Timeline audio concat or audio.parts: when the $0.01 job is wasted

If you need to join audio for one Timeline render, use audio.parts[] inside the render request, which adds no job and no charge beyond the $0.10 per started minute you already pay. Run the $0.01 Timeline audio concat job only when you need the joined audio as a reusable file, for example as an Avatar 1.0 audio input or for several renders.

Both routes are described on the Sume docs pages, and the Timeline audio page itself says: if you need a join only inside one render, use audio.parts[] and do not use the concat job. This post is the decision table, with the cent counted.

What each route gives you

Both do a sample-domain join with no silence at the seams and no re-synthesis. The differences are in what you get back and what you pay.

The cent that matters

The concat job costs one cent, so the money is not the issue. The issue is extra steps: a job to poll, a file to store, and an Idempotency-Key to manage. For a one-off render, the inline route has fewer moving parts. For a campaign of 24 renders that share the same 6-part voice and bed join, the concat route costs $0.01 once instead of describing 6 parts 24 times.

The returned segments[] are the useful part. They give the concat offsets you use to re-base video[].start in the render, so each clip begins where its audio does. With inline parts you do that arithmetic yourself.

A worked case

Take a 40-second Short with a 12-second intro bed, a 20-second main bed and an 8-second outro bed, all Lyria files you already hold. Inline, you list three parts in audio.parts[] with duration 12, 20 and 8, and the render is $0.10. With the concat job, you pay $0.01, wait for the job, then pass the returned audio_url as audio.url, for a total of $0.11 and one more call. The extra cent buys the file, so it is only worth it if a second render, an Avatar 1.0 clip or another market version will use the same join.

Another reason to prefer the job is auditing. Its result lists segments[] with the offset of each part, so a reviewer can see where each bed starts without opening the render.

Two limits to remember

The inline route requires that the declared part lengths cover the render: the Timeline page lists audio_parts_shorter_than_duration as the error when the sum is less than audio.duration_seconds. The concat route requires that every part share a channel layout (audio_parts_channel_mismatch), so check the layouts of mixed inputs before you submit.

A last case: a Lyria bed and a spoken line. Join them with neither route if you want the bed under the voice. Use the render's soundtrack field with duck_db (0 to 20), which is made for that, and keep the voice as the spine.