Subtitle Edit

the subtitle editor :)


OCR Burned-in Subtitle (Video OCR)

Extract hardcoded (burned-in) subtitles from a video into editable text lines using OCR.

How to Use

  1. Open Video → OCR burned-in subtitle… (a video file is required)
  2. Use the preview slider to find a frame that shows a subtitle
  3. Adjust the scan area rectangle so it covers where the subtitles appear (default: bottom third) — drag to move/resize, or use the preset buttons
  4. Pick an OCR engine and language (for Ollama and llama.cpp, also pick a vision model)
  5. Optionally click Test current frame to try the current frame before running the full pass
  6. Click Start OCR — lines appear in the list as they are recognized, already run through the OCR fix engine and spell check when a dictionary is selected
  7. Click OK to load the result into the main window

To fix an OCR mistake, press Enter (or F2) on a line — or right-click and choose Edit… — to open the text in a small edit window. The right-click menu also offers Italic and Delete (the Delete key works too). Double-click a line to show the frame it came from in the preview.

OCR Engines

Engines are listed best-first for burned-in subtitles:

How It Works

Frames are sampled from the scan area at a few frames per second with ffmpeg. Consecutive near-identical frames are collapsed so each on-screen subtitle is OCR’ed only once, then consecutive OCR results with near-identical text are merged into one line (the text variant shown the longest wins, weighted by the engine’s recognition confidence where available). Finally each line’s start and end are refined against the video’s own frames, so the times are precise to a few hundredths of a second even though the scan itself samples much coarser.

Fix OCR errors and spell check

With Fix OCR errors enabled and a Dictionary selected (the list shows the spell check dictionaries already downloaded; the language is pre-picked from the OCR language), every line runs through the OCR fix engine as it is recognized: common OCR mistakes are corrected from the language’s replace list, and the words are spell checked. In the list, words the dictionary knows are shown green and unknown words red — so the lines that need a human eye stand out while the OCR is still running.

Settings

Tips