video-ad-analyzer

What This Skill Does

The video-ad-analyzer is a sophisticated AI-driven tool designed to deconstruct video advertisements into actionable intelligence. By leveraging Google Gemini Vision and a suite of computer vision libraries, the skill provides deep insights into multimedia content. It performs frame extraction using smart sampling and scene change detection, OCR text detection via EasyOCR to capture overlays, and robust audio transcription using Google Cloud Speech-to-Text. Furthermore, it utilizes generative AI to provide detailed scene-by-scene descriptions, making it an indispensable asset for marketers, researchers, and content creators looking to reverse-engineer competitive ad strategies or audit their own video assets for performance.

Installation

To integrate this skill, ensure you have system-level support for video processing by installing ffmpeg and ffprobe. Once the environment is prepared, use the OpenClaw CLI by running: clawhub install openclaw/skills/skills/fortytwode/meta-video-ad-analyzer. You must also install the required Python dependencies: pip install opencv-python pillow easyocr ffmpeg-python google-cloud-speech vertexai google-api-python-client. Finally, configure your environment by setting the GOOGLE_APPLICATION_CREDENTIALS variable to point to your Google Cloud service account JSON file, ensuring the Speech-to-Text and Vertex AI APIs are enabled in your Google Cloud project console.

Use Cases

This skill is highly effective for competitive intelligence, enabling brands to analyze rival ad structures and messaging patterns. It is also perfect for accessibility compliance, allowing creators to generate automatic transcripts and scene descriptions for video content. Additionally, it streamlines media library management by auto-generating representative thumbnails and summarizing long-form video content into digestible text timelines, saving hours of manual review time.

Example Prompts

1. "Analyze this video ad at /path/to/ad.mp4. Tell me what the main call-to-action is and describe the scenes where the product appears."
2. "Extract all text overlays from this 30-second commercial and transcribe the voiceover to see if the visual text matches the audio script."
3. "Generate a detailed breakdown of the video timeline and provide a thumbnail image representing the first major scene."

Tips & Limitations

For optimal results, ensure your input videos are of clear quality; heavily compressed or low-resolution footage may reduce the accuracy of EasyOCR and scene detection. The skill relies on external API calls to Google Cloud, so verify your network connectivity and API quotas. Because it uses generative models, ensure your prompts are descriptive. Remember that processing speed depends on video duration; consider breaking very long videos (over 10 minutes) into smaller clips before analysis for faster results.