# Hugging Face portrait provenance

- Dataset: [TLeonidas/this-person-does-not-exist](https://huggingface.co/datasets/TLeonidas/this-person-does-not-exist)
- Pinned revision: `115c13c6c9cb623df34f650f66ffbbd18cbb0503`
- Archive: [`thisPersonDoesNotExist.zip`](https://huggingface.co/datasets/TLeonidas/this-person-does-not-exist/resolve/115c13c6c9cb623df34f650f66ffbbd18cbb0503/thisPersonDoesNotExist.zip)
- Archive bytes: 295,669,740
- Archive SHA-256: `8889bd8a069788734fcdd97fc6d817b21064a013ad7768659a02375b03981964`
- Dataset card: [README at the pinned revision](https://huggingface.co/datasets/TLeonidas/this-person-does-not-exist/resolve/115c13c6c9cb623df34f650f66ffbbd18cbb0503/README.md)
- Uploader license: [Apache License 2.0](./HF-LICENSE.txt), copied from the pinned repository.
- Upstream named by the uploader: [Kaggle dataset](https://www.kaggle.com/datasets/pablobedolla/this-person-does-not-exist-data). Its original redistribution permission has not been verified; the upstream listing reports an unknown license. The uploader's license label is not independent confirmation of rights to every image.

The uploader describes these as AI-generated profile pictures. Persona did not generate them and does not independently certify that a face cannot resemble a real person. The dataset has no verified age, gender or expression labels; those fields remain `unknown`.

## Import results

- Archive file members: 8,892
- Imported portraits: 8,891
- Native dimensions: 8,891 at 128x128 pixels
- Formats: {"PNG": 8891}
- Invalid images skipped: 0
- Exact decoded-pixel duplicates skipped: 1
- Non-image members ignored: 0
- Existing original portraits preserved: 12

Every imported image was fully decoded with Pillow. Original encoded bytes and native dimensions were preserved; images were not enlarged, cropped, retouched or recompressed. Files use stable names derived from their SHA-256. The importer detects exact decoded-pixel duplicates, not similar-looking faces. Full details are in [HF-IMPORT.json](./HF-IMPORT.json).

Reproduce with `python3 scripts/import-dataset.py` (Python 3 and Pillow required). Downloads are cached in `.cache/`; re-running replaces the Hugging Face manifest entries deterministically while preserving the original portraits. The ZIP is checksum-verified and checked for unsafe paths, symlinks, duplicate entries, excessive sizes and suspicious compression before any payload is imported.