I need an AI-driven workflow that ingests raw video (MP4/MOV) and automatically follows a chosen object through every frame, delivering a clean alpha-masked output I can drop straight into my edit. The goal is to save me from manual rotoscoping; accuracy and speed are both critical. Here’s how I picture the collaboration: you build or adapt a model—using tools such as PyTorch, TensorFlow, Detectron2, or a proven OpenCV pipeline—that detects the target, keeps the mask tight even during fast motion, and exports either a keyed video or a PNG sequence. A small sample clip will be provided for you to demonstrate proof of concept; once results look solid we’ll run the system on the full batch. Deliverables • Python or command-line script (with all dependencies noted) • Trained weights or clear instructions for training on new footage • One-click inference guide (README or short screencast) • Final masked sample showing accurate object tracking from start to finish Acceptance will be based on: 1. Mask edge quality—minimal bleed or chatter 2. Consistent tracking with no frame drops 3. Runtime compatible with a mid-range GPU (RTX 3060) If you have previous examples of AI powered roto or object tracking, I’d love to see them. Let’s make tedious masking a thing of the past.