Veo 3 Prompt Examples: Structure, Camera Control & Workflow

Veo 3 prompts work best when structured as: subject + action + environment + camera + lighting + style + technical parameters. Lead with the subject and primary motion, then layer cinematic controls — focal length, movement type, aperture — before aesthetic modifiers. This hierarchy mirrors how a DP blocks a shot, giving the model clear spatial and temporal instructions.

Core Prompt Anatomy for Veo 3

Every effective Veo 3 prompt contains six layers. Skip one and the model improvises — usually with generic defaults.

• Subject & Action: Who/what and what they're doing. Be specific: "a freckled woman in a yellow raincoat laughing while jumping over a puddle" beats "woman jumping."
• Environment: Location, weather, time of day, background activity. "Neon-lit Tokyo alley at 2 AM, steam rising from manholes" anchors the world.
• Camera Specification: Focal length (24mm, 50mm, 85mm), movement (static, dolly left, crane up, handheld), angle (low, eye-level, Dutch). Veo 3 responds to real lens vocabulary.
• Lighting & Color: Source (golden hour, practical neon, overcast), quality (hard, soft, dappled), palette (teal-orange, desaturated, high-key).
• Style & Texture: Film stock (Kodak Vision3 500T, Fuji Eterna), render mode (photorealistic, cinematic, stylized), grain/noise level.
• Technical Parameters: Duration (5s, 8s), resolution (1080p, 4K), aspect ratio (16:9, 2.39:1), seed for reproducibility.

Veo 3 vs. Veo 2 vs. Sora: Prompt Sensitivity Comparison

Veo 3's advantage is precise camera control. Where Sora excels at narrative coherence, Veo 3 executes a DP's shot list. Use Veo 3 when you need specific coverage: "50mm dolly right 2m, rack focus from foreground hands to background face, f/1.8 to f/5.6."

Camera Language Syntax That Works

Veo 3 interprets camera direction as a virtual cinematographer. These patterns produce repeatable results:

• Static: "locked off tripod, 35mm, eye-level" — zero drift.
• Dolly/Track: "dolly left 3 meters at 0.5m/s, 24mm, maintain subject distance" — parallax preserved.
• Crane/Jib: "crane up 4m over 3 seconds, 28mm, reveal cityscape behind subject" — vertical reveal with wide context.
• Handheld: "shoulder-mounted, 50mm, subtle micro-jitter, breathing rhythm" — organic energy without chaos.
• Focus Pull: "rack focus from 0.5m to 5m over 2 seconds, f/1.4 to f/8" — depth transition control.
• Zoom: "slow zoom in from 35mm to 85mm over 4 seconds, optical zoom only" — perspective compression, not digital crop.

Combine movements: "dolly left 2m while crane up 1.5m, 35mm, 3-second compound move" creates sophisticated camera choreography.

15 Ready-to-Use Veo 3 Prompt Templates

Copy, swap subjects/locations, and generate. Each follows the six-layer anatomy.

1. Cinematic Portrait: "Close-up of elderly watchmaker adjusting micro-screws, 85mm f/1.2, soft window light left side, dust motes dancing, shallow depth of field, Kodak Portra 400 aesthetic, 5s, 4K 16:9"
2. Action Reveal: "Skateboarder kickflips down concrete stairs, low angle 16mm, dolly tracking parallel 4m/s, golden hour backlight rim, motion blur on wheels, gritty 16mm film texture, 6s, 2.39:1"
3. Product Hero: "Luxury watch rotating on matte black turntable, macro 100mm f/8, ring light catchlights on sapphire crystal, slow 360° motorized rotation 10s, hyper-real 8K, commercial grade, 16:9"
4. Atmospheric Establishing: "Fog-shrouded lighthouse cliff at dawn, 24mm static, gradual light break through clouds, waves crash below, muted blue-gray palette, large format photography feel, 8s, 16:9"
5. Dialogue Two-Shot: "Two detectives in diner booth, 50mm eye-level, over-shoulder alternating, practical neon sign color spill, steam from coffee cups, 1970s crime drama grading, 10s, 2.39:1"
6. Macro Nature: "Water droplet falling from fern tip into pond, 100mm macro f/16, high-speed 120fps playback, crown splash formation, natural green bokeh, BBC Earth style, 4s slow-mo, 16:9"
7. Sci-Fi Corridor: "Astronaut walking infinite hexagonal corridor, 28mm one-point perspective, dolly forward steady, pulsed LED panels cyan/magenta, volumetric fog, Unreal Engine 5 cinematic, 7s, 2.39:1"
8. Dance Close-Up: "Ballerina pointe shoes on marley floor, 85mm low angle, rack focus toes to face, studio side-light dramatic falloff, motion blur on skirt, Degas painting come alive, 5s, 4:5 vertical"
9. Time-Lapse City: "Tokyo intersection day-to-night, 16mm locked off rooftop, 30-minute real-time compressed to 6s, light trails bloom, color temp shift 5600K to 3200K, Michael Mann aesthetic, 16:9"
10. Underwater: "Freediver descending blue hole, 20mm dome port, caustic light patterns on sand, particulate matter in water column, silent weightlessness, National Geographic style, 8s, 16:9"
11. Vintage Home Movie: "Family picnic 1975, Super 8mm emulation, gate weave, color fade, light leaks, kids running toward camera, handheld zoom jerky, nostalgic warmth, 6s, 4:3"
12. Fashion Editorial: "Model in flowing silk coat urban rooftop, 35mm slight low angle, wind machine coat billow, magic hour backlight translucent fabric, Vogue editorial grading, 5s, 9:16 Reels"
13. Horror Tension: "Child's bedroom night, 24mm static wide, door slowly creaks open 3 seconds, hallway light triangle expands, dust motes, silence then floorboard creak, James Wan lighting, 7s, 2.39:1"
14. Drone Landscape: "Icelandic braided river delta, drone 24mm equivalent, top-down orbit 360° 10s, glacial milk turquoise patterns, 4K HDR, planet Earth documentary, 16:9"
15. Abstract Experimental: "Ferrofluid spikes dancing to magnetic field, macro 60mm, black background, colored gel lighting shifting RGB, high contrast, synesthesia visualization, 5s loopable, 1:1"

Step-by-Step: From Concept to Final Clip

1. Define the shot purpose. Is it an establishing shot, emotional beat, product reveal, transition? Purpose dictates camera choice.
2. Write the subject-action core. One sentence, present tense, active verbs. "Chef tosses pizza dough" not "pizza dough being tossed by chef."
3. Add environment with sensory detail. Temperature, smell proxies (steam, dust), background motion. "Flour dust catching afternoon light" > "kitchen background."
4. Specify camera like a shot list. Focal length, movement, angle, focus plan. Use the syntax in Section 3.
5. Design lighting with motivation. Source, direction, quality, color. "Practical desk lamp 45° camera left, warm 2700K, hard shadow nose" gives the model physics.
6. Choose style reference. Film stock, director, photographer, or render engine. "Roger Deakins skyfall palette" transfers better than "cinematic."
7. Set technical specs. Duration, aspect, resolution, seed. Lock seed for iterative refinement.
8. Generate, evaluate, iterate. Change one layer per generation. Adjust camera first, then lighting, then style. Track changes in a spreadsheet.

Common Failure Modes & Fixes

Advanced Workflow: Batch Production for Narrative

For multi-shot sequences, treat Veo 3 like a virtual production unit:

• Master prompt sheet: Shared environment, lighting, style, technical specs across all shots. Only subject, action, camera change per row.
• Seed strategy: Use sequential seeds (1001, 1002, 1003) for visual consistency. Document which seed produced which shot.
• Reference images: Feed key frames as image-to-video prompts for character/location lock. Veo 3 accepts image conditioning.
• Continuity pass: Generate all wide shots first, then mediums, then close-ups. Match eyelines and screen direction in post.
• Upscale & interpolate: Run selects through Topaz Video AI or RIFE for 4K/60fps delivery. Veo 3 native 24fps often needs frame interpolation for smooth slow-mo.

The best Veo 3 prompts read like a cinematographer's shot list, not a creative writing exercise. Precision in camera and lighting language yields precision in output.

Closing Takeaway

Veo 3 rewards structural discipline. The six-layer prompt anatomy — subject, environment, camera, lighting, style, technical — isn't optional; it's the interface language. Start every session with a master template, swap variables systematically, and track seeds. When you prompt like a DP, you direct like one.