English Subtitles From Media

Treat readability as the primary constraint, not transcription accuracy. A subtitle that reproduces every word perfectly but flashes past too fast to read has failed at its one job. Optimize for what a viewer can actually absorb in the time a cue is on screen.

File format

• Default to SRT unless the target platform specifies VTT (most web video players, some streaming specs). SRT: a sequential index number, a timecode range in HH:MM:SS,mmm --> HH:MM:SS,mmm (comma as the decimal separator), the subtitle text, then a blank line before the next cue. VTT opens with a WEBVTT header line and uses a period instead of a comma in timecodes.
• Never nest HTML or Markdown inside cue text beyond what the target format explicitly supports (SRT/VTT both allow basic <i>/<b> tags for italics/bold — nothing else is guaranteed to render).

Line length and line count

• Keep each subtitle line to 42 characters or fewer, including spaces and punctuation. This is the de facto broadcast/streaming standard because it matches what a viewer can read in roughly two seconds without eye strain.
• Use at most two lines per cue. If content needs a third line, split it into a second cue instead.
• Break a two-line cue at a natural grammatical boundary: after a clause, before a conjunction, at a comma. Never split a noun from its adjective, a verb from its particle, or a name across two lines.

Reading speed

• Target roughly 17 characters per second of on-screen duration (close to 160–180 words per minute for average English text). A 60-character cue needs at least ~3.5 seconds on screen to be readable at a comfortable pace.
• Give every cue a minimum display duration of about 1 second, even for a single short word — anything faster reads as a flicker, not a subtitle.
• Cap a single cue's duration at roughly 7 seconds even if the speaker keeps talking uninterrupted; split long continuous speech into consecutive cues instead of one overlong block.

Sync and timing

• Start a cue in sync with the first audible syllable of the line, not earlier. End it when speech ends, or up to ~200ms after to avoid a cut that feels premature.
• Leave a small gap between consecutive cues — roughly 2 frames (about 80ms at 24fps) — so players register a genuinely new cue instead of visually merging two lines together.
• When dialogue overlaps between two speakers, prioritize the more plot-relevant line and compress or drop the secondary one rather than cramming both into an unreadable single cue.

Punctuation and style

• Use standard sentence punctuation — periods, commas, question marks — and an ellipsis (…) for a trailing-off or interrupted line, not a string of periods.
• Avoid ALL CAPS except for acronyms already capitalized in normal writing (NASA, DVD) or occasional deliberate emphasis in the source; caps used for volume or shouting read as jarring in subtitles and belong to sound-design cues instead.
• Use italics (where the format supports it) for off-screen narration, telephone/radio voices, internal thoughts, and song lyrics, so the visual style itself signals the source.

Sound cues and speaker labels

• Represent non-speech audio that's necessary to follow the scene in square brackets, capitalized: [music playing], [door creaks], [applause]. Give a sound cue its own cue block when it needs standalone timing, especially for accessibility/SDH-style captions.
• When the speaker isn't visually identifiable, prefix the line with their name in caps followed by a colon: JOHN: Wait, don't—. A change of speaker always starts a new cue; never merge two different speakers into one cue block even if their lines are short.

Numbers

• Spell out numbers one through nine in dialogue-style text; use numerals for 10 and above, following standard English style guides.
• Always use numerals — never spell out — for measurements, currency amounts, percentages, times, and dates, since those are scanned rather than read as prose.

QA pass

• Read every cue once, at the pace a first-time viewer would, and confirm it's fully readable before the cue disappears. If you stumble, the line is too dense or on screen too briefly — split or trim it, don't just extend the timing past the source audio.