as a WEBUI for the implementation of the SDXL as a Cog model. You can find details about Cog's packaging of machine learning models as standard containers here.
Quickstart
Generating Images
ComfyUI
You can use of ComfyUI with the following image for the node configuration:
Look in the training_out folder. Put the lora.safetensors
file in the comfy lora folder and rename it to what you want. Then put the embeddings.safetensors
file in the embedings folder, rename it to what you like. Udr them as found in the comfy image above.
Training a quick Model
- Load the
./example_datasets/zeke.json
sample Configuration File through the WEBUI. - Review the configuration parameters.
- Click the
Start training
button when ready.
Installation Guide
Windows11 Native
Windows Pre-requirements
To install the necessary dependencies on a Windows system, follow these steps:
-
Install Python 3.10.
- During the installation process, ensure that you select the option to add Python to the 'PATH' environment variable.
-
Install Git.
-
Install the Visual Studio 2015, 2017, 2019, and 2022 redistributable.
-
Install cuda toolkit 11.8.0
Steps:
- Install cuda toolkit 11.8.0
- Run ./setup.bat
- Run the following commands in windows terminal:
git clone https://github.com/bmaltais/cog-sdxl-webui.git
cd cog-sdxl-webui
.\setup.bat
.\webui.bat
Windows WSL2
Steps:
- Follow this guide: Windows WSL2 Installation Guide
- Run the following commands in WSL 2:
git clone https://github.com/bmaltais/cog-sdxl-webui.git
cd cog-sdxl-webui
sudo apt update -y && sudo apt install -y python3-tk
./setup.sh
./webui.sh
Linux
Prerequisite
Make sure to use Ubuntu 22.04 or adapt the installation to your distro
Steps:
- Install Docker.
- Install 11.8.0 cuda drivers from: https://developer.nvidia.com/cuda-toolkit-archive
- Run the following commands:
git clone https://github.com/bmaltais/cog-sdxl-webui.git
cd cog-sdxl-webui
sudo apt update -y && sudo apt install -y python3-tk
./setup.sh
./webui.sh
Training With Your Dataset
- Place 5 to 8 high-resolution images in the root of a zip file.
- Specify the zip file in the
Image file
field. - Adjust settings as needed.
- Start the training process.
Update Notes
2023-08-13
- Convert to run without docker... much easier overall
2023-08-11
- Initial release of Minimum Viable Product for training WEBUI.
2023-08-17
- ROI problem is fixed.
- Now BLIP caption_prefix does not interfere with BLIP captioner.
2023-08-12
- Input types are inferred from input name extensions, or from the
input_images_filetype
argument - Preprocssing are now done with fp16, and if no mask is found, the model will use the whole image
2023-08-11
- Default to 768x768 resolution training
- Rank as argument now, default to 32
- Now uses Swin2SR
caidas/swin2SR-realworld-sr-x4-64-bsrgan-psnr
as default, and will upscale + downscale to 768x768