The dataset is currently empty. Upload or create new data files. Then, you will be able to explore them in the Dataset Viewer.

  • One caption file per image
  • Filenames aligned one-to-one
  • No mixed identities or duplicate subjects

Dataset Creation

Curation Rationale

This dataset was curated to support:

  • Stable character identity during LoRA training
  • Reduced face and body drift
  • Photorealistic text-to-image generation

Images were selected to balance:

  • Facial angles (front-facing, ¾, side profile)
  • Upper-body and full-body coverage
  • Natural indoor and outdoor lighting
  • Casual, realistic environments

Annotations

Annotation Process

Captions were written manually using a simple, factual style focused exclusively on visible attributes.

No automated labeling, aesthetic scoring, or third-party annotation tools were used.

Who are the annotators?

The dataset was annotated by its creator.


Captioning Style

Captions describe only observable traits and conditions, avoiding artistic, emotional, or stylistic language.

Example caption:

Downloads last month
3