photo_prefer

# Photo Prefer Preference-selected image/caption pairs produced from the current personal aesthetic model. - Samples: 131211 - Aesthetic cutoff: 6.982475 - Score range: 6.982492 to 10.000000 - Caption model: gemma-4-31B-it - Prompt version: absolute_objective_en_v1 Each image in `data/` has a same-stem UTF-8 `.txt` caption. `manifest.jsonl` and `dataset.sqlite` contain the full score, caption, source mapping, and provenance. The downloadable dataset is distributed as a multipart, header-encrypted 7z archive. All numbered volumes must be present in one directory before extraction. Start extraction from the `.7z.001` file. The password is intentionally not stored inside this repository.