Synthesize Privacy-Preserving High-Resolution Images via Private Textual Intermediaries
Haoxiang Wang, Zinan Lin, Da Yu, Huishuai Zhang · 2025
Read Synthesize Privacy-Preserving High-Resolution Images via Private Textual Intermediaries by Haoxiang Wang, Zinan Lin, Da Yu, Huishuai Zhang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-07555.txt.
Description
Haoxiang Wang, Zinan Lin, Da Yu, and Huishuai Zhang generate private synthetic images by passing through text instead of wrestling with pixels, in a 2025 paper. Sharing sensitive photos is risky, and mathematically private synthetic copies, called differentially private, often come out blurry or structurally wrong at high resolution. Their method, SPTI, first has an image-to-text model describe each private photo in words.
A private text generator then rewrites those descriptions under a formal privacy budget, and a text-to-image model paints fresh pictures from the cleaned text. Nothing needs training; all parts are off-the-shelf models used as-is. On two standard photo sets, the synthetic images scored far closer to real ones than earlier private methods at the same privacy level. The authors offer the detour through language as a cheap, practical route to shareable private image sets.
Catalog record
| CREATOR | Haoxiang Wang, Zinan Lin, Da Yu, Huishuai Zhang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 40 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-07555