Multi-view clustering (MVC) remains challenging in preserving both view-specific information and cross-view sample relationships. We propose CVC-AFWM, a novel approach that addresses these limitations through two key innovations: an adaptive feature weighting mechanism that dynamically enhances discriminative features while suppressing noise, and a cross-view consistency strategy that aligns local feature similarities with high-confidence pseudo-labels to maintain semantic coherence. Extensive experiments demonstrate CVC-AFWM’s superior performance across multiple benchmark datasets, effectively balancing view diversity and consistency to achieve state-of-the-art clustering results.