Restaurant menu video music: a 30-second brief and cost

For a 30-second restaurant menu video, ask for a warm, mid-tempo instrumental of about half a minute, written to end cleanly where your last dish card ends. On Sume that is one Music Router generation at $0.125 and a Timeline render at $0.10, so the audio side of the clip costs $0.225.

Menu videos are short and silent-friendly, since many viewers watch without sound. That makes the bed a seasoning rather than the meal: it should lift appetite and pace, not dominate. The brief below pins the tempo to the cut rate of the dish cards and ends on a soft resolve so the clip does not stop mid-phrase.

The brief to send

The tempo is the part to match to your edit. With six dish cards in 30 seconds, one card every five seconds, a 96 BPM bed lands a beat group on each cut. The request below names the key moment, a small percussion lift when the signature dish appears at 0:15.

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: restaurant-menu-video-background-music-b-001" \
  -d '{
      "model": "sume/music-auto",
      "prompt": "Warm bistro background music, 96 BPM, F major. Nylon guitar, upright bass, soft brushed drums, a light accordion line. Relaxed and inviting, a small percussion lift at 0:15 for the signature dish, resolving cleanly at 0:30. Vintage-leaning, close and dry. Instrumental, no vocals. A 30-second track."
    }'

Why each axis is set that way

Getting the length right

Length lives in the prompt. Music Router has no duration parameter, and Sume rejects duration and duration_seconds, so write the 30 seconds into the request. A returned track that runs a few seconds long is handled in the render: audio.duration_seconds sets the film length, and soundtrack.fade_out_seconds (up to 10) fades the tail.

If the voice is silent, set audio.mode to silence with the declared length. That mode takes no spine file, which means duck_db is refused, so keep the bed at a sensible gain_db yourself. If a voice reads the prices, use a real spine and duck the bed 6 to 12 dB as a starting point.

What it costs

A 30-second clip is one started minute, so the render is a flat $0.10. Four menu variants for four locations cost four generations and four renders, $0.90 in all, because every generation is a separate accepted job.

Check before you publish

Run the menu clip with sound off first. If it works silent, the bed is a bonus and any acceptable track will do. If sound is the pitch, regenerate until the lift at 0:15 coincides with the hero dish.

Store the prompt with the menu version so the next seasonal edit reuses the same brief.

A menu changes often, so treat the brief as a template: swap the named moment and the dish, keep the tempo and instruments, and the family of clips will sound like one restaurant. Each swap is one more accepted generation at $0.125, which is small next to the time spent on a photo shoot.

Because the bed is fixed-price, there is no reason to settle for the first take. Generate two or three candidates for the hero clip, compare them on a phone speaker, and keep the one whose lift lands on the signature dish.

• Listen on a phone speaker; restaurant clips play there, not on studio monitors.
• If the venue plays licensed music in-store, keep the clip bed separate from the premises music.
• Send a stable Idempotency-Key per variant so a retry cannot bill twice.
• Check the usage terms for generated audio with Sume before the clip goes in a paid placement.