A video can look polished and still lose viewers because one line takes too long to get to the point. Start with the viewer, write one clear message, then shape the script around what they’ll hear and see.
These seven steps work for social clips, product ads, and longer tutorials, including videos made with AI tools.
Step 1: Define Your Video’s Purpose and Audience
Before you write a video script, decide what one thing the viewer should understand or do. This choice sets the angle, the detail level, and the ending.
Write a one-sentence brief. Try: “This video helps [specific viewer] solve [specific problem] by showing [one useful thing].” For a small shop, that could be: “This video helps people choosing a desk lamp see how its adjustable arm works.” It’s a clearer start than “We need a product video.”
Next, name the audience as a real group with a real need. “Everyone who likes home decor” is too broad to guide your words. “People working at a small desk who need focused light” gives you a better sense of what to show and what terms to use.
Pick one goal for the video. It might teach a task, explain a product, or ask viewers to take one next step. If the goal is a sale, don’t spend the opening on your company history. Show the problem your item addresses, then make the next action clear.
A useful starting order is purpose, format, and an outline. Use that order before polishing lines. It keeps the script tied to the job the video needs to do.
For an AI-generated video, write down what must stay true on screen. Note the product’s shape or color, the setting, and any character details the scene depends on. That gives you a check list when you review the generated shots, rather than relying on a vague prompt such as “make it look nice.”
Milestone: You should have a one-line audience, one main message, and one goal for the ending.
Step 2: Build a Hook, Main Message, and Call to Action
A clear video script has three parts: a hook, the main message, and a call to action. Put the viewer’s payoff first, then earn the ending by delivering what the opening promised.
Write the hook as one direct line. It can name a problem, show a result, or ask a question the video answers. For example: “Does your product photo look flat on a phone screen?” is more specific than “Want better content?” The first line should make sense even if a viewer has never seen your brand.
For a short social video, treat the first three seconds as a useful target for the hook, not a law. Then frame the problem early, around the first few seconds, and move through the proof in short beats. Hooked.so recommends beats of roughly four to seven seconds; use them as a drafting aid, then test your own audience’s viewing data.
Build the body around one message. Give the viewer a reason to believe it. That might be a close-up of a product detail, a quick demonstration, or a before-and-after that you can support. Don’t stack extra claims just because they fit in the caption. If a line needs evidence you don’t have, cut it or make it a question to investigate.
End with one action that matches the goal. “See the full details” suits a product explainer. “Save this for your next shoot” may fit a tip video. “Follow for more” is easy to type, but it’s weak when it doesn’t tell the viewer what they’ll get by following.
For a spoken character line in a generated clip, keep the dialogue short enough to fit the shot. Wan Studio’s guide to making a picture talk shows how a photo and written dialogue can become a talking-image clip. Treat the result as a draft to check, especially for mouth movement and timing.
Milestone: Read the hook, message, and CTA aloud. Each should sound like it belongs in the same video.
Step 3: Turn Your Ideas Into a Speakable Outline
A speakable outline gives you enough structure to stay on track without forcing every sentence. Start with the hook, then make a short list of points the viewer needs to follow.
For a tutorial, order the points as actions: show the item, explain the setting, demonstrate the task, then point out the result. For a story-led short, use a simple arc: show the situation, introduce a snag, then resolve it. You don’t need a dramatic tale. A customer trying to fit a lamp on a crowded desk can supply a small, clear story.
Put one idea in each bullet. If a bullet needs several sentences to explain, split it into smaller beats. Add a transition only when the viewer needs help seeing why the next point follows. “Now that the surface is clean, here’s how to keep it that way” feels more natural than a string of disconnected headings.
Choose between a full script and bullet points based on the speaker and the shot. A word-for-word draft can help a new presenter or a tightly timed ad. Bullets may sound more relaxed when you’re comfortable speaking to camera. You can also mix them: write the hook and product claims word for word, then outline the rest.
Draft in a shared document if other people need to review it. Google Docs is one simple option for a team that wants comments in one place. Keep the current draft easy to find, and label major revisions with a date or version name so the editor doesn’t cut from an old copy.
AI can help you brainstorm alternate hooks or spot sections that feel out of order. Don’t accept a generated script as fact. Check each claim, remove generic language, and make sure the lines sound like something your speaker would actually say.
When a scene depends on a steady character or a specific visual reference, plan the details in the outline instead of hoping they stay consistent. You can use Wan Studio’s Wan 2.7 model page to review its reference-driven workflow and model-specific options before you write prompts for a multi-shot scene.
Milestone: Your outline should show the order of ideas, the proof each point needs, and any lines that must stay exact.
Step 4: Calculate Script Length for Your Target Runtime
Match the script to the time you have, then leave room for pauses, visual beats, and edits. A page count alone won’t tell you how long a video will feel.
Use a simple estimate: choose a speaking pace, multiply it by the number of minutes, then treat the result as a draft range. For clear narration, start around 130 to 150 spoken words per minute as a planning assumption, not a guaranteed pace. At that rate, a one-minute voice track might hold about 130 to 150 words before pauses and visual moments.
Now subtract time for action. If a product demo needs five seconds to show a lid opening, those seconds aren’t available for narration. Leave a beat after a key result so viewers can see it. A short-form video with a quick visual reveal may need fewer spoken words than a talking-head clip of the same length.
Here’s a workable calculation:
- Set the intended runtime in seconds.
- Estimate the seconds used by pauses, demonstrations, or text on screen.
- Convert the remaining time to minutes.
- Multiply by your planned speaking pace to get a rough word budget.
For example, a 30-second video with six seconds reserved for visual action has 24 seconds of possible speech. At 140 words per minute, that’s about 56 spoken words. Read the draft at a natural pace before you lock it. The estimate helps you start; the timed run tells you what fits.
Short-form AI clips may be generated as separate shots rather than one continuous scene. Wan 2.5, for example, supports five- or ten-second clips, with synchronized sound and dialogue listed for that model. Plan the script in segments that fit the output you intend to make, then check the current model details before production.
If the script runs long, don’t speed up every line. Remove one supporting idea, shorten the setup, or let a visual carry information that doesn’t need narration. Keep the words that help the viewer understand the point.
Milestone: You should have a word budget and a rough time map for spoken lines, pauses, and action.
Step 5: Adapt the Script to Your Video Format
The format decides how much context the script needs and how the information should unfold. Keep the core message, but change the pace and detail for the screen where it will appear.
Short-form social video
For a TikTok, Reel, or Short, start with the most useful or surprising point. Don’t spend the opening introducing yourself unless your identity is part of the reason to trust the advice. Give each beat one job: name the problem, show evidence, then land the payoff. Use on-screen text to support the spoken line, not repeat a full paragraph.
A product-photo video can work as a small sequence: open on the product, show one useful detail, then end on the result or action. If the image becomes a moving scene, make sure the movement helps explain the product rather than distracting from it.
Product ad or promotional clip
Write the claim you can prove. Then decide what shot would make it clear. A skincare demo might show how a container opens or how a texture spreads, but it shouldn’t promise a result the business hasn’t substantiated. Keep the CTA tied to the viewer’s next step, such as viewing details or visiting a product page.
If you’re planning content with a hired creator, align the brief before filming, including expectations for delivery and revisions. Keep the script flexible enough for the creator’s natural voice.
Tutorial or video essay
A tutorial needs clear steps and enough time for viewers to see each one. Put the result near the start, then explain the method in order. A video essay can use a story arc or a central question, but each section should add a new point rather than restate the opening.
For all formats, write for the sound of the sentence. Replace long terms with plain words when the meaning stays intact. Read the lines out loud and cut phrases that look fine on screen but feel stiff in speech.
Milestone: You should know what changes for this format: the opening, the amount of context, and the kind of proof the viewer sees.
Step 6: Add Visual, Voice-Over, and Caption Cues
Use a two-column script to keep the spoken track and the picture in sync. Put dialogue or voice-over in one column, then describe the shot, action, or text on screen beside it.
Keep each row to one moment. For example:
| Audio or dialogue | Visual cue |
|---|---|
| “Need more light on your work?” | Open on a dim desk; show the lamp turning on. |
| “Move the arm to aim the beam.” | Close-up of the arm shifting over the page. |
| “See the size and finish online.” | Show the full product, then display the next action. |
The table is a planning sketch, not a promise about how a generated shot will turn out. Review the final video for timing, visual accuracy, and mouth movement. Wan Studio can generate clips with sound and dialogue using specific Wan models; Wan 2.5 and Wan 2.7 have their own supported durations and inputs, so don’t assume every model behaves the same way.
Mark voice-over separately from on-camera speech. Add cues such as “VO over close-up” or “speaker on camera” so the editor knows whether the line needs visible lip movement. If the line is meant to be spoken by a character, put the exact words in quotation marks in your prompt and check the generated lip sync. It can vary, so don’t skip review.
Plan captions while writing. Keep each caption brief enough to read at a glance, and don’t let it cover the product detail or face the viewer needs to see. Caption words should match the final spoken audio, including names and product terms.
Captions represent speech and relevant sounds, and synchronized captions should match the audio. Manually check captions and transcripts. Review the export with sound off, then listen with the screen out of view to catch gaps in the narration.
Some visual details may not be clear from dialogue alone. If viewers need a key action to understand the clip, describe it in narration or plan an audio description. This matters for a demonstration where the important step happens silently on screen.
Step 7: Rehearse, Edit, and Collaborate on the Final Draft
Read the script aloud, time a full run, then revise the lines that slow the message down. The page can’t tell you where the speaker will stumble.
Do one read without stopping. Mark words that feel awkward, lines that need a second breath, and places where the meaning isn’t clear on the first listen. Then record a rough run on a phone or webcam. You don’t need a final set to hear whether the hook lands or whether the product detail takes too long.
Cut filler that doesn’t help the viewer. That may mean removing a long greeting, a repeated point, or a company introduction that belongs elsewhere. Keep a pause when it lets the viewer take in a visual result. Tight writing isn’t the same as fast talking.
Use feedback that points to a moment. Ask a reviewer where they lost the thread, what they expected after the hook, and which line felt hard to follow. If you have retention data from a published video, look for the point where viewing drops, then review the line and shot at that time. Hooked.so recommends auditing the specific drop-off moment rather than rewriting the full script by default. That’s a useful rule: fix the weak beat first.
For team review, keep feedback in one shared draft. Assign one person to approve facts and claims, and another to check timing or production cues. The editor should be able to tell which version is final. A short note such as “replace the close-up in beat two” is more useful than a broad comment like “make it pop.”
When the draft is for an AI generator, make one change at a time. If a shot looks wrong, check whether the prompt is vague, the reference is missing, or the requested action is too complex for the shot. Wan Studio’s model pages describe distinct input and output options; use the model-specific details when planning a clip, then inspect every rendered version for character, product, sound, and caption accuracy.
Before export, check the hook, the ending, the spoken timing, and the captions. If one sentence is the problem, revise that sentence. A full rewrite can throw out good work that already fits.
Milestone: The final draft should be easy to read aloud, timed to the intended runtime, and clear about what each shot needs to show.
FAQ
How do you write a video script for beginners?
Start with one audience and one goal, then write a hook, a short outline, and one clear CTA. A beginner-friendly script uses plain words and focuses on one main idea. Read it aloud before filming. If a line feels hard to say or doesn’t help the viewer, shorten it or remove it.
How long should a script be for a one-minute video?
A one-minute script often needs about 130 to 150 spoken words as a starting estimate, depending on the speaker’s pace. Leave extra time for pauses, visual demonstrations, or on-screen text. Time a full read with the planned actions before you finalize the script, since word count alone doesn’t account for what viewers need to see.
Should I write a video script word for word?
Write the whole script word for word when timing or exact claims matter, or when the speaker needs more support. Use bullets for sections you can explain naturally. A mixed approach often works well: script the hook and required language exactly, then use an outline for the rest.
How do I make an AI video script sound natural?
Write short lines that a person would say aloud, then read them at the pace you expect in the video. Give each shot one clear action and keep character details consistent across prompts. Check the generated result for timing, lip sync, and visual changes. Model capabilities can differ, so confirm the options for the specific Wan model you use.
How should I add captions to a video script?
Plan captions alongside the dialogue, but match them to the final audio after editing. Keep each caption short enough to read on a phone and place it where it won’t hide a key visual. Check names and product terms by hand. For accessibility, also consider a transcript and whether important visual action needs audio description.
Conclusion
Write the message before polishing the lines, then rehearse and fix the exact moment that feels unclear. Pick one product or idea, draft a 30-second version, and map each spoken line to a shot. If you’re planning a photo-based scene, explore Wan Studio’s model options and use the output as a draft to review, not a finished script on autopilot.
