We propose a novel dual-conditional diffusion model that extracts high-frequency and semantic conditions to guide thermal image SR. It maintains semantic consistency between LR input and HR, facilitating effective feature interaction between different modalities.