Vertical Short Drama Script Format (2026): Episode Length, Word Counts, Hook Structure and AI-Generation Constraints

Three ways a vertical script differs from a feature or TV script

1. Scene density. A feature can spend three pages on one scene; a 90-second episode holds 2–4 scenes and keeps only the one or two beats that move the conflict.
2. Line length. Single lines of dialogue stay under about 20 words. A phone caption line holds 6–8 words; a long line wraps to three rows and the viewer reads captions instead of watching the face.
3. The hook is per episode. TV relies on one cliffhanger per hour; vertical drama needs one per episode, and the three hook types (identity, emotion, realisation) have to rotate. Two identical hooks in a row feel like a trick.

2026 length table

The conversion rests on one number: a typical spoken pace of about 3 words per second. A 90-second episode is roughly 270 words of dialogue and voice-over; action lines do not count. Work the formula backwards after every draft instead of guessing the running time.

Scene layout: an example you can copy

There is no single mandated template, but the layout below, organised by episode and scene, is what studios read fastest and what script-analysis tools split most reliably:

EPISODE 3

SCENE 1 · INT. VALE FAMILY CHAPEL – NIGHT
[ACTION] Claire kneels on the cushion, a half-burned marriage contract clenched in her hand. The door opens; the draft bends the candle flames.
MR. VALE (low): Did you burn the contract?
CLAIRE (not looking up): I did.
MR. VALE: Then you can explain it to the Carters yourself.
[ACTION] She lifts her head. No tears.

SCENE 2 · EXT. VALE HOUSE, FRONT GATE – NIGHT
[ACTION] Ethan Carter's car idles at the gate, headlights on her.
ETHAN (through the window): Get in. The wedding goes ahead.
[HOOK · realisation] On the passenger seat: the other half of the contract, intact.

• Scene heading in a fixed order: scene number, INT./EXT., location, time. Keep one spelling of every location for the whole season ("Vale Family Chapel", never "the chapel" one page and "Vale chapel" the next); it is what scene reference images and consistency are built on later.
• Action lines marked [ACTION] or indented, only what can be seen or heard, no interior monologue, under about 40 words each.
• Dialogue as CHARACTER (tone): line. Tone goes in parentheses; no "she said".
• Hook labelled with its type at the end of the episode so you can check the rotation yourself.

Hook structure: 5 seconds, 30 seconds, episode end

Three beats

• Conflict in the first 5 seconds. The first line or the first action is the conflict itself, not the setting. "Did you burn the contract?" beats "Morning at the Vale house."
• A turn about every 30 seconds. A 90-second episode turns at least twice: new information, a shifted allegiance, or escalated danger.
• The episode-end hook, rotating. Identity (someone is not who they said), emotion (a confession, refusal or rupture left hanging), realisation (viewer and lead discover an earlier judgement was wrong).

First-three-episodes beat sheet

The logic: free-to-watch feeds decide on completion of the first three episodes, paid apps convert at the paywall around episodes 8–12, and both need the first win no later than episode 5.

Submitting to a studio or app

• ☐ Episode count and length match the studio's current call; most requests specify a full-season outline plus the first 10–20 episodes in full.
• ☐ Every episode carries a labelled hook; the three types rotate.
• ☐ One spelling per location and per character name across the whole season.
• ☐ Single dialogue lines under about 20 words; action lines under about 40.
• ☐ Genre fits what the app is buying this quarter (billionaire romance, revenge, werewolf and second-chance stories have dominated the English-language catalogues; check the front page of the app you are pitching on the day you submit).
• ☐ Rights are clean: adaptation rights for any source novel, no real-person likenesses.

Extra constraints when the series is AI-generated

When the script feeds an AI video model instead of a crew, format is not the only test; the other is "can the model produce this picture". Patterns confirmed across a large volume of finished clips:

• Keep each episode to three or four physical locations: every extra set multiplies the cross-shot consistency problem.
• Convert dialogue at 3 words per second: a 15-second segment holds at most about 45 words; more than that means another segment.
• Write causes, not effects, in action lines: "a pendant lamp throws warm-white light from upper right", not "light brightens her forehead". It is a prompt-level rule, but building the habit at script stage saves storyboard rewrites.

A script in this layout goes into SceneMixer's analysis without reformatting: the "short drama" content type applies a per-episode running-time cap, rotates the end hooks between identity, emotion and realisation, and splits out characters, sets, props and the episode structure directly. How a scene then becomes a 15-second segment is covered in the continuity section of Character Consistency.

FAQ