Skip to main content
Mode: "precision" Best for: high-quality final delivery, talking-head videos, and content where accurate lip-sync is critical. For faster turnaround at lower fidelity, see Speed mode. For many videos in one job, see Video Translation batches.

How Precision Mode Works

Precision mode uses avatar inference and multiple models to re-render the speaker’s mouth movements to match the translated audio, producing significantly more realistic lip-sync than Speed mode. It requires longer processing time and is recommended for polished, client-facing, or broadcast-quality output. Precision shares its endpoints with Speed mode. Language listing, source video input, polling, brand glossaries, captions and listing or deleting translations work exactly as on the Speed mode page; only the differences are below.

Submit a Translation

Set mode to "precision" on POST /v3/video-translations. As in Speed mode, several output_languages return one ID per language in video_translation_ids.
Poll GET /v3/video-translations/{video_translation_id}, or pass callback_url for a webhook. running here means avatar inference is in progress. Precision mode takes longer than Speed mode, so plan polling intervals accordingly (e.g. every 30–60 seconds for longer videos).

Precision Mode Options

The full parameter list is in the create reference. In Precision mode:
  • speaker_num matters most here: accurate speaker separation directly improves the quality of avatar inference per speaker.
  • translate_audio_only: true skips avatar inference and only dubs audio, which negates the precision benefit.
  • srt supplies a custom subtitle file, and srt_role ("input" or "output") says which video it applies to. Both are Enterprise plan only.
  • brand_glossary_id applies a brand glossary so do-not-translate terms and forced translations are honored, as in Speed mode.
Captions arrive the same way as in Speed mode: see Captions.

Proofread Before Finalizing

Precision mode fully supports the proofread workflow: review and edit subtitles before committing to the full avatar inference render. This is especially valuable in Precision mode since generation takes longer and costs more. Create the session with "mode": "precision"; the remaining steps are identical.

List and Delete Proofread Sessions

GET /v3/video-translations/proofreads lists every proofread session in the workspace, newest first, paged with has_more and next_token. DELETE /v3/video-translations/proofreads/{proofread_id} removes one; it needs a Creator role or higher and cannot be undone.

When to Use Speed vs. Precision

Precision trades processing time for context- and gender-aware translation and high lip-sync quality. It suits faces with significant movement, side angles, or occlusions, and final delivery videos. The full comparison is on the Speed mode page. For high-volume jobs across many source videos, see Video Translation batches.