AG
Gemini & Antigravity v2.0 Agentic Engineering Guide
Multimodal & Veo 3 Multimodal Frontier Verified: 2026-10

Google Veo 3 Cinematic Video Generation Guide: Camera Directives & Production Prompts

Master Google Veo 3's generative video engine: 4K 60fps cinematic rendering, physical world realism, camera directives (Pan, Tilt, FPV), and structured prompt templates.

Author: Multimodal GenAI Lab • Google GenAI & Antigravity 2.0 Engineering Spec

Veo 3 Production Camera & Lighting Prompt Library

templates/veo3-cinematic-prompts.json

{
  "cinematic_fpv": {
    "prompt": "FPV drone cinematic shot accelerating through futuristic cyberpunk neon skyscrapers, hyper-realistic reflections on wet pavement, anamorphic lens flare, 4K 60fps, 35mm film grain, volumetric fog.",
    "camera_motion": "fast_forward_pitch_down",
    "aspect_ratio": "16:9",
    "seed": 42890
  },
  "macro_nature": {
    "prompt": "Extreme macro close-up of a morning dewdrop sliding down a translucent emerald leaf, refractive sunlight, cinematic slow-motion 120fps, shallow depth of field, photorealistic organic texture.",
    "camera_motion": "slow_push_in",
    "aspect_ratio": "16:9",
    "seed": 91823
  }
}

Video Generation with Director-Level Control

Early text-to-video systems suffered from motion flicker, inverted physics (liquids flowing upward, warped limbs), and an inability to understand camera angles.

Google’s Veo 3 overhauls spatio-temporal diffusion attention. It understands cinematic cinematography conventions and obeys physical laws, generating 4K 60fps video with realistic momentum, lighting reflections, and material physics.


Veo 3 Core Technical Innovations

1. Physical World Simulation

  • Fluid Dynamics & Gravity: Splashing water droplets, dispersing fog, and billowing clothing adhere to real-world physical constants.
  • Ray-Traced Reflections: Wet surfaces, polished chrome, and crystal glassware exhibit accurate Fresnel reflections and atmospheric refraction.

2. Director of Photography (DoP) Directives

Veo 3 interprets professional cinematography directives, giving creators precise camera trajectory control.


Camera Directives & Formula Reference

[Subject] + [Environment] + [Lighting / Texture] + [Camera Motion Directive] + [Quality Parameters]
Camera Motion DirectiveMovement Characteristics & Aesthetic Purpose
Drone FPV ChaseFast downward acceleration through complex geometry. High adrenaline, sci-fi chases.
Orbit Shot 360°Continuous circumferential rotation around subject. Epic character reveals, architectural grandeur.
Slow Push-InSubtle, slow focal compression toward subject. Builds dramatic tension and highlights micro-textures.
Crane Down to GroundSmooth descent from wide aerial vantage to intimate ground perspective. Seamless scene transition.
Whip Pan TransitionRapid lateral motion blur transition connecting contrasting visual environments.

Production Prompt Examples

Scenario 1: Cyberpunk Rain Night Pursuit

Cinematic 4K shot, high-speed camera tracking a sleek matte-black hypercar drifting around a wet neon-lit corner in Neo-Tokyo. Reflective asphalt puddles splashing dynamically under centrifugal force. Glowing holograms reflecting on the curved windshield. 35mm anamorphic lens, subtle optical distortion, shallow depth of field, 60fps cinema grade.

Scenario 2: Macro Biological Crystallization

Extreme macro slow-motion 120fps video of a crystalline ice crystal forming on a deep green pine needle during a silent blizzard. Shimmering internal refractions, sub-zero cold atmosphere, soft diffused twilight lighting, hyper-detailed natural geometry.

Generating via Python SDK

Invoke Veo 3 asynchronously using Google’s GenAI SDK:

from google import genai
from google.genai import types

client = genai.Client()

operation = client.models.generate_videos(
    model="veo-3.0-generate-preview",
    prompt="Cinematic drone shot flying through deep mist-covered Norwegian fjords at sunrise, 4K, 60fps.",
    config=types.GenerateVideosConfig(
        aspect_ratio="16:9",
        duration_seconds=8,
        fps=60
    )
)

print(f"Video generation task submitted: {operation.name}")