Turning a written shot list into a style-consistent storyboard
The situation
Your students can write a scene but cannot draw it, so boards either get skipped or come back as stick figures nobody can read. The goal is not pretty frames. It is that students make shot-by-shot decisions — angle, distance, order — and can defend them in a critique.
Steps
-
Break the scene into beats, not shots
Google (Gemini Pro) OpenAI for Education Anthropic-Claude for Education
Paste the scene and ask for the dramatic beats — the moments where something changes — rather than a shot list. Ask it to justify each beat from the text. Students then argue with the output, which is the actual exercise: the model's first pass is usually too even, giving equal weight to setup and climax.
What you only learn by doing it: Ask for beats before shots. If you ask for shots first, you get a generic coverage pattern — wide, medium, close, repeat — and students never make the editorial decision.
-
Lock a visual style before generating anything
Generate one reference image and iterate until it is right — lighting, palette, line quality, period. This single frame becomes the style anchor for everything downstream. Spend real time here; every later frame inherits its problems.
What you only learn by doing it: Students want to start generating frames immediately. The half hour spent on the style anchor saves the entire back half of the project.
-
Build the boards with character consistency managed
Load the cast once as reusable characters, then generate frames from the shot list. Boords is board-first: students still set camera direction, dialogue and notes per frame, and the AI fills the picture. LTX Studio is script-first and better when the piece is long enough that sequence matters more than individual frames.
What you only learn by doing it: Regenerate regions, not whole frames. Fixing one element of an approved frame keeps the rest — restarting the frame loses the composition the student already argued for.
-
Cut a timed animatic with scratch audio
Export the boards against scratch VO and temp sound. This is where pacing becomes visible: a board sequence that reads fine on paper falls apart when it has to hold real time. Students discover their scene is 40 seconds of story in three minutes of frames.
What you only learn by doing it: Have students record scratch VO themselves rather than using synthetic voice. Their own read exposes pacing problems that a perfectly-paced AI voice smooths over.
-
Critique the boards, not the pictures
Run the critique on the decisions: why this angle, why cut here, what does the audience know at this frame that they did not know at the last one. Grade the shot logic and the defense of it. The image quality is the least interesting thing on the screen.
What you only learn by doing it: Put the generated frames next to the student's own thumbnails if they made any. The comparison usually shows the thumbnails had better ideas and worse execution, which is a useful thing for a student to see about their own work.
Where this breaks down
Character drift is the failure that wrecks this. General image generators will give you twelve good frames of twelve different people, which is worse than stick figures because it hides the continuity problem behind polish. Use a tool that manages character references at the project level.
Two licensing traps to hand students in writing before they start: Recraft owns free-tier images outright, and LTX Studio's cheaper paid tier is personal-use only. A student who builds a festival submission on either is exposed.
Finally, watch what this does to a boards unit meant to build drawing confidence. If the learning objective is drawing, generated frames undercut it. If the objective is visual storytelling, they serve it. Decide which one you are teaching before you assign this.