Make a video from a script, step by step

What goes wrong

Before you start

• A script of 60 to 90 words. Spoken at a calm pace that is 25 to 35 seconds, which is four to six shots.
• One line per shot saying where it is, who is in it, one action, and the frame size (wide, medium, close).
• A NOLGIA account. New accounts start with 350 free credits, and the price of every shot is on the Generate button before you press it.
• Nothing to upload. Every shot here is Text to video.

At two in the morning the city belongs to the ones who are still awake. The dancer in the empty car park, rehearsing for nobody. The chef closing up, lit by the last blue flame. The two friends on a rooftop who missed the last train and did not mind. The taxi driver singing to an empty back seat. Nobody is watching. That is the point.

Turn the script into a shot list

Read the script aloud and mark where the picture changes. In ours it changes four times, one per sentence about a person, so the shot list has four rows. The opening line and the last two lines are narration over the first and last shots, not shots of their own.

Generate each shot in the video composer

Open the video composer. Under Start from, keep Text to video. Pick the model with the chip above the prompt: we used Kling v3 (Kling 3.0 in the menu) for the dancer, the chef and the taxi, because it keeps a moving body believable, and Seedance 2.5 for the rooftop. Type the shot's prompt, then press one chip in the Camera move row and pick a strength. The composer shows the exact sentence it will add to your prompt under the chips, and never rewrites what you typed.

• Set Length to 6 s and the shape to 16:9. Kling v3 renders 3 to 15 seconds and Seedance 2.5 4 to 30; the composer only offers lengths the model can make.
• The price is on the Generate video button. It is the live quote for this model, length and quality, and it is what you are charged.
• Press Generate once per shot. Each clip lands in your Library and in the Recent generations row under the form.

A multi-storey car park at night, empty, wet concrete floor reflecting magenta and cyan strip lights, thin haze. A young woman with cropped bleached hair wearing a black hooded tracksuit with a thin white stripe, the jacket zipped up to the chin, black cargo trousers and white trainers, dances alone in the centre of the frame, a slow controlled groove, arms cutting through the haze. Wide shot. Cinematic, anamorphic flare on the lights, fine film grain. No text, no logos.

SHOT 2 (Truck left, subtle): A small late-night restaurant kitchen, stainless steel, most lights off, one gas burner still lit with a blue flame. A chef in his forties with a shaved head, a black apron over a white tee, wipes down the steel pass, then reaches over and turns the burner off; the blue flame throws cool light on his face against warm amber pendant lights behind him. Medium shot, shallow depth of field, thin haze, cinematic. No text, no logos.

SHOT 3 (Crane up, medium, on Seedance 2.5): A city rooftop at night, two friends in their twenties sit side by side on a low ledge with their backs to the camera, a man in a red puffer jacket and a woman in a long camel coat; beyond them an anonymous city of dark rooftops and plain distant towers with small warm window lights, no famous landmarks, a hazy sodium glow in the sky, wind in their hair; one laughs and leans against the other. Wide shot from behind, haze, cinematic. No signs, lettering or logos anywhere.

SHOT 4 (Handheld, subtle): Inside a taxi at night, seen from the empty back seat: the driver, a man in his fifties with grey stubble and a flat cap, sings along to the radio and taps the wheel, rain on the windscreen, streaks of neon and traffic lights sliding across the glass and his face. Shallow depth of field, cinematic, film grain. No text, no logos.

Record the narration in the audio composer

Open the audio composer, choose Voiceover, paste the whole script as the prompt and pick a voice. We used ElevenLabs v3 with the voice Rachel at speed 0.95. Text to speech is priced by the length of the script, so the price on the button changes as you type. Our 66 words came back as 25.2 seconds of audio.

For music, the same composer makes a score from a description. We asked Lyria 3.5 for a slow synth pulse for a night-time city film and used the first 30 seconds under the cut. Keep the description about mood and tempo, not about the pictures.

A slow, moody synthwave pulse at 84 bpm for a night-time city film: warm analogue pads, a soft side-chained bass, distant reverb, no drums until the second half, no vocals, cinematic and intimate

Cut it together in Studio

Open Timeline and press New timeline; it asks for a project first, so name a new one. In the Media column on the left, open Library, find each shot and press its +: the clip lands at the playhead on a fresh track. Press End to move the playhead to the end of what is there before you add the next one, so the four shots run in sequence. Then press Home and add the narration and the score the same way; they land on audio tracks under the picture.

• Studio saves as you go. There is no Save button.
• A 2 minute score is longer than the cut. Put the playhead at the end of the picture, press C with nothing selected to split every clip under it, click the tail of the score and press Delete.
• Press Export video in the header when the cut plays the way you want. The MP4 lands in your Library, at no charge.

The finished piece

Twenty five seconds, four performers, one voice. Nothing in it needed a face to match from shot to shot, which is exactly why this shape is the fast one.

Which tool for which job

Where you still regenerate

• Signs you did not ask for. Kling v3 put lit signs with lettering on the rooftop skyline in three takes, even with "no signs" in the prompt. Seedance 2.5 rendered it clean once the skyline was described as what is there: dark rooftops and window lights.
• A walk-on instead of a dance. The first dancer take stepped up to the lens instead of dancing. Say the action again at the end of the prompt, and run it twice.
• Sound. Kling v3 renders its own sound in the composer; we made these shots without it, so the narration and the score are the whole soundtrack. If a shot's sound fights the score, mute its track in Studio.
• Length. A beat that needs more than one move is two shots, or the Extend option on the same model.

What it costs

Each shot is priced by model, length and quality, the narration by the number of characters in the script, the music per run. The cut and the export are free. The live rates are below.

Checklist

• Script read aloud and timed; one shot per change of picture.
• Every shot prompt names the place, the light, one action, the frame size and the wardrobe.
• A camera move chip pressed on every shot, strength chosen.
• Narration recorded separately, music described by mood and tempo.
• Shots trimmed in Studio, voice aligned to the first shot, own sound muted where it fights the score.
• Exported once, watched once end to end before sharing.

Try it yourself

The four prompts above, the camera chips named beside them, Kling v3 at 6 s and 16:9 (Seedance 2.5 for the rooftop), and the script pasted into the audio composer as Voiceover on ElevenLabs v3 reproduce this piece. The model rolls differently every time; the place and the wardrobe are what the words hold.