Remove old burned-in subtitles by cropping, then re-caption
Sume has no subtitle inpainting. If old captions sit in a bar at the edge, crop it with the video filter, test with the free check, then burn new cues.

If the old subtitles sit in a band at the top or bottom of the frame, crop that band off with Sume's video filter, then burn your new captions. This is the only subtitle-removal approach the docs support: there is no inpainting, and the filter list has no drawtext or subtitles filter. Cropping throws pixels away, so it works for a subtitle bar and not for text laid over a face.
The video filter takes ops for dim and crop plus an allowlisted filtergraph, with a source up to 300 seconds. A job is priced at $0.02, and a /check call is free.
When does cropping work?
Cropping works when the subtitles are in a band you can lose without losing the subject. It fails when the text sits in the middle of the frame, or the aspect ratio matters and you cannot afford to change it.
| Where the old text is | Crop works? | Alternative |
|---|---|---|
| Black bar under the picture | Yes, trim the bar | None needed |
| Bottom 10 to 15 percent of the picture | Usually, if the subject is higher | Check the framing first |
| Over the subject's body or face | No | Re-shoot or use a clean master |
| Moving or animated text | Only if it stays in the band | Use a clean master |
How do you run it safely?
The exact ops shape is in the video filter docs, which also describe the check call. The free check is covered in the dim-a-bright-clip post.
- Run
/checkfirst. It is free, so you can try a crop before paying for the job. - Look at the output frame, not just the status, to confirm no text remains in the kept area.
- Note that the picture gets smaller or changes shape. Decide whether to leave it that way or fit it to a target aspect ratio afterwards.
- Keep the original; you cannot undo a crop.
Then how do you add the new captions?
Post the cropped video to the caption endpoint with the new text as cues, or let speech-to-text run if the audio is clear. A standalone job is priced at $0.20 for videos up to 60 seconds under the current estimate. Burn into a clean area, since a caption that lands on remnants of the old text will look worse than either.
The new cues come from your own translation or edit, as in the SRT walkthrough. See Video captions for the style names.
What are the limits to remember?
The source clip for the video filter is limited to 300 seconds, and the filter list is an allowlist: blur and geometry operations are there, drawtext and subtitles are not. That means you cannot paint over text, only remove the area it sits in, or blur it. A blur leaves a visible smudge, so it suits a bar over a flat background and not a busy scene.
If the old subtitles were part of a video you do not own the clean master of, check whether you are allowed to alter it at all. Removing another party's on-screen text can touch their rights, so confirm permission first.
Sources
Related posts
More in Media tools
- Cut a clip into 30-second chunks for 30-second post-processing caps
Runway's SDR-to-HDR model and the Magnific upscaler take 30 seconds at most. Cut any clip into 30-second pieces with Sume video trim at $0.02 per cut.
- Diwali gift hamper hero still: Seedream, then upscale for $0.23
A Seedream 4.5 image is $0.033 and a Sume image upscale is $0.20. One print-ready Diwali hamper hero is $0.233; four tries and one upscale is $0.33.
- Douyin ad first frame: black at most 60%, check with stills
Douyin in-feed ads need a first frame that is at most 60% black. Pull stills from your render with Sume video inspect and measure the first frame before upload.
- Edits carousels: four matching images per call
Instagram Edits now supports carousels, per secondary reports. Sume image models with an n range of 1-4 return up to four matching images per request.
Written by Sume