A generative infrared and visible image fusion method inspired by human cognitive laws combines a multi-scale variational bottleneck encoder and a diffusion model guided by physical laws to achieve superior structural consistency and detail quality.