Frequently asked questions

In the Workspace, drop or pick a video, frame the area where subtitles appear, and press Start extraction. When it finishes, you can edit the text and timings in the result table, then download the SRT.

Common formats such as MP4, MOV, MKV, FLV and WebM. A few videos with unusual codecs may not be readable; convert them to a regular MP4 first.

It depends on your computer's performance and how dense the subtitles are — typically a tenth of the video's length or faster. Stay on this page while it runs; switching tabs or closing it stops the extraction.

The default Precise gives the most accurate timing — use it for a final subtitle file or line-by-line checking. Fast takes about half the time and is good for a first look at the content: timestamps are usually within half a second, occasionally close to a second, and about half of the subtitles shorter than 1.5 seconds are missed.

Possibly. Recognition runs on the CPU and is the main cost, but video decoding is handed to the GPU by the browser; if the chosen GPU doesn't support the video's codec, decoding falls back to the CPU and high-resolution videos get noticeably slower. As a rough guide, it's worth checking once processing takes more than 20% of the video length (a 10-minute video taking over 2 minutes). On Windows with both an integrated and an older discrete GPU, go to Settings → System → Display → Graphics, add Chrome or Edge, and set its option to "Power saving" (the integrated GPU), then restart the browser. To verify, type chrome://gpu in the address bar to see which GPU is in use. Rough hardware-decode coverage: Intel iGPUs support HEVC from 7th gen and AV1 from 11th gen; NVIDIA supports HEVC from GTX 10 series and AV1 from RTX 30 series; AMD supports HEVC from RX 400 series and AV1 from RX 7000 series. Any unsupported codec falls back to software decoding.

Pick the language the subtitles are written in, in the region panel. The default English / Chinese / Japanese / Latin alphabet set is the most complete and the safe pick when in doubt; subtitles only in Simplified Chinese or other Latin-alphabet languages can switch to English / Simplified Chinese / Latin alphabet — a smaller model about four times faster. Pick Korean or Russian for those languages. Your choice is remembered for next time.

Yes. After picking a video, choose “Use your own SRT” at the bottom of the subtitle region panel and select a .srt file. Recognition is skipped and the file is saved as this video's record, ready to search, edit or download again; the record is marked as imported.

Not always. Auto-detection only guesses the subtitle position from a sample of frames; with tickers, watermarks, in-scene text or unusually laid-out subtitles it can frame the wrong spot, frame too small, or find nothing at all. Before starting, always scrub to a frame with subtitles and confirm yourself that the box fully covers the longest subtitle line (both ends included); if it doesn't, drag to adjust or draw it again.

Pick the side the subtitle text hugs inside the box: most videos are centered; choose left or right for subtitles that consistently sit on that side. A wrong choice can drop short lines that don't cross the centre axis. Choose Unfiltered when the subtitle position varies inside the box; everything captured there, logos included, is then read.

Just large enough to cover where subtitles appear: too small and it clips characters; too large and other on-screen text (titles, watermarks) gets recognized along with the subtitles.

The most common causes are a box that doesn't cover the subtitles, subtitles that are too small, or too little contrast against the background. Enlarge the box, check the alignment, and scrub to a frame with subtitles before framing again.

Edit the text or times directly in the result table; re-recognize a single row with Re-recognize, or the whole video with Re-extract all. It's worth reviewing everything once before downloading.

A UTF-8 .srt subtitle file that you can load directly in a player or editor.

Yes. Drop several videos on the Batch page and they are processed one after another; the subtitle region can be auto-detected per video or reused from the workspace. A video that carries a text subtitle track (SRT, ASS or WebVTT inside an MKV, MP4 and so on) is read from that track with no recognition: the language you last picked in the workspace wins, then the interface language, and only then the track the file marks as default. When everything finishes you can download all SRTs at once. The batch list itself isn't kept after the tab closes, but finished results are saved to History.

Every finished extraction is saved automatically. You can search subtitle text or file names, file records into folders, export or import a subtitle library, and re-recognize the same video. Records live only in this browser and are removed when you clear browsing data.

Yes. A record stores subtitles and a file pointer, not the video itself. On the History page, choose "Link video files" and pick the folder holding your videos to link records to same-named files (when several share a name, the one matching the stored size wins); importing a library offers this once as well. Linked records can play, capture frames or re-recognize right away. This requires Chrome or Edge.

No. The video, its frames and the recognized text are processed only on your device; nothing is sent to any server.

Email contact@subku.app. If you can, include the browser you use, the video format and what happened — it makes the cause much easier to track down.

You can buy the author a coffee to help keep this free tool maintained.