back to portfolio ⚡ AutoRecap-MM Studio
🎬 AI Video Engine Python + FastAPI GPU Accelerated Myanmar Language

AutoRecap-MM

AI Myanmar Movie Recap & Dubbing Engine

A fully-automated video-to-Myanmar-narration pipeline — from YouTube URL to published recap in minutes. Powered by Faster-Whisper transcription, Gemini story understanding, Edge-TTS voice synthesis, and FFmpeg GPU rendering.

❯ initializing autorecap-mm-studio...
✦ Loading Faster-Whisper (GPU T4 / CPU int8)
✦ Connecting Gemini API key rotation pool
✦ Initializing Edge-TTS: ThihaNeural · NilarNeural
✦ FFmpeg NVENC encoder ready
✦ CLIP vision model loaded for viral detection
✓ Studio ready. Paste your YouTube URL to begin.
❯

7-Step Processing Pipeline

From raw video to broadcast-ready Myanmar recap — fully automated, GPU-accelerated end-to-end.

01
📥
Ingest
Accept a YouTube URL or direct video upload. yt-dlp fetches the highest available quality stream with metadata extraction — title, duration, thumbnails, and chapter markers.
yt-dlp YouTube URL Direct Upload
02
🎙️
Transcribe
Faster-Whisper runs on GPU T4 (or CPU int8 fallback) with Voice Activity Detection silence stripping. Outputs word-level timestamps and language confidence scores.
Faster-Whisper GPU T4 CPU int8 VAD Silence
03
🧠
Understand
Gemini Pro reads the full transcript — tracking characters, plot arcs, dialogue tone, and scene structure to build an internal story model before writing a single word of narration.
Gemini API Multi-Key Rotation Story Modelling
04
✍️
Translate & Dramatize
Gemini rewrites the transcript into natural, emotionally engaging Myanmar narration — not a literal translation. Sentence length is tuned to match TTS cadence. Pyidaungsu font compatible output.
Gemini API Myanmar Script Pyidaungsu Natural Narration
05
🔊
Voice
Microsoft Edge-TTS generates natural Myanmar speech. ThihaNeural is used for dramatic narration; NilarNeural for softer, feminine voiceover — selectable per project.
Edge-TTS ThihaNeural NilarNeural Myanmar Voice
06
⚡
Smart Sync
Video is retimed frame-accurately to match Myanmar voiceover duration. Scene cuts are preserved; long pauses are trimmed. GPU NVENC / QSV / AMF encoding for real-time output.
FFmpeg NVENC QSV / AMF Frame Sync
07
📤
Export
Myanmar subtitles burned directly onto the final render. CLIP Vision AI selects the most viral-potential moments for 9:16 Shorts clips. Fair-use transforms applied before delivery.
CLIP Vision AI 16:9 / 9:16 / 1:1 Subtitles Burned Fair-Use
Output Formats
16:9 Recap
9:16 Shorts
1:1 Square
Voice Models
ThihaNeural
NilarNeural
Edge-TTS
GPU Encoders
NVENC
QSV / AMF
CPU libx264
Runs On
Google Colab
Render / VPS
Android APK

3 Deployment Modes

Run anywhere — from a free Colab notebook to a self-hosted server or an Android phone.

📒
Google Colab
Free GPU access via Colab T4 runtime. Open the notebook, paste a YouTube link, and get a full Myanmar recap in under 5 minutes — no local install required.
Free GPU T4 Runtime Zero Setup
🌐
Self-Hosted Web Server
FastAPI server deployed on Render (or any VPS). Full REST API with queue management, webhook callbacks, and a clean web UI. GPU or CPU mode selectable at startup.
FastAPI Render REST API Queue Mgmt
📱
Android APK
WebView shell APK that wraps the hosted web UI into a native Android app. Distributable via APK sideloading — no Play Store needed. Runs on any Android 8.0+ device.
Android 8.0+ WebView Shell APK Sideload

Built With

A Python-first pipeline optimised for GPU throughput with graceful CPU fallbacks.

🐍 Python 3.11 ⚡ FastAPI server.py 🎙️ Faster-Whisper 🤖 Gemini API 🔊 Edge-TTS 🎬 FFmpeg 🖼 CLIP Vision AI yt-dlp GPU NVENC CPU int8 🇲🇲 Pyidaungsu Font 📒 Google Colab 📱 Android WebView