veo-video-generation

Veo 3.1 Video Generation

Generate videos via Google's Veo 3.1 API. Includes prompt engineering guide and executable script.

Usage

/veo-video-generation A slow dolly shot toward an artisan coffee cup, warm morning light

When invoked with arguments, generate the video directly. When invoked without, help the user craft a prompt using the guide below.

Quick Start

# Text-to-video
~/.claude/skills/veo-video-generation/generate-video.sh \
  --prompt "Slow tracking shot following a chef plating a dessert, warm kitchen lighting, cinematic"

# Image-to-video
~/.claude/skills/veo-video-generation/generate-video.sh \
  --prompt "Subtle breathing, slow blink, magical sparkles rise from hands" \
  --image ./character.png \
  --duration 8 --resolution 1080p

Requires: GEMINI_API_KEY env var. Cost: $0.35/sec ($2.80 for 8s).

Run generate-video.sh --help for all options.

Prompt Formula

[Cinematography] + [Subject] + [Action] + [Context] + [Style & Ambiance]

Always lead with cinematography — it's the strongest lever.

Component	Examples
Cinematography	Slow tracking shot, crane shot rising, extreme close-up, shallow depth of field
Subject	A tired corporate worker, a fluffy orange tabby cat, a young female explorer
Action	rubbing his temples, batting playfully at yarn, stepping cautiously forward
Context	in a cluttered 1980s office lit by fluorescent lights, in a mist-filled canyon at sunrise
Style & Ambiance	Retro film grain, warm magical atmosphere, cinematic color grading

Camera Movement Vocabulary

Movement	Use for
Dolly shot	Move toward/away from subject — builds tension or reveals
Tracking shot	Follow subject laterally — journey, progression
Crane shot	Vertical movement — reveals scale, establishes setting
Arc shot	Circle around subject — dramatic emphasis
Aerial view	Bird's eye — context, scale
Slow pan	Horizontal rotation — survey environment
POV shot	Character's viewpoint — immersion

Shot types: wide (environment), medium (conversational), close-up (emotion), extreme close-up (detail).

Audio Direction (Veo 3.1)

Veo 3.1 generates synchronized audio. Include cues:

# Dialogue — use quotation marks
A woman turns to camera and says, "We have to leave now."

# Sound effects
SFX: thunder cracks in the distance

# Ambient
Ambient noise: quiet hum of a starship bridge

# Music
SFX: A swelling, gentle orchestral score begins to play.

Silent Scenes

Front-load the constraint + describe closed mouth explicitly:

SILENT SCENE - no talking, no dialogue, no speech.
[Subject] has a warm, closed-mouth smile - lips stay together,
mouth does not open or move at any point.
Audio: soft ambient sounds only. No voices, no dialogue, no speech sounds.

Avoid implicit triggers: "talking", "chatting", "greeting". Use: "beckoning", "gesturing", "nodding".

Advanced Workflows

Timestamp Prompting (multi-shot in single generation)

[00:00-00:02] Medium shot from behind a young explorer approaching a canyon edge...
[00:02-00:04] Reverse shot of the explorer's face, eyes widening...
[00:04-00:06] Tracking shot following the explorer stepping forward...
[00:06-00:08] Wide crane shot pulling up and away, revealing the full canyon...

Character Consistency (up to 3 reference images)

Generate reference images first, then pass them to maintain consistent characters across shots. Use --image flag.

First/Last Frame Transitions

1. Generate start frame (e.g., with Gemini image gen)
2. Generate end frame
3. Animate transition with Veo describing the arc between states

API Parameters

Parameter	Options	Default	Notes
--prompt	Any text	Required	Video generation directive
--image	File path	None	.png/.jpg/.jpeg/.webp for image-to-video
--model	veo-3.1-generate-preview, veo-3.1-fast-generate-preview	veo-3.1-generate-preview	Fast variant same price
--duration	4, 6, 8	8	Seconds
--resolution	720p, 1080p	720p	1080p only with 8s duration
--aspect-ratio	16:9, 9:16	16:9	Landscape or portrait
--negative-prompt	Any text	None	What NOT to include
--no-audio	Flag	false	Disable audio generation
--output	Directory path	./output	Where to save video
--filename	String	Auto-generated	Output filename

Common Mistakes

• Vague prompts — "A chef working" vs "Slow tracking shot following a chef plating a dessert"
• Missing cinematography — Always start with camera movement/shot type
• String duration — API expects integer 8, not string "8"
• 1080p with short duration — 1080p only works with 8-second videos
• Wrong image field — Must be bytesBase64Encoded, not data or imageBytes
• Forgetting 2-day expiry — Generated videos deleted from Google servers after 2 days

Example Prompts

Cinematic product shot:

Slow dolly shot toward an artisan coffee cup on a wooden table,
steam rising in soft morning light through a window. Shallow depth
of field, warm golden tones, cozy cafe ambiance.
SFX: gentle coffee shop murmur, soft jazz piano in background.

Character animation:

Extreme close-up, shallow depth of field. SILENT SCENE - no dialogue.
A young woman with silver-streaked dark hair, warm closed-mouth smile,
performs a gentle beckoning motion toward the viewer. Tiny magical
sparkles rise from her open palm, drifting upward like stardust.
Hair drifts slightly in a soft breeze. Warm magical atmosphere,
soft diffused lighting.
Audio: soft celestial chimes, gentle twinkle effects. No voices.

First-person journey:

First-person camera perspective. The camera floats weightlessly
through a colorful asteroid field, gently drifting past crystal-studded
rocks catching starlight. A comet with a sparkling rainbow tail swoops
past, leaving trails of stardust. Distant nebulae swirl in cotton-candy
pinks and electric blues. The camera banks smoothly, revealing a
magnificent glowing planet ahead.
Audio: ethereal hum, gentle whooshes, crystalline twinkling.