The Anatomy of a Great AI Video Prompt

The instinct when you first write a video prompt is to describe a picture and hope it moves. That is why so many AI clips drift aimlessly, morph halfway through, or spend eight seconds doing nothing. A video prompt has to describe time as well as space: what the frame contains, how the camera behaves, and what changes between the first second and the last.

The Six Layers

Every dependable video prompt covers six things, in roughly this order. Skipping a layer hands that decision to the model, and motion decisions are the ones you least want it improvising.

• Subject and action: who or what, and the single movement they perform.
• Scene: the environment and one or two grounding details.
• Camera: the shot size and one clear movement.
• Light and style: the look, in photographic terms.
• Pacing: how time behaves — real time, slow motion, energetic.
• Format and audio: duration, aspect ratio, frame rate, and what it sounds like.

The Shape of a Prompt: A [shot size] of [subject performing one clear action]. [Scene with one or two grounding details]. Camera: [one movement]. Lighting: [direction and quality]. Style: [look and grade]. Pacing: [speed]. Audio: [sound design]. [Duration]-second clip, [aspect ratio], 24fps, smooth natural motion, temporally consistent. Avoid: flickering, morphing, warping, sudden jump cuts, distorted anatomy, on-screen text.

One Action, One Camera Move

Eight seconds is not long. A subject that walks, turns, sits and waves while the camera orbits, pushes in and cranes up will produce mush. Choose one action and one movement, and let them be simple: a barista pours, the camera pushes in slowly. Complexity is what breaks temporal consistency, not ambition.

Product Commercial, 8 Seconds: A close-up shot of a perfume bottle on a polished black marble plinth, a slow light sweep travelling across the glass. Dark seamless backdrop with a single soft reflection. Camera: smooth 180-degree orbit around the bottle. Lighting: single large softbox from above with crisp rim highlights. Style: high-end commercial, crisp and polished colour. Pacing: slow and deliberate. Audio: swelling cinematic score. 8-second clip, 16:9, 24fps, smooth natural motion, temporally consistent lighting. Avoid: flickering, morphing, warping, on-screen text or watermarks.

Describe Motion the Way an Editor Would

'Moves nicely' means nothing. 'Rises from waist height to eye level over four seconds' is a direction. Speed, distance and duration are the vocabulary that models trained on film descriptions understand, and they are also how you keep a clip from arriving at its most interesting moment after the cut.

Guardrails Belong in Every Video Prompt

Text-to-video fails in specific, repeatable ways: faces morph between frames, backgrounds flicker, limbs multiply, and models love to stamp fake subtitles across the bottom. A short avoid list costs a line and prevents most of it. Keep it constant across your prompts so you can compare results fairly.

• Say 'temporally consistent subject and lighting' — it is the single most useful realism phrase in video.
• Name the frame rate; 24fps reads as cinema, 60fps reads as a phone video.
• If a clip drifts, shorten the duration before rewriting the description.
• Generate three variations of the same prompt — video is far more variable than stills.

Six layers, one action, one camera move, and a fixed set of guardrails. That is the whole anatomy, and it travels across Manus, Sora, Veo, Kling and Runway without modification.