How to strip the captions out of an edit and rebuild it to new music
Upload the edit and the track you want under it. Vizard Agent measures the old text bars to the pixel, removes them and verifies the frame is clean, analyses the new track's transients, and re-times every cut so the edit lands on the new music rather than the old.
What is the short version?
Reusing an edit with a different song is not a soundtrack swap. The cuts were placed against the old track's beats, so putting a new one underneath means finding the new transients and moving every cut to them, and the burnt-in captions have to come out first.
- Go to Vizard Agent and upload the edit and the new audio.
- Say which captions should survive and which should go.
- Say that the cuts should match the new track.
What do you need before you start?
The edit and the track. Vizard Agent measures the existing overlays and removes them itself, so you do not need the original project file, which is usually the reason this job gets refused elsewhere. Saying exactly which text stays is the one instruction that matters.
- The finished edit. The rendered file is enough.
- The new track. The version you actually want, slowed or edited included.
- What stays. Often only the before-and-after labels.
- What goes. Every other burnt-in caption and bar.
- The shape. Usually unchanged from the original.
What do you type into Vizard Agent?
Be specific about which text survives. Vizard Agent measures and removes what you name, and "remove all the captions but keep the before and after labels" is precise enough to work from, whereas "clean it up" is not.
Prompt
Variants worth knowing:
- New labels instead of the old ones. Vizard Agent will design replacements.
- A tighter crop. Removing text sometimes leaves dead space worth cropping out.
- Transitions on the beat. Rebuilt against the new track rather than kept from the old.
What does Vizard Agent actually do?
Here is the order Vizard Agent worked in on a real rebuild. Almost half of it is measurement: finding the exact coordinates of every text bar before touching a pixel, then verifying the removal at the top and bottom of the frame separately.
- Inspects the uploaded video and audio and extracts frames to see the existing captions.
- Detects the scene cuts in the original edit and checks where the text bars sit.
- Examines the exact coordinates of every on-screen element and the overlay dimensions.
- Lays a measuring grid over the graphics to pin the top card and the bottom bar precisely.
- Tests the caption removal on sample frames and inspects the quality of each attempt.
- Verifies the cleaned top and bottom of the frame separately, re-measuring when the first coordinates were slightly off.
- Analyses the new track's tempo, structure and energy second by second and computes the beat-matched cut points.
- Designs new before-and-after labels and compares two layouts on frames.
- Renders all twelve shots with the new labels, tests a tighter crop to remove what the old graphics left behind, re-times every cut to the new track's transients and measures the final loudness.
Steps 4 and 6 are the ones that make a removal invisible. Guessing a bounding box leaves a faint edge of the old bar, and the only way to be sure is to measure it with a grid and then look at the cleaned frame at full size.
What does the result look like?
From the run this page is written from, probed on the delivered file: 1080x1920, H.264, 30fps, 17.53 seconds, AAC audio. Vertical, seventeen seconds, twelve shots with the original captions removed cleanly, new before-and-after labels, and every cut re-timed to the new track.
Seventeen seconds is short, and almost all the work is invisible. The edit looks like it was always cut to this song, which is the point: a rebuild that leaves the cuts where they were reads as a video with the wrong music playing over it.
When does this not work well?
Removing burnt-in text is a repair rather than an undo, and how cleanly it goes depends entirely on what happens to be behind the text. Vizard Agent measures the region and verifies the result at full size, and busy or moving backgrounds remain the hard case.
- Text over a moving background is harder. A static bar over a still area cleans up best.
- Rights follow the music. A slowed or edited track is still someone's song.
- The original edit sets the shots. This rebuilds the timing, not the footage.
- Removal can leave dead space. A crop often finishes the job better than a repair.
- A new track changes the length. Matching the cuts may mean losing or repeating a shot.
How do you fix a result that came back wrong?
Name the frame or the cut. Vizard Agent keeps the measured coordinates, the removal tests, the new track's transient map, both label layouts and the crop trials, so a further clean-up or a re-time is a targeted re-render rather than a rebuild.
- "There is still an edge at the top." Re-measured and cleaned, as this run did twice.
- "The cuts drift after the drop." Re-timed against the analysed transients.
- "Use the other label layout." Already designed and compared on frames.
How does Vizard Agent compare to doing it yourself?
By hand this means masking the old captions frame by frame, which is slow, and then dragging every cut point until it feels roughly right against the new song. Both jobs are the kind that look finished until you watch the export on a phone and see a faint bar at the top of every shot.
| By hand | Vizard Agent | |
|---|---|---|
| Finding the text bars | Eyeball a bounding box | Coordinates measured with a grid overlay |
| Verifying the removal | Watch it back once | Top and bottom checked separately at full size |
| Re-timing to new music | Drag until it feels right | Cuts computed from the track's transients |
| New labels | Reuse the old style | Two layouts designed and compared |
Common questions
Do I need the original project file? No. Vizard Agent works from the rendered video, which is what makes this possible at all.
Can it keep some captions and remove others? Yes. Name which ones survive and Vizard Agent measures and removes only the rest of them.
Will the removal be invisible? Over a static area, usually. Over a moving background it is harder, and Vizard Agent will suggest a crop where a repair cannot finish the job.
Does Vizard Agent detect the original cuts? Yes. It finds the scene changes in the rendered file first, so it knows where the existing edit points are before deciding where the new ones go.
Does it re-cut to the new song? Yes. Vizard Agent analyses the new track's transients and moves every cut to them.
Why measure with a grid instead of just cropping the text out? Because cropping loses picture on every shot, and a bounding box guessed by eye leaves a faint edge of the old bar behind. Measuring gives an exact region to repair and keeps the frame intact.
What if the new track is a different length? Vizard Agent re-times to it, which can mean a shot is held longer or dropped.
Can it design new labels? Yes. Vizard Agent designs them and compares the layouts on real frames before choosing one.
Does Vizard Agent check the finished edit? Yes. It reviews a control sheet of frames and verifies the sync and levels before delivery.