This paper proposes a lightweight information multi-distillation network (IMDN) for single image super-resolution (SISR). The IMDN is designed to achieve high performance with reduced computational cost and memory usage. The key components of the IMDN include a cascaded information multi-distillation block (IMDB), which extracts hierarchical features step-by-step and aggregates them using a contrast-aware channel attention mechanism. The IMDB is designed to retain partial information and further process other features at each step, enhancing the collected refined information. To handle SR of any arbitrary scale factor, an adaptive cropping strategy (ACS) is introduced, allowing the network to process images of any size with a single model. The ACS enables the network to scale the input image to the target size and then obtain image patches of appropriate size for lightweight SR model with downsampling layers. The proposed IMDN achieves competitive results with a modest number of parameters, demonstrating superior performance in terms of visual quality, memory footprint, and inference time compared to state-of-the-art SR algorithms. The IMDN is evaluated on several benchmark datasets, including Set5, Set14, BSD100, Urban100, and Manga109, and shows favorable performance across different scale factors (×2, ×3, ×4). The results indicate that the IMDN achieves a good balance between performance and computational efficiency, making it suitable for resource-constrained devices. The paper also explores the factors affecting the inference time, finding that the depth of the network is related to the execution speed. The proposed IMDN is a lightweight and efficient model that can be applied to various image restoration tasks.This paper proposes a lightweight information multi-distillation network (IMDN) for single image super-resolution (SISR). The IMDN is designed to achieve high performance with reduced computational cost and memory usage. The key components of the IMDN include a cascaded information multi-distillation block (IMDB), which extracts hierarchical features step-by-step and aggregates them using a contrast-aware channel attention mechanism. The IMDB is designed to retain partial information and further process other features at each step, enhancing the collected refined information. To handle SR of any arbitrary scale factor, an adaptive cropping strategy (ACS) is introduced, allowing the network to process images of any size with a single model. The ACS enables the network to scale the input image to the target size and then obtain image patches of appropriate size for lightweight SR model with downsampling layers. The proposed IMDN achieves competitive results with a modest number of parameters, demonstrating superior performance in terms of visual quality, memory footprint, and inference time compared to state-of-the-art SR algorithms. The IMDN is evaluated on several benchmark datasets, including Set5, Set14, BSD100, Urban100, and Manga109, and shows favorable performance across different scale factors (×2, ×3, ×4). The results indicate that the IMDN achieves a good balance between performance and computational efficiency, making it suitable for resource-constrained devices. The paper also explores the factors affecting the inference time, finding that the depth of the network is related to the execution speed. The proposed IMDN is a lightweight and efficient model that can be applied to various image restoration tasks.