Daily special video for 21 cents: TTS, one still, one render

A restaurant daily special as a 12-second video on Sume: one still, 180 characters of TTS and a Timeline render. Cost per day and per month, read 2026-10-10.

4 min readSume
All posts

A restaurant can post a daily special video for about $0.21 on Sume: one Nano Banana 2.1 still ($0.10), 180 characters of TTS ($0.0086) and one Timeline render ($0.10). Over a 26-day month that is $5.42.

This works because the specials change daily but the format does not. The still is the only visual, the voice reads the dish and price, and Timeline holds the still over the voice. It does not need a video model or a caption job if the voice carries the price. Rates are from the Sume pricing code (read 2026-10-10).

What is the endpoint chain?

Three writes, run in order only where one needs the other's output.

  • POST /v1/images: a plated dish from your description, or your own photo as the reference. Use 9:16 at 1K.
  • POST /v1/tts-1.0/generate: 'Today's special is the braised short rib with polenta. Twenty-four dollars until the kitchen runs out.' about 180 characters.
  • POST /v1/timeline-1.0/render: audio.url from TTS, one video[] still at 0, and duration_seconds from the TTS result. A 12-second voice is one started minute.
  • Send Idempotency-Key: special-2026-10-10 so a rerun the same day does not bill twice.

What does it cost over a month?

The per-day bill is flat, so extra days are linear.

Daily special cost (read 2026-10-10)
PeriodDaysCost
One day1$0.209
One week7$1.46
26 working days26$5.42
365 days365$76.12

What are the limits?

The model draws food that looks like food, not your food. If a regular must recognize the dish, upload a real photo and reference it. Check GET /v1/images/models for how many references the chosen model takes. Do not ask for the price inside the picture; the voice says it, and a caption cue can print it.

TTS reads digits as text. Write 'twenty-four dollars' in the script instead of '$24' and test one sample; a spoken price is the only part that must be right. For voice tuning, generation_config.speed runs 0.6 to 1.5 and volume 0.5 to 2.

A video is not a menu update. Keep the menu on your own site and use this for the post. For the render pattern see adding AI voiceover to a silent video.

When should I add motion?

Add a 3-second Wan 3.0 clip at 480p for $0.1875 on Fridays, when the specials post is worth more. Put it as slot one and the still as slot two. A weekly clip raises the month by about $0.75; the daily pattern stays the same.

A weekly routine

Keep a short template for the script and swap only the dish and price. The image prompt is the second variable. Everything else in the chain stays fixed, which makes it easy to run from a scheduler.

Because each request has an Idempotency-Key, a scheduler that fires twice by mistake returns the first job instead of making a second charge.

  • Monday to Thursday: still plus voice.
  • Friday: optional 3-second Wan clip.
  • Weekend: reuse the Friday render with a new voice line.

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume