Camera directives are the single most load-bearing tokens in generative video. While image models treat camera terms as composition guidelines, modern video diffusion architectures like Grok Imagine, Google Veo, and ByteDance Seedance interpret camera instructions as continuous spatio-temporal trajectories.
When you specify a camera move accurately, you constrain how background parallax, lighting reflections, and subject perspectives shift across every generated frame. When omitted, diffusion models default to subtle, unpredictable drift that undermines scene intentionality.
Below are ten production-grade camera moves distilled from thousands of generation benchmarks across our corpus, complete with exact prompt phrasing and usage parameters.
01. Tracking Shot (Dolly / Follow)
What it does: The camera physically moves alongside, in front of, or behind a moving subject while maintaining a relatively constant distance.
- Model interpretation: Establishes strong background motion parallax while locking attention onto the subject. Crucial for action sequences, street walks, and vehicle pursuits.
- Placement tip: Place at the beginning of the prompt. Models prioritize camera seating when it leads the prompt hierarchy.
- Prompt syntax:
Tracking shot, following closely behind a courier sprinting through a crowded neon-lit Tokyo market in heavy rain, low angle, splashing puddles, motion blur on background stalls, cinematic 35mm lens, 8s, 16:9.02. Pan (Horizontal Sweep)
What it does: The camera pivots horizontally from a fixed axis, sweeping across a landscape, interior, or group of characters.
- Model interpretation: Translates visual interest laterally across the screen. Excellent for environmental reveals, architectural surveys, and transition pacing.
- Prompt syntax:
Slow smooth pan from left to right across a fog-shrouded mountain valley at dawn, revealing an ancient stone castle perched on the cliff edge, volumetric sunbeams piercing the mist, 8k resolution, 10s, 16:9.03. Tilt (Vertical Pivot)
What it does: The camera tilts upwards or downwards from a stationary pivot point, shifting the vertical perspective.
- Model interpretation: Tilt up establishes grandeur, scale, towering height, or heroic posture. Tilt down grounds the subject, reveals ground details, or introduces vulnerability.
- Prompt syntax:
Slow cinematic tilt up from cracked cobblestones up the imposing gothic facade of a cathedral reaching into stormy twilight clouds, lightning illuminating stone gargoyles, 4k detail, 6s, 9:16.04. Push-In (Forward Dolly / Zoom In)
What it does: The camera moves directly forward toward a subject or focal point, tightening the composition.
- Model interpretation: Heightens dramatic tension, intensifies emotional connection, or draws immediate focus to a micro-detail.
- Prompt syntax:
Slow dramatic push-in on an astronaut seated inside an orbital capsule, helmet visor reflecting the glowing turquoise curve of Earth below, intense emotional focus, subtle cockpit instrument lighting, 8s, 16:9.05. Pull-Out (Backward Dolly / Zoom Out)
What it does: The camera retreats smoothly backward away from the subject, revealing the wider environment.
- Model interpretation: Shifts context from intimate or isolated to expansive. Perfect for final scene resolutions, conveying loneliness, or establishing environmental scale.
- Prompt syntax:
Smooth pull-out from a single lit candle on a wooden table, retreating through a vast abandoned banquet hall with towering vaulted arches, soft ambient moonlight through broken windows, 12s, 16:9.06. Orbit / 360 Arc Shot
What it does: The camera revolves continuously around the subject in a circular arc while keeping the subject centered.
- Model interpretation: Forces the diffusion model to compute 3D volumetric consistency and lighting shifts around the entire circumference of the subject.
- Prompt syntax:
Smooth 180-degree orbit shot around an ancient robotic warrior standing atop a windy sand dune, cloak billowing, sun setting on the horizon casting long shadows, cinematic grading, 10s, 16:9.07. POV (Point of View / First-Person)
What it does: The camera replaces the viewer with the eyes of an active participant in the scene.
- Model interpretation: The diffusion model suppresses visible protagonist head models and instead renders hands, tool interactions, or direct forward terrain motion.
- Prompt syntax:
First-person POV shot riding a vintage motorcycle down a winding coastal highway at sunset, hands gripping chrome handlebars, ocean spray on the lens, golden hour light, high temporal fluidity, 8s, 16:9.08. Close-Up & Extreme Macro
What it does: Tightly frames a subject’s face, eyes, or intricate object details, excluding environmental background.
- Model interpretation: Directs diffusion capacity toward micro-textures—skin pores, iris patterns, fabric weave, or water droplets—with shallow depth of field.
- Prompt syntax:
Extreme close-up macro shot of a honeybee landing on a dew-covered lavender flower, vibrating wings, pollen grains visible in sharp focus, creamy soft background bokeh, natural morning light, 6s, 16:9.09. Aerial & Drone Flyover
What it does: High-altitude perspective captured from an elevated drone or aircraft flying smoothly over terrain or urban architecture.
- Model interpretation: Maximizes horizontal perspective lines, expansive depth, and macro-scale motion across forests, oceans, or skyscrapers.
- Prompt syntax:
Cinematic aerial drone shot sweeping high over turquoise ocean waves crashing against jagged black volcanic sea cliffs, white sea foam, sea birds gliding below camera, crisp golden dawn light, 12s, 16:9.10. Static / Locked Tripod
What it does: The camera remains completely motionless on a fixed mount while all motion occurs organically within the frame.
- Model interpretation: Eliminates virtual camera shake and motion blur, concentrating model compute entirely on subject animation, cloth physics, fluid dynamics, or facial expressions.
- Prompt syntax:
Static locked tripod shot of a busy traditional ramen stall at night, steam billowing from boiling broth pots, chef slicing scallions, rain falling outside the open doorway, warm lantern illumination, 8s, 16:9.Benchmark Rules for Camera Prompting
- One Hero Motion: Avoid stacking contradictory moves (e.g. “drone orbit with handheld whip pan”). Pick one dominant camera trajectory and let the model execute it cleanly.
- Coordinate Camera Speed: Modifiers like “slow smooth glide” or “gentle push-in” prevent jarring acceleration jumps between diffusion frames.
- Lock the Subject: If the camera moves, keep subject motion predictable so temporal attention mechanisms maintain subject coherence without morphing.
Reading list: more guides, the techniques manual, or dive straight into the library.
