This repository contains glue code for automating captioning and keywording for PhotoPrism/local files. It's based on the Transformers library and requires a HuggingFace account/token to ensure models can be fetched automatically.
The PhotoPrism client implementation has been vendored separately.
- Blip (sentence-based captioning)
- Blip in VQA mode (better for keywording)
- Blip2
- All inference can run on the CPU; a generous amount of RAM is recommended.
- GPU/CUDA acceleration is automatically enabled.
Ensure you have a huggingface account and token! It's also recommended to use direnv or dotenv to automatically apply .env files.
python3 -m virtualenv .venv && source .venv/bin/activate && pip install -r requirements.txt
cp .env.example .env && vim .env && source .env
See example config for explanation on each envar.
source ./.venv/bin/activate && source .env
python labelizer.py local <path to file / folder / url> -o output.csv
# E.G.,
# python labelizer.py local $(find /some/photos/archive -name "*jpg" -print | head -20 | shuf | xargs -t)
Make sure you've entered your PhotoPrism credentials in .env;
source ./.venv/bin/activate && source .env
python labelizer.py photoprism # pass -r to skip updating PP photo details