NAMESAKES

NAMESAKES is a benchmark for studying identity memorization in text-to-image (T2I) models — whether a model reproduces a real person's likeness from their name, or fabricates a plausible face. It accompanies the paper NAMESAKES: Probing Identity Memorization in Text-to-Image Models.

📄 Paper: [arXiv link] 🌐 Project page: https://namesakes-web.github.io

Dataset Summary

NAMESAKES contains ~1.2K public figures sourced from English Wikipedia, spanning a wide range of fame levels (from very famous to relatively obscure). Each entry includes:

• name — the public figure's name
• doppleganger — a perturbed, less-famous name variant, used as a controlled unmemorized baseline (constructed by minimal edits over the Wikipedia name pool)
• views / log_views — Wikipedia pageview count (from Jan 1, 2026) and its log, used as a proxy for fame
• image — the figure's freely-licensed Wikipedia infobox photo
• aligned_image — the same photo after face detection and alignment (256px)
• license / license_url / artist — per-image licensing and attribution metadata
• image_name — the source image filename

Intended Use

NAMESAKES is intended for research on privacy, memorization, and interpretability in generative models, including auditing T2I models for memorized identities, evaluating unlearning, and studying fame-dependent memorization. It is not intended for, and must not be used for, identifying, surveilling, impersonating, or generating content targeting individuals, or any application that could harm the people depicted.

Licensing

This is a mixed-license dataset:

The individual images retain their original Wikipedia licenses (CC BY-SA, CC BY, Public Domain, or GFDL), indicated per-image in the license, license_url, and artist fields. Any use of the images must comply with their respective original license terms, including attribution. The NAMESAKES compilation — the selection of names, perturbed name variants, fame stratification, metadata, and benchmark structure — is released under CC BY-NC-SA 4.0 (non-commercial, share-alike, attribution). We ask that users respect non-commercial and ethical-use restrictions on the compiled benchmark, consistent with its protective research intent.

Demographic Composition

NAMESAKES is sourced from English Wikipedia and therefore skews toward populations over-represented there — it is majority White and male, with limited representation of other categories. This is a known limitation; see the paper's Limitations and Ethical Considerations sections.

Ethical Considerations & Opt-Out

All individuals are public figures per Wikipedia's notability criteria, and all images were published under reuse-permitting licenses. We honor removal requests from individuals included in NAMESAKES. To request removal, please contact the corresponding author.

Citation

@article{alper2026namesakes,
  title         = {NAMESAKES: Probing Identity Memorization in Text-to-Image Models},
  author        = {Morris Alper and Vasudha Varadarajan and Moran Yanuka and Angelina Wang and Hadar Averbuch-Elor},
  year          = {2026},
  eprint        = {2606.20155},
  archivePrefix = {arXiv}
}