How to storyboard a video with AI, panel by panel

Storyboarding a video for AI generation takes six steps: turn the idea into a shot list, compose one panel per shot around subject, camera, and motion, write captions as motion prompts, composite the board onto a single page, drive the clip from that page, and iterate on the board rather than the clip. The last step is where the money is. On PrismPoster, a board fix costs nothing and a still costs 12 credits, while a discarded 1080p clip costs 210 — the storyboard is what makes AI video cheap, not the generator.

Step 1: turn the idea into a shot list

Before you draw anything, write shots, not scenes. A shot is one camera position and one action; "the argument in the kitchen" is a scene, while "close-up: she sets the cup down too hard" is a shot. AI generation adds a constraint worth designing around: clips come in roughly 5-second units, so every shot should carry a single readable action that fits that window. For a 60-second video, that means a list of about twelve rows — subject, action, camera, duration. If a row needs two actions to make sense, split it into two shots now; the generator will not split it for you later.

Step 2: compose each panel around subject, camera, and motion

One panel per shot, answering three questions. Subject: who or what, and where in the frame — foreground or background, facing which way. Camera: how close and how high — a wide shot from below and a tight shot at eye level are different sentences in film grammar, and generators respect them. Motion: what changes during the five seconds — the subject moves, the camera moves, or both, and "both" is where generations most often fall apart, so spend both budgets rarely and deliberately.

Drawing skill is optional. Stick figures with an arrow communicate motion better than a beautiful panel without one, and panels can be uploaded photos or generated stills instead of drawings. If you think better on paper, print a storyboard template PDF and rough the sequence out by hand before you type a single caption. For shots that must end on a precise composition, add a second panel showing the end state — it maps directly onto end-frame generation later.

Step 3: write captions as motion prompts

The caption under each panel is not a description of the drawing; it is the instruction the video generation will follow. Write it as concrete verbs in present tense: "she turns toward the window; slow push-in" beats "a contemplative moment." Name the camera move with standard grammar — push in, pull back, pan left, orbit, handheld — and keep one primary action per caption. State what should move and, when it matters, what should hold still ("steam rises; she stays motionless"). Phrase everything positively: describe the motion you want rather than listing motions to avoid, because generative models follow affirmative instructions far more reliably than negations.

Step 4: composite the board onto one page

In PrismPoster's Storyboard, panels are raw by design — drawn, uploaded, or captioned, with no per-panel AI generation — and the working step is compositing them onto a single page in shot order. That page is the deliverable. It reads left to right like a comic: composition from the panels, motion from the captions, sequence from the order. If a panel looks wrong at page scale — an unclear silhouette, two panels that read identically — fix it now. This is the last stop where fixes are free.

Step 5: drive the clip from the board

The composited page goes to the video generation as its visual brief, captions riding along as the motion prompt. This is image-to-video, and it is the reason board-driven generation drifts less than prompt-only generation: video models follow pictures more faithfully than sentences, so a page that pins composition, wardrobe, and set leaves the model to solve only the motion. Two practical settings decide the outcome before you click. Aspect ratio comes first — PrismPoster generates and exports in 16:9 or 9:16, so board in the ratio you will release. And for shots with a defined destination, supply the end state too: end-frame mode sets the clip's final frame as well as its first. The mechanics of the generation itself are covered in the image-to-video guide, and if you are still choosing between starting from text or images, read text-to-video vs image-to-video.

Step 6: iterate on the board, not the clip

When a generation disappoints, the reflex is to regenerate. Resist it and diagnose instead: most bad clips are bad briefs. Wrong composition means the panel was ambiguous. Wrong motion means the caption was vague or asked for two things. Wrong look means the source still needs work, not the clip. Fix at the cheapest stage that explains the failure:

Where you fix it	What a retry costs	What you can change
The board — redraw, reorder, recaption	Nothing; panels are manual	Composition, camera, sequence, intent
The still — Image Studio	12 credits per image	Look, lighting, character, set
The clip — video generation	40 to 440 credits per 5 seconds	Motion only

The industry numbers make the case bluntly: measured regeneration rates across AI video tools run two to three attempts per usable clip, and the most common complaint on G2 for the biggest generators is credits consumed by generations that never became usable footage. A regeneration habit multiplies your real cost per clip; a board habit caps it. When you do regenerate, change exactly one thing — panel, caption, or still — so you learn what the model responded to.

When a script-to-frames tool fits better

Honest boundary: this workflow assumes you want to make the panels yourself, because the board's job is to control a generation. If what you actually want is AI to draw the board from a script, use an AI-native storyboard tool instead — Katalist starts at $19/month and auto-draws frames with recurring characters, and LTX Studio runs script-to-shots-to-video from $15/month, though its AI storyboards unlock on the $35 Standard tier, as of July 2026 (katalist.ai, ltx.studio). The trade is control for speed: reviewers of both report character drift between frames and credit burn on regenerations, which is exactly the failure mode the manual board avoids. The full segment breakdown is in the AI storyboard generator guide.

Frequently Asked Questions

Do I need to be able to draw to storyboard with AI?

No. Panels can be stick figures, uploaded photos, or generated stills — what the generation needs is clear composition and an unambiguous caption, not rendering skill.

How many panels does a 60-second video need?

About twelve, because AI video generates in roughly 5-second units. Use one panel per shot, and add a second end-state panel for shots that must land on a specific composition.

What should a storyboard caption say?

The motion, in concrete present-tense verbs: subject action plus camera move, one primary action per clip. Describe what you want to happen rather than what to avoid — generators follow positive instructions far better than negations.

Can I skip the storyboard and just prompt the video?

For a single clip, prompting works fine. Past three or four shots, prompt-only projects drift in character, set, and framing, and every drifted clip bills like a good one — the board is what makes a twelve-clip video look like one video. A worked example is the AI music video workflow.