🤖 AutoShortAi

Paste a video link. Get edited, HD-enhanced, ready-to-post vertical Shorts — cut, upscaled, voiced-over, and exported, all locally in your browser.

A local YouTube Shorts / Reels automation studio: powered by yt-dlp for fetching, ffmpeg for cutting and color, and AI face-restoration/upscaling (GFPGAN + Real-ESRGAN) for HD output — one Python script, no cloud, no subscription.

Features • Quick Start • How It Works • Roadmap • Contributing

---

Why AutoShortAi?

Most "video to Shorts" tools either lock you into a paid SaaS, upload your footage to someone else's server, or hand you one flat auto-cut file with zero control. AutoShortAi is different:

• 100% local — nothing you process leaves your machine. No account, no cloud storage, no subscription.
• One script, full pipeline — fetch → cut → live-edit → AI-enhance → export, all in a single Flask app with a modern browser UI (no build step).
• Real editing, not just cutting — per-clip speed, zoom, color grade, captions, logo, hide-regions, and AI voiceover before anything is written to disk.
• AI HD Enhance — optional GFPGAN face-restoration + Real-ESRGAN upscaling pass so clips stay sharp instead of pixelating when zoomed or upscaled.
• Built on yt-dlp — the same engine that powers most modern video-archival tools, so it isn't limited to a single site by design.

✨ Features

• 🔗 Paste a video link → auto-resolves the best available video + audio quality
• 🧠 Smart bot-detection bypass — tries cookies, browser sessions, and fallback clients automatically so fetches don't get blocked
• ✂️ Auto mode: splits the full video into fixed-length vertical (9:16) shorts
• 🎯 Manual mode: pick your own start/end ranges
• ⚡ Parallel clip cutting — cuts don't wait on each other
• 🖊️ Per-clip live editing: speed, zoom, mute/replace audio, text overlays, logo, blur/black/emoji hide-regions, rectangle/circle/arrow shapes
• 🎨 One-click color presets (Vivid, Cinematic, Moody, Vintage VHS, Warm Glow, Cool Blue)
• 🪄 AI HD Enhance — GFPGAN face restoration + Real-ESRGAN upscaling, denoise/detail pass (weights auto-download on first use)
• 🗣️ AI voiceover: Edge TTS (Hindi/English, multiple voices) + gTTS fallback
• 🎵 Drop-in background music library (audio/ folder)
• 💾 Smart export — stream-copies unedited clips instead of re-encoding for near-instant Save
• 📥 Batch download manager for all exported clips

🚀 Quick Start

git clone https://github.com/jastfan/AutoShortAi.git
cd AutoShortAi
pip install -r requirements.txt
python autojob.py

The app opens automatically at http://127.0.0.1:5789.

Tip: drop .mp3 files into audio/ before launching to use them as background music inside the editor. Note on AI Enhance: the first time you enable HD Enhance, it auto-downloads the GFPGAN and Real-ESRGAN model weights (one-time, a few hundred MB). This needs a working internet connection once; after that it runs fully offline.

Requirements

• Python 3.9+
• ffmpeg (bundled via imageio-ffmpeg, no separate install needed)
• A modern browser (Chrome/Edge/Firefox)
• ~2GB free disk space if you plan to use AI HD Enhance (model weights)

🛠 How It Works

1. Resolve — yt-dlp locates the best video/audio stream for the URL, with automatic cookie/session fallbacks if the first attempt is blocked.
2. Cut — the video is split into short segments in parallel using fast preview-quality encodes.
3. Edit — each clip opens in the browser UI for live-preview edits (nothing is re-encoded yet).
4. Enhance (optional) — GFPGAN + Real-ESRGAN run a face-restoration/upscale pass on the clip.
5. Export — on Save, edits are baked in; unedited clips are stream-copied for a near-instant export.

🌍 Beyond YouTube

Under the hood, AutoShortAi runs on yt-dlp, which supports 1,800+ sites — so pasting a link from Instagram, Facebook, or many other platforms will often work out of the box. The bot-detection bypass and cookie-fallback logic in this build are currently tuned specifically for YouTube; other platforms aren't yet tested to the same degree, and cookie handling for platform-locked content may need extra setup. Wider, tested multi-platform support is an active roadmap item — see below.

🗺 Roadmap

• Tested, documented Instagram + Facebook fetch flows
• Local file upload (edit your own PC videos without a URL)
• video2ppt-style module — extract slides/key-frames from lecture or webinar videos
• One-click auto-captions (Whisper-based)
• Docker image for one-command setup
• Multi-language UI

Have an idea? Open an issue or check good first issue to contribute.

🤝 Contributing

Contributions are welcome — see CONTRIBUTING.md. Bug reports, feature ideas, and PRs all help.

📄 License

MIT — see LICENSE.

---

If this saved you time, a ⭐ on the repo helps others find it.