Skip to content

in-transit

In today’s digital age, long-form video content struggles with information density, leading to viewer drop-off and knowledge loss.

Code Video Cerebral Valley

Core trick

Transcribes speech with timestamps, has Gemini segment the content by theme, then trims and joins segments through a video API.

Tags

productivityvoice

Description

In today’s digital age, long-form video content struggles with information density, leading to viewer drop-off and knowledge loss. In-Transit solves this by using AI to distill lengthy videos into concise summaries through a multi-stage pipeline: ElevenLabs delivers accurate speech-to-text transcription across accents and styles, Google Gemini 2.5 analyzes and segments content by themes and context, Google NanoBanana generates smooth visual transitions to maintain narrative flow, and Fal.ai enables advanced video processing for seamless editing and compression. With real-time processing of local uploads and direct MP4 URLs, the platform preserves the creator’s message while removing redundancy, allowing faster comprehension without losing critical information. This makes it particularly useful for educational content, business presentations, and informational videos where quick, efficient knowledge acquisition is essential.
rshiramss/in-transit★ 23 forksPythonCreated Sep 6, 2025 · last push Dec 22, 2025

README excerpt

# In-Transit 

An AI-powered video processing pipeline that transforms long-form content into engaging, condensed videos with seamless AI transitions. Built for the Google Nano Banana Hackathon in San Francisco (September 6-7, 2025).

##  Hackathon Achievement

This project was developed during the 48-hour Google Nano Banana Hackathon hosted by Google DeepMind and Cerebral Valley in San Francisco. The hackathon featured Gemini 2.5 Flash Image Preview (Nano Banana) and had a prize pool exceeding $400,000 with partnerships from Fal AI and ElevenLabs.

##  How It Works

In-Transit transforms lengthy videos into concise, engaging content through an intelligent 3-stage pipeline:

```mermaid
graph TD
    A[ Input Video] --> B[ Transcription Service<br/>ElevenLabs API]
    B --> C[ Original Transcript<br/>.srt file]
    C --> D[ AI Text Processing<br/>Gemini AI]
    D --> E[ Condensed Transcript<br/>25% most important content]
    E --> F[ Video Processing<br/>Nano Banana + Fal AI]
    F --> G[ Final Seamless Video<br/>with AI transitions]
    
    F --> H[ Frame Extraction<br/>Key transition points]
    H --> I[ Nano Banana<br/>Intermediate frame generation]
    I --> J[ Fal AI AMT<br/>Frame interpolation]
    J --> K[ Seamless Transitions<br/>Between segments]
    K --> G
    
    style A fill:#e1f5fe
    style G fill:#c8e6c9
    style I fill:#fff3e0
    style D fill:#f3e5f5
```

##  Features

###  **Seamless Mode** (Default)
- **AI-Powered Transitions**: Uses Google's Nano Banana

Team