First, decide which kind you have
If the words move with the beat, change colour word by word, and the background is abstract or generated, it is a lyric video — the text is the subject. Removing it is mechanically possible and practically pointless.
If the lyrics are a subtitle track laid over concert footage, a dance clip, or a music video with real photography, removal restores something that was there. That is the case worth paying for.
- Lyric video: the text is the design, so removing it leaves an empty frame — do not
- Karaoke burn-in: usually a wide band at the bottom, often with the active word highlighted
- Performance footage: lyrics over a stage, a crowd or a set — this is the case that works
Why per-word highlighting is worse than plain text
Karaoke highlighting changes the colour of the active word every beat. The text area therefore changes density and shape constantly, and a repair sized for one frame is wrong for the next. Box the whole line rather than the highlighted word, and expect the box to be doing more work than a static caption would.
The same applies to text that flies in, scales, or fades. Size the box for the full path the words travel, not for where they sit at the moment you paused.
Where it fails, and it fails visibly
Lyrics are usually centred, which puts them over faces. A name plate at the bottom of a frame sits in empty space; a lyric line sits across the singer. Our crowd test — pedestrians walking through a caption band — is the closest measurement we have, and it scored 19.8 dB with visible ghosts where each person crossed.
Text over a face fails the same way, and the result is worse to look at, because a face is the first thing the eye checks for damage. If the lyric line crosses a performer, cropping or re-cutting will look better than any reconstruction.