StoryLens / README.md
Marek4321's picture
Update README.md
66710f8 verified
---
title: StoryLens
emoji: 🎬
colorFrom: green
colorTo: blue
sdk: streamlit
sdk_version: 1.53.0
app_file: app.py
pinned: false
license: apache-2.0
---
# 🎬 StoryLens
Professional video ad narrative structure analyzer powered by AI. Diagnose your ad's storytelling effectiveness in 60 seconds.
## ✨ Features
- **Multi-source Input**: Upload video files (MP4, MOV, AVI, WebM) or paste YouTube URL
- **AI-Powered Analysis**: MiniMax Vision for frame analysis, OpenAI Whisper for transcription
- **Functional Role Classification**: 21 roles across 5 categories (Opening, Problem, Product, Persuasive, Closure)
- **Story Arc Detection**: Identifies patterns like AIDA, PAS, HFBA, and more
- **Industry Benchmarks**: Recommendations based on Meta research (WWW 2026)
- **Actionable Insights**: Prioritized recommendations with expected performance uplift
## πŸš€ Quick Start
1. Configure API keys in sidebar:
- **MiniMax API Key** + Group ID (for vision & LLM)
- **OpenAI API Key** (for Whisper transcription)
2. Select your **Industry** and **Campaign Goal**
3. Upload video file or paste YouTube URL
4. Click **"Analyze"**
5. Review narrative timeline, story arc, and recommendations
## πŸ”‘ API Keys Required
- **MiniMax**: Get from [MiniMax Platform](https://platform.minimax.io/)
- **OpenAI**: Get from [OpenAI Platform](https://platform.openai.com/)
## πŸ“Š Supported Industries
- Apparel & Accessories
- Beauty
- Food
- Beverages
- Other
## 🎯 Campaign Goals
- Retention (Dwell Rate)
- Click-Through (CTR)
- Conversions (CVR)
## πŸ“š Based On
Research paper: *MLLM-VADStory: Domain Knowledge-Driven Multimodal LLMs for Video Ad Storyline Insights* (Meta, WWW 2026)
## πŸ›‘οΈ Privacy
- API keys never stored permanently
- Videos processed temporarily only
- No data retention
## πŸ“„ License
Licensed under **Apache License 2.0**
## πŸ‘¨β€πŸ’» Author
**Marek Staniszewski**
**Heuristica**
🌐 [www.heuristica.pl](https://www.heuristica.pl)
πŸ“§ staniszewski@heuristica.pl
For questions, suggestions, or support, please contact the author.