Modern automobiles have pervasively deployed audiovisual systems, greatly enriching the users’ experience. Nevertheless, this has placed additional transmission demands on in-vehicle networks. Through a real-world measurement study conducted on a hardware-in-the-loop platform, we observe a low video bitrate with 7.63 Mbps (far away from smooth playing) with frequent resolution switching. We further demystify the root reason lying in the conflicts between bursty traffic delivery from the application-layer and sluggish resource response at the MAC-layer. Briefly, the demand for rapid data packet delivery is not adequately addressed by the MAC-layer. In this paper, we propose T-reSoNator, a middleware located in the media server for cross-layer adaptation. It fuses the Transformer-based learning model and Fast Fourier Transform method to smooth the aggressive video streaming over MAC-layer resources. Through massive evaluation, we conclude that T-reSoNatorachieves a 14.93 Mbps average video bitrate improvement with zero resolution switching, which is about 2 \(\times \) to the state-of-the-art methods.