Why AI Pre-Production Changed the Way Films Get Planned
Pre-production used to be the slowest, most paper-heavy phase of filmmaking. A writer finished a draft, a director marked it up, an assistant director built a shot list by hand, and a storyboard artist turned that list into sketches over several weeks. Every script revision restarted the entire loop, which is why so many ambitious ideas died quietly in a folder of half-finished PDFs.
Generative models collapsed that loop. A modern language model can read a screenplay, extract scenes, characters, props, and locations, then write shot descriptions that a text-to-image or text-to-video model renders in minutes. The output is not a finished film. It is a testable plan you can screen before anyone books a location, hires a crew, or opens a video editor.
The practical shift is this: the bottleneck moved from can we draw it? to did we describe it clearly enough? Directors now spend their time on intent — what a shot should feel like, what the audience should learn, where the cut lands — rather than on the logistics of producing boards. That is a better use of human attention, but it also means sloppy writing is punished faster. A vague sentence no longer produces a vague sketch; it produces a wildly wrong image that wastes a render cycle and derails a meeting.
This guide walks through a complete AI-assisted pre-production workflow: preparing a machine-readable script, breaking it into shots, generating storyboards, keeping characters consistent, animating boards into previsualization clips, and handing the package to a production team. It is tool-agnostic on purpose. The same logic works whether you are using Sora, Runway, Kling, Veo, Midjourney, or a local diffusion setup on your own machine.
The Three Pillars of an AI Pre-Production Workflow
Every AI-assisted pipeline rests on three artifacts. If any one of them is weak, the whole thing wobbles.
1. The Script
This is your source of truth. Everything downstream — shot lists, boards, prompts, animatics — is derived from it. If the script is inconsistent about character names, locations, or time of day, the AI will happily invent three different versions of your protagonist's apartment and you will spend hours cleaning up the mess.
2. The Shot List
A shot list converts narrative into coverage. It defines how many shots a scene needs, what each one shows, and roughly how long it lasts. In an AI workflow, the shot list doubles as a prompt sheet: each row contains the visual information the image or video model needs.
3. The Storyboard
Boards are the visual contract. They let a director, editor, client, or investor agree on a sequence before money is spent. AI boards are not meant to be beautiful in the way hand-drawn boards are beautiful. They are meant to be fast, specific, and replaceable.
A useful mental model: the script answers what happens, the shot list answers how we see it, and the storyboard answers what it looks like. Keep those three questions separate and your workflow stays clean.
Step 1 — Writing a Script That Machines Can Read
You do not need to write differently for humans and machines, but you do need to write consistently. The following conventions dramatically improve breakdown quality.
Standardize scene headings. Use INT. COFFEE SHOP - DAY or EXT. ROOFTOP - NIGHT - RAIN. Avoid creative headings like SOMEWHERE QUIET. Time of day, interior/exterior, and weather are prime inputs for lighting prompts.
Name every character consistently. Never alternate between "Maya," "the detective," and "she" in action lines if you want a consistent character sheet. Use proper names in the first mention of each paragraph.
Write action in the present tense and in single, literal sentences. "Maya slides the folder across the table" is far easier to visualize than "Maya, having finally decided to trust him, produces the folder and hands it over rather than keeping it." The second version is better prose; the first is better pre-production.
Break long paragraphs into beats. Each distinct physical action tends to become its own shot. If a paragraph contains four actions, expect four shots, and the breakdown tool will likely merge or drop some unless you separate them yourself.
Specify visual anchors. Props, wardrobe, color, and location details that recur across scenes should be described identically every time. If the red umbrella matters in Act One, describe it the same way in Act Three.
A Quick Pre-Flight Checklist
Before running any breakdown, verify:
- Every scene has a proper slug line with time of day.
- Every character has one canonical name.
- Every recurring prop or location has one canonical description.
- No paragraph contains more than two distinct actions.
- Dialogue-heavy scenes include at least one line of physical business so the coverage has something to cut to.
Fixing these five items takes twenty minutes and saves hours of regeneration later.
Step 2 — Turning a Script into a Shot List
Once the script is clean, the breakdown step extracts structure. A language model reads the draft and returns a scene-by-scene list of beats: who is present, what changes, what the audience needs to notice, and what emotional turn the scene delivers.
From those beats you build coverage. Standard patterns apply, and they are worth encoding as reusable templates.
Master shot. Establishes geography. Usually the first shot of a scene, often wider and static or slowly moving.
Medium shot. Carries dialogue and performance. The workhorse.
Close-up. Delivers emotion or information. Use for reveals, decisions, and reactions.
Insert. A detail: a phone screen, a hand, a ticking clock. Inserts are cheap to generate and invaluable in the edit.
Reverse or over-the-shoulder. The other side of a conversation.
Transition shot. A movement, a doorway, a vehicle leaving. These give the editor room to breathe between scenes.
Writing Shot Descriptions That Render Well
Each shot row should read like a compact technical brief. A reliable template:
SHOT 14 — INT. COFFEE SHOP — DAY. Medium close-up of Maya, 30s, dark curly hair, olive jacket, seated at a window table. She opens a red folder, eyes scanning. Warm afternoon light from the left, shallow depth of field, slight handheld drift. Approx. 4 seconds.
That single sentence contains subject, framing, wardrobe, action, lighting direction, lens feel, camera behavior, and duration. Six of those inputs map directly onto prompt fields in most image and video tools. The seventh — duration — matters for the animatic stage.
Number every shot. Use a consistent scheme, such as S02_SH14, so scene and shot numbers survive reordering. You will thank yourself during the fifteenth revision.
Step 3 — Generating Storyboards from Shot Descriptions
The shot list is now a prompt sheet. The storytelling work is done; the remaining task is rendering it consistently.
Start with a style reference. Generate one or two frames that establish your visual language — palette, contrast, texture, level of realism. Once you like them, reuse that style description word-for-word in every subsequent prompt. Consistency comes from repetition, not from vocabulary variety.
Lock your aspect ratio early. Boards for a vertical social piece and boards for a widescreen film are different compositions. Changing ratio halfway through means re-rendering everything.
Batch by scene, not by shot. Generate all shots from one scene in a single session while the style reference is loaded. This reduces drift.
Generate three variants per shot, keep one. Cheap iteration beats careful perfectionism. The goal is a readable board, not a portfolio piece.
Annotate as you build. Add camera movement notes, sound cues, and duration to the board itself. A pretty image with no directorial instruction is decoration, not pre-production.
A realistic pace for a short film: 40 to 80 shots, three variants each, one afternoon of generation and one afternoon of selection and annotation. That is a week of work compressed into a day, and it leaves more time for the part that actually needs a human — deciding whether the scene works.
Step 4 — Keeping Characters, Locations, and Props Consistent
Visual consistency is the single hardest part of AI pre-production. Faces drift, jackets change color, and the same apartment acquires a new kitchen between scenes. These habits fix most of it.
Build Character Sheets First
Create one reference image per principal character before boarding anything. Front-facing, neutral expression, plain background, full wardrobe description. Then attach that reference to every shot featuring them, using image reference, character reference, or style-reference features depending on your tool.
Freeze Location Plates
Generate one establishing plate per location and treat it as canon. Every interior shot in that location should reference the plate, including the direction of windows, the color of walls, and the position of major furniture. Blocking that contradicts the plate will confuse audiences in ways they cannot articulate.
Use Naming Discipline
Name your files S02_SH14_maya_closeup_v03. Names like final_final2 guarantee that someone edits the wrong frame.
Accept Controlled Imperfection
Boards are provisional by nature. A slightly different nose between shots is acceptable in a board that will be reshot on set with real actors. Spend your consistency budget on the things that affect decisions: wardrobe color, location layout, time of day, and the presence or absence of key props.
When Consistency Still Fails
If a character keeps morphing, the problem is usually the description, not the model. Reduce adjectives, remove contradictory details ("messy but immaculate"), and describe one distinguishing feature rather than five. Specificity beats richness.
Step 5 — From Static Boards to Animated Previs
A storyboard shows composition. An animatic shows rhythm. Once your boards are approved, image-to-video tools can turn each frame into a short moving clip, usually three to six seconds, which you then assemble into a timed sequence.
Keep motion prompts minimal. One camera move and one subject action per clip. "Slow push in, she turns her head" works. "Slow push in while she turns, stands, walks to the door and glances back" will produce mush.
Match the move to the intent. A dolly-in builds tension or intimacy. A handheld drift suggests unease. A static frame with movement inside it suggests observation. Choose deliberately; do not let the tool decide.
Cut to temporary music early. Timing decisions made against silence are almost always wrong. Lay in a scratch track, then trim clips to the beat.
Add placeholder dialogue and sound effects. Even robotic text-to-speech reveals whether a scene is too long.
Watch it end to end without pausing. This is the real test. If the animatic is confusing, the finished film will be confusing, and fixing it now costs nothing.
An animatic is also the best communication tool you have. Clients who cannot read a shot list will immediately understand a moving sequence — and they will tell you exactly which shot feels wrong.
Step 6 — Reviewing, Versioning, and Handing Off
Pre-production is iterative, and iteration without version control becomes chaos. A few lightweight habits prevent that.
Keep the shot list as the master document. Every change starts there and propagates outward. If a shot is cut, cut it in the list first.
Version the whole package, not individual files. Tag a review round: boards_v04 includes the shot list, boards, animatic, and notes. This makes it possible to compare two complete states of the project.
Record decisions, not just changes. "Cut shot 22 because it repeats the information in shot 19" is far more useful to a future collaborator than a silent deletion.
Export a production-ready handoff. A practical package includes: the shooting script, the numbered shot list with durations, the board frames with annotations, the animatic with timecode, and a one-page visual reference sheet for wardrobe and locations.
Archive the prompts. The prompt text behind every board is a legitimate production document. When a reshoot is needed months later, regenerating a matching frame is trivial if you kept the prompt, and impossible if you did not.
Choosing Tools: Decision Criteria That Actually Matter
Tool choice should follow workflow, not the other way around. Rank candidates against these criteria.
Controllability. Can you attach a reference image, lock a seed, specify a camera move, and set an aspect ratio? Tools that ignore references are unusable for character work.
Consistency features. Character reference, style reference, and image-to-image conditioning are the features that determine whether you can build a coherent board set.
Iteration speed. A tool that produces mediocre results in eight seconds is usually more useful in pre-production than one that produces beautiful results in ten minutes, because pre-production is about volume and speed of decision-making.
Clip length and motion quality. For animatics, five-second clips with coherent motion beat fifteen-second clips full of warping artifacts.
Export and integration. Check resolution, file formats, and whether you can batch-download a scene's worth of frames without clicking forty times.
Collaboration and permissions. If a team reviews boards, you need sharing, comments, and clear ownership of assets. Also verify how your tool handles commercial usage and training data policies before uploading sensitive scripts.
Cost model. Compare pricing against how many shots you realistically generate per project, including discarded variants. Most people underestimate retries by a factor of three.
A practical approach: pick one text-to-image tool for boards, one image-to-video tool for animatics, and one editing application for assembly. Specializing beats forcing a single tool to do all three jobs badly.
Common Mistakes and Frequently Asked Questions
Mistake: boarding every line of dialogue
Coverage should serve the scene, not transcribe it. A three-minute conversation does not need sixty close-ups. Board the emotional turns and let the actors and editor handle the rest.
Mistake: chasing photorealistic boards
Beautiful boards create attachment. Rough boards invite revision. If your boards look like final frames, stakeholders will debate the color grade instead of the story.
Mistake: ignoring the edit
Shots exist to be cut together. Board a sequence, then watch it in motion before approving it. Individual frames lie; sequences do not.
Mistake: mixing aspect ratios and frame rates
Standardize before you generate, not after. Retrofitting composition is painful.
Mistake: treating AI output as final
Boards are a hypothesis. The moment they stop being revised is the moment they stop being useful.
FAQ: Do I still need a human storyboard artist?
For stylized pitches, complex action, and anything where drawing style is part of the sell, absolutely. For volume coverage, rapid iteration, and internal planning, AI boards are usually faster and cheaper. Many teams use both: AI for the bulk, a human artist for hero frames and pitch decks.
FAQ: How many shots should a short film have?
A rough rule: one shot per five to eight seconds of finished runtime for dialogue-driven scenes, and more for action. A three-minute short typically lands between 40 and 70 shots. If you exceed that significantly, you are probably over-covering.
FAQ: Can an animatic replace a live-action test shoot?
No, but it can tell you whether the test shoot is worth doing. Animatics answer structural questions about pacing and clarity. They cannot answer questions about performance, lighting, or whether an actor can actually walk and talk at the same time.
FAQ: What if the AI renders something completely wrong?
Rewrite the shot description. Nine times out of ten the failure is ambiguity, not model limitation. Remove adjectives, add spatial relationships, and specify direction of light and camera.
FAQ: How do I present AI boards to a skeptical client?
Frame them as planning documents, not artwork. Show the shot list alongside the frames, explain the timing, and emphasize that boards are provisional. Clients respond well to clarity and badly to overselling.
FAQ: Where should a beginner start?
Pick one scene — ideally a two-character dialogue with a clear emotional turn. Write it cleanly, break it into five to eight shots, generate boards, cut an animatic, and watch it. The entire exercise takes an afternoon and teaches more than a month of reading about tools.
The workflow is not complicated. It is simply disciplined: write clearly, break down honestly, board quickly, animate briefly, revise ruthlessly, and hand off completely. Do that consistently and the gap between an idea and a shootable plan shrinks from weeks to days.
