Generate images and videos, analyze multimedia files, and process documents using Google Gemini’s multimodal AI capabilities.

What This Skill Does

Challenge: Marketing requires processing diverse media (audio, images, videos, PDFs) and creating visual content. Using separate tools for each format is inefficient and expensive.

Solution: The AI Multimodal skill integrates Google Gemini API for audio transcription (9.5h), image analysis, video processing (6h), PDF extraction, Imagen 4 image generation, and Veo 3 video generation (8-second clips with audio). One API, 2M token context window.

Activation

Implicit: Activates automatically when agents need image analysis, audio transcription, video processing, or visual content generation.

Explicit: Activate by name when needed: “Activate ai-multimodal skill”

Capabilities

1. Audio Transcription and Analysis

Transcribe audio files with timestamps, speaker detection, and non-speech analysis.

# Transcribe audio
python scripts/gemini_batch_process.py --files meeting.mp3 --task transcribe

# Analyze podcast for topics
python scripts/gemini_batch_process.py --files podcast.mp3 --task analyze --prompt "Identify key topics and timestamps"

Formats: WAV, MP3, AAC (up to 9.5 hours)

Output: Markdown with timestamp format [HH:MM:SS -> HH:MM:SS]

Important: Files >15 minutes may be truncated due to output token limits. Split into 15-minute chunks for full transcription.

Guide: references/audio-processing.md

2. Image Understanding and Analysis

Analyze images for content, text extraction, object detection, and visual Q&A.

# Analyze screenshot
python scripts/gemini_batch_process.py --files screenshot.png --task analyze --prompt "Describe UI layout and identify all interactive elements"

# Extract text from image
python scripts/gemini_batch_process.py --files poster.jpg --task extract --prompt "Extract all text in structured format"

Formats: PNG, JPEG, WebP, SVG (up to 3.6k images)

Capabilities: Captioning, classification, OCR, object detection, visual Q&A

Guide: references/vision-understanding.md

3. Video Analysis

Process videos for scene detection, temporal Q&A, and context-aware transcription.

# Analyze marketing video
python scripts/gemini_batch_process.py --files promo.mp4 --task analyze --prompt "Identify scenes, key messages, and suggested timestamps for cuts"

# Analyze YouTube URL
python scripts/gemini_batch_process.py --files https://youtube.com/watch?v=xyz --task analyze

Formats: MP4, MOV, WebM (up to 6 hours)

For Long Videos: Extract audio with FFmpeg, split into 15-minute chunks, transcribe separately.

Guide: references/video-analysis.md

4. Image Generation with Imagen 4

Generate marketing images from text descriptions.

# Generate social media image
python scripts/gemini_batch_process.py --task generate --prompt "Modern SaaS dashboard screenshot, clean interface, blue and white color scheme, professional"

# Batch generate
python scripts/gemini_batch_process.py --task generate --prompt "Product hero image" --count 4

Models:

  • imagen-4.0-generate-001 (standard quality)
  • imagen-4.0-ultra-generate-001 (highest quality)
  • imagen-4.0-fast-generate-001 (fastest)

Aspect Ratios: 1

, 16
, 9
, 4
, 3

Guide: references/image-generation.md

5. Video Generation with Veo 3

Generate 8-second video clips with audio asynchronously.

# Generate video clip
python scripts/gemini_batch_process.py --task generate-video --prompt "Product unboxing, smooth camera motion, professional lighting, cheerful mood"

Model: veo-3.1-generate-preview

Duration: 8 seconds with audio

Output: MP4 file

Guide: references/video-generation.md

Prerequisites

API Access:

  • GEMINI_API_KEY from Google AI Studio
  • Python 3.8+ with google-genai, python-dotenv, pillow

Installation:

pip install google-genai python-dotenv pillow

Verify Installation:

python scripts/check_setup.py

Configuration

Environment Variables (.env):

GEMINI_API_KEY=your_key_here

Available Scripts:

  • gemini_batch_process.py - Main CLI for all tasks
  • media_optimizer.py - Compress/resize media for API limits
  • document_converter.py - Convert PDF to markdown
  • check_setup.py - Verify API keys and dependencies

Best Practices

1. Choose Right Model For Task

  • gemini-2.5-flash for speed (transcription, analysis)
  • gemini-2.5-pro for complex reasoning
  • imagen-4.0-generate-001 for standard images
  • imagen-4.0-ultra-generate-001 for hero images only (higher cost)

2. Optimize Media Before Upload Use media_optimizer.py to compress files >20MB before processing.

3. Split Long Audio/Video 15-minute chunks prevent transcription truncation. Use FFmpeg to split.

Common Use Cases

Use Case 1: Transcribe Podcast With Topics

Scenario: Transcribe 45-minute podcast episode with timestamps and extract topics.

Workflow:

  1. Split audio into 3 × 15-minute chunks (FFmpeg)
  2. Transcribe each: python scripts/gemini_batch_process.py --files chunk1.mp3 --task transcribe
  3. Combine transcriptions
  4. Extract topics: --task analyze --prompt "Identify top 5 topics with timestamps"

Outcome: Full transcription with topic summary.

Use Case 2: Generate Social Media Images

Scenario: Generate 4 variations of product announcement image.

Workflow:

  1. Define prompt with brand context: “Modern tech product announcement, gradient blue background, clean typography, professional, 16
    ”
  2. Batch generate: python scripts/gemini_batch_process.py --task generate --prompt "..." --count 4
  3. Review outputs
  4. Select best variation for posting

Outcome: 4 unique images matching brand guidelines.

Troubleshooting

Issue: API returns 401 Unauthorized Solution: Verify GEMINI_API_KEY in .env file. Get new key from Google AI Studio if needed.

Issue: Transcription truncated for long audio Solution: Split audio into 15-minute chunks using FFmpeg or media_optimizer.py.

Issue: Image generation doesn’t match brand Solution: Include specific brand colors, style keywords, and reference existing assets in prompt. Use ai-artist skill to optimize prompts.

  • /design/good - Generate high-quality images
  • /design/video - Plan video content
  • /content/enhance - Improve content with AI analysis