The fastest way to make AI video look expensive is to stop describing scenes and start describing shots. Modern models — Seedance 2.5 especially — were trained on enough real footage that they respond to the same vocabulary a director of photography uses on set. This guide covers the moves that work, and how to phrase them.

The core moves

Dolly. The camera physically travels toward, away from, or alongside the subject. “Slow dolly-in on her face” creates building intensity; “dolly-out to reveal the empty hall” delivers context as a punchline. Use slow or creeping to control pace — unqualified dollies tend to come back too fast.

Tracking shot. The camera follows a moving subject, keeping it roughly centered. “Handheld tracking shot following the cyclist through traffic” adds documentary energy. Say smooth tracking shot or steadicam if you want the polished version.

Crane / jib. Vertical travel. “Crane up from street level to rooftop height” is the classic establishing move. The reverse — “crane down behind her shoulder” — pulls the viewer from overview into intimacy.

Orbit. The camera circles the subject. “Slow 180-degree orbit around the sculpture” is reliable; full 360s work best on centered, stationary subjects.

Static. Underrated. “Locked-off static shot, subject moves through frame” reads as confident and composed — and it is the single most effective way to avoid the floaty, drifting look that screams AI.

Lens language

Camera position is half the story; the lens is the other half.

  • “Shallow depth of field” isolates the subject against creamy bokeh. Add “85mm portrait lens” for maximum compression.
  • “Wide-angle, 24mm” exaggerates space and makes interiors feel vast. Great for architecture, risky for faces.
  • “Rack focus from the glass to her eyes” shifts attention mid-shot. Name both focus targets explicitly.
  • “Anamorphic lens flare, 2.39:1” buys instant blockbuster texture. Use sparingly.

Putting it together

Keep one move per shot, and give it its own sentence at the end of the prompt:

An old fisherman mends a net on a dock at dawn, gulls overhead, mist on the water. Warm low light, muted colors. Slow dolly-in from wide to medium shot, shallow depth of field.

If you need a sequence — establish, approach, close-up — generate each shot separately and cut them together in the editor. Models hold one continuous camera idea per clip far better than three chained ones, and you get editorial control as a bonus.

Print this list, or just keep it open in a tab. Ten minutes of deliberate camera language will do more for your renders than an hour of adjective tuning.