AI Video Studio

A local, GPU-accelerated AI video creation studio. Generate talking-head videos, voiceovers, images, music, and full ad campaigns — all running on your own hardware with no cloud APIs.

• Agentic Mode — Describe an ad and the system auto-generates a script, character image, voiceover, background music, and lip-synced video
• Image Generation — Text-to-image via Stable Diffusion XL (1024x1024, customizable)
• 4K Upscaling — Upscale any image to 4K using Real-ESRGAN
• Text-to-Speech — Natural voiceovers with Edge TTS or CoquiTTS
• Voice Cloning — Clone any voice from a sample WAV file
• Music Generation — AI-generated background music from text prompts
• Script Generation — Auto-generate ad scripts, stories, and more via local LLM (Ollama)
• Lip Sync — Sync a face image to audio using SadTalker
• Image Animation — Animate still images into short video clips
• Manual Compose — Step-by-step pipeline: character → voiceover → lip sync
• Timeline — Scene-based timeline for organizing generated content
• Real-time Progress — WebSocket-based progress updates during generation

├── frontend/          React + Vite UI (port 5173)
├── backend/           Node.js/Express API server (port 3001)
├── ai-services/       Python GPU services (port 5000)
│   ├── tts_service.py
│   ├── image_gen_service.py
│   ├── upscale_service.py
│   ├── music_service.py
│   ├── lipsync_service.py
│   ├── animate_service.py
│   ├── script_service.py
│   └── api_server.py
├── outputs/           Generated files (audio, video, images)
├── uploads/           User-uploaded files
├── start.bat          Launch frontend + backend
└── start-ai.bat       Launch Python AI services

• GPU: NVIDIA GPU with 8GB+ VRAM (RTX 3060 or higher recommended)
• CUDA: CUDA toolkit installed
• Node.js: v18+
• Python: 3.10+
• Ollama: Running locally for script generation

cd backend
npm install

cd frontend
npm install

cd ai-services
python -m venv venv
venv\Scripts\activate
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
pip install diffusers transformers accelerate safetensors
pip install TTS edge-tts
pip install realesrgan
pip install audiocraft
pip install sadtalker

Models are downloaded automatically on first use (~5-10 GB total).

1. Run start-ai.bat — starts the Python AI server (loads models into GPU)
2. Run start.bat — starts the backend API + frontend UI
3. Open http://localhost:5173 in your browser

1. Select the Agentic tab in the sidebar
2. Describe your ad (e.g., "A tech startup ad for a productivity app")
3. Set product name, tone, aspect ratio
4. Click Create — the system handles everything:
   • Generates a script
   • Creates a character image
   • Produces a voiceover
   • Generates background music
   • Lip-syncs the character to the audio

All endpoints available at http://localhost:3001:

Built by Colin McDonough — LinkedIn · GitHub