Semantic communication (SemCom) powered by generative artificial intelligence enables highly efficient and reliable information transmission. However, it still necessitates the transmission of substantial amounts of data when dealing with complex scene information. In contrast, the stacked intelligent metasurface (SIM), leveraging wave-domain computing, provides a cost-effective solution for directly imaging complex scenes. Building on this concept, we propose an innovative SIM-aided multi-modal SemCom system. Specifically, an SIM is positioned in front of the transmit antenna for transmitting visual semantic information of complex scenes via imaging on the uniform planar array at the receiver. Furthermore, the simple scene description that contains textual semantic information is transmitted via amplitude-phase modulation over electromagnetic waves. To simultaneously transmit multi-modal information, we optimize the amplitude and phase of meta-atoms in the SIM using a customized gradient descent algorithm. The optimization aims to gradually minimize the mean squared error between the normalized energy distribution on the receiver array and the desired pattern corresponding to the visual semantic information. By combining the textual and visual semantic information, a conditional generative adversarial network is used to recover the complex scene accurately. Extensive numerical results verify the effectiveness of the proposed multi-modal SemCom system in reducing bandwidth overhead as well as the capability of the SIM for imaging the complex scene.