How to storyboard a film with AI

A storyboard is a sequence of frames, one per shot, that shows what a film will look like before it is made. Directors have drawn them for a century because they are the cheapest place to find out that a scene does not work. With image models, a storyboard stops being a sketch and becomes something much more useful: frames good enough to be the actual first frames of your shots.

This guide covers how to go from a shot list to a finished AI storyboard: writing one frame per shot, keeping the boards consistent, reviewing them as a sequence, and handing the approved frames to a video model.

Why storyboard when you could generate video directly

• It is cheaper. A still generates faster and costs less than a clip. You can try many versions of a frame for the price of one video take.
• Problems are easier to see. Composition, eyelines, light and continuity all show up in stills. Motion hides them until the edit.
• It locks the look before motion. When you animate from an approved frame, the video model inherits the character, costume, set and light, instead of inventing them.
• It tests the story. Reading frames in order with the sound off tells you whether the film makes sense visually.

Start from a shot list, not a script

A storyboard frame illustrates a shot, so you need the shots first. If you do not have a shot list yet, make one: number each shot and give it a size, an angle, a movement, a subject and one action. The AI filmmaking workflow shows a worked example.

Before drawing any frames, finish your look development: a style line you will paste into every prompt, and approved reference images for each character and location. Storyboarding without them produces a set of frames that each look good and do not look like the same film.

Writing a prompt for each frame

Translate each shot-list line into an image prompt. A frame is a single moment, so describe the peak of the action rather than the whole movement: not "she walks to the window" but "she stands at the window, one hand on the frame, looking out". If the shot moves, draw the frame where it starts; you can add a second frame for where it ends.

Medium close-up, eye level. An elderly lighthouse keeper in a heavy navy wool sweater stands at a salt-streaked window, one hand on the frame, looking down toward the sea. Lit by a single oil lamp to her right, warm on her face, cold blue dusk outside. Muted teal and amber palette, 35mm lens, shallow depth of field, soft film grain.

• Attach the character reference to every frame the character appears in.
• Name the light source and direction in every interior. It keeps light consistent across a scene.
• Set the aspect ratio in the model settings to match your final film, so compositions are designed for the real frame.
• Leave out camera movement words from still prompts. They confuse image models and add nothing to a single frame.

The grid method: several panels in one image

Another approach is to ask an image model for a single image laid out as a grid of panels, such as a 2x2 or 3x3 sequence of one scene. Because the panels are generated together, they tend to share the same character, costume and light more closely than separate generations would. The trade-off is resolution: each panel is a fraction of the image.

The grid works best for quick coverage of a scene, like an establishing shot, a medium and a close-up of the same moment, or for exploring how a sequence could be broken into shots. Split the panels into separate frames afterwards, keep the good ones, and regenerate the strongest panels at full size as proper frames. On Pindow, the Prompt Engine includes grid and sequence templates for this, and the Grid Extract action on an image splits a grid into individual frames on the canvas.

Keeping the boards consistent

Consistency is the difference between a storyboard and a mood board. Check each new frame against the references and against the frames on either side of it. The details that most often slip between frames are hair, costume colours, props changing hands, time of day and the direction of the light.

If a character will not hold, work through the consistent characters method: stronger references, a locked description and shot choices that stay close to the reference angle. For locations, generate a wide establishing frame first and use it as a reference for closer frames in the same place.

Reviewing the storyboard as a sequence

Lay the frames out in order, left to right, and read them like a comic. Better still, drop them into an editor with rough timings from the shot list and play them as an animatic, with temporary music or a read of the dialogue. Ask:

• Can someone follow the story with no dialogue?
• Is the screen direction consistent? If a character exits frame right, do they enter the next shot from frame left?
• Do eyelines match? When two characters look at each other across a cut, their gazes should meet.
• Is there variety in shot size? A string of mediums feels flat; a wide, a medium and a close-up give a scene shape.
• Is every shot necessary? Cut anything that does not move the story.

Fix problems now by regenerating frames. It is the last cheap moment in the process.

Boarding a dialogue scene

Conversations are where storyboards save the most trouble, because a dialogue scene is really a small system of shots that have to agree with each other. Board it in three layers. First a wide or two-shot that establishes where both people are. Then an over-the-shoulder from each side. Then a closer single on each character for the lines that matter most.

Draw a line between the two characters on your establishing frame and keep every camera position on one side of it, following the 180-degree rule. If character A looks right to left in their single, character B must look left to right in theirs. Check this on the boards, where it takes seconds to fix, rather than in the edit, where a crossed line means regenerating a whole side of the conversation.

From storyboard to video

Once the boards are approved, each frame becomes the first frame of its shot. Give it to an image-to-video model with a prompt about motion only: what moves, how, and what the camera does. For shots where the motion must end on a particular composition, generate the end frame as a second storyboard frame and use a model that accepts both. First frame and last frame in AI video explains how to design a pair the model can bridge.

On Pindow, the storyboard frames, the references and the video generations can all live on one canvas, so turning an approved frame into a video shot is a right-click away, with the frame carried over as the start.

Common storyboarding mistakes

Frames that are too busy

A frame packed with detail animates poorly and reads badly at a glance. Give each frame one clear subject and one clear point of interest.

Every frame at the same size

If every board is a medium shot, the edit will be monotonous. Plan sizes deliberately, using the shot size vocabulary.

Describing motion in a still

Image models cannot render a pan or a push in. Describe the moment, and save movement for the video prompt.

Fixing it later

Problems in the storyboard become more expensive in video and harder still in the edit. Solve them while they are stills.