Camouflaged object detection (COD) plays a crucial role in computer vision, and its core is to accurately identify objects that highly resemble their background in texture and color. This task has caught extensive attention due to its broad application prospects. In recent years, diffusion models have achieved outstanding performance in image generation and denoising, but their application in COD still faces many challenges. Existing COD methods based on diffusion models mainly rely on static spatial-domain conditional features to guide the denoising process. However, this kind of method results in obvious deficiencies in the global feature modeling capacity of conditional features and fails to effectively perceive variations across diffusion time-steps. To this end, this study proposes a brand-new method called the time-frequency enhanced diffusion model for COD (TFD-COD). Specifically, the time embedding mechanism is introduced into the backbone network, enabling the sensitive perception of subtle variations of diffusion time-steps during the conditional feature extraction process. Additionally, a frequency-spatial dual branch attention (FSBA) module is designed, where the frequency branch accurately captures global structural information via Fourier transform and self-attention, while the spatial branch models local context via multi-scale convolutions and self-attention. The two branches collaborate for synergistic enhancement of global and local features. Furthermore, TFD-COD introduces a pyramid filtering fusion (PFF) strategy, which employs shallow features to guide the upsampling of deep features to achieve the deep fusion of multi-level features. Finally, these carefully designed conditional features are fed into the diffusion model to guide efficient denoising. Experimental results show that TFD-COD outperforms existing methods significantly in COD tasks.