AI Video Prompt Glossary: Camera Moves, Shot Types and Lighting

This glossary covers the words that change what the camera does in an AI video: how it moves, how close it sits, where it looks from, and how the scene is lit. Each entry gives a plain definition, when the term earns its place, and a phrase you can paste into a prompt and adapt. Pick one or two terms per prompt, not one from every section.

Before you start

On SpicyGen, camera angle, lighting, setting and mood are written into the prompt, not set with separate controls. When you animate an image from your library or a preset look, the first frame has already decided the shot size and angle, so your video prompt mostly decides the movement. When you generate from a text description alone, every term below is in play.

A prompt order that works in both cases: shot size and angle, the subject described as an adult (a woman in her thirties), the action, the camera move, then the light. The examples below say "her" or "him" to stay short; your full prompt names the adult subject once.

Camera moves

Push in. The camera travels toward the subject, so the frame tightens. Use it to pull attention onto a face or a reaction. slow push in on her face as she looks up

Pull out. The reverse. The camera backs away and reveals more of the room, a natural ending for a clip that opens tight. camera slowly pulls out to reveal the candlelit bedroom

Pan. The camera turns left or right from a fixed spot, like turning your head. It follows someone crossing the frame or carries the eye across a set. slow pan right following him across the loft

Tilt. The camera pivots up or down without changing position. A tilt from shoes to face is the classic introduction. camera tilts up from her heels to her face

Orbit. The camera circles the subject while keeping them centered. It shows an outfit or pose from several sides, but it asks a lot of the model, so keep the subject still. slow partial orbit around a woman in her thirties standing by the window

Tracking shot. The camera moves alongside or behind a moving subject and keeps pace. Use it for walks. tracking shot following her down the hotel corridor

Handheld. A slight, organic sway, as if someone is holding the camera. It reads as candid, like a phone video. handheld camera, gentle natural sway

Static shot. The camera does not move at all. It is the most underrated choice here: when the action is the point, a locked camera lets the model spend its effort on face and body. static camera, locked-off tripod shot

Shot sizes

From tightest to widest. The size tells the model how much of the person and the room belongs in frame.

Extreme close-up. One feature fills the frame: eyes, lips, a hand around a glass. Use it for texture and detail. extreme close-up of her eyes as she blinks slowly

Close-up. The face from chin to forehead. As SpicyGen's user guide puts it, a tight crop on the face suits eye and lip movement. close-up of a man in his forties, a slow smile spreading

Medium shot. Waist up. It leaves room for hands, posture and breathing, the safest default for most actions. medium shot of her sitting on the edge of the bed

Full shot. Head to toe, with some air around the body. Use it when the action involves the whole figure, like turning or sitting down. full shot of a man in his forties leaning against the doorframe

Wide shot. The setting dominates and the person takes up a fraction of the frame. Use it to establish place. wide shot of a penthouse at night, a woman in her thirties at the far window

Camera angles

Eye level. The lens sits at the subject's eye height. Neutral, and what most models assume when you say nothing. eye-level shot

Low angle. The camera sits below the subject and looks up. It makes someone look confident and a little larger than life. low angle looking up at her on the staircase

High angle. The camera sits above and looks down. It softens the subject and suits anyone seated or lying back. high angle looking down at him reclining on the sofa

Overhead. Directly above, pointing straight down, also called a top shot. Strong for someone lying on a bed or floating in a pool. overhead shot of a woman in her thirties lying on white linen sheets

Lens and depth terms

Shallow depth of field. Only a thin plane of the scene is sharp and the background melts into blur. It isolates a person from a busy room and looks filmic. shallow depth of field, background softly blurred

Rack focus. Focus shifts during the shot from one distance to another. It works as a reveal when it works at all. rack focus from the wine glass to her face

Focal length. Wide lenses (around 24mm) stretch space; long lenses (85mm and up) compress it and flatter faces. Treat the numbers as a mood cue rather than precise optics. 85mm portrait lens look

Lighting terms

Soft window light. Diffuse daylight from one side with gentle shadows. The most forgiving light for skin, and an easy default. soft window light from the left, early morning

Rim light. A light behind the subject that traces a bright edge along hair and shoulders, lifting them off a dark background. strong rim light outlining her hair in a dark room

Practical lights. Light sources you can see in the frame: lamps, candles, a neon sign, a phone screen. They make a room feel lived in and give the light a logical source. lit only by bedside lamps and a few candles

Golden hour. The warm, low sun just after sunrise or before sunset, with long shadows and amber skin tones. golden hour sunlight through sheer curtains

Low key. Mostly shadow, with one controlled pool of light. Moody, with a touch of noir. low key lighting, single warm lamp, deep shadows

Pacing words

AI video tends to move too much, too fast. Pacing words are the brakes.

Slow. Applies to a camera move or an action. Put it directly in front of the thing it modifies. slow push in

Gradual. Asks for change spread across the whole clip instead of crammed into the first second. gradual pull out over the entire shot

Subtle. Keeps motion restrained. Good for breathing, a faint smile, hair stirring in a draft. subtle movement, she breathes and smiles faintly

Which terms models follow, and which they skip

This varies between models and runs, so treat it as experience, not a rulebook.

Usually followed:

• Static, push in and pull out. One-axis moves are the easiest for a model to hold.
• Slow pans and tilts, when the frame has something to move toward.
• Lighting words. Golden hour, candlelight, low key and neon tend to land more consistently than any camera term.
• Pacing words, as long as they sit right next to the move or action they modify.

Often ignored or loosely followed:

• Rack focus and exact focal lengths. You get the mood more often than the mechanics.
• Full 360 degree orbits, which tend to stall halfway or bend the background. Ask for a partial arc.
• Angle changes in image to video. An eye-level starting image rarely turns into an overhead view halfway through. Start from an image that has the angle you want.
• Named tricks such as dolly zoom, whip pan or crash zoom.
• Stacked instructions. "Push in while panning left and tilting up" usually collapses into one of the three, or into a wobble.

One camera move, one subject action

The most reliable structure is one camera move plus one subject action, each with its own pacing word. A model can only keep so much motion coherent in one clip. If the camera and the person both do a lot, faces drift and hands go strange.

A prompt built that way:

Medium shot, eye level. A woman in her thirties in a silk robe sits on the edge of the bed and slowly turns her head toward the camera. Slow push in. Soft window light from the left, shallow depth of field.

And one that asks for too much:

Orbit around her while pushing in and tilting up as she stands, walks to the window, turns around and laughs.

That is three camera moves and four actions. Something will get dropped, and you don't get to choose what. If you need two moves, make two clips.

The SpicyGen user guide says the same: keep actions simple and realistic, be specific about movement and expression, and avoid complex multi-person scenes. If you start from a preset motion action, let it be your one action and add only a camera move and a lighting note. For the action half of the prompt, see our guide to writing image to video prompts.

When a clip misses, change one term at a time. Swap push in for static camera and run it again before touching the light, and you'll learn quickly which words a model listens to. The cinematic video generator is a good place to run those side-by-side tests.