Style transfer methods usually use pre-trained VGG or more complex models as encoders. This results in significantly slow processing of high-resolution images. To address this issue, we present a degree-controllable detail attention-enhanced lightweight fast style transfer (DcDae) framework, which adopts a small, shallow, and compact architecture for efficient forward inference. Moreover, we utilize global semantic invariance loss to retain the semantic information of content images, along with a detail attention-enhanced module to preserve their intricate details, and style discriminators to improve color and texture matching. Importantly, this is the first method that allows for modulation of detail preservation and style transfer based on subjective evaluation. Compared to state-of-the-art models, our DcDae not only yields visually superior results but is also 17 to 250 times smaller and 0.26 to 6.5 times faster, achieving a maximum processing speed of 0.38 s for 4K high-resolution images.