How I Used VideoDB API to Build a Smart Video Scene Finder and Shorts Maker

📌 Tired of manually scrubbing through videos to find key scenes?
This AI-powered project automates the whole process — extracting specific scenes using semantic search and instantly converting them into vertical, mobile-optimized content (9:16).
✨ Imagine typing "fight scene" or "man laughing" and instantly getting that clip, converted into vertical format and ready for Reels — no scrubbing, no editing.

👋 Hey, Devs!

As a part of ongoing AI Demos x VideoDB Hackathon, I have built a cool side project that solves a real pain for video creators: finding specific scenes in long videos and turning them into Shorts with zero editing. Think AI-powered semantic search, scene extraction, and auto vertical cropping — all baked into a single app.

This blog walks you through how I did it using the awesome VideoDB API and Python. Let's get into it.

---

🧠 Behind the Idea

Most creators waste hours reviewing long videos to find viral moments. This tool solves this by:

1. 🔍 Auto-detecting specific moments like “person talking” or “car chase”

2. 🧠 Using semantic search instead of timestamp scrubbing.

3. 🌐 Converting selected moments into ready-to-publish vertical videos

4. ⚡ Boosting productivity for video editors, marketers, and influencers.

---

⚙️ What My App Does

• User uploads a video (via link)

• Enters a scene query, e.g., "person talking"

• The backend hits VideoDB API and gets scene timestamps

• Extracts clips using ffmpeg

• Clips are resized/cropped into 9:16 using smart centering (OpenCV)

• Outputs ready-to-publish reels in seconds!

  No manual work. Just magic.

---

⚙️ Tech Stack

Tech	Role
Python & Flask	Core backend logic
VideoDB API	AI-driven semantic scene search
OpenCV	Smart cropping & resizing
HTML, CSS, JS	UI for uploading & previewing
ffmpeg	Format conversions

---

📁 Project Structure

---

🔍 Scene Detection Using VideoDB API

We utilise video_processor.py to interact with VideoDB:

• Send video file + semantic query

• Receive timestamps of matching scenes

• Process only the relevant chunks

• Zero manual editing!

---

📱 Vertical Cropping Magic

Traditional resizing distorts aspect ratio.
We use vertical_converter.py to:

• Auto-detect region of interest

• Use OpenCV to smart crop central subjects

• Output 9:16 videos without losing focus

---

💻 Demo Snapshot

Here’s a quick preview of the UI:

• Upload your video

• Type your scene query

• Get vertical video in a minute!

Watch demo video here

---

🚀 Try It Yourself

Clone the repo:

git clone https://github.com/pradyot29/AI-powered-Scene-Extractor-and-Vertical-Video-Converter.git
cd AI-powered-Scene-Extractor-and-Vertical-Video-Converter
pip install -r requirements.txt

Set your.env file with your VideoDB API key.

VIDEODB_API_KEY=your_videodb_key_here
SECRET_KEY=your_flask_secret_key_here

Then launch the app:

python app.py

---

🔚 Final Thoughts

This project is useful and very practical if you’re:

• A YouTuber or Instagram Reels creator

• A news editor trying to repurpose content

• A startup marketer automating video campaigns

• A developer looking to build on AI + video editing tools

⭐ Don’t just edit faster. Edit smarter with AI.

---

📢 Let’s Connect!

If you liked this project:

• ⭐ Star the GitHub repo Github-pradyot

• 💬 Drop a comment below or share this blog!

• 🤙 Connect with me on LinkedIn.

• 🌐 Want to build and win big? Register fast; AI Demos x VideoDB hackathon is live now.

---