Quick Start

• First Time Setup (if you’re using Peter’s GPU no need to do this)
• Start the instance if its not running
• Click Connect, then connect to “JUPYTER 8000”
• Open Dreambooth-Stable-Diffusion folder and then double click dreamboth_runpod_Joepenna.ipynb
• Upload your training images into /training_images
  • Make sure your images are square.
    • https://www.birme.net/?target_width=512&target_height=512
      • use this to generate your model square images
• Under Training replace:
  • project_name
  • class_word
  • token
• The training takes around 60-90 minutes
• Once training is complete, run “Copy and name the checkpoint file”
• Remember to pause (STOP) the GPU instance

Videos

• https://www.youtube.com/watch?v=7m__xadX0z0
  • great video walking you through the process
  • runpod offers cheap servers
  • create a model with your faces with 30+ photos
• https://youtu.be/dfMLrytpfAU?t=1110
  • 18:39 best video, just follow along

Resources

• https://github.com/AUTOMATIC1111/stable-diffusion-webui
  • vince showed this is the right approach but build face model using dreambooth
• https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/4990
  • at the bottom is a script that works on macos (if its even necessary)
• dreambooth extension for automatic1111
  • install from extensions in webui
  • restart the process via terminal
  • dreambooth tab now available
• How to generate images that don’t suck

Examples