Burned-in (hardcoded) subtitles are part of the pixel layer: they survived export and live on the picture itself. The moment you dub such a video into another language, two caption tracks fight for the bottom third of the frame — which is why localizing subtitled source footage has always ended in either a compromise or an afternoon of rotoscoping.
The three real options
1. Blur or stamp over it (delogo)
Every NLE and most online tools can box-and-blur a region. It's fast, and it looks it: a frosted band crawling across your video. On flat backgrounds it's almost acceptable; over a moving subject it's a watermark you installed yourself.
2. Manual object removal, frame by frame
Inpainting tools genuinely rebuild backgrounds — one frame at a time. The catch is temporal consistency: per-frame fixes shimmer, and 33 seconds at 30fps is 990 frames. Realistic for one shot; hopeless for a course.
3. Detect-then-repair as a pipeline step
What production dubbing infrastructure does today: OCR localizes the text band across the clip, a temporal pass separates static text from moving background, and an inpainting model (LaMa-class, running on the band, not the frame) repaints pixels using motion from neighboring frames. Clothing texture and gradients re-emerge instead of a smear — and the pass runs once, reused by every target language.
Order matters: erase before you dub
Dub first and the new captions bake on top of the old ones. The clean sequence is remove burned-in text → transcribe → translate → re-voice → burn fresh styled captions. If your dubbing tool can't do step one, the workflow breaks into a separate cleanup service, a re-upload and a second billing cycle.
The honest cost line
Removal is GPU work — minutes of a modern card per minute of video — which is why it's priced as an add-on (about half a credit per finished minute) rather than free. Clean sources skip it automatically, and line re-dubs reuse the erased master so you're never billed twice.
Voxtakes removes burned-in subtitles inside its 9-language dubbing pipeline — see the demo or apply for the open beta.