appenz / sdshell

Simple script to reformat and rename for Stable Diffusion Dream Booth learning

sdshell

This is a simple tool to format images for for learning with the dreambooth trainer for Stable diffusion. Specifically it resizes the images to 512 x 512 by finding the centered largest square and scaling up or down as necessary. It also renames them to a consisten name of "NAME (seq).jpg" where NAME is the parameter passed via --iname and is a sequence number. This is useful for taking a directory full of images and preparing them for learning with a Dream Booth script.

Running

python3 sdshell.py --srcdir <dir> --traindir <dir> --iname <conceptname>

Command line options

Options:
  --srcdir TEXT          Source directory for the raw, unformatted images.
  --traindir TEXT        Storage directory for the resized training images.
  --trainwidth INTEGER   Optional training image width, needs to be same as height. Default 512.
  --trainheight INTEGER  Optional training image height. Default 512.
  --iname TEXT           Name of the new concept that the model should be
                         trained for.
  --help                 Show this message and exit.