How to Make a Cartoon With AI — From Cast to Finished Episode

If you want an AI-generated episode that looks like one show instead of a pile of clips, start in a cartoon or anime style. Bold shapes, flat color, and a simple silhouette give the model something it can repeat. Photoreal faces have too many soft variables — pores, stubble, the exact fold of a collar — and those variables drift. Cartoons compress a character into a few signals, which is why they're the right first project for almost everyone.

To make this in Paintbrush, choose your project style, create the cast and settings, and review their reference images. Use those same assets in each scene and assign character voices before generating the episode shot by shot.

This is a soup-to-nuts pass for a 60–90 second cartoon short. The same pipeline scales to a series once the cast exists. For a deeper style-specific walkthrough, see the anime guide.

Step 1: Name the show's look in one sentence

Not "cartoon." Saturday-morning thick outlines and saturated primaries is a different show from a muted indie short with soft fills. Write one sentence at the project level — "chunky outlines, limited pastel palette, no gradients, sitcom blocking" — and leave it there. Do not renegotiate the aesthetic in scene 6 because a single clip looked cooler another way. That clip will be the one that doesn't belong.

Step 2: Design a cast of two or three

Write each character as a silhouette test: if you saw only the hair and the coat in black, could you tell them apart? Give them different heights, palettes, and one signature prop. Skip photoreal skin detail. Generate the reference sheet (front, side, and back) and reject it if the side view invents a new haircut. That sheet is the show bible.

Cast voices with the same contrast. A low narrator and a bright kid lead will carry a scene even when the drawing is simple.

Step 3: Draw the map, not a new background every shot

Two locations are enough for a first episode: home base and the place the problem happens. Describe materials and time of day, not a travelogue. "A small diner at night, red booths, rain on the glass, neon pie sign" will hold. "A beautiful city" will not. Reuse the diner until the story leaves it.

Step 4: Stage like TV, not like a movie poster

Cartoons read on medium shots and close-ups. Write action you could act on a mark: enter, point, drop an object, turn to the door. @mention the character and the setting; don't restate the design. Keep two people in frame. A group shot is three clips of pairs, not one crowded generation.

Chain shots that stay in the same room. Hard-cut when you change location or jump time. Three to five seconds per clip is the cartoon tempo — hold a pose, cut, hold the reaction.

Step 5: Record the track as you generate

Write the line in the scene audio field. Let duration follow speech. If a joke needs a pause, split the pause into its own short clip of a reaction face rather than asking one generation to hit the setup and the laugh. That's how 2D shows are boarded. The tool should follow the board, not the other way around.

Step 6: Edit like a cartoon, not like a film

In the editor, cut on the action. Add a simple music bed and captions. Don't grade every clip toward photoreal — you'll fight the style. If one shot slips off-model, regenerate that shot against the same references. Don't "fix" it by prompting a new outfit.

Episode two is supposed to be faster

The first short pays for the bible: cast, diner, look. The second episode should mostly be new scenes on those assets. That's the difference between making a cartoon and making a clip. If you redesign the lead because you saw a prettier generation, you don't have a show. You have a mood board with a laugh track.

The takeaway

Make the show rules first — look, cast, two rooms — then generate short, staged clips that obey them. Cartoon language is how you get consistency "for free" from models that were not trained to be your continuity supervisor. The reference sheets are that supervisor. Use them.

To keep your cast recognizable from shot to shot, read how character references work.