Make a short film with AI

One idea enters the canvas, GPT-6 Astra turns it into three shot prompts in the language of cinema, and Veo 3.1 Lite shoots them with their sound. The full workflow is visible without an account, with the edited film.

Interactive demo

Real workflow, real results. Pan and zoom freely.

Use this workflow

Real workflow: one idea in a sentence, three Text nodes on GPT-6 Astra each writing one shot, three Video nodes on Veo 3.1 Lite, and the Edit node joining the twenty-four seconds.

At a glance

What the workflow produces: 3 videos of 24s in 720p, 16:9
Canvas structure: 9 nodes wired by 9 connections
Models wired in: GPT-6 Astra, Veo 3.1 Lite
Cost of one full run: about 180 credits, roughly $2.16

There is only one place where you write in this workflow: the opening sentence. « Le dernier train de la nuit s'arrête dans une petite gare de campagne déserte, sous la pluie, et une seule personne en descend. » Everything else follows from it, all the way to the edit. That is what makes the canvas interesting: text is not a setting here, it is the first link.

The three Text nodes do not hold a fixed text, they hold a MODEL. GPT-6 Astra receives the same idea through a cable in all three, and each asks it for a different shot: the opening, the moment the train arrives, the end. What comes back is not a summary, it is a shooting prompt: lens, camera move, light, and what is heard. The prompts shown in the demo are the ones it actually wrote.

Those three prompts then go into three Veo 3.1 Lite nodes, eight seconds each, with sound. It is the cheapest model in the catalogue that outputs audio with the image, and a silent short film does not hold: the rain, the brakes and the hiss of the doors come from the video model, with nothing to add. The Edit node joins the three shots. The whole film costs a hundred and eighty credits, about two dollars twenty.

How it works

1
Write the idea in one sentence
A place, a moment, an event. « The last train of the night », not « a story about a train ». The more concrete the sentence, the more the three shots it produces hold together.
2
Let the instructions do the cutting
Opening, middle, end. They are what turns an idea into a film rather than three shots that look alike. They can be rewritten: four shots, a cutaway, a close-up.
3
Read the prompts before shooting
A shot prompt is judged in ten seconds: is there a lens, a move, a light, a sound. If one is missing, complete the instruction rather than the prompt, and rerun that node.
4
Generate everything, then watch the edit
"Generate all" walks the chain in order: the three texts, then the three shots, then the edit. A film is judged edited, never shot by shot.

Variations and use cases

The same canvas, tuned for different needs: each variation is two or three lines of prompt away.

The teaser for a story not yet written

Twenty-four seconds are enough to know whether an idea stands up. It is far faster than a synopsis, and far more convincing when you have to show it to someone else.

A mood film for a brand

The idea describes the place and the moment rather than the product. The three shots give an atmosphere, and the product lands on top at the edit, or in a fourth shot added to the canvas.

A music video

One sentence per verse, three Text nodes per sentence, and the canvas grows to the length of the track. The Edit node takes twenty-four clips, enough for several minutes.

Testing a directing intention

Keep the idea, change the instructions: "handheld camera" everywhere, or "long locked-off shots". The same story changes nature entirely, and you see it in one generation.

The animated storyboard of a real shoot

Before sending out a crew, these twenty-four seconds tell the cinematographer what you are after. A prompt written by Astra already carries the lens and the move, it reads like a director's note.

The film, then its score

Sound already comes out of the shots, but if you want music underneath, the add sound to a video workflow makes it and the Edit node lays it under the three clips.

Common mistakes

Writing an abstract idea

"Loneliness" gives nothing, "the last train of the night in a deserted station" gives a film. The model needs a place, a moment and an event to write a shot.

Rewriting the model's prompt by hand

If it does not suit you, fix the instruction and rerun: the next prompt will be better for the same reason. Fixing the output by hand throws away the benefit of the node on the following generation.

Asking for dialogue

Veo can make people speak, but over eight seconds one line eats the shot and rings false. The three demo prompts explicitly forbid dialogue, and that is what makes them cinematic.

Judging a shot on its own

The second shot always looks plain in isolation. The edit is what tells you whether the film holds: watch the twenty-four seconds before rerunning anything.

Recommended models

Frequently asked questions

Do the demo prompts really come from the model?

Yes, as they are. They were produced through the same path the product uses when you click Generate, and copied without being edited. That is also why they are long: that is the shape a well-written shot prompt takes.

Why three Text nodes and not one?

A model asked for three shots at once returns a block of text, and a block cannot be cabled into three Video nodes. One shot, one node, one cable: that is what makes each shot rerunnable on its own.

Can I change the video model?

Yes, the node offers about fifteen. Veo 3.1 Fast gives a richer image for three times the price, Kling V3 accepts fifteen seconds per shot. The prompt Astra wrote stays valid for all of them.

How much does a short film cost?

A hundred and eighty credits for the twenty-four seconds: twelve per text, forty-eight per eight-second shot with sound. The price shows on each node before you click, and rerunning one shot costs only that shot.

What is the maximum length of a shot?

Eight seconds on Veo 3.1 Lite, which also offers four and six. For longer shots, Kling V3 goes to fifteen seconds and Wan 3.0 to thirty, in the same place in the node.

Is the edit really done in the canvas?

Yes, and in your browser: the Edit node assembles the three clips without sending anything anywhere and without costing a credit. The demo film comes out of it as it is.

Can I write the idea in another language?

Yes. GPT-6 Astra answers in the language of the request, and the prompts it writes are understood by the video models whatever the language. The demo is in French.

Can I do it from a phone?

Yes. The canvas runs on mobile, and this is a workflow where you type a sentence and then wait. The create on mobile guide explains the gestures.

Make a short film with AI

Use this workflow

Free account, no card required. The workflow opens pre-filled in your canvas.