How do you write a script for a short video?

Start from the time budget, not the idea

A short video script is a timed plan: what is said, what is shown and when. On TikTok, Douyin, Reels or YouTube Shorts the viewer decides in the first second or two whether to stay, and the whole video often lasts under a minute. That leaves no room for a warm-up, so the length decides the script before the idea does.

Spoken English in a clear explainer style usually runs at about 2.5 to 3 words per second. Faster creators push past 3; calm narration sits near 2.5. Multiply the target length by that rate and you have your word budget. Then remove about 2 seconds of words, because a short almost always needs a beat of silence for a reveal, a reaction or the last frame. Other languages run at different rates, so time your own read instead of converting word counts. As a rough guide, natural Mandarin narration runs about 4 to 5 characters per second and Japanese about 5 to 6 characters (kanji and kana together).

The math for 30 seconds: 30 × 2.5 = 75 and 30 × 3 = 90. Take away 2 seconds of silence (5 to 6 words) and you land at roughly 70 to 85 words. If your draft has 130 words, it is a 45-second video, not a 30-second one. Do not fix it by talking faster; cut a point.

Check the real number by reading the draft aloud with a timer. For example, cv.cm's word-count tool shows the word count and an estimated speaking time, and its text-to-speech tool reads the text with your device's built-in voice so you can hear the pace before you record or generate anything. Note that YouTube Shorts can now run up to 3 minutes, but a longer limit is not a reason to write longer; write the shortest version that still delivers the payoff.

Video hook formulas with example lines

A video hook is the first line plus the first frame. Its job is to make the viewer ask a question that the rest of the video answers. A hook that promises something the video does not deliver gets swipes later instead of now, so keep it honest. These formulas cover most short-form hooks:

The line is only half of the hook. The first frame matters as much. In our tests with AI-generated shorts, the opening frame decides much of the early drop-off: it should already be the best-looking frame of the video, bright, with the subject's face fully visible and lit, already moving or reacting. Slow reveals, a figure looking down, dim light or a waking-up intro lose viewers before the first line ends. Put the concept into the setting and props instead of a slow opening, and keep a close or medium shot for the first seconds so the subject fills most of the frame.

Five structures for short videos

Choose one structure per video. Mixing two in 30 seconds usually means neither lands. Each structure below lists its beats in order.

Two mechanics help in every structure. First, a pattern break every few seconds: a new shot, a new angle, an on-screen number or a sound. The script should mark where those happen. Second, a loop: if the last line leads back into the first, a viewer who keeps watching flows into the start again without a visible seam, and many viewers watch part of the video twice. A simple loop is to end mid-sentence on a phrase that the hook completes.

A worked 30-second script with a beat sheet

Below is a complete problem–solution script for a 30-second vertical short. Each beat has a timestamp, the voiceover (VO), the on-screen text and the visual. The voiceover is 79 words, which fits the 70 to 85 word budget for 30 seconds with a short pause at the end.

TITLE: Why bread goes stale faster in the fridge
LENGTH: 30 s, 9:16, voiceover + captions

0.0-2.5 s  HOOK
  VO:     You are storing bread in the worst place in your kitchen.
  SCREEN: The fridge = stale bread
  VISUAL: Close-up, hand pulls a loaf out of a bright open fridge, taps it, it sounds hard.

2.5-9.0 s  PROBLEM
  VO:     Cold air does not keep bread fresh. It makes the starch firm up faster,
          so a chilled loaf turns stale in days.
  SCREEN: Cold = faster staling
  VISUAL: Medium shot, loaf cut in half, crumb looks dry, a slice snaps instead of bending.

9.0-19.0 s  FIX
  VO:     Keep the bread you will eat in two days on the counter, in a paper bag or bread box.
          Slice the rest and freeze it.
  SCREEN: 2 days: counter / rest: freezer
  VISUAL: Top shot, bread into a paper bag; cut to slices going into a freezer bag.

19.0-26.5 s  PROOF
  VO:     A frozen slice goes straight into the toaster and tastes close to fresh.
  SCREEN: Freezer -> toaster, no thawing
  VISUAL: Close-up, frozen slice into toaster, pops up golden, butter melts on it.

26.5-30.0 s  PAYOFF + LOOP
  VO:     So next time you reach for the fridge...
  SCREEN: (none)
  VISUAL: Same open fridge as the first frame, hand stops, turns to the bread box.
          Last frame matches the first frame for a clean loop.

Notice what the script does. The hook names a mistake the viewer probably makes. The problem beat gives one reason, not three. The fix is two concrete actions with a number in them. The proof is visual, so it needs few words. The last line stops mid-thought and the image returns to the opening fridge, so the replay starts with the answer to its own question.

From script to shot list and AI video prompts

Script to video AI works best when you convert the script yourself instead of pasting it into a generator. Each beat becomes one or more shots, and each shot becomes one prompt. Our guide to storyboard prompts covers the full method; for short videos, three rules matter most.

1. Write only what a camera would see. Video models tend to render metaphors literally: a line like "his words cut like a knife" can put a real knife in the shot. Keep the spoken script as expressive as you like, but make the visual lines plain: subject, action, material, light, framing.
2. Match shot length to what the model can produce. Many video models accept only fixed durations, such as whole seconds or tiers like 5, 10 or 15 seconds. Plan beats so they can be cut from those lengths. A 6.5-second beat can be trimmed from a 10-second clip; it cannot be made from a 5-second one.
3. Generate the hook shot first and judge it hardest. If the opening frame is not bright, close and already in motion, regenerate it before you spend time on the rest.

Here is the first beat of the worked script written as a video prompt:

Vertical 9:16, close-up. A hand opens a brightly lit white fridge and pulls out a loaf of
sourdough bread, then taps the crust twice with a knuckle. The crust looks dry and hard.
Cool white fridge light on the bread, warm kitchen light behind. The camera is static,
slightly below eye level. Duration 3 seconds. No text on screen.

Keep captions and on-screen text out of the video prompt and add them in the editor. Models often misspell generated text, and you will want to change the wording later anyway. For prompt structure in general, see our guide to video prompts.

Using an AI video script generator without sounding generic

An LLM such as ChatGPT, Claude, Gemini, DeepSeek or Qwen can draft a script in seconds. Asked plainly, it produces the same script everyone else gets: a vague hook, three soft tips and "follow for more". The fix is to give it the material only you have and hard constraints it must obey.

Write a 30-second vertical video script, 70-85 spoken words.
Structure: problem-solution. Beats with timestamps: hook 0-2.5 s, problem, fix, proof, payoff.
For each beat give: VO line, on-screen text (max 6 words), and one visual a camera could film.
Audience: renters with small kitchens.
Facts to use (do not add others): [paste your notes or source]
My own detail to include: [what you tried and what happened]
Rules: no rhetorical questions after the hook, no "in this video", no "follow for more",
no similes in the visual lines. Give 5 alternative hook lines at the end.

• Feed it facts, not a topic. The model fills gaps with plausible guesses; give it your notes or a source and tell it not to add claims.
• Add one personal detail: a number you measured, a mistake you made, a place you filmed. That detail is what makes the script yours.
• Ask for several hooks and choose one yourself. Then rewrite it in the words you would actually say.
• Read the draft aloud. Cut any sentence you would not say to a friend, and any word that exists only to sound impressive.
• Check every fact the model kept. A script read to thousands of people is a bad place for a hallucinated statistic.

Common script mistakes and limits

• Too many words. The most common problem. If the voiceover must be sped up to fit, the script is too long.
• A greeting or intro. "Hey guys, today I want to talk about..." spends the two seconds that decide whether anyone stays.
• A hook the video does not pay off. Viewers who feel tricked swipe away and are less likely to see your next video.
• One beat with three ideas. Each beat should carry one idea and one visual.
• No visual plan. A script with only voiceover forces the editor or the model to invent images, which often produces generic stock-like shots.
• Ignoring retention data. In YouTube Studio, Shorts analytics show how many viewers stayed to watch past the first seconds. Compare that number across hooks before you decide which formula works for your audience.

A script is a starting point. The best short-video scripts are usually rewritten after the first edit, when you can see which lines the visuals already explain and can be cut. If AI-generated people or realistic scenes appear in the final video, add the platform's AI label; our guide to labeling AI content explains the rules on each platform.