GenAI Video Generation Skill

GenAI Video Generation Skill

Generate videos using the genai-cli video command with Veo 3.1.

Uses fast mode by default ($0.15/​sec) for quick generation. Use --quality-mode for higher quality output ($0.40/​sec).

Quick Start

# Basic video generation (async - returns job ID)
uv run genai-cli video "sunset over ocean, gentle waves"

# Wait for completion (blocking)
uv run genai-cli video "sunset over ocean" --wait

# Image-to-video (use existing image as starting frame)
uv run genai-cli video "camera pullback revealing the scene" --input ./​image.png --wait

# First/​last frame interpolation
uv run genai-cli video "smooth morph transition" --input ./​start.png --last-frame ./​end.png --wait

CLI Reference

uv run genai-cli video [OPTIONS] PROMPT

Options:
  --aspect, -a      Aspect ratio: 16:9 or 9:16 (default: 16:9)
  --duration, -d    Duration in seconds: 4, 6, or 8 (default: 8)
  --resolution, -r  Resolution: 720p, 1080p, or 4k (default: 720p)
  --seed, -s        Random seed for reproducibility
  --audio           Generate video with audio
  --wait, -w        Wait for completion (blocking mode)
  --input, -i       First frame image for image-to-video
  --last-frame, -l  Last frame image (requires --input)
  --negative, -n    What NOT to include in video
  --ref             Reference images for style (up to 3, 16:9 only)
  --ref-type        Reference image type: style (aesthetic) or asset (character/​object)
  --extend, -e      Video file URI to extend
  --person-gen      Person generation: allow_all or allow_adult
  --quality-mode    Use quality model (slower, higher quality, $0.40/​sec)
  --enhance         Auto-improve prompt for better video quality
  --fps             Frames per second (e.g., 24 for cinematic, 60 for smooth)
  --compression     Compression quality: optimized (smaller) or lossless (larger)
  --variations, -v  Number of video variations to generate (1-4)
  --mask, -m        Mask image for targeted editing (requires --input)
  --mask-mode       Mask mode: insert, remove, remove-static, outpaint
  --output, -o      Output directory (default: ./​genai_outputs/)
  --json            Output as JSON

Model Modes

Fast Mode (Default)

Uses veo-3.1-fast-generate-preview model at $0.15/​second. Optimized for speed and business use cases.

uv run genai-cli video "sunset over ocean" --wait

Quality Mode

Uses veo-3.1-generate-preview model at $0.40/​second. Higher quality output for premium content.

uv run genai-cli video "cinematic sunset scene" --quality-mode --wait

Prompt Enhancement

Auto-improve your prompts for better video quality:

# Let the model enhance your simple prompt
uv run genai-cli video "sunset" --enhance --wait

Frame Rate Control (FPS)

Control frames per second for different effects:

# Cinematic 24fps (film-like)
uv run genai-cli video "dramatic scene" --fps 24 --wait

# Standard 30fps
uv run genai-cli video "product demo" --fps 30 --wait

# Smooth 60fps (action/​sports)
uv run genai-cli video "slow motion water" --fps 60 --wait

Compression Quality

Control file size vs quality tradeoff:

# Optimized - smaller file size (default behavior)
uv run genai-cli video "social media clip" --compression optimized --wait

# Lossless - larger file, no quality loss (for archival/​editing)
uv run genai-cli video "product demo" --compression lossless --wait

Multiple Variations

Generate multiple video variations from the same prompt:

# Generate 3 different versions
uv run genai-cli video "abstract animation" --variations 3 --wait

# Generate 4 variations for A/​B testing
uv run genai-cli video "hero scene" --variations 4 --wait

Modes

Async Mode (Default)

Returns immediately with a job ID:

uv run genai-cli video "flying through clouds"
# Output: Job ID: models/​veo-3.1.../​operations/​abc123

Blocking Mode (--wait)

Waits for video to complete and downloads it:

uv run genai-cli video "flying through clouds" --wait
# Output: ✓ Saved: genai_outputs/​video_20260114_123456_1.mp4 (2.1 MB)

Image-to-Video

Use an existing image as the starting frame for your video:

# Camera pullback from an image
uv run genai-cli video "Dramatic camera pullback revealing the full scene" \
  --input ./​my_image.png \
  --aspect 9:16 \
  --wait

# Animate a still image
uv run genai-cli video "Gentle movement, hair blowing in wind" \
  --input ./​portrait.png \
  --wait

First & Last Frame Interpolation

Generate videos by specifying both the starting and ending frames. The model interpolates smooth motion between them:

# Morph between two images
uv run genai-cli video "Smooth transition with dramatic lighting" \
  --input ./​start_frame.png \
  --last-frame ./​end_frame.png \
  --wait

# Scene transformation
uv run genai-cli video "Day transforms to night, cinematic transition" \
  --input ./​day_scene.png \
  --last-frame ./​night_scene.png \
  --aspect 16:9 \
  --wait

Requirements:

• Both images must match the video's aspect ratio
• --last-frame requires --input (first frame) to be set
• Duration defaults to 8 seconds for interpolation

Reference Images for Style/​Assets

Use up to 3 reference images to guide the video. Two reference types are available:

Style References (Default)

Transfer aesthetic elements like colors, lighting, and visual style:

# Style-guided generation (default)
uv run genai-cli video "A person walking through a forest" \
  --ref ./​style1.png --ref ./​style2.png \
  --ref-type style \
  --aspect 16:9 \
  --wait

Asset References

Maintain character or object consistency across scenes:

# Character consistency
uv run genai-cli video "The character walks through a city" \
  --ref ./​character_front.png --ref ./​character_side.png \
  --ref-type asset \
  --aspect 16:9 \
  --wait

# Object consistency
uv run genai-cli video "The product rotates on display" \
  --ref ./​product_view1.png --ref ./​product_view2.png \
  --ref-type asset \
  --aspect 16:9 \
  --wait

Note: Reference images only work with 16:9 aspect ratio and require duration of 8 seconds.

Video Extension

Extend a previously generated Veo video:

# Extend an existing video
uv run genai-cli video "Continue the scene with more action" \
  --extend "files/​video-file-id" \
  --wait

Note: Extension requires duration of 8 seconds.

Negative Prompts

Specify what NOT to include in your video:

uv run genai-cli video "A serene beach scene" \
  --negative "people, crowds, text, watermarks" \
  --wait

Person Generation Control

Control how people are generated in videos:

# Allow all person generation (text/​extension mode)
uv run genai-cli video "People walking in park" \
  --person-gen allow_all \
  --wait

# Allow adult-only generation (image/​interpolation/​references)
uv run genai-cli video "Professional meeting" \
  --input ./​office.png \
  --person-gen allow_adult \
  --wait

Audio Generation

Generate videos with synchronized audio:

uv run genai-cli video "Ocean waves crashing on beach with seagulls" \
  --audio \
  --wait

Resolution & Duration

High Resolution (1080p/​4K)

# 1080p requires 8s duration
uv run genai-cli video "Cinematic landscape" \
  --resolution 1080p \
  --duration 8 \
  --wait

# 4K also requires 8s duration
uv run genai-cli video "Ultra HD nature scene" \
  --resolution 4k \
  --duration 8 \
  --wait

Duration Options

# 4 seconds
uv run genai-cli video "Quick action shot" --duration 4 --wait

# 6 seconds
uv run genai-cli video "Medium scene" --duration 6 --wait

# 8 seconds (default, required for 1080p/​4K/​extension/​references)
uv run genai-cli video "Full scene" --duration 8 --wait

Camera Movement Prompts

For image-to-video, describe the camera movement:

Video Masking (Targeted Editing)

Use masks to edit specific regions of videos:

Insert Mode

Add new objects into a masked region:

uv run genai-cli video "add a butterfly flying" \
  --input ./​scene.png \
  --mask ./​butterfly_area_mask.png \
  --mask-mode insert \
  --wait

Remove Mode

Remove tracked objects from video (tracks object through frames):

uv run genai-cli video "remove the car smoothly" \
  --input ./​street_scene.png \
  --mask ./​car_mask.png \
  --mask-mode remove \
  --wait

Remove-Static Mode

Remove static objects (watermarks, logos, fixed elements):

uv run genai-cli video "clean background" \
  --input ./​video_frame.png \
  --mask ./​watermark_mask.png \
  --mask-mode remove-static \
  --wait

Outpaint Mode

Expand the video beyond its original frame:

uv run genai-cli video "expand the landscape view" \
  --input ./​cropped_scene.png \
  --mask ./​outpaint_mask.png \
  --mask-mode outpaint \
  --wait

Mask Requirements:

• Mask must be same dimensions as input image
• White areas (255) = region to modify
• Black areas (0) = region to preserve
• --mask requires --input (first frame image)

Output

Videos are saved to ./​genai_outputs/ by default:

• video_YYYYMMDD_HHMMSS_1.mp4

Prompt Tips

1. Describe movement: "Camera slowly pulls back", "Gentle zoom in"
2. Include atmosphere: "Moody lighting", "Dust particles in air"
3. Use audio cues: With --audio, describe sounds in prompt
4. Be cinematic: "Dramatic", "Cinematic", "Film-like"
5. Use negative prompts: Exclude unwanted elements with --negative

API Constraints

Prerequisites

API key must be configured:

uv run genai-cli auth set-key