FullFight.AI

A smart video tool that uses AI to detect and compile only the most intense, action-packed scenes from anime episodes, so fans can jump straight to the good stuff.

Comment

Inspiration

Frequently when I watch anime, amazing fights get interrupted by flashbacks I'm not interested in. This makes me skip through the episode just to find the next fight scene.

What It Does

The user uploads a video (anime episode), and the AI model skims through the content to detect and extract only the fight scenes. The result: a clean, action-packed highlight reel without any filler.

How We Built It

We used a Random Forest classifier to classify four key features from the video:

  • Audio RMSI level
  • Mean brightness per frame per second
  • Motion flow
  • Dialogue emotion detection

These features are analyzed across the episode to detect fight scenes. Once identified, the system generates subclips from the original video containing only those scenes.

Challenges We Ran Into

  • Integrating the ML model with frontend functionality and calling the correct timestamps for trimming
  • Training and evaluating the model to ensure high accuracy

Accomplishments We're Proud Of

  • A fully trained machine learning model
  • A clean and intuitive UI
  • End-to-end system working: model interprets the video, trims the correct scenes, and stitches them together seamlessly

What We Learned

  • How to debug with discipline through many failed attempts
  • How to collaborate effectively using different tools and workflows

What's Next for FullFight.AI

  • Support for multiple episodes in one upload
  • Implementing neural networks for improved accuracy and performance

Built With

Share this project:

Updates