Open data

The data behind 60505 films

Everything this archive knows, free to take and use for anything, with no key, no registration and no rate limit. Released into the public domain under CC0.

What is in it

Films	60505 playable, 84291 hours
Period	1884 to 2028
With a year	58864
With a director	17624
With cast	14710
With a synopsis	22772
Matched to Wikidata	19313
Evidenced public domain	5323

Files

films.json — every film with full metadata films.ndjson — one film per line, streamable channels.json — channel definitions and counts datapackage.json — Frictionless Data descriptor with a field-by-field schema catalog.json — the compact form the player itself loads

Per-channel slices are at /dataset/channel/<id>.json, for example noir.json.

Using it

# every film noir with English subtitles, as a table
curl -s https://movierewind.tv/dataset/films.ndjson \
  | jq -r 'select(.channels[]=="noir" and .english_subtitles)
           | [.year, .title, .runtime] | @tsv'

# what a record looks like
curl -s https://movierewind.tv/dataset/channel/silent.json | jq '.films[0]'

How the metadata was made

Every film is probed for availability and embeddability before it is published, and re-checked on a schedule, more often for material likely to be pulled. Feature length is required. Where a recorded runtime exists the print is compared against it, and a disagreement is noted on the record rather than hidden.

Credits, runtimes and languages come from Wikidata. Synopses come from Wikipedia or from the hosting item's own description, always with a link back. No field is generated. For an obscure 1930s film a language model would write something plausible and wrong, so where the archive does not know something, the field is empty.

If you use this

Take it and do what you like, no attribution required. It is compiled and kept current by one person, and the archive itself is free with no ads and no account, so if the data is useful to you, chipping in is the thing that keeps it being maintained.

Caveats worth knowing

Titles are normalised from upload titles, which are frequently bad, and corrected against Wikipedia where a match was found. Roughly half the catalogue has no Wikidata match, so its metadata is thinner. A YouTube-hosted film can disappear at any time; the last_verified date tells you how recently we confirmed it played. Rights basis is recorded where it is evidenced and absent where it is not, and absence is not a claim either way.