Word-by-word captions you can still edit

The caption look where the whole line holds on screen and the spoken word lights up. Almost every tool that makes it hands back a finished MP4 with the captions painted into the picture. This one hands back an FCPXML that Final Cut Pro or DaVinci Resolve turns into one editable title per word. Here it is, playing.

Why neither editor can do this on its own

Final Cut Pro and DaVinci Resolve both transcribe and both build subtitles, and neither of them builds this. Their captions are cue-level objects: one block of text, one span of time, one colour. There is nowhere in either application to say that the third word of a line should be a different colour for a third of a second.

So the look gets built by hand, and by hand it means one title clip per word, each one dragged to the frame the word is spoken on and styled to match its neighbours. For a sixty-second clip that is a couple of hundred title clips. This is why the effect is nearly always bought as a plugin, a Fusion macro, or a burned-in render from a separate app.

What the file actually contains

The same thing, generated. The karaoke export writes one title per word, and each one carries the whole line split into at most three runs: the words already spoken, the word being spoken, and the words still to come. The first two are one colour, the middle one another. Play it back and the line holds still while the highlight steps along it.

That is also the honest description of the animation. The highlight steps word to word rather than sweeping across each word, because the colour of a title in FCPXML is a fixed attribute of that title: changing a colour means a new title, and the smallest unit worth having a title for is a word. A continuous wipe needs a Motion or Fusion template, which is a different piece of work.

Every title is a real title. In Final Cut it is a Basic Title you can restyle; in Resolve, FCPXML import produces Text+ titles, which are Fusion nodes and take anything Fusion does. Nothing is painted into the picture, so a wrong word is a text edit rather than a re-render.

Burned-in captions and what they cost you

The usual way to get this look is a web tool that transcribes your video and gives you back an MP4 with the captions already in the pixels. It is genuinely faster if the video is finished. It stops being faster the moment anything changes.

• A misheard name means re-uploading the video and re-rendering it, rather than fixing a word on your timeline.
• The captions are locked to the resolution, the crop and the colour grade that were in place when they were burned in.
• You are re-encoding a finished export, so the picture goes through another generation of compression for the sake of some text.
• The timing lives in that tool's project rather than in your file, so a recut starts the whole loop again.

Exporting captions as titles instead keeps all of that on your timeline, where the rest of the edit already is.

Why the word timing has to be real

A highlight that is a tenth of a second late is more distracting than no highlight at all, because the eye is being told to look at the wrong word. Cue-level timing cannot produce it: an SRT says when a line starts and ends and nothing about the words inside it, so anything built from an SRT alone is dividing the line up evenly and hoping.

Transcription here is followed by forced alignment, which takes the words it already has and finds each one in the audio, giving every word its own start and end. That is where the highlight timing comes from, and it is the same timing a split or joined caption line inherits in the editor.

Premiere Pro is the exception

Premiere cannot read FCPXML. The two formats share a lineage and nothing else, so there is no karaoke export for Premiere and this page would be lying if it offered one. What you can download for Premiere is a marker XML, which drops a named sequence marker on the first frame of every caption line, plus the SRT. The markers are a timing guide for building the look with Essential Graphics or a template, not the look itself.

Placement, before you download it

Vertical captions have to clear the interface: the caption bar, the handle, the buttons down the right. The style panel sets the caption's height as a percentage of the frame rather than in pixels, and a conventional lower third is around minus 30 percent. The preview holds a 16:9 frame, so read it as the proportion of the way up or down the frame rather than as a crop of your own vertical sequence, and check it against the picture once it is on the timeline.

Frequently asked questions