Veo 3.1 Video Prompter
Transform ideas into professional Veo 3.1 prompts using cinematic structure, audio direction, and multi-shot choreography.
When to Use
Invoke when user:
- Says "create a video prompt" or "generate a Veo prompt"
- Wants to "make a video of..." or "animate this..."
- Asks for help with "video generation" or "AI video"
- Needs "Veo 3" or "Veo 3.1" prompt assistance
- Wants to create "multi-shot" or "cinematic" video sequences
Core Prompt Formula
[Cinematography] + [Subject] + [Action] + [Context] + [Style & Audio]
Every prompt should address these five elements for maximum control.
Prompt Density: Finding the Sweet Spot
Prompts fail in two directions:
- Too sparse: Model fills gaps unpredictably, you lose creative control
- Too dense: Model can't execute all instructions, produces confused output
The Priority Framework
Tier 1 - MUST INCLUDE (model needs these):
- Shot size (wide/medium/close-up)
- Subject identity (who/what is in frame)
- Primary action (what happens)
- One dominant mood/style word
Tier 2 - SHOULD INCLUDE (significant impact):
- Camera movement OR angle (pick one, not both)
- Lighting quality (natural/dramatic/soft)
- One audio layer (dialogue OR SFX OR ambient)
- Setting/environment
Tier 3 - NICE TO HAVE (diminishing returns):
- Secondary audio layers
- Specific lens type
- Color palette details
- Film stock/grain texture
- Background action
Rule of thumb: Include all Tier 1, most of Tier 2, and 1-2 from Tier 3.
Density Comparison
TOO SPARSE (model guesses too much):
"A professor talking about philosophy"
TOO DENSE (model overloaded):
"Medium close-up shot at eye level with a 50mm lens at f/1.8 creating shallow depth of field with bokeh highlights, of a 52-year-old female professor with silver-streaked auburn hair pulled back in a loose bun, wearing an olive tweed jacket with leather elbow patches over a cream silk blouse with a small pearl brooch, standing in a contemporary lecture hall with tiered mahogany seating and brass fixtures visible in the soft background, natural diffused daylight streaming through floor-to-ceiling windows on the left side creating soft rembrandt lighting on her face with a gentle fill from reflected light on the right..."
OPTIMAL (directed but breathable):
"Medium close-up of a professor in her 50s, tweed jacket, standing in a university lecture hall. She gestures while speaking: 'Kant asked one question: could everyone do this?' Warm natural window light from left, soft academic atmosphere. SFX: marker on whiteboard."
Calibration Signals
Signs your prompt is too sparse:
- Results vary wildly between generations
- Key elements missing or wrong
- Mood/tone inconsistent with intent
Signs your prompt is too dense:
- Model ignores some instructions entirely
- Unnatural or frozen-looking motion
- Conflicting elements appear (e.g., both day and night)
- Audio doesn't match visual action
Iteration Strategy
- Start with Tier 1 only - generate test
- Add Tier 2 elements that matter most to your vision
- Add ONE Tier 3 detail if something specific is missing
- Remove any element the model consistently ignores
See references/prompt-calibration.md for detailed examples and troubleshooting.
Cinematography Elements
Shot Composition
- Wide shot, medium shot, close-up, extreme close-up
- Single shot, two shot, over-the-shoulder shot
- High angle, low angle, eye level, worm's eye, bird's eye
Camera Movement
- Dolly (in/out), tracking shot, crane shot
- Pan (left/right), tilt (up/down), zoom
- Steadicam, handheld, aerial, POV
Lens & Focus
- Shallow depth of field, deep focus
- Wide-angle lens, telephoto, macro lens
- Soft focus, rack focus, bokeh
Audio Direction
Veo 3.1 generates synchronized sound. Direct it explicitly:
Dialogue (use quotes):
"A man says, 'The storm is coming.'"
Sound Effects (label with SFX):
"SFX: Thunder rumbles in the distance, rain patters on glass"
Ambient Noise:
"Ambient noise: busy café chatter, clinking cups, soft jazz"
Music:
"A swelling orchestral score begins to play"
Timestamp Prompting
For multi-shot sequences within one generation (max 8 seconds):
[00:00-00:02] Medium shot of a detective at his desk, lighting a cigarette.
SFX: Match strike, paper rustling.
[00:02-00:04] Close-up of his eyes narrowing as he reads a letter.
Ambient: Rain against the window.
[00:04-00:06] Reverse shot of a shadowy figure in the doorway.
A woman's voice: "You shouldn't have looked."
[00:06-00:08] Wide shot as the detective stands, reaching for his gun.
SFX: Chair scraping, thunder crack.Style Keywords
Visual Aesthetic:
- Photorealistic, cinematic, documentary, animation
- Retro (sepia, grainy film, 1980s vaporwave)
- Noir, epic fantasy, sci-fi, romantic, horror
Mood & Lighting:
- Warm golden hour, cool blue tones, moody shadows
- Harsh fluorescent, soft morning light, dramatic chiaroscuro
- Neon-lit, candlelit, overcast diffused
Film Grain Tip:
Add "slightly grainy, film-like" to avoid overly clean AI look
Output Formats
Quick Prompt: Single sentence for simple shots Structured Prompt: Multi-line with all five elements Timestamp Sequence: Choreographed multi-shot within 8s Storyboard Mode: Multiple prompts for full narrative
Example Prompts
Action Shot:
"Tracking shot following a parkour athlete sprinting across rooftops at sunset, warm orange light, urban cityscape background, cinematic, shallow depth of field. SFX: footsteps on concrete, wind rushing past."
Dialogue Scene:
"Medium two-shot in a dimly lit bar, a woman in red leans toward a man in a suit. She says quietly, 'I know what you did.' Ambient: jazz music, glasses clinking. Moody noir aesthetic, warm tungsten lighting."
Nature Documentary:
"Slow-motion close-up of a hummingbird drinking from a flower, macro lens with shallow focus, lush green garden background, soft morning light. SFX: gentle buzzing, birdsong."
Technical Specs
- Duration: 4, 6, or 8 seconds
- Resolution: 720p or 1080p
- Aspect Ratio: 16:9 (landscape) or 9:16 (portrait)
- Frame Rate: Configurable (default: 24 FPS)
Advanced API Options
When using Veo through API (not Flow), these additional parameters are available:
| Parameter | Description | Default |
|---|---|---|
negativePrompt | Elements to exclude from the video | - |
seed | RNG seed for reproducible results (same prompt + seed = same video) | Random |
enhancePrompt | Let the model rewrite your prompt for better results | false |
generateAudio | Generate synchronized audio | true |
personGeneration | Control person generation: dont_allow or allow_adult | - |
referenceImages | Up to 3 asset images OR 1 style image for consistency | - |
Negative Prompts
Explicitly exclude unwanted elements:
"A forest at sunset" + negativePrompt: "people, animals, buildings"
Seed for Consistency
Use the same seed to reproduce similar results:
First generation: seed=12345 → video A Same prompt + seed=12345 → nearly identical video
Useful for:
- Iterating on a specific "look"
- Creating variations with controlled changes
- A/B testing different prompts
Reference Images
Maintain visual consistency across shots using reference images:
Asset References (up to 3):
- Character appearances
- Locations/settings
- Props or products
Style References (1):
- Overall aesthetic
- Color palette
- Visual treatment
References
references/prompt-calibration.md- Finding the right detail levelreferences/cinematography-glossary.md- Full camera termsreferences/prompt-examples.md- 20+ categorized examplesreferences/advanced-workflows.md- Image-to-video, first/last frame