Agent-Skills.md

Agent Skills: Nano Banana - Video Generation

Generate videos using Veo 3.1 — text-to-video, image-to-video, frame interpolation, and video extension

UncategorizedID: flight505/nano-banana/video

Repository

flight505/nano-banana
flight505License: MIT
2

Install this agent skill to your local

pnpm dlx add-skill https://github.com/flight505/nano-banana/tree/HEAD/skills/video

Skill Files

Browse the full folder contents for video.

Download Skill

Loading file tree…

skills/video/SKILL.md

Skill Metadata

Name
video
Description
"Generate videos using Veo 3.1 — text-to-video, image-to-video, frame interpolation, and video extension"

Nano Banana - Video Generation

Overview

Generate videos using Google's Veo 3.1 models via the google-genai SDK. Supports four generation modes: text-to-video, image-to-video, frame interpolation, and video extension.

Key Features:

  • Four generation modes from a single script
  • Fast (default) and standard quality models
  • Aspect ratio control (16:9, 9:16)
  • Resolution up to 1080p
  • Configurable duration (4, 6, or 8 seconds)
  • Automatic audio stripping (opt-in to keep)
  • Reference images for style guidance

When to Use This Skill

Use this skill when you need:

  • Animated content from a text description
  • Image animation — bring a still photo or illustration to life
  • Smooth transitions between two frames (interpolation)
  • Extending an existing video clip
  • Short-form video for reels, demos, or presentations

Not for: Static images (use image skill), technical diagrams (use diagram skill), or diagram source rendering (use kroki skill).

Quick Start

Text-to-Video (default)

python3 ${CLAUDE_SKILL_DIR}/scripts/generate_video.py "A drone flyover of a coastal city at golden hour" -o flyover.mp4

Image-to-Video

Animate a still image with a motion prompt:

python3 ${CLAUDE_SKILL_DIR}/scripts/generate_video.py "Camera slowly zooms in while clouds drift" \
  --input photo.png -o animated.mp4

Frame Interpolation

Generate a smooth transition between two frames:

python3 ${CLAUDE_SKILL_DIR}/scripts/generate_video.py "Smooth morph between scenes" \
  --input start.png --last-frame end.png -o transition.mp4

Video Extension

Continue an existing video clip:

python3 ${CLAUDE_SKILL_DIR}/scripts/generate_video.py "The camera keeps panning right revealing more of the landscape" \
  --extend clip.mp4 -o extended.mp4

With Reference Images

Guide the visual style with up to 3 reference images:

python3 ${CLAUDE_SKILL_DIR}/scripts/generate_video.py "A cat walking through a garden" \
  -o styled.mp4 --reference style1.png --reference style2.png

Models

| Model | ID | Speed | Best For | |-------|-----|-------|----------| | Veo 3.1 Fast | veo-3.1-fast-generate-preview | Fast | Default — quick iterations, drafts | | Veo 3.1 | veo-3.1-generate-preview | Standard | Final output, higher quality |

Use -m to select a model:

python3 ${CLAUDE_SKILL_DIR}/scripts/generate_video.py "A sunset timelapse" -o sunset.mp4 -m veo-3.1-generate-preview

Resolution and Duration Constraints

| Resolution | Supported Durations | Notes | |------------|-------------------|-------| | 720p (default) | 4, 6, 8 seconds | All durations, required for extension | | 1080p | 8 seconds only | Higher quality, duration locked |

# 720p, 4 seconds (fast preview)
python3 ${CLAUDE_SKILL_DIR}/scripts/generate_video.py "Quick test" -o test.mp4 --resolution 720p --duration 4

# 1080p, must be 8 seconds
python3 ${CLAUDE_SKILL_DIR}/scripts/generate_video.py "Cinematic landscape" -o landscape.mp4 --resolution 1080p --duration 8

Constraint rules:

  • --resolution 1080p requires --duration 8
  • --extend requires --resolution 720p
  • Maximum 3 --reference images

Aspect Ratio

| Ratio | Use Case | |-------|----------| | 16:9 (default) | Landscape, presentations, YouTube | | 9:16 | Portrait, mobile reels, TikTok, Stories |

# Vertical video for social media
python3 ${CLAUDE_SKILL_DIR}/scripts/generate_video.py "A vertical reel of a coffee being poured" \
  -o reel.mp4 --aspect-ratio 9:16

Audio Handling

By default, audio is stripped from generated videos using ffmpeg. This avoids unexpected AI-generated audio.

# Default: audio stripped
python3 ${CLAUDE_SKILL_DIR}/scripts/generate_video.py "A beach scene" -o beach.mp4

# Keep generated audio
python3 ${CLAUDE_SKILL_DIR}/scripts/generate_video.py "A concert crowd cheering" -o concert.mp4 --audio

Requirements for audio stripping:

  • ffmpeg must be installed (brew install ffmpeg on macOS)
  • If ffmpeg is not found, the video is saved with audio intact and a warning is printed

All CLI Flags

| Flag | Default | Description | |------|---------|-------------| | prompt (positional) | required | Text description of the video | | -o / --output | required | Output .mp4 file path | | -m / --model | veo-3.1-fast-generate-preview | Model ID | | -i / --input | — | Input image (image-to-video / interpolation) | | --last-frame | — | Last frame image (interpolation, requires --input) | | --extend | — | Video .mp4 to extend | | --reference | — | Reference image (repeatable, max 3) | | --aspect-ratio | 16:9 | 16:9 or 9:16 | | --resolution | 720p | 720p or 1080p | | --duration | 8 | 4, 6, or 8 seconds | | --audio | off | Keep generated audio | | --timeout | 360 | Max wait in seconds | | --api-key | — | Override GEMINI_API_KEY |

Configuration

API Key Setup

export GEMINI_API_KEY='your-key-here'

Get a key at https://aistudio.google.com/apikey.

Or add to a .env file in your project:

GEMINI_API_KEY=your-key-here

Run /nano-banana:setup for guided configuration.

Tips for Better Videos

Be Descriptive About Motion

# Vague — model guesses camera movement
"A mountain"

# Specific — clear motion intent
"A slow aerial drone shot pulling back from a snowy mountain peak, revealing the valley below, golden hour lighting"

Specify Camera Movement

Use terms like: pan left/right, zoom in/out, dolly, orbit, tracking shot, crane shot, steady cam, timelapse.

For Image-to-Video, Describe the Animation

# Good: describes what should move
python3 ${CLAUDE_SKILL_DIR}/scripts/generate_video.py "The clouds slowly drift across the sky while the water gently ripples" \
  --input landscape.png -o animated.mp4

Troubleshooting

"No Gemini API key found"

Set the GEMINI_API_KEY environment variable or create a .env file. Run /nano-banana:setup for help.

"Constraint validation failed"

Check the resolution/duration constraint table above. Common issue: using --resolution 1080p with --duration 4.

Generation Timeout

Default timeout is 360 seconds (6 minutes). Video generation typically takes 1-4 minutes. Increase with --timeout 600 for complex prompts.

"ffmpeg not found" Warning

Install ffmpeg for automatic audio stripping: brew install ffmpeg. Without it, videos are saved with AI-generated audio.

Video Appears Black or Corrupted

Try a simpler prompt, or switch to the standard quality model (-m veo-3.1-generate-preview). Some complex scenes may not render well at lower quality settings.

Comparison: video vs image vs diagram

| Aspect | video | image | diagram | |--------|---------|---------|-----------| | Output | .mp4 video | .png image | .png diagram | | Models | Veo 3.1 | Gemini Flash/Pro | Gemini Pro | | Duration | 4-8 seconds | Instant | 1-2 passes | | Editing | Extend existing | Edit existing | Edit existing | | Best For | Animation, motion | Photos, art | Architecture, flowcharts |