Storyboarding an AI video means deciding every shot before you generate anything: what the viewer sees, for how long, from which angle, and how it connects to the shots either side. It matters even more with AI than with filming, because generation is slow, costs credits, and gives you short clips that do not naturally join up. A good storyboard turns a pile of random clips into a sequence that reads as one piece.
Why AI video needs more planning, not less
It is tempting to skip planning and "see what the model comes up with". That usually produces a folder of attractive clips that do not fit together: characters change, lighting jumps, camera moves fight each other, and nothing lines up with the narration.
Several features of AI video make planning essential:
- Clips are short. Most generators produce clips of a handful of seconds. A two-minute video needs dozens of them, and each is a separate generation.
- Every generation is independent. The model does not know what came before unless you tell it with references.
- Iteration is expensive. Regenerating costs time and often credits. A plan reduces wasted attempts.
- The edit is driven by the voice. In most narrated videos, the voiceover sets the timing. If you generate before you know the timing, you will have clips that are too short or too long.
Start from the script and the voice
Storyboard from a finished script, ideally with the voiceover already recorded or generated. Then you know exactly how long each line lasts. Our guide to writing for AI voiceover covers scripts that read well aloud.
Break the script into beats
A beat is one idea, one moment the viewer needs to see. Go through the script and mark where the idea changes. That is where the picture should change too.
For example, this narration:
"In 1858, London's river was so polluted that Parliament could barely sit. The smell drifted through the windows of Westminster. Engineers were given a simple brief: fix it."
breaks naturally into three beats: the polluted river, the smell reaching Parliament, and the engineers receiving the brief.
A useful rule of thumb is one beat for every sentence or two, but let the meaning decide, not a timer. A dense sentence may need two shots; a short linking line may share a shot with its neighbour.
Write a shot list
For each beat, write a row in a shot list. A spreadsheet works well. Here is a format we use:
| Shot | Time | Narration | What we see | Framing | Camera | Characters / refs | Notes |
|---|---|---|---|---|---|---|---|
| 01 | 0:00–0:04 | "In 1858, London's river…" | Murky Thames, Victorian boats, haze | Wide | Slow push in | None; style ref A | Establishing shot |
| 02 | 0:04–0:08 | "The smell drifted…" | Parliament window, curtain soaked in liquid | Medium | Locked off | Style ref A | Detail tells the story |
| 03 | 0:08–0:12 | "Engineers were given…" | Engineer at desk with plans | Medium close | Slight drift left | Engineer sheet, style ref A | Match lighting to 02 |
The columns that matter most:
- Time. Taken from the recorded voice, so you know how long each clip must run.
- What we see. A plain description, written as you would brief a camera operator.
- Framing. Wide, medium, close. Vary it: several wides in a row feel flat.
- Camera. Locked off, push in, pull out, pan, drift. Be explicit. If you do not specify a locked camera, many video models will add movement anyway.
- References. Which character sheet and style image to feed the generator. Our guide to AI character consistency explains how to prepare these.
Sketch or generate storyboard frames
A written shot list is often enough, but storyboard frames make problems visible early. You do not need to draw well: stick figures, arrows for camera moves and a few words are fine.
Alternatively, generate quick still images for each shot using the same references you will use for the final. These frames do three jobs:
- They show whether the sequence reads visually before you commit to video.
- They reveal continuity issues, such as a character facing left in one shot and right in the next.
- They can become the start frames for image-to-video generation, giving you far more control over the final clip than text alone.
Build an animatic
An animatic is your storyboard frames laid on a timeline against the voiceover, each held for the duration of its beat. It is the cheapest possible version of the finished video, and it is where most structural problems show up.
Watch the animatic and ask:
- Does each picture appear when the narration mentions it?
- Is any shot on screen too long with nothing happening?
- Are there several similar shots in a row?
- Does the story make sense with the sound off?
Fix those in the storyboard, not after generation. Adding a gentle zoom to each frame in the animatic, as described in our Ken Burns effect guide, also gives you a useful sense of pacing, and for some shots the animated still may turn out to be good enough for the final cut.
Plan continuity
Continuity is what makes separate clips feel like one film. Decide these before generating:
- Screen direction. If a character walks left to right, keep them moving that way unless you show a turn.
- Lighting and time of day. Note it per scene and repeat the same wording in every prompt for that scene.
- Colour palette. One consistent grade or style reference across the whole video.
- Props and costume. List them per scene, so they do not appear and vanish.
- Frame rate and aspect ratio. Decide once. Many AI video tools output 24 fps; if the rest of your project is 25 or 30 fps, plan the conversion. Our guide to frame rates explains the options.
Generate in order of risk, not order of story
Once the storyboard is locked, generate the hardest shots first: those with characters interacting, complex motion or crucial story moments. If they fail, you may need to rethink the scene, and it is better to know before spending effort on the easy shots.
Name every output after its shot number (shot_03_v2.mp4) so the edit stays organised, and keep a note of which prompt and references produced each accepted clip. A consistent project folder structure makes this far easier on longer projects.
Common mistakes
- Generating before the voice is finished. Timing changes, and clips no longer fit.
- Prompting for motion in every shot. Constant movement is tiring. A still or very slow shot often lands better.
- Too many wide shots. Mix framings to guide attention.
- Forgetting cutaways. Details such as hands, objects and signs are easy to generate and give the edit flexibility.
- No versioning. Without shot numbers and notes, you cannot reproduce a clip that worked.
Quick checklist
- Finish the script and record the voice first.
- Break the script into beats where the idea changes.
- Write a shot list with timing, framing, camera and references for every beat.
- Make rough frames and an animatic against the voiceover.
- Fix pacing and continuity in the animatic, not after generation.
- Lock frame rate, aspect ratio, palette and style references.
- Generate the riskiest shots first and name outputs by shot number.
FAQ
How detailed should a storyboard be for a short video?
For a 30 to 60 second short, a shot list with timings, a one-line description and camera direction per shot is often enough. Longer or character-driven pieces benefit from frames and an animatic.
Do I need to be able to draw?
No. Stick figures, arrows and notes are fine, or you can generate rough frames with an image tool. The goal is clarity, not artwork.
Should I storyboard before or after writing the script?
After. The script, and ideally the recorded voice, sets both the content and the timing of each shot.
How many shots do I need per minute?
It depends on the style and pace. A calm documentary might change picture every six to eight seconds; a fast Short might change every two or three. Let the narration's beats decide.