Veo Build

Veo Video Generation and Editing

This skill provides comprehensive workflows for using Google's Veo models (Veo 2 and Veo 3) via the google-genai Python SDK.

Quick Start Setup

All Veo operations require the google-genai library and an authenticated client with Vertex AI enabled.

from google import genai
from google.genai import types
import os

PROJECT_ID = os.environ.get("GOOGLE_CLOUD_PROJECT")
LOCATION = os.environ.get("GOOGLE_CLOUD_REGION", "us-central1")

client = genai.Client(vertexai=True, project=PROJECT_ID, location=LOCATION)

Reference Materials

• [Generation (Veo 3)](references/generation.md): Text-to-Video, Image-to-Video.
• [Editing (Veo 2)](references/editing.md): Inpainting, Masking.
• [Advanced Controls](references/advanced.md): Frame Interpolation, Video Extension, Reference Images.
• [Prompting Guide](references/prompting.md): Camera angles, visual styles, and best practices.
• [Source Code](references/source_code.md): Deep inspection of SDK internals (models.py, types.py).

Available Workflows

1. Video Generation (Veo 3)

Create new videos from text or image prompts.

• Text-to-Video: Create videos from detailed text descriptions.
• Image-to-Video: Animate static images.
• Prompt Engineering: Optimization keywords for camera, lighting, and style.

2. Video Editing (Veo 2)

Modify existing videos using masks (Inpainting).

• Remove Objects: Erase dynamic or static objects.
• Insert Objects: Add new elements into a scene.

3. Advanced Controls (Veo 3)

Specialized generation tasks for precise control.

• Frame Interpolation: Generate video bridging two images (first & last frame).
• Video Extension: Extend the duration of an existing video clip.
• Reference-to-Video: Use specific asset images (subjects, products) to guide generation.

Veo 3 Advanced Controls

This guide covers advanced generation techniques: Frame Interpolation, Video Extension, and using Asset Reference Images.

Frame Interpolation

Generates video content that bridges a first_frame and a last_frame. Useful for creating transitions or animating between two states.

from google import genai
from google.genai import types

client = genai.Client(vertexai=True, project="YOUR_PROJECT", location="us-central1")

operation = client.models.generate_videos(
    model="veo-3.1-generate-001",
    prompt="a hand reaches in and places a glass of milk", # Optional guidance
    image=types.Image.from_file(location="cookies.png"), # First frame
    config=types.GenerateVideosConfig(
        last_frame=types.Image.from_file(location="cookies-milk.png"), # Last frame
        duration_seconds=8, # Duration of the generated bridge
        aspect_ratio="16:9",
        generate_audio=True,
    ),
)

Video Extension

Extends an existing video clip in time. Requires veo-3.1-generate-preview.

operation = client.models.generate_videos(
    model="veo-3.1-generate-preview",
    prompt="a butterfly flies in and lands on the flower", # Describes the *new* content
    video=types.Video(uri="gs://bucket/short_clip.mp4", mime_type="video/mp4"),
    config=types.GenerateVideosConfig(
        duration_seconds=7, # How many seconds to ADD to the video
        output_gcs_uri="gs://bucket/extended_clip.mp4",
        generate_audio=True,
    ),
)

Reference-to-Video (Asset Images)

Use up to 3 specific reference images ("assets") to guide the generation. This helps preserve the identity of subjects, objects, or scenes across the video.

Asset Types

• Subject: A person or character.
• Object: A product or item.
• Scene: A background or environment.

Code Example

# Load images from GCS or local files
ref_image_1 = types.VideoGenerationReferenceImage(
    image=types.Image.from_file(location="man.png"),
    reference_type="asset",
)
ref_image_2 = types.VideoGenerationReferenceImage(
    image=types.Image.from_file(location="woman.png"),
    reference_type="asset",
)

operation = client.models.generate_videos(
    model="veo-3.1-generate-preview", # or veo-3.1-fast-generate-preview
    prompt="a woman and a man drinking a cup of coffee in a cafe",
    config=types.GenerateVideosConfig(
        reference_images=[ref_image_1, ref_image_2], # List of up to 3 references
        aspect_ratio="16:9",
        duration_seconds=8,
        person_generation="allow_adult",
        generate_audio=True,
    ),
)

Veo 2 Video Editing

This guide covers video editing tasks (inpainting) using the Veo 2 model. These operations require a source video and a mask image.

Model

• veo-2.0-generate-preview

Concepts

• Inpainting: Modifying a specific region of a video defined by a mask.
• Mask: An image where white pixels (255) represent the area to edit, and black pixels (0) represent the area to keep.
• Modes:
• REMOVE: Dynamic inpainting. Removes an object selected in the first frame mask throughout the video.
• REMOVE_STATIC: Static inpainting. Applies the mask to every frame (good for watermarks or fixed camera obstructions).
• INSERT: Adds new content into the masked area based on a prompt.

Remove Object (Dynamic)

Removes an object identified by the mask in the first frame.

from google import genai
from google.genai import types

client = genai.Client(vertexai=True, project="YOUR_PROJECT", location="us-central1")

operation = client.models.generate_videos(
    model="veo-2.0-generate-preview",
    source=types.GenerateVideosSource(
        video=types.Video(uri="gs://bucket/source_video.mp4", mime_type="video/mp4")
    ),
    config=types.GenerateVideosConfig(
        mask=types.VideoGenerationMask(
            image=types.Image.from_file(location="mask.png"),
            mask_mode=types.VideoGenerationMaskMode.REMOVE,
        ),
        enhance_prompt=True, # Recommended
    ),
)

Remove Static Object

Removes a static element (like a logo) that is in the same place in every frame.

config=types.GenerateVideosConfig(
    mask=types.VideoGenerationMask(
        image=types.Image.from_file(location="static_mask.png"),
        mask_mode=types.VideoGenerationMaskMode.REMOVE_STATIC,
    ),
    # Optional prompt to guide what replaces the background
    # prompt="a mountain landscape", 
)

Insert Object

Inserts a new object into the masked area defined by the prompt.

operation = client.models.generate_videos(
    model="veo-2.0-generate-preview",
    source=types.GenerateVideosSource(
        prompt="a sheep", # The object to insert
        video=types.Video(uri="gs://bucket/truck.mp4", mime_type="video/mp4")
    ),
    config=types.GenerateVideosConfig(
        mask=types.VideoGenerationMask(
            image=types.Image.from_file(location="mask.png"),
            mask_mode=types.VideoGenerationMaskMode.INSERT,
        ),
        output_gcs_uri="gs://bucket/output_with_sheep.mp4",
        enhance_prompt=True,
    ),
)

Veo 3 Video Generation

This guide covers creating videos from scratch using Text-to-Video and Image-to-Video with the Google Gen AI SDK.

Models

• veo-3.1-generate-001: High quality, standard latency.
• veo-3.1-fast-generate-001: Lower latency, optimized for speed.

Text-to-Video

Generate a video purely from a text prompt.

from google import genai
from google.genai import types

client = genai.Client(vertexai=True, project="YOUR_PROJECT", location="us-central1")

operation = client.models.generate_videos(
    model="veo-3.1-generate-001",
    prompt="a cinematic wide shot of a detective interrogating a rubber duck in a dark room",
    config=types.GenerateVideosConfig(
        aspect_ratio="16:9",      # "16:9" or "9:16"
        resolution="1080p",       # "720p", "1080p", or "4k" (4k adds latency)
        duration_seconds=6,       # 4, 6, or 8
        number_of_videos=1,       # 1 or 2
        person_generation="allow_adult", # "allow_adult" or "dont_allow"
        enhance_prompt=True,      # Let the model rewrite/improve your prompt
        generate_audio=True,      # Generate synchronized audio
        output_gcs_uri="gs://your-bucket/output.mp4" # Optional: Save directly to GCS
    ),
)

Image-to-Video

Generate a video starting from a static image context. The model animates the image based on the prompt.

operation = client.models.generate_videos(
    model="veo-3.1-generate-001",
    prompt="zoom out of the flower field, play whimsical music",
    image=types.Image.from_file(location="path/to/image.png"),
    config=types.GenerateVideosConfig(
        aspect_ratio="16:9",
        duration_seconds=6,
        resolution="1080p",
        generate_audio=True,
    ),
)

Prompt Engineering Parameters

When constructing prompts, consider these dimensions for better control:

Camera Control

• Angles: Eye-Level Shot, Low-Angle Shot, High-Angle Shot, Bird's-Eye View, Close-Up, Wide Shot, Over-the-Shoulder Shot, Drone Shot.
• Movement: Pan (left/right), Tilt (up/down), Zoom (In/Out), Dolly (In/Out), Truck (Left/Right), Handheld, Shaky Cam.

Visual Style

• Styles: Photorealistic, Cinematic, Vintage, Claymation style, Stop-motion animation, Film noir style, Cyberpunk.
• Lighting: Golden hour glow, Volumetric lighting, Film noir style, High-key lighting.

Audio Hints

• Mention sound effects or dialogue directly in the prompt (e.g., "Sound of waves crashing", "The person says: 'Hello world'").

Vertex AI Veo Prompting Guide

This guide provides a comprehensive overview of how to write effective prompts for Veo, Google's text-to-video and image-to-video generation model.

Core Components

A well-structured prompt typically includes: 1. Subject: The main character, object, or focus. 2. Action: What the subject is doing. 3. Scene/Context: The environment, setting, and background. 4. Cinematography: Camera angles, movements, and lens effects. 5. Visual Style: The overall aesthetic, lighting, and mood. 6. Ambiance: Sensory details. 7. Audio: Sound effects or speech.

Detailed Options

Cinematography (Camera & Lens)

Camera Angles

• Eye-Level Shot: Neutral perspective.
• Low-Angle Shot: Subject appears powerful.
• High-Angle Shot: Subject appears small/vulnerable.
• Bird's-Eye View / Top-Down: Directly from above.
• Dutch Angle / Canted Angle: Tilted to convey unease.
• Close-Up / Extreme Close-Up: Emphasizes emotions/details.
• Wide Shot / Establishing Shot: Shows context.
• Over-the-Shoulder: Behind one person looking at another.
• Point-of-View (POV): Character's visual perspective.

Camera Movements

• Static/Fixed: No movement.
• Pan (Left/Right): Horizontal rotation.
• Tilt (Up/Down): Vertical rotation.
• Dolly (In/Out): Camera moves closer/further.
• Zoom (In/Out): Lens focal length change.
• Truck (Left/Right): Camera moves laterally.
• Pedestal (Up/Down): Camera moves vertically.
• Crane / Aerial / Drone Shot: High altitude, sweeping.
• Handheld / Shaky Cam: Realism or unease.
• Whip Pan: Fast blur pan.

Lens & Optical Effects

• Wide-Angle (e.g., 24mm): Broader view, exaggerated perspective.
• Telephoto (e.g., 85mm): Compressed perspective, isolation.
• Shallow Depth of Field / Bokeh: Blurred background.
• Deep Depth of Field: Everything in focus.
• Lens Flare: Bright light source effect.
• Rack Focus: Shifting focus between subjects.
• Vertigo Effect (Dolly Zoom): Disorienting distortion.

Visual Style & Aesthetics

• Photorealistic / Cinematic: High fidelity.
• Vintage / Film Noir: Sepia, grainy, high contrast black & white.
• Animation Styles: 3D cartoon, Claymation, Stop-motion, Anime.
• Artistic: Impressionist (Van Gogh), Surrealist.
• Lighting: High-key (bright), Low-key (dark/moody), Golden hour, Volumetric (God rays), Backlighting (silhouette).

Temporal Elements

• Pacing: Slow-motion, Fast-paced action.
• Evolution: Time-lapse, Hyperlapse.
• Rhythm: Pulsating light, Rhythmic movement.

Best Practices

• Be Specific: Avoid "A man walking." Use "Eye-level medium shot of a young man in a soaked trench coat..."
• Negative Prompts: Define what to exclude (e.g., "Negative prompt: blurry, distorted, text, watermark").
• Iterate: Use Gemini to rewrite prompts for better detail.

Audio (Preview)

Specify sound effects or speech clearly.

• "The audio features water splashing."
• "The man says, 'Where is the rabbit?'"

Google GenAI SDK Source Code (Veo)

Use web_fetch to retrieve raw code for deep inspection of SDK internals, especially for video generation parameters.

Base URL: https://raw.githubusercontent.com/googleapis/python-genai/main/google/genai/

Key Modules for Veo

Models (Generation Logic)

• File: models.py
• URL: https://raw.githubusercontent.com/googleapis/python-genai/main/google/genai/models.py
• Purpose: Contains generate_videos and generate_images methods. Check this for parameter handling.

Types (Configuration)

• File: types.py
• URL: https://raw.githubusercontent.com/googleapis/python-genai/main/google/genai/types.py
• Purpose: Definitions for GenerateVideosConfig, PersonGeneration, VideoCompressionQuality, etc.

Operations (Async Handling)

• File: operations.py
• URL: https://raw.githubusercontent.com/googleapis/python-genai/main/google/genai/operations.py
• Purpose: Handling LROs (Long Running Operations) returned by video generation.

Client

• File: client.py
• URL: https://raw.githubusercontent.com/googleapis/python-genai/main/google/genai/client.py