This project implements a text-conditioned U-Net that segments drywall defects and features using natural language prompts. Given an image and a prompt like "segment crack" or "segment taping area", the model produces pixel-level binary masks for construction quality assurance.