video-ad-deconstructor

# Video Ad Deconstructor

AI-powered deconstruction of video ad creatives into actionable marketing insights.

## What This Skill Does

- **Generate Summaries**: Product, features, audience, CTA extraction
- **Deconstruct Marketing Dimensions**: Hooks, social proof, urgency, emotion, etc.
- **Support Multiple Content Types**: Consumer products and gaming ads
- **Progress Tracking**: Callback support for long analyses
- **JSON Output**: Structured data for downstream processing

## Setup

### 1. Environment Variables

```bash
# Required for Gemini
GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json
```

### 2. Dependencies

```bash
pip install vertexai
```

## Usage

### Basic Ad Deconstruction

```python
from scripts.deconstructor import AdDeconstructor
from scripts.models import ExtractedVideoContent
import vertexai
from vertexai.generative_models import GenerativeModel

# Initialize Vertex AI
vertexai.init(project="your-project-id", location="us-central1")
gemini_model = GenerativeModel("gemini-1.5-flash")

# Create deconstructor
deconstructor = AdDeconstructor(gemini_model=gemini_model)

# Create extracted content (from video-ad-analyzer or manually)
content = ExtractedVideoContent(
    video_path="ad.mp4",
    duration=30.0,
    transcript="Tired of messy cables? Meet CableFlow...",
    text_timeline=[{"at": 0.0, "text": ["50% OFF TODAY"]}],
    scene_timeline=[{"timestamp": 0.0, "description": "Person frustrated with tangled cables"}]
)

# Generate summary
summary = deconstructor.generate_summary(
    transcript=content.transcript,
    scenes="0.0s: Person frustrated with tangled cables",
    text_overlays="50% OFF TODAY"
)
print(summary)
```

### Full Deconstruction

```python
# Deconstruct all marketing dimensions
def on_progress(fraction, dimension):
    print(f"Progress: {fraction*100:.0f}% - Analyzed {dimension}")

analysis = deconstructor.deconstruct(
    extracted_content=content,
    summary=summary,
    is_gaming=False,  # Set True for gaming...