Black Friday audio ad read: a TTS voice over a music bed

A 30-second ad read is about 400 to 450 characters of script, a voice take from Sume TTS, a music bed from the music router and one Timeline 1.0 render to mix them. Black Friday is 27 November 2026, so there are weeks to test versions. Each version costs about a quarter, which makes A/B copy tests cheap.

Budget the script

At a normal ad pace of about 2.5 words a second, 30 seconds is 75 words, about 450 characters with spaces. These are working assumptions; generate once and measure the actual duration, then trim the script until it fits. Sume prices TTS at $0.0475 per 1,000 characters, so a 450-character read costs about $0.021.

The three jobs

Music bed prompt

Describe a bed that stays out of the voice's way: mid-tempo, sparse, no vocals. Ask for a length in the prompt, since there is no duration field. Example: "Upbeat retail pop, 108 BPM, C major, plucked synth bass, soft claps, bright keys. A 40-second track, builds at 0:20 into a short stinger. Instrumental, no vocals." Over-generate length slightly and let fade_out_seconds end the mix cleanly.

Mix

In Timeline 1.0, put the voice take on the audio spine, the bed on soundtrack, duck_db between 6 and 12, and fade_out_seconds of 1 to 2 seconds. Plan first with the unbilled POST /v1/timeline-1.0/plan. Detail in Timeline 1.0 and the music docs.

Limits

Ad claims, discounts and dates are your responsibility; generate only the copy you have approved. The ducking depth is a target against a voice at speech level, so a quiet take will duck less than you expect. Music is not guaranteed to land its stinger on a time you ask for. For a looped version, see looping a bed under a long video.