<p>We propose <b>GUIDE</b>, a single-image framework for physically based human material estimation that supports robust and controllable relighting. Single-image human PBR estimation is highly ill-posed due to material–illumination ambiguity and diverse human appearances. Existing approaches either lack effective priors or over-rely on imperfect ones, leading to unstable optimization, biased material estimation, and illumination leakage into albedo. GUIDE addresses these issues by injecting semantics and priors in a controllable and uncertainty-aware manner. We introduce a Dual-Gated DINO Conditioning module that performs feature-wise linear modulation with gated residual injection, enabling stable semantic guidance and preventing over-conditioning for dense prediction. We further propose a Correctable Prior Mixture-of-Experts that refines noisy priors via confidence-gated correction and fuses them through spatially adaptive mixture-of-experts routing, suppressing unreliable regions and reducing prior-induced bias. Extensive experiments on synthetic and real-world datasets show that GUIDE consistently outperforms state-of-the-art methods in both material accuracy and relighting quality.</p>