Movie Scene Map plots 15,565 real filming locations across 166 countries on one interactive map: the studios, castles, streets, and landscapes where films and TV series were actually shot. Click a point and you get a photograph plus the productions shot there. It covers 9,287 films and series, 2,153 video games, 407 anime, and 365 manga, and it is free with no account, no ads, and no paywall. The HN posting this week gave it a fresh wave of attention, but the part that interests me is not the map. It is how the data works.
The data model is the interesting part
There is no scraping and no generated data here. The backbone is filming location statements on Wikidata, joined to each place’s coordinates, its photograph on Wikimedia Commons, and its Wikipedia article. When a location is only supported by a sentence or a category in a Wikipedia article, rather than a structured Wikidata statement, the site labels it “per Wikipedia” and keeps the two evidence tiers separate. That separation is a small design decision that says a lot: weaker evidence is shown, but never blended with stronger evidence. Most crowdsourced map projects would have mixed them on day one and called it curation.
The honesty extends to coverage gaps. A production earns a page only once enough Wikipedia editions cover it, and an empty country means sparse Wikidata coverage, not that nothing was filmed there. The site maintains a public changelog of every rebuild, including mistakes, and a gaps page naming productions where a single Wikidata edit would add them to the map.
Games, anime, and manga get a different verb
A game is rendered and an anime is drawn, so none of them was filmed anywhere. The atlas handles this by placing them where their stories are set, using Wikidata’s narrative location property and Wikipedia’s setting categories, and every page about one says “set in” rather than “filmed in”. It is a subtle distinction, but it is the difference between a data project that respects its sources and one that quietly misrepresents them.
An open data pipeline you can build on
Two things make this useful beyond browsing. First, the entire atlas downloads as GeoJSON or CSV under CC0, so you can take the 15,565 locations and do whatever you like with them. Second, the site exposes a read-only MCP endpoint that answers the same questions live for AI assistants. A map with an MCP endpoint is a quiet signal of where things are heading: public datasets increasingly want to be queryable by agents, not just browsed by humans. If you are building anything that needs filming locations, screen tourism recommendations, or place-based media history, the data layer is already there and free.
Corrections flow upstream too. If a film you know is missing, the fix is to add a filming location statement to its Wikidata item with a source, and it appears at the next rebuild, both on this map and in every other project reading the same data. That is the open data loop working as intended: improve the source once, benefit everywhere.
The counts move with every rebuild, which is worth internalizing if you quote them. The 15,565 figure on the front page is a snapshot of Wikidata at the last rebuild, not a fixed number, and the changelog records each shift. For anyone who has managed a derived dataset, this is the correct behavior: the published number always has a provenance trail instead of quietly drifting.
The technical stack underneath
The map renders on OpenFreeMap with OpenMapTiles data and OpenStreetMap underneath, which means no map API billing and no vendor dependency for the base layer. The same maker runs sibling atlases with the same architecture: a World Train Map, a World Beach Map, The Castle Map, and a Sunshine Map. It is a nice demonstration that a one-person open data project can ship a production map site with zero recurring map costs if you pick the free infrastructure stack.
There is more than one way to use the site, too. Beyond pan-and-zoom browsing, it has a “Surprise” button that drops you at a random location, a FilmGuessr game that shows you a filming spot and asks you to name the production, and day-trip guides for filming locations near 400 cities. The day-trip angle is the one with real use: pointing the tool at your own city surfaces locations you have probably walked past without knowing what was shot there.
The location statistics are fun on their own. California leads with 1,747 productions, Los Angeles County with 1,436, England with 945. For New Zealand readers: 124 places on the map, which puts the country roughly level with China and ahead of Mexico. The most-covered productions, ranked by Wikipedia language editions, are the ones you would guess: Game of Thrones, Forrest Gump, Titanic, The Godfather.
Try it
If you want a starting point, search a favorite film and check whether its famous scenes were shot where you assumed. Half the fun is finding out they were not. And if you spot a gap, the five-step Wikidata fix is linked right from the missing page. It is a rare thing to see a consumer-facing site that treats data provenance as a feature rather than an afterthought, and it makes a decent template for anyone publishing derived datasets.