Geotag.World: Reading, Editing and Stripping Photo Location Data

Overview

Every photo can carry GPS and EXIF data that shows where it was taken. People share photos without knowing it is there.

The Challenge

Geotag.World is a CodesSavvy product, not a client engagement. It reads a photo's GPS and EXIF data, shows the place on a map, and lets you edit or strip the data. The tools that find the location, view EXIF, check privacy, remove GPS, strip EXIF and strip C2PA (AI content credentials) are free. Adding or editing GPS costs credits. The hard part was not the first version. It was keeping a multi-language, paid, browser-heavy product correct as it grew.

Our Solution

The first commit was on 22 Sep 2025. Supabase credits arrived in Feb 2026. Atomic deduction and tighter row-level security followed on 1 May 2026, and a MapLibre map on 14 May. Payments moved from Lemon Squeezy to Polar in May 2026. HEIC support followed in Jul 2026, decoded in the browser with libheif (heic2any, with heic-to on WebAssembly as the fallback). The hero A/B test ran from 18 Jul to 11 Aug 2026. Five languages (English, Spanish, French, Indonesian, Chinese) came on 20 Jul, then bulk mode on 27 Jul. Five problems shaped the build. (1) A language regression: the blog data had no locale field, so 37 English posts were served under four other language paths, with about 89% of the English body text appearing word for word on the Spanish page (measured on the Spanish blog pages in the 5 Sep 2026 audit). A Search Console URL audit of 245 sitemap URLs found 125 of 196 non-English URLs never crawled. We made the blog English-only, added noindex on the other locales, removed them from their sitemaps, moved hreflang into the head, added a crawlable footer language row, cut the sitemap from 245 to 93 URLs (the count at the time of the fix on 5 Sep 2026) and added end-to-end guards. Fixed on 5 Sep 2026. (2) C2PA credentials live in the JPEG APP11 segment, and the EXIF library only removed APP1. We wrote a byte-level JPEG segment stripper that drops APP1, APP11 and APP13 without re-encoding, and fixed the gap in two code paths. (3) A paid export could charge without delivering a file. We reworked it to deliver first, then charge, and added an end-to-end suite that runs against real Supabase accounts. (4) Firefox and Safari lost the photo after checkout. Photos are now stored as raw bytes, not Blobs, in IndexedDB, and we added Firefox and WebKit Playwright projects, plus a test that checks the stored bytes, because the first regression test passed on the broken build. (5) We do not redirect by browser language, because automatic redirects by browser language are discouraged in Google's guidance on localized versions. A suggestion banner is used instead.

Results

Tech Stack