How to train Stable Diffusion on your photos on a remote GPU using RunPod and Dreambooth

This is the helper to accompany the video tutorial.

  1. !git clone https://github.com/JoePenna/Dreambooth-Stable-Diffusion/

  2. To resize: https://www.birme.net/ - to resize all files in one click (512 x 512)

  3. To convert JPGs to PNGs: https://www.iloveimg.com/jpg-to-image

  4. Easiest way: doesn’t work anymore because Google limited number of downloads from my GoogleDrive

  5. Harder way:

    https://huggingface.co/CompVis/stable-diffusion-v-1-4-original

    %pip install gdown

    !gdown https://drive.google.com/uc?id=

    TROUBLESHOOTING:

    1. Make sure you have 120 GB of memory on your RunPod

    2. Make sure it’s an even number of photographs you’re using (16, 18 or 20 photographs are good amount)

    3. If you use community GPU sometimes, you may lose GPU; in that case, stop and rent another one and start over

    4. CUDA out of memory: Run Terminal

      ps aux

      then look which processes have launcher and webgui and kill them

      kill 11 25 26

    5. Fix the path with your new model name

    HELPFUL WEBSITES:

    Which Celebrity Look Alike Application:

    https://www.starbyface.com/