Make a short AI movie

A short AI film is the most ambitious thing on this site, and the one where expectations need managing hardest. You can make a two-to-three-minute voiced, scored short entirely with AI — but it's a project, not a click. The thing that makes or breaks it isn't any single tool; it's *consistency* across shots. Get your character to look the same in every scene and it feels like a film; let them morph between shots and it feels like a glitchy slideshow. This workflow is built around protecting that consistency. Budget a couple of hours and expect to regenerate scenes.

Scripting first gives every later step a target: the voice knows what to say, the video knows what to show, the edit knows the order.

What you'll walk away with

• A 2–3 minute short-film script with scene headings and dialogue
• Distinct AI voices cast for each character
• Generated scenes with the visual consistency that sells the illusion
• A finished, color-graded short ready to share

Step by step, with the judgment that matters

1. Write the screenplay with ChatGPT. Ask for a 2–3 minute script from your logline, with scene headings, brief visual descriptions, and dialogue. Keep the cast small — every additional character is another face to keep consistent later. Fewer, stronger scenes beat many thin ones.

2. Cast voices in ElevenLabs. Pick a distinct voice per character and generate their lines scene by scene. Consistency applies to voices too — once a character has a voice, never switch it. Match the voice to personality, not just gender.

3. Generate scenes in Kling — and copy-paste your character description. This is the whole game. Use the *identical* character description (hair, clothing, age, build) in every scene prompt. Any drift in wording produces a different-looking person. Ask for cinematic lighting and natural camera movement, and expect to regenerate shots that stray.

4. Edit in CapCut. Assemble scenes in order, sync each character's dialogue to their shot, add ambient music, and color-grade everything to one consistent look. A unified grade hides a lot of shot-to-shot variation and is what makes disparate clips feel like one film.

A real run

For a 2-minute short about a lighthouse keeper, ChatGPT writes four scenes with one character. You lock a description — "grey-bearded man, sixties, navy wool sweater, weathered face" — and paste it verbatim into every Kling prompt. Two scenes come out consistent; one drifts younger, so you regenerate it. ElevenLabs gives the keeper one gravelly voice throughout. In CapCut you sync the voiceover, add a low ambient drone, and grade everything cool and desaturated. About two hours, and it holds together as a film.

Where it goes wrong

• The character changes between shots. Your descriptions varied. Keep the character description word-for-word identical across every scene prompt.
• It feels like disconnected clips. Missing a unifying grade. Color-grade all scenes to one look in the edit — it's the cheapest way to make them feel like one piece.
• "Is this beginner-friendly?" It's the most involved recipe here. Budget the time and expect to regenerate several scenes to nail the look.

The honest cost

$10 to about $40. This is the priciest workflow — Kling's free credits run out fast when you're generating and regenerating multiple scenes. ChatGPT, ElevenLabs, and CapCut stay mostly free; the video generation is where the money goes. Runway works as a substitute for the scene step if you prefer its controls.