Blind Face Restoration (BFR) focuses on the intrinsic challenge of transforming low-quality facial images with unknown and varied degradation into high-quality counterparts. To rectify the issue of varying degradation levels may lead to sub-optimal restoration or over-correction, this paper introduces a novel approach, Degradation Level Predictable Face Restoration (DLP-FR), to leverage a fusion of a degradation assessment framework and Stable Diffusion. The core of DLP-FR lies in its two primary components: a degradation probability predictor that quantifies the degradation severity of the input image and a multi-modal prompt-guided Stable Diffusion process to dynamically adapt the restoration efforts based on the predicted degradation level. Degraded datasets derived from the CelebA-Test are specifically crafted for the model to encompass a broad spectrum of degradation severity. Abundant experiments indicate that DLP-FR significantly outperforms existing state-of-the-art methods, allowing us to comprehensively demonstrate DLP-FR’s superior performance in handling various levels of image degradation.