AI Video Prompt Builder (Sora, Veo, Runway)

An image prompt describes one frozen frame. A video prompt has to describe what changes over four to ten seconds: who moves, how the camera moves, how fast time passes and what the viewer hears. Leave those out and the model picks them for you, which is why many first attempts drift, float or cut to something unexpected.

The building blocks. This builder asks for a subject and action, a setting, time of day and lighting, a shot type (wide, medium, close-up, extreme close-up, over-the-shoulder or POV), a camera move (static, pan, tilt, dolly in or out, tracking, crane, handheld, orbit or zoom), a lens, a pace (slow motion, real time or timelapse), a film look, and audio. Lens choice matters more than people expect: a 24mm lens exaggerates space and speed, a 35mm lens feels like natural human vision, and an 85mm lens compresses the background and isolates a face.

One shot per prompt. Current models handle one continuous shot best. If you need a cut, generate two clips and edit them together.

Example. Subject: "A barista pours latte art. She looks up and smiles." Setting: a small corner cafe. Lighting: early morning window light. Shot: close-up. Movement: dolly in. Lens: 50mm. Dialogue: "Your order is ready." SFX: milk steamer hiss. The Veo tab starts with "Close-up shot, camera dolly in, 50mm lens", then the action, the setting line, the style line and the three audio cues. The Sora tab lists the same action as two beats under Actions.

Change one thing at a time. If motion looks wrong, adjust only the camera move or pace and regenerate, so you can see which word caused the change.

Keep dialogue short. A clip of 8 seconds fits roughly one short sentence of natural speech, about 15 to 20 words. Longer lines get rushed or cut off.

Match the move to the story. A dolly in builds attention on a face, a tracking shot follows action, a crane reveals a location, and handheld adds urgency.