This paper presents a latency-aware semantic communication framework using pre-trained generative models. The transmitter decomposes the input signal into multiple semantic modalities, including a textual prompt and conditioning signals, and transmits each modality with appropriate coding and communication schemes based on the communication intent. For the textual prompt, a re-transmission scheme is used to ensure reliable transmission, while other modalities use adaptive modulation and coding to adapt to varying wireless channels. A semantic and latency-aware scheme allocates transmission power to different modalities based on their importance, subject to semantic quality constraints. At the receiver, a pre-trained generative model synthesizes a high-fidelity signal using the received multi-stream semantics. Simulation results demonstrate ultra-low-rate, low-latency, and channel-adaptive semantic communications. The proposed framework is suitable for applications requiring communication of large multi-modal data with stringent latency and reliability requirements, such as the wireless metaverse, extended/mixed reality, and holographic teleportation.This paper presents a latency-aware semantic communication framework using pre-trained generative models. The transmitter decomposes the input signal into multiple semantic modalities, including a textual prompt and conditioning signals, and transmits each modality with appropriate coding and communication schemes based on the communication intent. For the textual prompt, a re-transmission scheme is used to ensure reliable transmission, while other modalities use adaptive modulation and coding to adapt to varying wireless channels. A semantic and latency-aware scheme allocates transmission power to different modalities based on their importance, subject to semantic quality constraints. At the receiver, a pre-trained generative model synthesizes a high-fidelity signal using the received multi-stream semantics. Simulation results demonstrate ultra-low-rate, low-latency, and channel-adaptive semantic communications. The proposed framework is suitable for applications requiring communication of large multi-modal data with stringent latency and reliability requirements, such as the wireless metaverse, extended/mixed reality, and holographic teleportation.