Retry one scene and keep the voice track: Format previous_run_id

How do you change one scene of a finished AI video without losing the voice-over? Start a new Format run with previous_run_id set to the finished run and an instruction that names the scene and says what to keep. The new run continues from the old one instead of starting from nothing.

New voices arrive constantly, such as ElevenLabs' Eleven v4 (read 2026-10-04), and it is tempting to redo everything to try one. A targeted retry keeps the approved narration and spends credits only on the scene that changed.

The request

The Format runs guide shows the shape: you post the same Format run create call with previous_run_id and a new instruction. An example instruction in the docs is "Retry the selected scene only. Keep every other scene and the voice track unchanged." Be as specific about the scene as you can: its number or the line it speaks.

When a continue is refused

The run is checked before work starts, so a refusal costs nothing.

Habits that make retries cheap

Wait for the earlier run to reach a terminal state before you continue it. Keep one instruction per retry, so you can tell which change produced which result. If the voice itself is what you want to change, say so explicitly, because the default instruction above tells the run to leave it alone.

For a line-level change inside a scene, regenerate that line's audio as a TTS job and re-time the scene from its word timestamps, as in cut a voice-over into sentence clips.