How to make a multi-scene AI video

Every film in Film Studio goes through the four steps on DreamBench’s AI video generator page. Here they are with the buttons you press.

Step 1: Describe the video

1. 1Open Film Studio and write your idea in the Your story box: a line or a page.
2. 2Under Who’s in it?, add pictures of the people, products or places that must appear, up to 8. Give each a name and tag it as a character, creature, location or prop, then cite it in the story with @ and its name.
3. 3Pick one model, the length, the shape and the resolution. The composer shows the price in credits before anything starts.
4. 4Leave the switch on Review each step and press Write the script. Switch it the other way and the button reads Make the film, which runs from your story to the finished film without stopping.

Step 2: Read the script

1. 1When the film’s card says Your turn, open it. The script shows a title, a one-line summary and every scene and shot in a sentence, with who is in each.
2. 2To change one shot, press Change under it and write a note, such as “make this a close-up of her hands”. The rest of the script stays word for word.
3. 3To change the whole story, press Rewrite with a note.
4. 4When it reads right, press Approve the script. Or let it run to the end from here without stopping again; Stop here keeps the script and charges nothing after it.

Some films first show your idea beside a few twists on it, under Pick the idea. Pick one, or keep your own, and press Write the script.

Step 3: Meet the cast

1. 1Read the cast: a reference sheet for every character, place and prop the script needs, drawn from your pictures where you gave them.
2. 2If a sheet isn’t right, set how many new takes you want and press Draw 3 takes, then pick one and press Use this take. Or press Use my picture to draw it from a photo of yours.
3. 3Press Approve the cast. Every scene is then shot with these sheets and cut into one film.

On a film of two minutes or more, the button reads Shoot the pilot: a few of its shots, about a tenth of the film, are shot first. Watch them under Shots, then press Shoot the rest, or Stop here, having paid for the pilot alone.

Step 4: Watch it, then fix any scene

1. 1Each shot shows on the board as it lands, ready to play.
2. 2Watch the finished film, then press Open shot editor and click a shot that misses: say what to change, and regenerate it.
3. 3Change as many shots as you like, regenerate them together, then confirm the version of each that goes in the film.

With Review each step on, each step is paid as it starts: the script when you press Write the script, the cast when you approve the script, the shots when you approve the cast. Together they cost what the film costs without stopping, and stopping early costs nothing more.

The plan is built from the brief alone. A good one names the story’s beats, the cast, the places and the look; leave one out and the writer guesses it.

• Story beats: what happens first, what changes, how it ends. Three to five moments is plenty for a minute. “A boy loses his kite, chases it across town, finds it in a stranger’s garden” plans better than “a film about a kite”.
• A cast: who is in it, each with a few details that stay true in every scene: age, hair, one item of clothing. Give a name to each, so the script can follow them.
• Places: where each beat happens, and the time of day. Fewer places hold together better; a new place is a new scene.
• A look: one sentence on the style, such as “handheld, warm evening light, 35 mm film grain” or “soft watercolor, pastel colors”. It goes with every shot.
• The frame and the sound: say 9:16 or 16:9, and put any spoken line in quotes after its speaker.

For a face or a product, a picture beats a paragraph: add it under Who’s in it? and cite it, as in “@mia opens the box”. How it reaches each shot is reference to video. The AI video prompt generator can turn a rough idea into a fuller brief.

Every join is a hard cut between two separate generations. Neither shot saw the other, so only what both were told connects them. DreamBench’s rule for every cut has two halves.

What carries across a cut

• The place and where everyone stands in it: someone on the left stays on the left.
• Faces, hair and clothes, and poses: a person sitting is still sitting.
• What hands hold, and anything moved: a cup half drunk stays half drunk, a door left open stays open.
• The light: the same time of day, the same direction and color.
• Momentum: a runner going left to right keeps going left to right, at the same pace.

What changes at a cut

The framing has to change clearly: at least two steps of shot size, from wide to close-up for instance, or a quarter-turn of the camera, or a cutaway to something else, or a new scene. Two shots at nearly the same size and angle read as a jump, as if frames were lost.

The same face from shot to shot is character consistency; a steady face inside one shot is temporal consistency. Multi-shot video explains cuts inside and between clips.

Sound across cuts

On models that make sound, each shot’s sound is generated with its picture. Keep a line of dialogue inside one shot and cut at the end of lines, the way a film cuts between speakers. More in native audio.

A scene is one place and moment; a shot is one piece of video between two cuts. A scene is usually two to four shots. Film Studio cuts a new film into shots of 3 to 12 seconds, so a length sets the fewest shots a film can have. On Veo 3.1, whose shots are 4, 6 or 8 seconds, it takes more.

The writer sets the shot count by the beats, up to 40 shots: a glance needs a short shot, a walk across a room a longer one. Fewer, longer shots slow the pace; many short ones quicken it.

A five-minute film is ten to fifteen scenes and at least 25 shots, on Ultra or Studio. Four things make it go well.

• A full brief: every scene in order, the cast, the places, the look and the ending. Under Film Studio’s composer, a film of two minutes or more offers Insert a brief template.
• A pilot: with Review each step on, approving the cast shoots the film’s opening first, about a tenth of it. If the look is wrong, stop there, having paid for the pilot alone.
• A draft that costs less: on Seedance 2.5, make it at 480p, fix the shots that miss, then press Upscale to 1080p under the finished film.
• Parts for anything longer: on a finished film, Continue in a new film opens the composer with its cast, its settings and a brief that starts where it ends.

The same method for a lesson, step by step, is in how to make an educational video with AI.

One model makes every shot of a film: each model reads a reference in its own way, and a film that switched models would change its look at the cut. Choose by how your cast reaches each shot, and by the sound.

Each model page has its limits and examples: Seedance 2.5, Seedance 2.0, Veo 3.1, Kling 3.0, Kling 4.0 and Wan 3.0. Film Studio writes the shot prompts itself, but the prompt guides explain what each model does well, which helps you write the brief: the Seedance 2.5 prompt guide, the Veo 3.1 prompt guide and the Kling 3.0 prompt guide.

Every shot is its own generation, so one bad shot never means a new film. Fix it at the earliest step you can: words cost less to change than pictures, and pictures less than shots.

A redo costs the seconds you regenerate plus a small edit fee, shown before you press.

Each names its beats, cast, places and look. Copy one, change the details, and read its script before you approve it.

A 30-second short film: the lost kite

A short film in three scenes, 16:9. A boy of about eight with a red scarf and a yellow raincoat flies a blue kite on a windy hill; the string snaps and the kite drifts over the rooftops. He chases it through narrow streets, past a baker and a cyclist. He finds it caught in an apple tree in an old woman’s garden; she climbs a ladder, hands it down and he grins. Overcast autumn light, handheld camera, soft film grain. Wind, footsteps, one line from the woman: “Hold on tighter next time.”

A 1-minute product film: a steel water bottle

A 1-minute film for @bottle, 9:16. Morning: a trail runner with short black hair and a green jacket fills @bottle at a mountain stream. Midday: she climbs a rocky ridge, @bottle clipped to her pack, and drinks at a lookout. Sunset: at the summit she opens @bottle and offers it to a friend. End on @bottle in close-up on a rock, the label facing the camera. Crisp natural light, wide landscapes cut with close details, the sound of wind and water.

A 2-minute story: two neighbors and a piano

A two-minute story in six scenes, 16:9. @ana, a retired music teacher in a gray cardigan, plays an upright piano every evening. Upstairs, @leo, a teenage boy with headphones, bangs on the floor. One evening the piano stops, and after a week of silence he goes down to check on her. She is unwell; he makes tea and plays the piano badly. Weeks later they play a duet, neighbors listening in the stairwell. Lamp-lit rooms, slow camera, the piano in every scene.

The @ handles are pictures added under Who’s in it?. Without pictures, Film Studio designs the cast from the words.

A brief written shot by shot, as made on DreamBench

You don’t need Film Studio to make a multi-scene AI video. Done by hand in any video model, the method is the one a film crew uses:

• Write a shot list: what each shot shows, its framing, its length and who is in it.
• Make a reference picture of each place and a character sheet of each person, then use the same ones for every shot.
• Write each shot’s prompt with the characters described in the same words every time, and generate the shots one by one on the same model.
• Check each shot against the sheets and generate again the ones that drift.
• Join the shots in a video editor, in order, with hard cuts, and fix the sound at the joins.

Runway publishes such a workflow for one 15-second Seedance 2.5 scene: an empty location first, then “a three-panel character sheet: front, back and a close-up” for each performer (Runway). Kling’s Video 3.0 has a Multi-Shot switch that “automatically plans the shot transitions” inside one generation of 3 to 15 seconds (Kling AI).

When doing it by hand is the better choice

• A single scene under 15 seconds: a model’s own multi-shot mode cuts inside one generation, with every frame made together.
• Control of every prompt and every take, in the model maker’s own app.
• Your own edit: transitions, titles and music, which Film Studio doesn’t add.

When Film Studio is the better choice

• A film longer than one generation: the script, the sheets, the shots and the cut come from one brief.
• The same cast in every shot: the sheets are drawn once and go with each shot.
• Fixing as you go: a shot at the script, a sheet at the cast, a shot in the finished film.

Film Studio’s price depends on the length, the model, the resolution and the sound, and is shown before you start; plans are on the pricing page.

• Perfect matches at every cut: shots made apart can still differ in a small detail, such as a ring or which hand holds the cup.
• Hands and bodies in contact, fine work with the fingers and words that must read are the shots models get wrong most often.
• Film Studio has no start or end frame to set for a shot, no way to extend one clip from its last frame, and no upload of your own voice or soundtrack to lip-sync.
• A film is one model and one shape throughout; to try another model, make it again.

A film of generated shots can pass for real footage. Say it is AI-made where it could mislead, and never put a real person in it without their consent. A picture that appears to show a real person under 18 is refused.