Doodle Reel example output generated by Doodle AI — A real run of this skill. The input is a synthetic portrait, not a customer photo.

The photo used as input for Doodle Reel — A real run of this skill. The input is a synthetic portrait, not a customer photo.

Ready to use

Doodle Reel

A short clip starring your doodle character

Use this skill

Your photo: Required
Output: 5-15s clip with sound, landscape 3:2
Category: freeform
Cost: 5 credits

video
reel
character

What it does

What you get back

Reference-to-video keeps a character recognisable across a clip that was never drawn: you supply doodles that say who they are, and describe what they should do. Expect one continuous camera move, full-body or waist-up staging, a simple flat background and a sparse playful sound bed. Costs one credit per second of clip, renders in the background, and arrives in the chat about a minute after you ask.

When the agent picks it

Use when the user wants a NEW short clip starring their doodle character — "make a video of her on a scooter", "a clip of this character waving hello from a beach". The reference doodles fix the likeness, and the shot itself is new. Not for animating one exact picture: that is doodle-motion.

The full skill file

Every skill is one SKILL.md — this is the exact file the agent reads.

src/mastra/skills/doodle-reel/SKILL.md

Doodle Reel

Film a NEW shot starring a character the user already has doodles of. The references define who the character IS — hair, face, colours, outfit — not the composition of the shot.

When this is the right skill

Pick this when the request describes something that is not already in the picture:

“make a video of her riding a scooter”
“a clip of this character waving from a beach”
“put my doodle in a coffee shop, five seconds”

Choose doodle-motion instead when the user wants THAT exact drawing to move (“animate this”, “make the one you just made wave”). Getting this backwards produces a technically fine clip of the wrong thing, which is the most common mistake with this model.

What to film

One action, described plainly. Walks in and waves. Spins once. Sips coffee.
One continuous camera move — a gentle hand-held drift or a slow push in. Not both, and never a whip pan or an orbit.
Staging that survives a phone screen: full body or waist up, generous empty space, one flat background, at most two doodled props.

How to run it

Gather the reference URLs. Best case is several doodles of the same character from this conversation; one is enough. Put the clearest full view first — references are cited in order.
Call generateVideo with skill: "reel", those URLs as referenceUrls, and the user’s action in description. If they named a length, pass seconds — otherwise it renders 5 seconds at 1 credit per second.
If the user has no doodle of the character yet, make one first with an image skill, show it, then offer the reel.
The tool returns status: "queued" with a jobId: the clip is RENDERING. Say it will land here on its own in about a minute and that it has sound, then stop.

What not to do

Do NOT re-call the tool for the same request, or call it to check progress. The finished clip is delivered to the app on its own.
Do NOT send more than ten references — the extras are dropped, and the first ones are the ones that shape the character anyway.
Do NOT describe or evaluate the clip. You have not seen it.
Do NOT film a real, named person. This skill is for the user’s own doodle characters.