How to write prompts for video generation

How to write prompts for video generation

Models have become smarter: they are better at keeping context, understanding the physics of movement, and remembering the characters in the frame. But the quality of the result still depends on the formulation of the request.

The formula for a short working prompt:

Object + environment + movement + style + camera

The working structure of the prompt:

1. Main object or action
   Who or what is in the centre of the frame? The more specific the better.
   Not «robot», but «industrial robot with scratches on the body».
   Not «a man walks», but «a girl in a red cloak walks on the wet pavement».
2. Scene and surroundings
   Where the action takes place: interior, street, time of day, weather, era.
   Example: «Post-apocalyptic city, dusk, light fog, broken shop windows».
3. Dynamics and movement
   Video is about movement. Make it explicit:
   «the camera slowly zooms in», «the drone flies over the canyon», «the character turns his head and looks into the lens».

4. Modern models have a better understanding of complex choreography, but if the movement is not described - the result will be static.

5. Style and visual language
   Realism, animation, documentary style, retrofuturism, 70s cinema. You can add atmosphere: «gloomy», «ironic», «epic», «warm».

6. Important: 2026 models respond worse to overloaded reference lists. 1-2 clear references work better than 10.

7. Pace and rhythm
   If the model supports duration and speed control:
   «slow cinematic movement»,
   «dynamic montage»,
   «abrupt frame shifts.».
8. Camera and light (for advanced modes)
   Panorama, tracking, zoom, handheld effect.
   Lighting: natural, studio, backlight, neon.
   Plan: close-up, medium, general.

Try Sora 2 to generate
https://pr-cy.ru/ai-video-generator/

This post has video content attached. To familiarise yourself with the content of the video - you need to go to the source at the link below.