README.md

Dreambooth Stable Diffusion Demo w/ a GUI ::: Based on JoePenna's Implementation 🤗

Setup/Configure/Regularization/Fine-Tune: with this DreamBooth stable diffusion demo.

demo-video.mp4

BEFORE continuing. Please Verify the following pre-requisites:

- the system is running a NVIDIA gpu
- the gpu has ~24 GBs for fine-tuning
- python is installed
- NVIDIA drivers are up to date
- the system has a compatible version of the CUDA toolkit installed
- the system has a compatible version of the cuDNN toolkit installed

Please make sure to install python: Python Official Website

1. And add it to PATH
2. Install the packages with:

   • ./setup.sh
     

3. Run the UI:

   • python webui.py
     

• GUI Features a Model & Dataset Configuration tab, and Model Training tab

Additional planned features:

• The Pruning Feature from the repo (12GB to 2GB)
• Resolve Gradio Components not updating Components for real-time logging feed
• Add realtime graphs for tracking loss & other metrics
• Separate Mode to run the Diffusers Dreambooth with UI (resolves most gpu memory constraints) <---> (In Next Update)
  • Option 1: fast-stable-diffusion
  • Option 2: ShivamShrirao-diffusers
• Insert support for jupyter notebook & google colab
  • Notebook finished & tested successfully
  • Adding https://github.com/localtunnel/localtunnel to view the GUI from browser if using a cloud computing instance
• Add model resume training from checkpoint
• Merge multiple datasets into one
• Custom Preset Configurations to Add/Delete/Load
• Custom Training Job Scheduler
• Progress Bar for image generation & training
• An Advanced Image Cropping Tool using YOLOv7
• Multi-GPU support:
  • Model Parallelism
  • Data Parallelism
• Utilizing Other Sampling Algorithms
• Windows Support