How much narration per product in a gift guide video, and its cost?

Write about 150 characters of narration per product in a gift guide video, plus roughly 300 characters for an intro and an outro. Eight products come to 1,500 characters, which Sume TTS bills at $0.08 as one job (the rate is $0.0475 per 1,000 characters, rounded up to the cent). Eight separate jobs, one per product, cost $0.08 plus $0.01 to join them: $0.09. The money is trivial. The real choice is how you get each product's voice to line up with its picture.

TTS is metered by characters, and spaces and punctuation count, so the script is the price list. A product line of 150 characters is roughly two short sentences: what it is, and who it is for.

Write each product to a fixed shape

Give every product the same two beats. Beat one names it and says the one thing it does. Beat two says who should get it or why it is a good gift. 'A cast-iron skillet that goes from stovetop to table. For the friend who hosts every Sunday.' is 91 characters; add a price phrase and you are near 150. Equal shapes make equal lengths, and equal lengths make the picture cuts feel steady.

Do not read prices as digits if you can avoid it. Spoken numbers cost characters and are easy to mishear. Show the price on screen with caption cues, which are exact strings you control, and keep the voice for the reason to buy.

One job or one job per product

A single TTS job is the cheapest to run and gives you one file, but you then have to find where each product starts in it. Request sentence segments, or measure the file and time it by hand. One job per product costs a cent each at this length (150 characters is $0.01 after rounding up), and each file's length is exactly that product's slot. Join them with Timeline 1.0 audio, a flat $0.01 job that returns segments[] with each part's start offset, and use those offsets as the start of each video[] slot.

Because every job rounds up to the cent, short per-product jobs cost slightly more than one long job once the guide is long. At 12 products the split route is 3 cents more. At five or six products the difference is a cent or two. Pay it when you want exact slots, or when a retake of a single product should cost one cent and not a whole re-read.

curl -X POST https://api.sume.com/v1/timeline-1.0/audio \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: giftguide-voice-join-1" \
  -d '{
    "operation": "concat",
    "parts": [
      { "url": "https://media.sume.com/artifacts/artf_demo/intro.wav" },
      { "url": "https://media.sume.com/artifacts/artf_demo/p1.wav" },
      { "url": "https://media.sume.com/artifacts/artf_demo/p2.wav" }
    ],
    "output": { "format": "wav" }
  }'

What it costs by guide size

• A TTS job is never below $0.01, so very short lines are billed at the minimum.
• Render the finished guide with Timeline 1.0 at $0.10 per started minute; a 61 to 120 second guide is two started minutes, $0.20.
• Add a music bed for $0.125 and duck it under the voice with duck_db.

Match the voice to the picture

Plan the cut from the script. Eight products at about 8 to 10 seconds each is a guide of 70 to 80 seconds, which is two started render minutes. If you want it under one minute, drop to six products or cut each line to about 110 characters. Because the rate is per character, a shorter script is also the cheaper one, but the saving is a few cents; the saving that matters is the viewer's attention.

Put the strongest product first and the best budget pick last. People skip forward, and the first and last items get the most plays. Write that order into the script before you generate any audio.

Check the read before you render

Listen to the joined audio once without the pictures. Wrong stress on a product name is the usual fault. Fix it by respelling the name in the script for that one job and re-run it; with per-product jobs the retake is a cent, plus the join and the render. Keep the original wav files: wav keeps the joins sample-exact, while mp3 adds a little encoder padding at each edge.

Then match the pictures. Each product needs a slot at least 0.2 seconds long, and the first slot must start at 0. If the voice runs a second past a product, extend that slot's duration rather than cutting a word.