Storybook: A Cartoon From One Sentence

Everyone has a cartoon in them. The bedtime story that got a laugh. The explainer your team keeps re-explaining. The owl who opens a food truck and immediately regrets it. The trouble is what happens between the idea and the video: either a week inside an animation tool, or a video model that gives you six seconds of beautiful soup where nobody's face survives the cut.

Storybook takes a different route. You describe the story in a sentence or two, pick a look and a length, and it writes the script, paints the cast, the sets and the props, records the voices, and animates every shot. But it doesn't ask a video model to dream up the frames. It works the way a puppet theatre works: the characters are cut-out figures, and for every shot the AI writes stage directions. Who stands where, who says what and when, where the camera looks, what gets picked up. Our own animation engine then performs those directions, exactly, frame by frame, in your browser, and the result exports as a real MP4 or opens in the Video Editor for polish.

That one decision, staging instead of dreaming, is why the characters look the same in shot nine as in shot one, why a shot can be corrected with a note instead of a re-roll, and why the whole thing runs on your own Gemini key without us in the middle.

What you get

Type "Two rival sky pirates get stuck sharing one tiny hot-air balloon and slowly become friends." Choose a format (landscape, vertical, square), a look (flat vector, paper cut-out, Saturday-morning, storybook ink, chibi), and a length from a 30-second short to a four-minute episode. Then:

The script. Scenes, shots, dialog, and a cast with clear descriptions, written for the medium: cut-out puppets that walk, turn, gesture and talk. You read it before a single image is painted.
The cast, sets and props. One full-body sheet per character, painted and cut out on your machine, with a talking mouth and blinking eyes placed on it (and a simple face editor for when the AI guesses the mouth wrong). One painting per set. One picture per hero prop, the locket or the last piece of sushi, so it's a real object, not a coloured box.
The voices. Every line recorded with a distinct voice per character, and lip sync taken from the audio itself.
The shots. Each one staged and previewed live in the studio. Don't like a beat? Type a director's note, "make Rico jump for joy after he speaks", and the shot is revised, not started over.
The cartoon. Download the episode, save it to your library, or open it in the Video Editor with every shot as its own clip and its dialog baked in, so trims and reorders keep the lip sync.

Why it's staged, not dreamed

Ask a video model for a cartoon and you're negotiating with a slot machine. The face drifts. The hat changes colour. Shot two doesn't start where shot one ended. And when it's almost right, your only tool is to pull the lever again.

Storybook's animation engine is ours. It has a stage, characters that walk and gesture and talk, props they can hold, a camera, and a clock that counts frames. The AI's job is to direct: a few dozen stage directions per shot. Before a shot is kept, the engine performs it once from start to finish and checks it. A line that was never spoken, a character who never made it on stage, a camera that wandered off the edge of the background: each of those goes back to the AI as a specific note, and the AI fixes it. If a shot still won't come right, the engine falls back to a simple, clean staging so the episode always exports.

Because it's staged, a shot plays exactly the same every time you render it. Because it's staged, you can direct it: your note becomes a revision of the same shot, not a new roll of the dice. And because the characters are cut-outs the engine places, not pixels a model imagines, consistency is built in rather than hoped for.

Continuity is geometry, not luck

Within a scene, each shot records the picture it ends on: where every character stands, which way they face, where the camera is and how close. The next shot is written to open from exactly there. The studio then checks the cut with a ruler: if the opening doesn't match the previous ending, it tells you by how much and offers to re-stage the shot to match. Scene changes cut freely, as they should.

Props are real, and characters hold them

The first cartoons people made with Storybook had a tell: the hero object, a brass locket that drives the whole plot, was a drawn rectangle in four different shots, a different size each time. So props became real. The script lists them, they're painted once as cut-outs, and the studio tells the animator how big each one should be next to a character.

Then the second tell showed up: a camera held in a character's hands drifted away from her hand as she walked, because it was being moved beside her instead of with her. Now a character holds a prop. It becomes part of them, riding the walk, the bob, the lean and the turn, and can be raised to the mouth, the eyes or the chest with a single direction. Chopsticks lift a piece of sushi to a mouth that actually moves. We also gave walks a natural pace and taught the camera to follow a walker without outrunning them. Small things, but they're the difference between a cartoon and a slideshow.

Your cast, across stories

The expensive part of a cartoon is a cast that looks the same everywhere, and it used to die with its project. Now you save a painted character to your gallery and cast them in the next story: same look, same voice, same description, and the script is written around them.

You can also start from a picture. Attach a photo of your dog, your kid's drawing of a dragon, your company mascot, or a room, and tag it as a character, a set or a prop. Storybook describes it, writes the script around it, and paints it in the chosen look, using your picture as the reference. Style wins over the photo, so a flat-vector cast doesn't get a photorealistic member.

Music, without the fight

Add a bed for the whole episode: generate one from the script's own one-line music brief, upload a track, or pick audio already in your library. In the exported file it fades in, dips under every line of dialog, and loops with a crossfade rather than a hard restart. In the Video Editor it lands on its own audio track for you to shape.

Measured, not vibes

If you've read about Managed Workflows or the highlight-reel pipeline, you know the house rule: what the AI makes should be something you can see and direct, and the way we improve it should be something we can measure.

Storybook's shot staging has a test suite. Twenty scripted scenes, from a two-character argument to a three-shot sequence that has to stay continuous, a crossing walk with three characters, a bun rolling off a counter, and a character eating from a plate. Each is staged by the real AI, performed by the real engine, and graded automatically: every line spoken, nobody talking over themselves, the opening matching the previous ending, the prop held through the walk, the food reaching the mouth. Every time a real cartoon went wrong, it became a test. And every fix the suite has demanded so far traced back to our instructions, not the AI's judgment, which is exactly what you want a test suite to tell you.

Your key, your cost

Every AI call, the script, the images, the voices, the staging, the music, runs from your browser on your own Gemini API key. Our servers store the project and your media, nothing more; they never see a prompt or an image. As with the rest of the AI tools, we don't mark up what the models cost. Storybook is part of the Premium plan.

Practical details

Browser: Chrome or Edge. Rendering and export happen in the browser, so it wants a laptop or desktop rather than a phone.
Style: cut-out and puppet cartoons, motion comics. Not hand-drawn frame-by-frame animation, and we say so up front.
Lengths: 30 seconds to about four minutes, in 16:9, 9:16 or 1:1.
You review before you spend: the script, the cast and the sets are all on the page before any art is painted, and every step tells you how many calls it will make on your key.
Everything is yours: sprites, voices and renders are ordinary media in your library, and the cartoon opens in the Video Editor like any other project.