AddSR is a novel method for blind super-resolution based on adversarial diffusion distillation (ADD) that enhances restoration quality and accelerates inference speed. The method incorporates ControlNet and CLIP to receive image and text information for multi-modal restoration. A prediction-based self-refinement (PSR) strategy is introduced to provide high-frequency information for the student model with minimal additional time cost. Additionally, a timestep-adaptive ADD (TA-ADD) is developed to address the perception-distortion imbalance issue. TA-ADD balances GAN loss and distillation loss based on student and teacher timesteps, improving the generative ability of the student model at small inference steps while reducing it at larger steps. Experiments show that AddSR achieves better restoration results and faster speed than previous SD-based state-of-the-art models, with AddSR-4 achieving 7× faster speed than SeeSR. AddSR also demonstrates superior performance on various degradation cases and real-world data, generating high-quality images with fewer inference steps and less time. The method is effective and efficient, making it suitable for real-world applications.AddSR is a novel method for blind super-resolution based on adversarial diffusion distillation (ADD) that enhances restoration quality and accelerates inference speed. The method incorporates ControlNet and CLIP to receive image and text information for multi-modal restoration. A prediction-based self-refinement (PSR) strategy is introduced to provide high-frequency information for the student model with minimal additional time cost. Additionally, a timestep-adaptive ADD (TA-ADD) is developed to address the perception-distortion imbalance issue. TA-ADD balances GAN loss and distillation loss based on student and teacher timesteps, improving the generative ability of the student model at small inference steps while reducing it at larger steps. Experiments show that AddSR achieves better restoration results and faster speed than previous SD-based state-of-the-art models, with AddSR-4 achieving 7× faster speed than SeeSR. AddSR also demonstrates superior performance on various degradation cases and real-world data, generating high-quality images with fewer inference steps and less time. The method is effective and efficient, making it suitable for real-world applications.