The script is the spine. Every animation cue in the finished video is anchored to a word in this take, so a weak script cannot be rescued by motion later.
Rule 1 — write for the EAR, not the page
The reader cannot scroll back. Anything that needs re-reading is lost.
| Don't | Do |
|---|---|
| "Navigate to Settings › General › Storage" | "Buka Settings, General, Storage" — the way a person says it |
| subordinate clauses stacked three deep | one idea per sentence |
| "approximately 23 gigabytes" | "23 giga" |
| reading out a file path or a URL | show it on screen, say the shape of it |
| a paragraph of caveats | one caveat, in the place it matters |
Read it aloud. If you run out of breath, the sentence is too long. If you can't tell where the emphasis goes, neither can the voice model.
Rule 2 — the hook is complete in the first two seconds
Not "starts in" — complete. The viewer must know, inside two seconds, what problem this is about and that it applies to them. Three shapes that work:
- The situation, named precisely. "Storage HP lo penuh padahal udah dihapus semua?"
- The event, in progress. "Your phone just got stolen." — present tense, happening now.
- The cancelled purchase. Name the thing they were about to buy, then remove the need for it.
What fails: throat-clearing ("In this video we'll look at…"), a promise with no stake ("5 tips for…"), and anything that needs a sentence of setup before the point lands.
Then pay it off. If the hook promises N things, the body delivers N things. "Feels rushed" almost never means the pace is too fast — it means the hook promised more than the body showed. Count them.
Rule 3 — length follows content, under 2:59
179 seconds is a hard ceiling, because longer videos get deprioritised for people who don't follow you yet.
Budget at roughly 3 words per second of finished narration:
| Shape | Target | Words |
|---|---|---|
| single-concept explainer | 30–45 s | 90–135 |
| step-by-step / listicle | 90–150 s | 270–450 |
| absolute ceiling | 179 s | ~500 |
Over budget? Cut words, never steps. A tutorial that drops step 4 to fit is a tutorial that fails the viewer at step 4. If the content genuinely needs more than 2:59, it is two videos.
Rule 4 — structure a step-by-step so it can be followed
Each numbered step needs, in this order:
- the number and a two-word name — becomes a title card
- one line of "why" — why this step exists, not just what it does
- the exact path, said the way a person says it
- the payoff — what changes, with a number if you have one
Never invent a number to make arithmetic work. If the steps don't add up to a total, don't state a total — restate the promise instead. A viewer who checks and finds the sum wrong stops believing the rest.
Rule 5 — one take, one voice, no per-scene stitching
Write it as one continuous block of prose. Not a list of scenes, not one line per shot. Per-scene narration recorded separately produces audible seams at every join and a flat, list-like delivery — and it is the single most common reason a finished video sounds like a robot reading a document.
Sentence-level variety is what makes a take sound human: mix a 3-word sentence against a 15-word one, and let punctuation do the work. A question mark and a full stop produce different prosody in every decent voice model; a comma splice produces none.
Rule 6 — end on ONE action
One instruction, one destination. "Save this and do it tonight" beats "follow, share, comment and check the link". If there is a link, say where it is, once.
Output shape
{
"slug": "kebab-case-name",
"mode": "tutorial",
"scenes": [{ "id": "vo", "text": "the ENTIRE narration as one continuous string." }]
}
One scene. One string. The voice step turns it into a take plus word-level timestamps, and those timestamps become every cue in the video — which is why the script is finished before anything is animated.
Before you hand it on
- Read it aloud, timed. Does it fit the budget?
- Does the hook land inside two seconds, and does the body pay off everything it promised?
- Is every number one you can defend?
- Is there exactly one call to action?
- Does any sentence require the viewer to have caught the previous one perfectly? Rewrite it.