Traditional control methods such as PID controllers have been widely used for quad-copter control. However, these approaches often struggle to maintain stability and adapt to variable environmental conditions or vehicle dynamics. Additionally, they may not fully exploit the capabilities of modern drone hardware. Therefore, there is a pressing need for innovative control algorithms capable of overcoming these limitations. Deep Reinforcement Learning (DRL) has the potential to enhance the stability and maneuverability of drones, particularly in dynamic environments or adverse weather conditions. However, DRL requires hundreds of hours of training in simulation environments like NVIDIA Isaac Sim. In order to increase the sample efficiency of DRL, Dreaming Falcon presents a solution using Model-Based Reinforcement Learning inspired by the Dreamerv3 paper. Our algorithm consists of 2 major components: a world model which will dynamically learn vehicle and environmental conditions, and a behavior model which will find the optimal policy based on defined reward function. A data buffer is implemented to train the world model in real-time using batch-processing. This approach can learn flight dynamics while also optimizing control actions simultaneously. Preliminary results show that our algorithm can learn a world model from data generated by a 3-DOF simulation. Our work will allow EVTOLs to learn and adapt to changes in the vehicle and adverse environments.