Burned-in captions and subtitles
Select the entire caption band, including outlines and shadows. Watch the tutorial comparison for two-line captions and bouncing word highlights.
sel_areaAI video text cleanup
Remove Text and Captions from Video Online with AI — No Blur, No Crop
Upload your video
Drag & drop a file here, or click to browse
Secure processing — temporary retention follows your plan and task status
MP4, MOV, AVI, MKV, WebM · Up to 2GB
See the difference →
Use UnmarkAI as an AI caption remover to remove captions from video you own or are authorized to edit, including burned-in subtitles and animated word highlights. This video text remover also handles titles, date stamps, labels and promotional copy. AI reconstructs the selected region without blur or cropping; output keeps the original audio and resolution.
10 free points after sign-in · Up to 2 GB · Preview before downloading. Processing and preview require login. Choose the scope around the words you want to replace, then review the complete result.
What text can you remove from a video?
Start with the position and movement of the words. Separate SRT or VTT subtitle tracks can be disabled or removed directly; they do not need AI pixel reconstruction. The types below describe selection workflows, with recorded examples where available.
Moving creator captions
When the caption block changes between the bottom and top, compare the vertical creator example using full-frame detection.
all_areaTitles and headlines
Use a bounded selection for a fixed title. The city street example covers both headline lines; inspect the partly hidden sign and window edges. Edit source text layers directly when available.
sel_areaPromotional and price text
Separate an upper offer label from changing lower copy. The UGC product example uses two regions on the same clip.
multi_areaSticker labels
Select the whole label and its movement path. The recorded sticker example uses a mostly plain wall; labels over products were not tested.
sel_areaDates and timestamps
A fixed corner date or recording timer fits a single box. Include the full footprint of every digit as it changes. The lab security-cam example uses an opaque OSD box over soft background.
sel_areaScrolling tickers and credits
Cover the complete path when it fits a bounded region, and inspect each text change. Measured example coming soon.
sel_area / all_areaHow to remove text from a video in 4 steps
Confirm that the words are part of the picture, work from the best authorized source available, and review the complete result before publishing.
- 01
Confirm the text is hardcoded
Turn subtitles off in the player. If the words disappear, remove or replace the subtitle track; if they remain, they are part of the video pixels.
- 02
Upload the best source available
Use an owned, licensed, or explicitly authorized source with as little prior compression as possible. More real source detail gives reconstruction a stronger reference. Sign in to process and preview your upload.
- 03
Choose the text removal scope
Use sel_area (Box Select) for one stable caption band, multi_area (Multi-Box) for up to three non-overlapping boxes, or all_area (Full Frame) when text changes positions. Include outlines, shadows and the full bounce of highlighted words; check the estimate before processing.
- 04
Preview, compare and download
Watch the complete preview alongside the source. Scrub through caption changes, scene cuts and moving text. Check for residue, flicker and softened detail; adjust your selection if needed, then download. Output keeps the original audio and original resolution.
Real examples: remove text from video before and after
Compare six cases: tutorial captions, product copy, vertical creator captions, title text, a sticker label and a security-cam timestamp. Each 12-second comparison keeps the input and actual API output synchronized.
Demo overlays added to licensed footage; outputs come from the real API with no manual cleanup. These are examples, not a quality or speed guarantee.
Tutorial captions
Clear instructional captions before translating or updating a tutorial.
| Scene | Tutorial captions |
|---|---|
| Mode | sel_area |
| Selected area / source frame | 24.3% |
| Duration | 12 s |
| Processing seconds | 101.87 |
| Cost (USD API) | $0.045 |
| Recorded result review | The two-line words and yellow highlights are no longer readable in the reviewed caption transitions. Sawdust and wood grain in the lower band look softer than the input; the central tool and gloves remain outside the selected region. |
| Limitations | Wood grain and sawdust are moving behind the caption band. Reconstruction may soften those details; this clip does not establish performance on every tutorial. |
Measured 2026-10-06 · Recorded USD API charges; web points are separate. Processing time varies by clip and queue.
Watch case & request details →UGC product copy
Replace old promotional text before making a new campaign or language version.
| Scene | UGC product copy |
|---|---|
| Mode | multi_area |
| Selected area / source frame | 12.2% |
| Duration | 12 s |
| Processing seconds | 102.04 |
| Cost (USD API) | $0.090 |
| Recorded result review | The upper label and changing lower copy are removed in the reviewed frames. The face and moving product remain visible; the lower clothing area becomes smoother at some moments. |
| Limitations | Hands and the product move near the lower region. Check product contours and clothing texture before using cleaned footage in an advertisement. |
Measured 2026-10-06 · Recorded USD API charges; web points are separate. Processing time varies by clip and queue.
Watch case & request details →Vertical creator captions
Prepare a clean master for new Reels, TikTok or Shorts captions.
| Scene | Vertical creator captions |
|---|---|
| Mode | all_area |
| Selected area / source frame | 100.0% |
| Duration | 12 s |
| Processing seconds | 170.31 |
| Cost (USD API) | $0.090 |
| Recorded result review | The lower two-line captions and upper caption block are removed across the reviewed pages. Glasses, hair and the laptop remain visible; skirt folds and wall texture inside reconstructed areas can soften. |
| Limitations | Full-frame detection may also affect text you want to keep. Review the face, glasses, laptop and background; use a selected region when only one band needs cleanup. |
Measured 2026-10-06 · Recorded USD API charges; web points are separate. Processing time varies by clip and queue.
Watch case & request details →Title and headline text
Reuse authorized B-roll without an old episode title or headline card.
| Scene | Title and headline text |
|---|---|
| Mode | sel_area |
| Selected area / source frame | 13.5% |
| Duration | 12 s |
| Processing seconds | 133.67 |
| Cost (USD API) | $0.05625 |
| Recorded result review | Both headline lines are no longer readable in the reviewed frames, and the facade, street lamp and windows continue through the region. Lettering of the hotel sign that was partly covered looks slightly distorted compared with an unobstructed sign. |
| Limitations | The hidden area is a building facade with a hotel sign behind the headline. Reconstructed sign lettering and window edges can look slightly warped; this clip does not establish results on faces or fast camera motion. |
Measured 2026-10-20 · Recorded USD API charges; web points are separate. Processing time varies by clip and queue.
Watch case & request details →Sticker and label text
Remove an outdated promo sticker from authorized product footage before a new campaign.
| Scene | Sticker and label text |
|---|---|
| Mode | sel_area |
| Selected area / source frame | 8.0% |
| Duration | 12 s |
| Processing seconds | 105.61 |
| Cost (USD API) | $0.05625 |
| Recorded result review | The sticker, its outline and both text lines are gone in the reviewed frames. The wall inside the box is slightly smoother than the surrounding plaster texture, visible mainly when stepping frame by frame. |
| Limitations | The background behind the sticker is a mostly plain wall, which is an easier case. Stickers over faces, products or detailed scenery are harder and were not tested here. |
Measured 2026-10-20 · Recorded USD API charges; web points are separate. Processing time varies by clip and queue.
Watch case & request details →Date and time stamp
Clear a burned-in camera-style date/time stamp from your own footage before reusing it.
| Scene | Date and time stamp |
|---|---|
| Mode | sel_area |
| Selected area / source frame | 3.4% |
| Duration | 12 s |
| Processing seconds | 83.26 |
| Cost (USD API) | $0.05625 |
| Recorded result review | The opaque black OSD box and green CAM-03 time are gone in the reviewed frames. Soft out-of-focus lab tones continue through the region with no obvious box outline or residual digits at normal playback speed. |
| Limitations | The stamp sits over soft, out-of-focus lab background. This does not establish results when a stamp crosses faces, fine texture, or high-contrast edges. |
Measured 2026-10-20 · Recorded USD API charges; web points are separate. Processing time varies by clip and queue.
Watch case & request details →Chinese burned-in captions across scene cuts — 10-second comparison
Chinese burned-in captions across scene cuts — 10-second comparison
A side-by-side comparison of Chinese burned-in captions, moving subjects and several shot changes. Source is on the left; cleaned output is on the right.
10 s · 1080 × 960 · H.264 · 24 fps
Play the clip at normal speed, then scrub through caption changes and cuts. Compare the cleaned right half with the source on the left for text residue, flicker, repeated texture, and softened edges near the caption band.
AI caption remover for burned-in captions and subtitles
First check whether the words are burned into the pixels or stored in a subtitle track. Burned-in text can use one box, up to three non-overlapping boxes, or full-frame processing; removable SRT or VTT tracks do not need AI reconstruction.
Hardcoded captions are encoded into the picture. SRT, VTT, and other switchable subtitle tracks are separate data and should be removed or replaced without reconstructing the video.
Turn subtitles off first
If the words disappear in the player, edit the subtitle track instead of spending points on video cleanup.
Use a box for stable text
A fixed caption band, timestamp, or lower third usually needs only a tight region around its letters, outline, and shadow.
Treat moving text as a harder case
Animated words, scene cuts, and text crossing faces or fine detail require closer full-clip review.
Can I remove two-line, word-highlight captions from a video?
Yes, burned-in two-line captions with an active-word highlight can be processed like other video text. Cover the entire animated word area, including both lines, the active word’s bounce and outline. The 12-second tutorial and vertical creator examples below show actual API results. Check every word change for remnants and flicker; separate subtitle tracks should be removed directly.
Choose the cleanup scope before processing
The editor exposes three scopes instead of forcing every clip through the same workflow. It also shows the point estimate before the task starts.
- Box Select — Target one predictable caption band, timestamp, or label. The standard web rate is 5 points per started 30 seconds.
- Multi-Box — Mark up to three separate, non-overlapping text regions. The standard web rate is 10 points per started 30 seconds.
- Full Frame — Process the complete frame when a bounded selection is not suitable. The standard web rate is 10 points per started 30 seconds.
Rates can change; check current point usage
Prepare a clean source master for the next edit
Text removal is most useful when it unlocks a specific revision. Keep required notices and ownership marks, and remove only the copy approved for replacement.
- Replace burned-in captions — Clean an old caption band before adding corrected or translated subtitles.
- Refresh campaign copy — Update an expired offer, product label, lower third, or approved sponsor line when the source project is unavailable.
- Clean owned screen recordings — Remove temporary meeting labels, timers, notifications, or review notes from recordings you control.
After cleanup, continue with video translation
Caption replacement and localization. Prepare a source master for corrected captions, a new language, or a different subtitle style.
What makes video text removal easier or harder?
| Text or scene | Selection approach | What to inspect |
|---|---|---|
| Static text on a stable background | A tight sel_area box | Letter edges, shadows and repeated texture |
| Two-line or word-highlight captions | Include both lines and the full animation footprint | Each highlighted word and caption change |
| Separated labels and promo copy | multi_area for up to three non-overlapping regions | Every region, especially near hands and product contours |
| Moving text or changing positions | A region covering its path, or all_area when needed | Entrances, exits and text you intended to keep |
| Faces, hair, fabric or fine patterns | Keep the selection as small as practical | Soft detail, invented texture and altered contours |
| Camera motion, scene cuts or occlusion | Review the complete processed interval | Flicker, smearing and texture that moves differently |
Cropping changes the composition, while blur or mosaic leaves a visible patch. Inpainting attempts to reconstruct the selected pixels, but the result still needs review because hidden detail cannot always be recovered reliably.
Check the result before publishing
- Play the complete clip at normal speed, then scrub through word changes and cuts.
- Look for residual letters, shadows and outlines at the selection edges.
- Check for blur-like softness, smearing, repeated texture and invented detail.
- Watch for flicker, texture drift or a repaired area that lags behind the scene.
- Compare faces, hands, hair, clothing and product contours with the source.
- Check that text outside your approved cleanup scope remains present.
- Listen to the original audio and confirm the downloaded video has the original resolution.
Files, limits and pricing for the video text remover
| Item | Available on the web |
|---|---|
| Accepted video files | MP4, MOV, AVI, MKV, WebM |
| Maximum file size | 2 GB |
| Selection modes | Box Select (sel_area), Multi-Box (multi_area, up to three non-overlapping boxes), Full Frame (all_area) |
| Audio and resolution | Output keeps the original audio and original resolution. |
| Processing and preview | Login required. Compare the result before downloading. |
| 10 free points after sign-in | USD 0 |
| Box Select | 5 points per started 30 seconds |
| Multi-Box / Full Frame | 10 points per started 30 seconds |
| One-time point packs | $4.90 / 100 points; $9.90 / 250 points; $19.00 / 800 points; $49.00 / 2600 points; $99.00 / 5600 points |
For developers and teams
View API documentationAutomate video text removal
The public API supports signed uploads, full-frame, selected-area or multi-region jobs, cost estimates, safe retries, polling, and signed webhooks.
Use text removal on video you are authorized to edit
Process only video you own, license, or have explicit permission to edit. Do not remove attribution, provenance, ownership marks, safety labels, evidence, or legally required notices.
Read the Privacy Policy · Acceptable UseFrequently asked questions
How do I remove text from a video online for free?
Sign in to start with 10 free points. Upload an authorized video, choose Box Select, Multi-Box or Full Frame, check the point estimate, process and compare the preview before downloading. Processing and preview require login.
Can I remove text from a video without cropping it?
Often, yes. Box Select, Multi-Box, and Full Frame keep the existing frame dimensions while AI reconstructs the processed area. Review the full preview because motion, compression, and hidden detail affect the result.
Do I need AI to remove an SRT or VTT subtitle track?
No. If subtitles are stored as a separate SRT, VTT, or player track, disable or replace that track. AI cleanup is for words already burned into the video pixels.
What makes video text removal harder?
Fast motion, scene cuts, large text, heavy compression, flashing light, and text over faces or fine patterns leave less reliable background information. Use a tight selection and inspect the entire interval where the text appears.
What should I check before publishing the cleaned video?
Watch the full result at normal speed and scrub through every caption change and scene cut. Compare the selected area with the source for residue, flicker, repeated texture, soft edges, or changes near faces and objects.
What text should not be removed?
Keep attribution, provenance, ownership marks, safety labels, evidence, and legally required notices. Process only footage you own, license, or have explicit permission to edit.
What kinds of text can this video text remover process?
Burned-in captions, titles, lower thirds, date stamps, promotional copy, labels and sticker text can be selected for reconstruction. Use a region that covers the whole text footprint. The recorded examples show tutorial captions, promotional copy, vertical captions, titles, sticker labels and a security-cam timestamp.
Can it remove moving text or captions that change position?
The 12-second vertical creator example uses all_area for captions that move from the lower band to the upper band after eight seconds. Where possible, use a bounded region covering the text path. Review every position change and any text you want to keep.
Can I remove text from only one part of the frame?
Yes. sel_area targets one region, and multi_area supports up to three non-overlapping boxes. Use all_area for full-frame text removal when a bounded selection is unsuitable. Compare the whole result before downloading.
How long does it take, and how much does it cost?
Recorded 12-second API runs include USD 0.045–0.09 charges (caption and promo cases) and USD 0.05625 for later title, sticker and timestamp cases (15 billable seconds each). Processing times vary by clip. These are measured runs, not guaranteed turnaround times. Web points are separate: Box Select costs 5 points per started 30 seconds, Multi-Box and Full Frame cost 10. Check the pricing page and task estimate.
Will the video quality or audio change?
Output keeps the original audio and original resolution. AI reconstructs the selected pixels, so fine texture, motion and hidden details can still show artifacts. Compare the complete preview with the source before publishing.
Can I remove text from TikTok, Instagram Reels or YouTube Shorts videos?
Use an uploaded video you own, license or have explicit permission to edit. The vertical creator example demonstrates two-line highlighted captions for a Reels, TikTok or Shorts editing workflow. Inspect the complete result before adding replacement captions.
Remove text from your video — preview before you download
Sign in, upload your video and select the text. Compare the complete result before downloading with the original audio and resolution.
Remove text from your video →