Free AI video prompt generator

Think in shots, not stories

Video models generate short clips, usually a few seconds long. A prompt that describes a whole story (“she arrives, argues, leaves, it starts raining”) gets compressed into confusing motion. Treat each prompt as one shot: one place, one main action, one camera move. If you need a sequence, write several prompts and edit the clips together.

The builder asks for exactly what a director would specify for one shot:

• Scene: the setting, time of day and weather.
• Action: who moves and how — “walks toward the camera and smiles” rather than “is happy”.
• Shot type: wide establishing, medium, close-up, extreme close-up, POV or over-the-shoulder.
• Camera movement: static, dolly-in or out, pan, tracking, orbit, crane, handheld, drone or zoom.
• Look: style, lighting and mood.
• Sound and dialogue: used by models that generate audio.

How the output changes per model

The same shot is written differently for each model. With the scene “a rainy neon street in Tokyo at night” and the action “a woman in a yellow raincoat walks toward the camera and smiles”:

Hailuo:
[Push in] A woman in a yellow raincoat walks toward the camera and smiles. Setting: a rainy neon street in Tokyo at night. Medium shot. Cinematic film still, neon lighting, mysterious mood.

Runway:
Slow dolly-in: A rainy neon street in Tokyo at night. a woman in a yellow raincoat walks toward the camera and smiles. Medium shot. Cinematic film still, neon lighting, mysterious atmosphere.

Hailuo receives bracketed camera commands such as [Push in], [Pan right] or [Pedestal up] at the start. Runway gets the movement as a leading label. Pika gets a compact sentence plus a negative prompt line. Kling and Wan get labeled sentences for scene, camera, lighting and style, with duration, ratio and a negative line at the end; Kling also gets a sound line. Veo 3 and Sora 2 have their own dedicated pages: the Veo 3 prompt generator and the Sora prompt generator.

Camera movement vocabulary that models understand

Choose one move per clip. Combining several moves in a few seconds usually produces jittery or unnatural camera paths.

Duration, sound and what to avoid

The duration menu offers 5 to 20 seconds, but each model and plan has its own limits; the number in the prompt is a hint, and the real length is set in the model's interface. Sound descriptions matter only for models that generate audio. In the avoid field, list concrete problems such as subtitles, text overlays, distorted faces. For a model-specific walkthrough, see the Kling prompt generator.

A practical way to choose a model is to write the shot once, switch the model menu, and test the same idea in two tools with the same settings. Models differ in how they handle faces, fast motion and physics, and your own shot is the only reliable benchmark.