Proteins, the building blocks of life, perform their biological roles by changing shape in response to various signals. These changes, known as conformational changes, are crucial for processes like substance synthesis and transport. However, predicting these dynamic changes has remained a challenge, even for advanced artificial intelligence systems like AlphaFold3. While AlphaFold3 has revolutionized the field by accurately predicting the static, folded structure of proteins from their amino acid sequences, it typically generates only a single conformation per protein. This limitation restricts its use in understanding how proteins change shape during function, which is vital for applications like drug design. AlphaFold3 uses a type of AI called a diffusion generative model, which starts with a random, noisy arrangement of a protein's atoms and gradually reduces the noise to find a stable, low-energy structure. This process is akin to moving atoms down an energy gradient to reach a final, folded state. However, this method tends to settle on the lowest energy conformation, often missing other possible shapes that a protein might adopt. This single-conformation focus limits the AI's ability to model the dynamic nature of proteins, which can switch between multiple shapes to carry out their functions. To address this issue, researchers at the Institute for Molecular Science (IMS) and the Graduate University for Advanced Studies, SOKENDAI, introduced a novel approach. They modified AlphaFold3 by adding a repulsive force between predicted structures. This repulsive force biases the AI away from previously predicted conformations, encouraging it to explore a wider range of possible shapes. This technique, named AF3-ReD, enables the AI to sample multiple conformational states that were previously difficult to predict. In one example, the method successfully predicted both open and closed conformations of the F1β subunit of ATP synthase, a protein involved in energy production. The development of AF3-ReD opens new possibilities for studying protein dynamics and designing drugs. By generating diverse protein conformations quickly and accurately, researchers can better understand how proteins function and evolve over time. Additionally, since diffusion generative models are also used in designing new proteins and drugs, applying the repulsive bias approach could lead to more diverse and effective designs in the future. This advancement marks an important step toward leveraging AI to fully capture the dynamic complexity of proteins.