To create an autonomous drone agent capable of executing a perfect "flyover and frame" maneuver. The goal is for the drone to identify a specific target feature in a scene (like a QR code or alphanumeric character) and autonomously plan a flight path to capture a perfectly framed image of that target. This open-source project moves beyond traditional computer vision by combining two cutting-edge AI approaches: 1. Vision (Perception): Instead of standard object detection, I utilized a diffusion-inspired architecture for vision encoding. To train this data-hungry model efficiently, I generated vast amounts of synthetic data, creating photorealistic training environments without needing thousands of hours of real-world flight time. 2. Action (Motion): The flight path isn't hard-coded. An agent was trained using Reinforcement Learning (RL), allowing it to learn the optimal flight dynamics via trial and error in simulation to achieve the best possible framing shot. This results in a sophisticated proof-of-concept demonstrating how generative AI architectures and reinforcement learning can converge to solve complex, real-world robotics challenges.