AddThisFeature

Translated Subtitles

Let viewers pick their own language for a video's captions.

involved Accessibility & Internationalization

What it adds

Additional caption tracks in other languages, generated from the source track and selectable in the player.

What your agent is told to do

5
  1. 1

    Translate cue by cue from the corrected source caption track, not from the raw audio, so timing and sentence boundaries are already settled before translation begins.

  2. 2

    Reuse the app's existing AI Translation path and its language list rather than introducing a second translation configuration. This feature adds tracks and player selection, not a new translation engine.

  3. 3

    Offer only the languages the audience actually uses. Generating twenty tracks per video costs real money and buries the two that matter in a long player menu.

  4. 4

    Default the player's caption selection to the viewer's own language preference where a track exists, using the app's existing Locale and Language Switcher setting rather than guessing from the browser alone.

  5. 5

    Do not present machine translations as authoritative. Label them, and allow a human-reviewed track to replace one without regenerating the rest.

Edge cases it handles

8
  • Translated text is routinely longer than the source. Keep the original cue timings but split overlong cues across the available window rather than letting text overflow the screen or get truncated.
  • When a viewer requests a language with no track, fall back to the source language with a visible note rather than showing an empty caption area or silently turning captions off.
  • Right-to-left languages need correct text direction, alignment, and punctuation placement in the caption renderer. Left-aligned mirrored text is worse than no track.
  • Editing the source track must invalidate the translations derived from it, so a corrected name in the original does not persist as the wrong name in five other languages.
  • Cache translations per cue and re-translate only the cues whose source text changed. Re-running the whole track on every edit is slow and expensive for no benefit.
  • Proper nouns, product names, and code identifiers should be protected from translation, or the captions will helpfully translate the name of the app.
  • Track language codes must include region where it matters, so two variants of the same language do not collide in the player menu.
  • Generation runs as a background job with progress, and a failure on one language must not block the others.

Definition of done

9
  • Viewers can select from the available caption languages in the player.
  • Cue timings are preserved and overlong translations are split rather than clipped.
  • A missing requested language falls back to the source track with a visible notice.
  • Right-to-left languages render with correct direction and alignment.
  • Editing the source track invalidates the derived translations.
  • Regeneration re-translates only the cues whose source changed.
  • Machine-translated tracks are labelled and can be superseded by a reviewed version.
  • The feature matches the existing design system.
  • No existing functionality is broken.

Related features

How it works

  1. 1

    Copy the link

    Grab the Markdown instruction URL for this feature.

  2. 2

    Give it to your AI

    Paste it into Claude Code, Cursor, v0, Lovable — whatever you build with.

  3. 3

    It inspects, then implements

    Your agent reads your existing app first, then adds the feature to fit it.

Works with your stack

These instructions are written to adapt. They tell the agent to detect your framework, match your existing design system, and reuse what you already have — rather than assuming a particular stack.

Need it tighter than that? Customize the feature and tell it exactly what you're running.