Voiceover-only Short: silent B-roll plus a TTS spine in Timeline

A voiceover-only Short is easiest to build narration first: generate the speech, measure it, then lay clips over it. In Sume the narration file becomes audio.url with duration_seconds set to its real length, and every clip is a video slot, so picture follows voice and cannot run past it.

YouTube's editor accepts voiceover in its timeline per Enhance your Shorts (read 2026-10-03). The render rules are from the Timeline 1.0 doc.

What order should I work in?

Voice first, because it fixes the length. Then pick clip count from that length, and send the plan call before you pay. The steps:

• Write the script and generate the narration; note its exact duration.
• Cut the slots: for a 24-second voice and six beats, about 4 seconds a slot, each at least 0.2.
• Use fit: cover for clips that are already vertical and blur for landscape clips.
• Call /v1/timeline-1.0/plan (unbilled) and read billable_minutes.
• Render with an Idempotency-Key.

What if the clips are shorter than the voice?

The Timeline doc says video coverage may trail the narration by at most 0.5 seconds, so plan slots that add up to the voice length. A gap between two slots is filled by holding the previous frame and returns timeline_gap_filled; when the clips end before the audio, the last frame is held and the result carries video_coverage_shorter_than_audio. Both are soft warnings in the compiler, not failures. Read warnings[] on the result instead of assuming the picture covers the narration.

How much does it cost?

Timeline is $0.10 per ceil output minute, so a Short up to 60 seconds is $0.10 for the render. The narration and the clips are separate lines: check the live catalog for the speech model you use and for each video model, because they change.

What Sume does not do

Sume does not choose clips from a script or sync cuts to words on its own. You decide each slot's start and duration.