April 7, 2026
Pin the Past, Stir the Comments
AI helps add 10k more photos to OldNYC
AI puts 10k more vintage NYC pics on the map — fans cheer, skeptics squint
TLDR: OldNYC added 10,000 more historic photos and uses AI and open maps to pin down locations, bringing the archive to 49,000. Commenters cheer the smarter search but warn against AI-altered images, flag a few wrong pins, and crack jokes about the map stopping at New Jersey.
History nerds had a mini panic—then a sigh of relief. When OldNYC said “AI added 10,000 more photos,” some readers braced for fake, touched‑up images. “oh god no, the photos are probably all hallucinated…” But the twist: the pictures aren’t altered at all. The AI is just doing the boring-but-brilliant work of reading old captions and pinning photos to the right spot on the map.
The site now shows 49,000 vintage shots, thanks to OpenAI’s GPT‑4o parsing tricky descriptions and open maps data stepping in where long-gone streets used to be. The dev says GPT helped place about 6,000 images, pushing location coverage to 87%, with 96% of pins landing in the right place. There’s also cleaner text: AI-powered OCR (turning scanned captions into text) makes typewritten notes easier to read.
Cue the drama: preservation purists shouted guardrails—“As long as they’re not GenAI altered photos, I’m cool”—while map nerds grabbed magnifying glasses. One local found a mispin and posted receipts. And, of course, the city banter: a reader cackled at the dot parade that “stops at Jersey City.” In short: big win for finding the past, with internet doing what it does best—cheering, nitpicking, and joking.
Key Points
- •OldNYC expanded from ~39,000 photos (2016) to 49,000, adding 10,000 historic images.
- •Geolocation improved by using OpenAI’s GPT-4o to extract locations from descriptions, enabling ~6,000 additional photos to be mapped.
- •The project replaced Google Maps geocoding with OpenStreetMap and NYPL historical street datasets for better historical accuracy.
- •OldNYC now locates roughly 87% of photos with usable location data, with ~96% of mapped images in the correct location.
- •OCR is highlighted as dramatically improved versus a 2015 Ocropus pipeline; detailed methods are not provided in the excerpt.