Inspiration : Every creator struggles with repurposing long-form videos into short-form clips (Shorts/TikToks/Reels) because manual trimming, captioning, and hook-finding takes hours. This solves an immediate, expensive pain point.
What it does : A Python script/web app where a creator drops a YouTube video URL. The tool:
Downloads the transcript and analyzes it using an LLM to find the top 3 most engaging "viral" hooks. Clips those segments from the source video using ffmpeg. Auto-generates burning stylized captions (OpenAI Whisper) and crops the video to vertical 9:16. Automatically drafts optimized titles, descriptions, and tags for YouTube Shorts.
How we built it : To score high on Technical Execution and Functionality, build a modular script with a clean, simple UI (Streamlit or Gradio is fastest for hackathons).
Tech Stack Recommendation Backend: Python YouTube Transcript / Data API: youtube-transcript-api or YouTube Data API v3 AI/LLM Processing: OpenAI API or Anthropic Claude API (for hook detection & metadata generation) Video Processing: moviepy or ffmpeg-python Transcription: whisper (OpenAI) UI (Bonus points for usability): Streamlit (lets you build a web app in 50 lines of Python)
Challenges we ran into : Day 1 (Core Logic): Get the YouTube URL, fetch the transcript, send it to an LLM with a prompt asking for timestamp ranges of high-retention moments.
Day 2 (Automation/Video): Write the ffmpeg wrapper to slice the video clips based on those timestamps, crop them to 9:16, and overlay auto-generated captions. Day 3 (Polish & Integration): Generate the optimized SEO metadata. Build a quick Streamlit interface so judges can paste a link and see results instantly. Day 4 (Submission Prep): Clean up the GitHub repository, write a killer README, and record a snappy 2-minute demo video.
Log in or sign up for Devpost to join the conversation.