Learn AI Filmmaking in 90 Days: A Realistic Plan

Ninety days will not make anyone a filmmaker. It is enough time, though, to stop being a person who watches tutorials and become a person who has finished a film. That distinction is the whole plan. The schedule below assumes five to seven hours a week — an hour on two weeknights and a longer weekend session — and it ends with one short film of two or three minutes, exported and shown to somebody. If you have more time, make the film more ambitious, not the syllabus longer.

A warning before the calendar. The most common way this fails is not lack of talent; it is the tutorial loop. Watching is comfortable, and it feels like progress. The plan therefore front-loads making, and every two weeks produces something you can watch, however small.

Weeks 1–2: learn to look

Before generating anything, spend two weeks learning what a shot is. Watch ten short films and take notes on shot size, camera movement, light direction and where the cuts fall. Pause on frames and describe them in writing. This is the vocabulary you will later use in prompts, and it is much easier to acquire from finished films than from a tool's help centre.

At the same time, read the documentation for the tools you plan to use. Runway's Academy collects courses and a prompt guide in one place, which makes it a reasonable first stop. Do not subscribe to anything yet.

The first deliverable: a page of notes on three films, and a list of eight shots you want to be able to make.

Weeks 3–4: stills and the look

Image generation is the cheapest place to learn composition, and the fastest way to build a lookbook: six to ten frames that define palette, contrast, texture and lens character for the film you eventually want to make. Runway's image prompting guide sets out the elements worth controlling — subject, scene, composition, lighting, colour, style — and makes a good checklist for a daily exercise: one image, one variable changed, ten minutes.

Keep every frame you like and note the prompt that produced it. By the end of week four you should have a lookbook and a rough sense of which phrasing changes which visual property. This is the foundation for the prompting practices that carry the rest of the plan.

Weeks 5–6: motion and the single shot

Now add movement. Take one still from the lookbook and animate it. Then take a shot you wrote yourself and generate it: five seconds, one subject, one action, one camera move. Study the failures. Learn how the model handles hands, faces in profile, fast movement and objects leaving frame. These failure modes are not trivia; they determine what kind of film you can write.

Runway's camera reference is useful here because it pairs terms such as push in, tracking, orbit and rack focus with example outputs, so you can see what each word does before you use it in a shot list.

The deliverable for week six: a fifteen-second sequence of three shots that cuts together and makes sense.

Weeks 7–8: sound and the cut

Editing is where generated footage becomes film, and it is the skill most beginners skip. Spend two weeks in an editor. Much of DaVinci Resolve's training material is free, covers editing, colour and audio, and is designed for new users, which suits this stage. Cut your fifteen-second sequence again, then cut a one-minute trailer from material you generate specifically for it. Add ambience, effects and a temporary voice track — ElevenLabs' documentation covers text-to-speech if you want a synthetic voice rather than your own.

Learn three things: how to cut on action, how to build a soundtrack in layers, and how to match two shots so they feel like the same scene. These are worth more than another model.

Weeks 9–10: make the short

Write two pages. One location, one or two characters, one turn, two to three minutes. Make a shot list, then make the film in passes: blocking, performance, texture. Give yourself a hard deadline at the end of week ten: picture lock, no exceptions.

This is the stage where the first-film workflow matters more than any prompt. Work in the order it sets out — script, list, lookbook, test shot, passes, assembly — and resist the urge to keep improving one shot while the rest of the film waits. Expect the first assembly to be worse than you hoped. That is normal and not a signal to restart.

Weeks 11–12: finish and show it

Grade for consistency, design sound properly, set titles in the editor rather than in a prompt, and export. Then disclose. If you plan to submit to an AI-focused festival, read its rules early: the Runway AI Festival's film track, for instance, asks entrants to describe their AI techniques in a short written note, and sets a three-to-fifteen-minute runtime with a linear narrative. If you publish online, platforms such as YouTube publish guidance on disclosing realistic generative content. Do this plainly, in a sentence, and move on.

The final deliverable is not the film. It is the film plus a page of notes: what worked, what you would do differently, and the three skills to practise next. That page is what turns ninety days into a practice.

How to practise deliberately

Three habits carry the plan:

• One variable at a time. Change the camera or the light or the wardrobe, never all three, so you learn what caused the result.
• Keep a log. Model, settings, references, prompt and one line of judgement for every approved take. Future you will need it.
• Finish on schedule. A finished mediocre film teaches more than an unfinished good one. The deadline is part of the curriculum.

The traps

There are six reliable ways to lose the ninety days: hopping between tools instead of learning one, expanding the story every time it gets hard, skipping sound, treating generation as the whole job, waiting for the perfect model, and hiding the use of AI until someone asks. The last one is the easiest to avoid and the most damaging if you don't. Say what you made and how you made it; the work can stand on that.