How to find subtitles that match your copy of the video

Why the wrong file is so easy to get

Subtitle files are shared by name, and the name is the name of a film, not of a copy of a film. The same title exists as a Blu-ray, a web stream, a TV broadcast, a DVD, in a theatrical and an extended edition, in a 24 and a 25 frames-per-second transfer. All of them are “the film”. Each has its own timeline, and a subtitle file made against one of them is wrong for the others by a constant, by a stretch, or by a jump — the three repairs the rest of this site covers. The cheapest repair is to not need one.

Read the release name

The file name of a video from anywhere but a shop is usually a recipe written in a fixed order. Take a typical one apart:

Film.Title.2019.1080p.BluRay.x264-GROUP

• Title and year identify the film. Obvious, but the year matters when a title has been remade.
• Resolution and codec (1080p, x264, x265, HEVC) describe the picture and have no effect on timing at all. People filter on these first because they are the biggest words, and they are the only ones that do not matter.
• Source (BluRay, WEB-DL, WEBRip, HDTV, DVDRip) is the master the copy was made from, and it decides the cut and very often the speed. A Blu-ray and a web stream of the same film differ by a studio card, a recap, an extra trailer; a PAL DVD or European broadcast runs about four percent faster than the film was shot. This is the word to match.
• Edition words (Extended, Directors.Cut, Unrated, Remastered, PROPER, REPACK) mean the timeline differs from the plain release, sometimes by whole scenes. If your file has one, the subtitle file must have the same one.
• Group (-GROUP at the end) is who made the copy. Two files from the same group and source are the same cut in practice, so a subtitle file that names your group is as close to a guarantee as the name can give.

Order of matching, then: source and edition first, group second, resolution and codec never. Subtitle sites usually list the release each file was made for; read that field rather than the download count. A file labelled for 720p.BluRay.x264-GROUP fits your 1080p.BluRay.x264-GROUP exactly; a file labelled 1080p.WEB-DL probably does not, however many stars it has.

Better than the name: search by the file

Names lie, and renamed files lie most. The stronger method uses the video file itself. Your player or subtitle tool computes a checksum from the file’s size and bytes at its start and end, and asks a subtitle database for files that other people have already paired with that exact checksum. A hit means someone had the same file you have and found subtitles that fit it. VLC has a subtitle-search extension that does this from the menu bar, and the desktop tools from the big subtitle databases do it too; look for “search by hash” or “by file” rather than “by name”.

A hash match is the best evidence you can get without watching. It is not a proof: people occasionally upload a file that was nearly right and leave it. So do the one check below before you settle in.

Check the runtime before you press play

Whatever found you the file, one comparison tells you in ten seconds whether it was timed for your cut. Load it into the studio below. Beside the format and cue count it shows Runs to: the end time of the last cue. Put that next to the duration your player reports for the video.

• Within a minute or so of the video’s length. Same cut. The last line of dialogue sits near the end of any film, so the numbers agree when the timelines agree. If the subtitles are still off when you play it, it is a plain constant delay from wherever the two copies start, and one number fixes it.
• About four percent short or long — roughly five minutes on a two-hour film. The subtitles were timed at the other framerate. Do not shift them; a framerate preset rescales the whole file in one click.
• Off by some other number of minutes. A different cut. Either keep looking for a file that names your edition, or take this one and fix it at the break.

The check works because the last cue is an anchor you get for free: you did not have to measure it, the file’s author did. It is the same idea as the two-anchor repair, read backwards. If the file turns out to drift, it also gives you the late anchor for the drift repair — and if you still need to measure, do it by nudging in VLC, not by scrubbing.

When nothing matches

Sometimes there is no file for your release: a new web cut, an obscure edition, a language nobody has done yet for that source. Then the question is which wrong file is the least wrong, and the order is simple. Prefer the same source and edition, whatever the group. Failing that, prefer the file whose Runs to figure lands closest to your video’s length. A file for the same cut that is merely early or late costs one number; a file at the other framerate costs one preset; only a file for a different cut costs real work. Pick the cheapest repair, not the most downloaded file.

One more thing a downloaded file can be wrong about that has nothing to do with timing: its text encoding. Subtitle sites hold decades of uploads saved in regional Windows code pages, and a file that plays in perfect sync while showing café or question marks for accented letters needs a conversion to UTF-8, not a repair.

If your subtitles are already in the video as a track and you are reading this because that track is wrong, the same logic applies: an external file for the same release is often better than extracting and repairing the embedded one, and every player will load it from the folder beside the video.

Match the source and edition, ignore the resolution, trust a hash match over a name, and read Runs to against the video’s length before you watch. Most of the time that ends the problem without a repair. When it does not, it has already told you which repair you need.