Why cropping is the wrong default
Cropping changes the aspect ratio. If the clip is going back to the same platform at the same size, you now have to letterbox it or scale it — and scaling a 1080p source up to fill a 1080p frame softens the whole image, not just the caption area.
Reconstruction only touches the caption band. The rest of the frame is untouched, so sharpness, grain and colour stay consistent across the shot.
Captions versus subtitles
People use the words interchangeably, but the practical difference matters. Subtitles usually sit in a fixed band at the bottom and change with dialogue. Captions in social video are often animated word-by-word, centred, sometimes with a coloured background box.
Word-by-word captions that jump around are the hardest case. Select a box wide enough to cover the whole caption zone, not just where the text happens to be on the current frame.
- Fixed caption band: automatic mode works well
- Animated or bouncing captions: manual box covering the full zone
- Captions with a solid background box: manual mode, box slightly larger than the text
Frequently asked questions
Do I lose video quality by removing captions this way?+
Less than cropping and upscaling. The frame keeps its original resolution, and only the masked area is regenerated. There is still one re-encode, so keep the source file if you plan more edits.
Can it handle captions that move around the frame?+
Partially. Manual selection defines a fixed rectangle, so choose one that covers the full range of movement. Text that travels outside that box will be missed.
What if the caption has a coloured background box?+
Include the box in your selection, with a small margin. The model removes the whole element rather than trying to separate text from its background.
Does it remove captions from TikTok and Reels exports?+
Yes — those are ordinary MP4 files with burned-in text. Vertical video is supported; keep the selection inside the frame and the output stays vertical.