I need to build an AI-powered video creator that leans heavily on computer-vision techniques. The goal is a tool that takes raw clips or even still images and, through smart scene understanding, automatically assembles polished videos—essentially an AI video maker that handles the tedious parts of editing and enhancement for me. Core objectives • Train or adapt a computer-vision model that can detect key moments, faces, objects, and scene changes, then use that information to sequence footage intelligently. • Implement automated video generation logic—transitions, basic trimming, and soundtrack alignment—so the user can go from assets to finished output with minimal clicks. • Include real-time or near-real-time preview plus the option for manual tweaks (simple timeline edits, text overlays, color adjustments). • Deliver clean, well-commented code—Python preferred—with clear instructions for installing dependencies (OpenCV, PyTorch/TensorFlow, FFmpeg, or any libraries you choose). • Provide a short demo video and sample input–output files showing the system working end-to-end. Acceptance criteria 1. Feed the system a folder of mixed MP4 clips and JPG/PNG images; receive a coherent, 1080p final video without crashes. 2. Key events (e.g., faces speaking) appear in the final cut ≥90 % of the time in the demo set. 3. All external model weights and training notebooks are included so I can retrain or fine-tune later. 4. Setup works on Windows or Linux in under 30 minutes following your README. Once these milestones are met, we can discuss further enhancements such as advanced color grading, subtitle generation, or social-media-ready aspect-ratio exports.