AI Short-Drama Storyboard Template: Writing a 15-Second Segment a Video Model Will Execute

SinCoSphere EditorialUpdated 2026-09-125 min readGuide

A storyboard a 2026 video model can execute has three parts: a segment header (location, time, opening state with each person placed by a scene anchor), 3–5 timestamped shots ("Shot 2 [3.5–8.0 s]: framing / who is in frame and what they do / Speaker (voice hint): "line" / light source and direction"), and a closing beat — a small action after the last line so the model does not fill spare seconds by repeating dialogue. Size dialogue at about 2.5 words per second plus a beat; write light and motion as causes, not results; name nobody who is outside the frame. Wan and Veo both document timestamped multi-shot prompts.

Segment and shot: the two units you are writing

A segment is one video-model call; a shot is a framing change inside it (definitions). Write long segments with short shots: joins between segments are where continuity breaks, while cuts inside one call are rendered with the same references loaded. The model sets the segment length, so choose it first:

Model	Seconds per call	Vendor guidance on shots
Wan 3.0	2–30	"Shot 1 [0–3 s]" timestamps; 4–6 s per shot
MiniMax H3	4–15	Shots written in the prompt, up to 7,000 characters per text item
Kling 3.0	3–15	Omni guide examples with up to five shots in 15 s
Veo 3.1	4, 6 or 8 (8 with reference images)	[00:00-00:02] timestamp prompting
Vidu (viduq3)	3–16	Subjects cited as @name

Vendor guidance points the same way: Wan suggests 4–6 seconds per shot and Kling's examples fit up to five shots into 15 seconds, so 3–5 shots per 15-second segment is a reasonable starting range. Test denser cutting on your model before you rely on it.

The template, field by field

Field	What to write	Why
Location and time	Place, time of day, weather	Keeps the scene plate and light consistent
Opening state	Framing; who is where, by scene anchor; what they hold; which side of whom (the axis)	The model draws this first; carry-over from the previous segment goes here
Shot N [start–end s]	A timestamp range	Wan and Veo both document timestamped shots
Framing and camera	Wide, medium or close; lens height; at most one camera move	One move per shot keeps motion readable
In frame	Names of the people visible	A named absent person tends to be drawn
Action	Cause and contact: what moves, what it touches, where it ends	A result written into the action becomes the start state
Dialogue	Speaker (voice hint): "line"	Quotation marks mark speech in Wan's and Google's guides
Sound	Ambience and effects, or "No dialogue." / "No background music."	Wan documents these control phrases
Light	Source, direction, colour temperature	A landing spot is painted as a bright patch
Closing beat	A small action after the final line	Spare seconds are otherwise filled by repeating the line

Worked example: one 15-second segment

Location: Sam's diner, counter — night, rain on the windows. Opening state: medium shot; Nora sits at the end of the counter by the window, both hands around a white mug; Sam stands behind the counter by the coffee machine, a towel in his hands. Axis: Nora on the window side, Sam on the kitchen side; every camera stays on the customer side of the counter.

Shot	Time	Framing	In frame and action	Dialogue and sound	Light
1	0.0–3.5 s	Medium, lens at counter height	Sam and Nora. Sam drops the towel over his shoulder and walks along the counter toward Nora.	No dialogue. Rain on the glass.	One warm-white pendant above the counter, soft from above; cool light from the window side.
2	3.5–8.0 s	Close-up, Nora	Nora. Her thumbs tighten on the mug.	Nora (low, slow): "I waited three years for this." Lips close after the last word.	Same pendant; window light on the far side of her face.
3	8.0–11.0 s	Over Nora's shoulder onto Sam	Sam; Nora's shoulder in the foreground. Sam stops across the counter, eyes on the mug.	No dialogue.	Same.
4	11.0–15.0 s	Close-up, Sam	Sam.	Sam (quiet): "It's still hot." Then he sets both palms flat on the counter.	Same.

Check the arithmetic: six words in 4.5 seconds and three words in 4 seconds both fit the timing table below with room for the closing action. Every spoken line ends with the lips closing and one small action; every silent shot says "No dialogue." Nora's face in shot 3 is not described, because she is seen from behind.

Dialogue timing table

Words in the line	Speech at 2.5 words/s	Shot length to write
1–3	about 1.2 s	2.5 s
4–5	about 2 s	3.5 s
6–7	about 2.8 s	4.5 s
8–9	about 3.6 s	6 s
10 or more	4 s or more	Split into two shots

The rate comes from a commonly cited average of about 150 words per minute for conversational English (a National Center for Voice and Speech figure, cited by Baruch College); the extra second or so covers the breath before the line and the reaction after it. Add about 0.75 s for each additional sentence in the same line. Too many seconds and the model repeats or stretches the line; too few and the last words land on the next shot. If you burn in captions, check reading time as well: Netflix's English style guide caps adult subtitles at 20 characters per second and 42 characters per line, so a 4.5-second shot can hold one full two-line subtitle (up to 84 characters) inside the reading-speed limit. More: dialogue timing.

Adjust the template to the model

Wan 3.0: use its multi-shot form — an overall description, then "Shot 1 [0–3 s]" and the shot content; cite references as Image 1, Video 1, Audio 1. If your API call enables automatic prompt rewriting (prompt_extend), switch it off for storyboards so it does not rewrite cut points (prompt extend).
Veo 3.1: clips are 8 seconds when reference images are used; Google's own example fits four 2-second shots without dialogue into one clip, while lines timed with the table above leave room for two or three shots. Google's guide uses [00:00-00:02]-style timestamps, quotation marks for speech, "SFX:" and "Ambient noise:" labels, and describes exclusions positively ("a street with no cars") rather than as a bare "no X".
MiniMax H3: up to 15 seconds and 7,000 characters per text item; a first frame cannot be combined with reference images.
Kling 3.0: up to 15 seconds; Custom Multi-Shot lets you set each shot's content and duration. Native audio is an option priced per second (9 credits against 6 without it at 720p), so decide per segment whether generated sound is needed.
Sora 2: not covered here. OpenAI's deprecations page schedules the Sora 2 models and the Videos API for removal on 2026-09-24, with no replacement named; move storyboards written for it to one of the models above.

The general rules behind these adjustments are in the prompt-writing explainer.

Eight contradictions to delete before rendering

"Takes two steps" and "feet planted" in the same shot.
A facial expression asked of someone speaking toward the back of the frame; the model turns them to camera.
Light written as a landing spot ("light on her forehead"); you get a bright patch there.
Leftover image-model tokens such as "volumetric light", "8K" or "masterpiece"; write the physical cause instead ("a layer of smoke hangs in the air").
The result inside the action shot ("is hit, falls, spits blood"); the end state becomes the start state.
Large rigid-body motion (a door kicked open, a table flipped) finished inside one short shot with a person; give it its own shot with a start and an end state.
Facial features for the person seen from behind in an over-the-shoulder shot; an extra face appears.
Naming who "exits frame" or "is not present".

Pre-render checklist

Shot times add up to a segment length the model supports.
Every line has a speaker, a voice hint and seconds from the table.
Every silent shot says "No dialogue."
The opening state matches the previous segment's end state.
Names appear for visible people and nobody else.
Each person keeps one look for the whole segment.
Light is a source with a direction and a colour temperature.
The reference files listed match the people and places in the text.
The last beat after the final line is a small action, not a new line.

Keeping the cast stable across segments is covered in the character-consistency checklist.

Key figures and sources

Figure	Value	Source	Date
Average conversational speaking rate (US English)	About 150 words per minute (National Center for Voice and Speech figure; second-hand, via Baruch College)	tfcs.baruch.cuny.edu	2026-09-12
Netflix English subtitle limits	42 characters per line, 2 lines, up to 20 characters per second (adult)	partnerhelp.netflixstudios.com	2026-09-12
Wan 3.0 multi-shot format	2–30 s per call; 4–6 s per shot with timestamp annotations	www.alibabacloud.com	2026-09-12
Veo 3.1 clip lengths	4, 6 or 8 s; 8 s required with reference images or 1080p/4K	ai.google.dev	2026-09-12
MiniMax H3 limits	4–15 s per call; up to 7,000 characters per text item	platform.minimax.io	2026-09-12
Kling VIDEO 3.0 duration	3–15 s per generation	kling.ai	2026-09-12

FAQ

How many shots should one segment have?

Start with 3–5 shots per 15-second segment. Alibaba's Wan guide suggests 4–6 seconds per shot and Kling's Omni examples fit up to five shots into 15 seconds; denser cutting needs testing on your model before you rely on it.

Can a segment open on a silent establishing shot?

Yes, but keep it short and write "No dialogue." explicitly. A long silent opening gives the model room to start speech early or to shift the voice track; if you need a long establishing beat, make it its own segment.

What about languages other than English?

Speech rates differ by language and by speaker. Time a native speaker reading the lines aloud and rebuild the table from that rate; for captions, use the reading-speed limits of that language's subtitle style guide rather than the English figures.

Should the storyboard be prose or structured data?

Whatever your tool expects. What matters is one shot per line or row, timestamps that add up, and the same field order every time, so that you and the model can find the speaker, the action and the light at a glance.

Where does camera movement go?

In the framing field, one move per shot (push in, pan, follow). Stacking moves in a short shot usually produces drift, and a move written into the action line is often read as character motion.

Sources

Storyboard script Prompt-writing explainer Character consistency Storyboard tools