Text-to-video: build a shot from a clear scene brief.

Published by ClapGen · Updated 25 October 2026

Text-to-video starts with words rather than a supplied opening image. The prompt needs to establish the visible scene as well as the action you want to communicate. The public release currently uses simulated requests. Live generation and paid subscriptions are unavailable.

1. Choose text-to-video for the right starting point

Choose this workflow when you want to describe a scene that does not yet have an approved image. You can define the subject, environment, framing, light and movement in one shot brief.

If a particular character or product already has a fixed appearance, a starting image or reference-capable workflow may communicate it more clearly. Text alone is not an identity lock.

In Video Studio, choose a model that exposes the Text workflow. No uploaded image is required for that workflow; adding a reference means you should check which reference workflow accepts it.

2. Write the scene before decorating the style

Start with a concrete subject and action. Then add the location, light, shot size and viewpoint. Say what should be visible at the end so the action has a clear direction.

Original example brief: An orange delivery bicycle rolls slowly past a closed flower stall at dawn. Wide street-level view, locked camera, soft cool light and realistic materials. The bicycle enters from the left and leaves the right side of the frame.

This is a writing example, not evidence of a generated clip. Its useful parts are the identifiable subject, one action, a stable viewpoint and a visible ending. Replace those choices with the needs of your own shot.

3. Make camera and subject movement compatible

A static camera holds its position while the subject moves. A pan turns the camera. A tracking move travels through the scene, often alongside a subject. An orbit asks the viewpoint to move around it. Pick the movement that helps the viewer understand the action.

For the bicycle brief, a useful controlled variation is to replace the locked camera with a slow tracking move beside the bicycle. Keep the location, action and light the same while comparing that one change.

Avoid asking the same shot to be both completely static and an orbit, or asking a brief action to contain several incompatible changes of viewpoint. More instructions can make a request less clear when they conflict.

4. Match the brief to available settings

Choose a duration offered by the selected model and allow enough time for the action you described. Choose the aspect ratio for the intended composition, then select an available resolution. These controls vary by model and workflow.

The prompt is creative direction, while the controls specify supported request settings. Writing a different duration, resolution or camera capability into the prose does not add an unsupported option.

Check audio separately if the model offers it. A text-to-video brief does not by itself guarantee generated dialogue, music, exact spoken wording or a particular sound track.

5. Review one problem at a time

For a future live clip, ask whether the scene contains the intended subject, whether the action completes and whether the camera moves as intended. Then inspect anatomy, contact, object detail and background stability.

If the scene is too crowded, reduce secondary actions. If movement is unclear, describe the starting and ending positions. If the camera direction is wrong, remove competing camera language before changing the rest of the prompt.

Keep the brief and selected settings with your notes. Repeating a request can change the output; it is not a guarantee of reproducing the same shot.

Plan several shots without losing continuity

Write a separate brief for each planned shot and record the details that should carry over: the character, clothing, visible props, time of day and setting. A sequence needs those decisions even when each shot looks convincing on its own.

If exact appearance is important, move from an approved character or scene image into a compatible image/reference workflow. Do not expect repeatedly naming a fictional character in prose to preserve every facial feature.

Further help with visual direction

The main video guide compares the available starting points. Google DeepMind's prompting guide also explains visual choices such as shot framing, camera movement, style and lighting. Its examples describe Google's tools; the controls offered by the selected ClapGen model remain the relevant limits here.