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… See the full description on the dataset page: https://huggingface.co/datasets/blameitondan/Rex.
