Make a 40-second story film with one AI agent (no video model)

One coding agent calls an image model and a music model as tools, then writes the code that animates the pictures and renders a short film. No video model, no pipeline.

| 3 min read
Clay robot directs from a director's chair with a megaphone while the orange clay cat films with a vintage camera, three storyboard pictures on easels between them

⏱ About an hour · 💰 No video-model fees · 💻 Needs a computer or a cloud agent

What you'll get: a vertical short film of your own story, 30 to 45 seconds, with music, in any look you like: felt toys, clay, anime, watercolor.

Watch first

This is the film we made with it. Six pictures, one music track, and code that does all the moving.

Why this works

Video models are expensive, slow, and hard to steer: ask for a small change and you get a new clip with a different face. This recipe splits the job instead.

  • The image model makes the pictures: one still per story beat, each edited from the first one, so your characters stay the same.
  • The music model makes the score, as long as the film.
  • Your agent is the director. It writes the code that moves the camera, fades between shots, adds the words and the magic effects, then renders the film frame by frame.

Because the motion is code, every fix is an edit. When we wanted the honest chores wiped off the resume with a wand, that took minutes and cost nothing. No new generation, no new face.

There's no extra framework or pipeline in between: one agent, calling models as tools.

Copy this into your AI agent

Open Claude Code, Codex, Antigravity, or another agent that can read a link and run code on a computer or in its cloud workspace. Paste this:

Review this skill: https://github.com/meeow-ai/meeow_skills/tree/main/skills/stills-to-film
Copy it into your skills folder and adapt it to my setup (which image and music models I can use).
Then tell me plainly whether I can make a short film with it and what I'd need first.
Ask me before installing anything or spending money.

The skill lists everything it needs. Your agent checks what you already have and tells you what's missing before it touches anything. If it has its own image or music tools, it can use those instead.

Then give it your story. A shape that works for short gags: a problem, a helper, a transformation, a test, a twist, and a payoff line.

How it works

Five steps: you tell your story in beats, your agent asks an image model for one picture per beat and a music model for a score, writes the code that moves everything, and renders your film

What it looked like when we made it

The film above, Resume Magic #2, came out of one Claude Code session with this recipe:

  • 8 picture requests for 6 pictures. Two were retries: the first interview shot put both characters side by side, and in the first shocked face the sunglasses stayed on.
  • 1 music track, a 46-second jazzy caper.
  • About 3 minutes to render 44.6 seconds of film.
  • No video model, so no video fees.

The fixes after the first cut were all edits: a longer reading time for the resume, the wand wiping each chore away, gold highlights on the silliest corporate words.

If you get stuck

  • Faces change between shots. Make the first picture of each character, then ask for every later shot as an edit of it ("keep this exact character, now she is…").
  • Two characters stuck side by side. Spell out where the camera is: "seen over her shoulder, the boss across the desk facing the camera".
  • The words fly past. Give each line on screen about 3.5 seconds, and hold the last card 2 to 3 seconds longer.

Next

Already have a film idea with recurring characters? Make the first picture of each one, keep it, and build every episode from it.

Film made on 22 October 2026. We haven't run this exact prompt in other agents yet.