Automated defect detection for wind turbine blades remains challenging under uncontrolled outdoor illumination, where backlight and low-light condi tions suppress texture cues, reduce local contrast, and increase false detections. We address this problem from both data and model perspectives. First, we construct a Complex-Lighting Wind Turbine Blade Defect Dataset (CL-WTBD) by integrating heterogeneous UAV-based inspection sources and harmonizing their annotations into a unified taxonomy. CL-WTBD is balanced across normal light, backlight, and low-light domains, providing a controlled benchmark for evaluating illumination robustness in blade defect detection. Second, we propose UniIA-DET, an illumination-adaptive detection framework built on top of an off-the-shelf object detector. Instead of relying on a standalone image enhance ment module or aligning features only in latent space, UniIA-DET introduces an Illumination-Adaptive Prompting Module (IAPM) before the detector backbone. The IAPM uses content-driven dynamic routing to compose an input-specific correction from a learnable prompt pool, which is additively injected into the raw image to calibrate exposure and contrast before feature extraction. The module is optimized jointly with the detector using the detection objective, allowing the correction process to be guided by classification and localization requirements rather than by perceptual reconstruction criteria. Experiments on CL-WTBD show that the proposed module improves robustness under harsh illumination. On the YOLOv8n instantiation, the main benefit of IAPM appears under harsh illumination: averaged over three seeds, it improves the Backlight mAP@50 from 0.352 to 0.392 and the Low-Light mAP@50 from 0.392 to 0.446, raising the macro-domain mAP@50 from 0.424 to 0.466, while the global mAP@50 changes from 0.387 to 0.398, within the run-to-run variation. Additional experiments with RT-DETR and Mamba-YOLO indicate that IAPM can be attached to different detector families without modifying their original architectures. These results demonstrate the practical value of detector-oriented input-space illumination correction for UAV-based wind turbine blade inspection.1