AI Storyboarding Workflow: Design Shots and Boards That Work

Start With the Decisions You Need to Unblock

Most storyboard sessions fail for a boring reason: nobody agreed on what question the boards were supposed to answer. A board built for a client pitch answers "does this feel right?" A board built for a camera crew answers "where does the dolly go and where does the light sit?" A board built for an editor answers "do we have enough coverage to make this cut?". Those are three different documents wearing the same name.

So before you open any tool, write down three to five decisions the boards must unlock. Useful examples:

• Which opening image earns the audience's attention in the first eight seconds?
• Does the protagonist read as sympathetic before the turning point, or only after?
• Can the chase sequence be covered from two positions, or do we genuinely need four?
• Is the colour palette warm enough to sell the nostalgia the script promises?
• Does the client understand that the second half is a single continuous take?

Once that list exists, every frame becomes either useful or noise. That single habit does more for board quality than any generator setting.

AI image and video tools have collapsed the cost of a frame from an afternoon of drawing to a few seconds of waiting. That changes the bottleneck. It is no longer "can we picture it?" but "can we choose well?" Selection, sequencing, and continuity are now the scarce skills, and they are exactly the skills a shot list and a review loop protect.

What AI Does Well in Storyboarding — and Where It Fails

Generative tools are excellent at some parts of pre-production and mediocre at others. Knowing the boundary saves days.

Task	AI is strong	AI is weak
Mood and tone exploration	Fast, abundant, surprising	Can drift into generic "cinematic" looks
Coverage variants	Trivial to produce 20 alternates	Rarely respects exact set geography
Lighting and palette studies	Very good for mood reference	Less reliable for practical light positions
Character casting look	Great for wardrobe and type	Hands, props, and jewellery break easily
Precise camera math	Approximates compression and distortion	Cannot guarantee a true 32mm field of view
Animatics and timing	Image-to-video gives instant motion studies	Performance and dialogue sync remain crude
Text in frame, signage	Occasionally readable	Usually garbled; add in post

A practical rule: use AI for shots that must be felt, and use simple hand sketches, floor plans, or overhead diagrams for shots that must be shot precisely. A rough rectangle with two stick figures and an arrow will communicate a dolly move to a crew better than a beautiful generated frame that gets the geography wrong.

Stage One: Turn the Script Into Beat Cards

Mark the beats

Read the scene and mark every moment where something changes: new information lands, power shifts between characters, the location changes, or the emotional temperature moves. Each change becomes one beat card. A three-page dialogue scene usually produces four to seven beats; an action sequence can produce fifteen.

Use a consistent card template

A beat card should be short enough to read in ten seconds. A workable template:

• Beat ID: SC04-B2
• Location and time: Rooftop, pre-dawn
• Characters present: Mira, Detective Osei
• What changes: Mira admits she lied about the file
• Emotional temperature: Cold, controlled, one crack
• Estimated runtime: 35–50 seconds
• Must-see image: Mira's hands, file folded shut, city behind her
• Constraint: Practical wind, no crane access

That last field matters more than people expect. Constraints written at the beat stage prevent you from boarding a shot you cannot physically get.

Tier the beats

Not every beat deserves a fully rendered frame. Split them into tiers. Tier A beats carry the story and get three to five finished frames each. Tier B beats get one or two. Tier C beats get a single thumbnail or a written description. This keeps a hundred-shot project from turning into a hundred-frame illustration job.

Stage Two: Build a Shot List That Survives the Edit

Columns that earn their place

A shot list you will actually use has these columns: shot ID, beat ID, shot size, angle, movement, lens intention, subject action, dialogue or audio, estimated duration, and coverage group. Anything else tends to be decoration.

The 3-1-1 coverage rule

For each Tier A beat, generate three shot-size variants, one movement variant, and one insert. That gives you a wide, a medium, a close, something that moves, and a detail the editor can cut to when the performance needs air. It is enough coverage without drowning the edit in choices.

Sequence by screen order, shoot by logistics

Keep two views of the same list. The board and shot list follow screen order because that is how the story is judged. A separate shooting plan groups shots by location, lighting setup, and cast availability. Mixing the two is the fastest way to confuse a crew.

Write the button, not just the setup

Every shot needs an intended exit — the last image that hands off to the next shot. If you cannot describe the exit frame, you have a mood image rather than a shot. Boarding exits is what makes an animatic readable without sound.

Designing Shots: Lens, Framing, and Movement

Focal length as an emotional dial

Wide lenses stretch space and pull the viewer into the room; long lenses compress and isolate. Rough guidance that translates well into prompts:

• 14–24mm: disorientation, scale, interiors that feel too small for the emotion
• 28–35mm: documentary realism, walking alongside a character
• 50mm: neutral, close to human attention
• 85mm: intimacy and flattery, shallow separation from background
• 135mm and beyond: surveillance, voyeurism, heat haze, compressed crowds

Most image models respond better to descriptions of compression and distortion than to exact millimetre numbers, but including both costs nothing: "long-lens compression, subject isolated from a soft city background, roughly 135mm character."

Composition patterns worth naming

Give your frames vocabulary so the team can request changes precisely: centre-punch for confrontation, negative space for loneliness, frame-within-frame for entrapment, low horizon for optimism, high angle for vulnerability, leading lines for momentum. Naming the pattern turns vague notes like "make it more dramatic" into a specific instruction.

Movement vocabulary

For each moving shot, define a start frame and an end frame. Static lock-off, slow push-in, pull-back reveal, lateral tracking, handheld drift, crane rise, orbit, whip pan. Then note the pace: creeping, steady, urgent. A push-in that takes twelve seconds reads as dread; the same push at two seconds reads as aggression.

Blocking, eyeline, and the line

Decide where each character stands, which way they look, and where the imaginary 180-degree line sits. Screen direction matters: if a character exits frame left in one shot, they should enter frame right in the next unless you are deliberately disorienting the audience. Write the line into your notes so nobody has to guess during editing.

Stage Three: Prompt Patterns That Produce Usable Frames

The six-slot formula

Reliable prompts are structured, not poetic. Fill six slots:

1. Subject and action: who, doing what, wearing what
2. Environment and time: where, weather, hour
3. Framing and lens: shot size, angle, compression
4. Lighting and palette: source, direction, colour temperature
5. Mood and genre: emotional register, reference era
6. Technical: aspect ratio, texture, grain, detail level

Example: "Mid-fifties lighthouse keeper, hands wrapped around a rusted lantern, coat soaked; storm-battered granite stairwell at dusk; medium close-up, slightly low angle, mild wide-angle distortion; single warm bulb overhead, cold blue window light from the left; melancholy, 1970s coastal drama; 2.39:1, fine grain, natural skin texture."

Change one slot per experiment

When you want alternates, change exactly one slot per batch and keep everything else identical. Three variants with different lighting will teach you something. Three variants with different lighting, framing, and wardrobe teach you nothing except that the model is random.

Negative prompts and guardrails

Keep a reusable negative list: extra fingers, duplicated limbs, warped text, watermark, lens flare overload, plastic skin, excessive sharpening, anachronistic clothing, cluttered background faces. Reuse it across the whole project so failures stay consistent and diagnosable.

Respect the delivery ratio

Board in the ratio you will deliver. A composition that sings at 2.39:1 can collapse when cropped to vertical. If you also need social cuts, generate designated vertical frames for the three or four shots that carry the story rather than hoping a crop will rescue the rest.

Keeping Characters and Style Consistent

Build character sheets first

Generate a front, three-quarter, and profile view of each principal against a plain backdrop before boarding any scene. Lock down wardrobe, hair, age markers, and any distinguishing detail — a scar, a specific watch, a crooked collar. Then describe the character with the exact same wording in every subsequent prompt. Consistency comes from repeated language more than from any single setting.

Anchor with hero frames

Pick one strong frame per character and per location and treat it as the anchor. Where your tool supports image references or seeds, reuse them. Where it does not, paste a short fixed description block at the start of every prompt. It is unglamorous and it works.

Write a look bible

A look bible is a short document with a fixed style signature. For example: "35mm anamorphic feel, warm tungsten practicals, teal shadows, gentle halation, natural skin texture, light grain, no digital sharpening." Store it once, paste it everywhere, and when a frame drifts off-style you can see immediately which clause was dropped.

Run a continuity checklist

Before a sequence leaves the board stage, check: hair and wardrobe state, prop positions, time-of-day light direction, screen direction, eyeline match, and the geography of the set. Six items, ninety seconds, and it prevents the most embarrassing notes in a review.

Review Loops, Versioning, and Common Mistakes

Two passes, two questions

Pass one is the thumbnail test: shrink the whole scene to small tiles. Can you follow the story without reading a caption? If not, the problem is sequence, not rendering. Pass two is the detail review: continuity, craft, and whether any frame is doing work that a different frame does better.

Version like an editor

Name files predictably: project_scene_shot_v03. Keep one contact sheet per scene so a reviewer sees the sequence instead of a folder. Archive rejected frames but tag why they were rejected — "wrong eyeline", "too soft", "reads as comedy". Rejection reasons become your prompt improvements.

Mistakes that cost the most time

• Boarding every beat at equal detail and running out of energy before the important scenes.
• Chasing beautiful frames that contradict the established set geography.
• Letting prompts drift so that each scene looks like a different film.
• Producing hundreds of images with no decision log, so nobody remembers why anything was chosen.
• Skipping the animatic, then discovering in the edit that two shots do not cut together.
• Ignoring the delivery ratio until the final week.
• Treating one tool's aesthetic as the house style instead of defining your own.

Choosing Tools and Building Your Own Stack

Judge tools against your actual bottleneck, not against demo reels. Useful criteria:

• Reference and character consistency: can it reuse a subject reliably across many frames?
• Sequence view: does it show a scene as an ordered board, or only as a gallery?
• Motion support: can it turn a still into a short animatic clip for timing?
• Export quality: storyboard PDFs, contact sheets, and shot lists that a crew can read.
• Collaboration: comments, versions, and permissions for a small team.
• Cost predictability: per-project budgeting beats surprise overages.
• Privacy: where do unreleased scripts and client footage end up?

A lean stack usually includes one general image generator for look exploration, one storyboard-oriented assistant that organises shots into sequences, one image-to-video tool for animatics, and one asset manager or shared drive with strict naming. You do not need more than that; you need discipline about what goes where.

FAQ

Do I still need to draw?

Only where precision beats beauty. Rough shapes, arrows, and floor plans still communicate camera geography better than a rendered frame. Everything else can be AI-assisted.

How many frames should a scene get?

A rough starting point is one frame per four to six seconds of finished runtime for a well-covered scene, plus extra detail on the three or four shots that carry the story. A two-minute scene usually lands somewhere between twenty and thirty-five frames.

Can an AI storyboard be shot-accurate?

For mood, lighting intent, and coverage, yes. For exact lens fields of view, prop placement, and stunt geometry, treat boards as intent rather than blueprint, and supplement with diagrams.

How do I keep characters consistent across scenes?

Fixed character sheets, identical descriptive wording, anchor frames, and a written style signature. Repetition of language is the real mechanism.

What about dialogue and performance?

Storyboards show where a line lands, not how it is delivered. Mark the line's function in the beat card — pressure, confession, deflection — and leave the performance to the rehearsal room.

Is a fully AI-generated board acceptable to clients?

It is acceptable when the client understands it is a visual plan, not a promise of final rendering. Label the boards clearly, present them in sequence, and note which frames are mood reference only.

What is the single highest-leverage habit?

Keeping a decision log. One line per rejected frame explaining why. It turns a pile of images into an actual pre-production process, and it makes the next project noticeably faster.