Comedy skit music: one bed and two stings with AI music

For a 45-second comedy skit, generate three short pieces: one light, bouncy bed of about 45 seconds and two stings of two to three seconds each, one for the setup and one for the punchline. On Sume that is three Music Router generations at $0.125 each ($0.375) and one Timeline render at $0.10, so $0.475 for the audio and assembly.

Comedy depends on timing, and a sting is only a very short piece of music. The router has no length field, so you ask for that shortness in the prompt. The bed below is the main brief; the two stings use the same seven axes with a two-second length and a single named moment.

The brief to send

The skit bed stays out of the way of the dialogue and keeps a dry, playful tone. Stings are requested separately, each as its own prompt, for example a 2-second brass hit with a descending tuba tail for the fumble, and a 3-second bright pizzicato flourish for the punchline.

curl -X POST https://api.sume.com/v1/music-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: comedy-skit-music-and-stings-one-bed-two-001" \
  -d '{
      "model": "sume/music-auto",
      "prompt": "Playful sitcom background, 120 BPM, C major. Pizzicato strings, muted trumpet, light woodblock, walking upright bass. Cheerful and slightly mischievous, a small stop-and-restart break at 0:25 for the twist, then back to the groove. Dry 1960s TV-comedy production. Instrumental, no vocals. A 45-second track."
    }'

Why each axis is set that way

Getting the length right

Each sting is a separate generation, and the router refuses duration and duration_seconds, so describe the length in words: a 2-second sting, a single hit, ending on a clean stop. Returned files can run longer than asked, and a sting that rings past its beat is easy to trim in the edit by cutting the file where the laugh lands.

In Timeline, the dialogue is the spine. Give the bed soundtrack.url with duck_db so it dips under speech. Place stings in your edit at the exact frame; there is no billing difference, because the render is billed on output minutes, not on the number of audio layers.

Before paying for a render, call POST /v1/timeline-1.0/plan with the same body. It is unbilled and reports the billable minutes, which for this 45-second video is 1. Then send the real POST /v1/timeline-1.0/render with an Idempotency-Key, wait up to 30 seconds in sync mode, and poll GET /v1/jobs/:id/status if it is still queued or processing. Statuses are queued, processing, completed, failed and canceled, and a 429 queue_full means try again shortly.

What it costs

A three-generation score costs more than a single bed, but the pieces are small. Because each accepted generation is a fixed $0.125, a sting costs the same as a full track. Many creators therefore ask for a pack: one bed, two stings, and a spare sting in case the first one does not land. Scaling is simple arithmetic. One finished video here is $0.225 of audio and assembly, so ten of them are $2.25 and fifty are $11.25, before any retries. Failed jobs are not billed as accepted generations, but a regeneration you ask for is a new job, so budget one or two extra takes for a hero clip.

Check before you publish

Write the skit script first, mark the two sting points, and only then generate the music. The timestamps give the prompt something to anchor on.

Store each prompt beside the cut; a series of skits works best when the stings repeat as a recognisable signature.

• Test each sting against the real laugh point; a sting that arrives a beat early kills the joke.
• Keep the bed quieter than you think; dialogue clarity matters more than groove.
• Use a unique Idempotency-Key per piece so a retry does not create a duplicate generation.
• Confirm usage terms for generated music with Sume before posting paid comedy content.