Most AI video projects do not fail at the render button. They fail three steps earlier, when nobody bothered to decide what the scene actually needs. A model can produce a gorgeous four-second clip of a woman walking through rain, but if your film has six shots of that same woman and each one looks like a different person in a different city, the clip is worthless. Planning is not bureaucracy. It is the difference between a folder of pretty fragments and a sequence that an audience can follow.
This guide lays out a neutral, tool-agnostic workflow for AI storyboarding and shot planning. It borrows from traditional film pre-production, adapts it to generative pipelines, and focuses on the decisions that actually change your output: how to break a script into shots, how to keep characters stable, when to use which generation method, and how to run review loops without drowning in versions.
Why Shot Planning Decides Whether an AI Video Works
Generative video tools are extraordinary at local coherence and terrible at global coherence. They can maintain a face for a few seconds, keep a camera move smooth, and light a scene beautifully. What they cannot do reliably is remember that the antagonist drives a green truck, that the kitchen window faces east, or that the hero's jacket was buttoned in the previous shot.
That gap is where planning earns its keep. A shot list is essentially a memory system. It externalises the details a model will forget and a human will misremember after the fifteenth iteration. When you write down framing, lens, movement, wardrobe, location, time of day, and emotional beat for every shot, you convert a vague creative intention into a set of constraints that any generator can be held to.
Planning also protects your budget of time. Every ambiguous prompt costs you a round of generation, a review, and a revision. Teams that plan well do not generate fewer images because they are timid; they generate fewer because each attempt is aimed at a specific, already-defined target. Aimless exploration feels productive and quietly consumes entire days.
Finally, planning makes collaboration possible. A director, an editor, a sound designer, and a client can all argue about a storyboard far more cheaply than they can argue about a finished sequence. Frames on a wall are negotiable. Rendered minutes are expensive.
The Pre-Production Stack Every AI Video Project Needs
Think of pre-production as four stacked layers, each one narrower than the last. Skip a layer and the layers above it wobble.
Layer one: the script and its beats
Start with the written scene and mark the emotional or informational beats. A beat is a change: a decision, a revelation, a reversal, a physical action that alters the situation. A ninety-second piece might have four beats. A three-minute piece might have eight. Beats, not minutes, are the true unit of structure.
Layer two: the visual language
Before generating anything, write a short lookbook in plain language. Palette, contrast, texture, era, film stock feel, aspect ratio, and reference moods. Include a rule or two that you will enforce ruthlessly, such as "no camera movement faster than a slow push except in the chase beat." Constraints create style; unlimited options create mush.
Layer three: the shot list
The shot list is the working document. Each row should carry a shot number, a beat reference, a description of action, framing, camera movement, duration estimate, location, characters present, and any continuity notes. This is the file you will actually read while generating.
Layer four: the storyboard and animatic
Frames turn the list into something you can see. Timing those frames against dialogue, music, or a temp track turns the storyboard into an animatic, which is the cheapest possible version of your edit.
Turning a Script Into a Structured Shot List
The fastest way to build a shot list is to read the script three times with three different questions in mind.
The first pass asks: what must the audience understand? These become your coverage shots, the ones that carry information. A hand opening an envelope. A face reacting. A clock. Coverage shots are functional and should be shot plainly; cleverness here usually hurts.
The second pass asks: where is the emotion? These become your performance beats, the shots where the camera should linger. Give them more duration in your estimate and resist the urge to cut them short in the edit.
The third pass asks: what changes physically? Movement, entrances, exits, object handoffs. These become your continuity-critical shots, and they need the most explicit notes because models handle object permanence poorly.
Write shot descriptions in a fixed order: subject, action, location, time of day, lighting mood, camera framing, camera movement, duration. Fixing the order trains you to notice when something is missing. A shot description without a duration estimate is a shot you have not finished thinking about.
Useful duration heuristics for AI-heavy projects: three to five seconds for reaction and insert shots, five to eight seconds for dialogue exchanges, and eight to twelve seconds for establishing or movement shots. Shorter clips generate more reliably, so build sequences from more, shorter shots rather than fewer, longer ones. If a shot needs to be twenty seconds, ask whether it is genuinely one shot or two shots that should be cut together.
Finally, number shots and never renumber casually. Shot numbers are the shared vocabulary between your list, your frames, your renders, your audio stems, and your editor's timeline. Renumbering mid-project is how files get lost.
Keeping Characters, Props, and Locations Consistent
Consistency is the single most common complaint about AI video, and it is largely solvable with documentation rather than better models.
Create a character sheet for every recurring figure. Include age range, build, hair, skin tone, distinguishing features, wardrobe per scene, and a short paragraph of personality that informs posture and expression. Then generate a small library of reference images: a clean front view, a three-quarter view, a profile, and one expressive close-up. These references become your anchor. Whether you are working with a tool that accepts image references, trains a personal style, or uses identity-conditioned generation, the sheet is what keeps you from improvising.
Do the same for locations. A location sheet should state orientation and geography: where the door is, which way the windows face, where the light comes from at each time of day. This sounds obsessive until the third shot where a character walks toward a window that was behind them in shot one. When you have a location sheet, you can write prompts that describe the camera's position relative to fixed landmarks, which dramatically improves spatial coherence.
Props deserve their own short list, especially anything an audience will recognise again: a ring, a phone model, a car, a mug. Give each prop a one-line physical description and reuse that exact phrasing. Small phrasing consistency pays off surprisingly well because generators respond to the same token patterns in the same way.
A practical trick for wardrobe continuity: assign each character one signature garment per act and describe it with three adjectives maximum. Between shots, list the garment's state, such as sleeves rolled or jacket on. State tracking beats vague memory every time.
Choosing the Right Generation Method for Each Shot
Not every shot should be generated the same way. Matching method to intent is the most underrated skill in AI video production.
Text-to-video for exploration and simple motion
Text-to-video is ideal for mood tests, establishing shots, and any moment where exact framing is flexible. It is fast, it surprises you, and it is a poor choice for continuity-critical beats because you have the least control over composition.
Image-to-video for composition control
When framing matters, generate or shoot a still first and animate it. Image-to-video lets you lock composition, subject placement, and lighting before spending compute on motion. This is the default for dialogue beats, inserts, and anything that must match a storyboard frame. It is also the best method for consistency, because your reference frame already contains the correct face and wardrobe.
Motion and depth tools for technical camera moves
When the shot's entire point is a specific move, such as a slow dolly-in on a face or a lateral parallax across a room, use tools designed for camera control rather than hoping a prompt produces it. Describe direction and speed precisely: the move should be a character in your shot list, not a footnote.
Hybrid and multi-pass approaches
Long or complex shots often work best as two passes. Generate a wide, stable version for geography, then generate a tighter version for performance, and cut between them. Alternatively, animate a still with a subtle motion pass, then enhance detail in a second pass. Two clean passes beat one ambitious mess.
Stylised and animated treatments
If your project uses illustrated, painterly, or 3D-stylised rendering, choose the treatment per project, not per shot. Mixing visual treatments inside a single sequence reads as an error rather than a choice, unless the transition is deliberately motivated by the story.
Camera Language: Framing, Lenses, and Movement
AI video tools respond to film vocabulary, but only if you use it consistently. Build a small internal dictionary and stick to it.
For framing, decide between wide, medium, and close, and add an intent word such as intimate, observational, or oppressive. For lens feel, describe perspective rather than focal length numbers, since numbers are inconsistently interpreted: wide-angle distortion, normal perspective, compressed telephoto. For movement, use directional language: slow push in, pull back, pan left, tilt up, tracking alongside, handheld drift, static locked-off.
A useful rule is one camera idea per shot. Push in and pan and rack focus at the same time and you will get mush. Reserve compound moves for moments that justify spectacle.
Also plan your coverage pattern before generating. A simple and reliable pattern is master, over-the-shoulder, close-up, insert, cutaway. Generating a full coverage set for each scene, rather than cherry-picking individual beautiful shots, is what makes editing possible later. Editors need options; a sequence of five unrelated masterpieces has nowhere to cut.
Finally, write down your axis. If two characters face each other, note which side of the line you are on for each shot. Violating the axis in AI video is easy to do accidentally because each shot is generated independently, and it produces a subtle, disorienting unease that audiences feel without naming.
From Frames to Animatic: Assembling the Storyboard
Once frames exist, resist the urge to polish them. A storyboard is a decision document, not an art portfolio. Rough frames are actually better during early review because they invite structural feedback instead of aesthetic nitpicking.
Lay out frames in shot order with shot numbers and durations visible. Add dialogue, voiceover, or scratch narration as text under each frame. Then time it. An animatic with a temp track and rough timing will reveal problems that no single frame can show you: a beat that drags, a transition that is confusing, a scene that peaks too early.
A practical animatic checklist: does each scene have a clear entry and exit? Does every cut have a reason, whether that is information, emotion, or physical continuity? Is there anything you can delete without loss? Is there any moment where the audience has to re-read the image to understand it?
Do this before generating final video. Deleting a shot at the storyboard stage costs seconds. Deleting it after generation, audio, and grading costs an afternoon. The whole point of the animatic is to be wrong cheaply.
Version Control, Review Loops, and Keeping Sanity
AI video production generates a lot of near-identical files. Without a naming convention, you will lose the good one.
Adopt a rigid scheme: project, scene, shot, version, and a one-word note. Something like scene03_shot07_v04_tighter. Never save anything as final, because there is always a later final. Keep the version that was approved and the version that is currently in the edit clearly separated.
Set review cadence deliberately. Daily micro-reviews of individual shots are useful. Weekly sequence reviews are essential, because the sequence is what the audience experiences. Reviewing shots only in isolation is how you end up with a perfectly crafted sequence that has no rhythm.
Also decide who has approval authority before you start. Nothing stalls a project faster than three people with veto power and no tie-breaker. For client work, define which review round includes structural changes and which rounds are limited to fixes. Unlimited revision scope is a planning failure disguised as generosity.
Common Mistakes and How to Fix Them
Here are the failure patterns that appear again and again, with the correction for each.
Generating before the shot list exists. Fix: write the list first, even a bad one, and revise it. A rough list beats a blank page.
Chasing a single perfect shot for hours. Fix: set an attempt limit, such as six generations per shot. If it is not working, the problem is usually the shot's conception, not the prompt. Split it, simplify it, or change the method.
Inconsistent character descriptions across shots. Fix: copy and paste the exact character phrasing from the sheet. Never retype it from memory.
Overlong clips. Fix: cut the shot into two. Short clips generate more reliably and give the editor more control.
No coverage for a scene. Fix: plan at least three angles per emotional beat, or accept that you will be stuck with what you have.
Mixing too many visual treatments. Fix: lock the look for a sequence, then vary only within it.
Ignoring audio planning. Fix: note dialogue, ambience, and music intent in the shot list. Sound design shapes pacing and often dictates whether a shot can be shorter than you assumed.
No shared naming convention. Fix: agree the scheme before generation begins, not after a hundred files appear.
Skipping the animatic. Fix: build one, always. It is the highest-value hour in the entire pipeline.
Treating the first good frame as the standard. Fix: compare candidates on continuity, not beauty. A slightly less striking frame that matches the previous shot is usually the right choice.
FAQ: Practical Questions About AI Storyboarding
How detailed should each prompt be? Detailed enough to encode everything the shot needs, and no longer. A reliable structure is subject, action, environment, lighting, camera, style, and duration. If you find yourself writing a paragraph of atmosphere, move half of it into your lookbook instead where it applies to every shot.
Do I need to draw to storyboard with AI? No. You need to describe. Rough composition sketches help, but written framing plus generated frames is enough for most productions.
What is the minimum viable pre-production for a short piece? A beat sheet, a character sheet, a lookbook paragraph, and a shot list with durations. That is roughly two hours of work and it will save you a full day.
How many versions should I keep per shot? Keep the approved version, one alternative, and delete the rest once the sequence is locked. Archive rather than delete if storage is cheap.
How do I handle a character who must age or change across the story? Build separate character sheets per state and treat the change as a new entity. Ask for explicit transition shots that show the change happening.
When should I stop planning and start generating? When every shot in your list has a framing, a duration, and a continuity note. If you cannot fill those three fields, you are still designing, not planning.
Can I plan shots for vertical formats the same way? Yes, but re-decide your framing rules. Vertical emphasises faces, hands, and depth stacking rather than wide geography, so your coverage pattern should shift toward close-ups and foreground layering.
What is the most common reason a planned sequence still feels off? Pacing. If the sequence feels flat, shorten reaction shots, lengthen movement shots, and check that your cuts land on beats rather than on convenience. The storyboard gave you the structure; the edit gives it rhythm.
Good planning is not the enemy of spontaneity. It is what makes spontaneity affordable. Once the shot list and the animatic exist, you can experiment freely inside each shot, because you already know what the sequence needs from it. That is the whole craft: constrain the intention, liberate the execution.
