AutoShorts: Local-First AI Short-Form Video Clipper & Generator ⚡
How to extract viral vertical short clips (9:16) from podcasts, streams, and YouTube videos on your desktop using DeepSeek, Claude, and local FFmpeg.
AutoShorts replaces expensive cloud video clipping SaaS subscriptions with a local-first desktop application. Powered by Tauri 2 + Rust + React + SQLite, it ingests video files, extracts audio, generates transcriptions with word-level timestamps, and leverages DeepSeek or Claude to score retention moments, outputting cropped 9:16 portrait MP4s with zero cloud rendering costs.
The best short-form clips meet three criteria: 1) Strong contrarian opening hook in the first 2.5 seconds, 2) High information velocity with zero filler words, and 3) An unresolved open loop or actionable takeaway before the loop restart.
Key Architecture & Capabilities
- Multi-LLM Ranking: Seamlessly switch between DeepSeek V3 (ultra-low cost cloud analysis) and Claude 3.5 Sonnet (premium narrative hooks).
- 100% Offline Mode: Run fully air-gapped using local Ollama (Llama 3.2 3B / Qwen 2.5) and local OpenAI Whisper.
- Native FFmpeg Portrait Auto-Cropping: Converts 16:9 widescreen video into 9:16 vertical clips with speaker centering.
- Local SQLite Database: Stores candidate timestamps, video metadata, and transcripts on your machine without cloud leaks.
1-Click Viral Podcast Moment Detector Prompt
Paste this prompt into Claude or DeepSeek alongside a timestamped transcript to extract top-ranked 9:16 clip candidates:
Act as an expert Short-Form Video Producer and Viral Content Strategist. Analyze the following timestamped transcript and identify the top 5 highest-retention vertical clip candidates (duration 30 to 60 seconds each): For each candidate, provide: 1. Exact Start & End Timestamps [MM:SS - MM:SS] 2. 3-Second Viral Hook Headline (The exact on-screen caption hook) 3. Viral Score (1-100) based on emotional resonance, contrarian insight, or practical utility 4. Visual Pacing Note (Where to insert B-roll, zooms, or kinetic text emphasis) 5. Outro Hook (Why the viewer will share, save, or loop the video) Here is the transcript: [PASTE TRANSCRIPT HERE]
Local-first AI viral short video clipper built with Tauri 2 + Rust.
Rust toolchain, Node.js 20+, FFmpeg, Ollama (optional for 100% offline)# 1. Clone the autoshorts repository
git clone https://github.com/JayWebtech/autoshorts.git
cd autoshorts
# 2. Install frontend dependencies
npm install
# 3. Launch desktop app in development mode
npm run tauri dev
# 4. Or build production release binary
npm run tauri build