Simple illustration of multi-subject finetuning in SD and evaluation via trained models. See blog post
Installation:
- you'll need to be able to run SD itself, and enough GPU memory to perform SD finetuning.
- Some dependencies are potentially not captured by
requirements.txt
. For the rest,pip install -r requirements.txt
. - to build/install the underlying module,
pip install -e .
from the repository directory. - Update paths in cfg.py to your local env.
Actual experiments / data from the blog post are not posted here for privacy reasons, but I've included three notebooks demonstrating manual scoring, evaluator model training, and evaluator model based scoring.