How to Write a Video Generation Prompt: a Guide with Examples

An AI model generates exactly what you describe — so 80% of your video's quality is decided by the prompt. Here is how to write prompts that produce cinematic clips instead of an "animated screensaver".

The anatomy of a good prompt

The working formula: what happens + where + how the camera moves + light and mood.

Compare:

• Weak: "quad bikes ride along a river"
• Strong: "Three quad bikes race through the shallows of a mountain river at sunrise, backlit spray, a drone camera flies alongside at wheel level, then rises into a wide canyon shot"

The second version gives the model an action, a location, camera work and light — and every element shows up in the finished clip.

Long clips: break the story into phases

A 15–30 second video described in one sentence comes out monotonous. Write the story as timed phases instead:

• Phase 1 (0–5s): sunrise over the canyon, mist, static wide shot
• Phase 2 (5–15s): the quad bike column bursts into frame, side tracking shot, spray
• Finale (15–20s): the leader breaks away, the drone rises, the rider raises a fist against the sun

Each phase should change something meaningful: the action, the camera, the light, or a new element entering the frame. Keep one location — hard cuts between unrelated scenes inside a single generation look like a glitch.

Camera vocabulary

Models respond remarkably well to cinematography terms: aerial shot, tracking shot, close-up, dolly-in, golden hour, shallow depth of field. One or two of these do more than a paragraph of adjectives.

Sound is part of the prompt

Modern models (like Wan 3.0, which powers Tegas) generate audio together with the picture: engine roar, water, character speech with lip sync. Describe the soundscape in its own line — "roaring engines and rapids, driving rock" — and the clip arrives with sound that matches the action.

Common mistakes

1. Overload: five actions in a five-second clip. Rule of thumb: one action per 5 seconds.
2. Asking for text and logos: models draw captions poorly, and provider filters reject recognizable brands. Add titles in post instead.
3. Abstractions: "a beautiful atmospheric video about freedom" contains zero visual facts. Write what can be seen.

The no-prompt way

In Tegas you can skip prompt writing entirely: the built-in Tessa assistant asks a couple of questions about your idea, writes the story phase by phase, picks camera work and sound, and fills the generation form. Press "Help me with the prompt" on the Text-to-Video page — and compare its version with yours. It is the fastest way to learn.