Due to stringent requirements on data rate and reliability, image transmission over wireless channels remains challenging for both classical layered designs and joint source–channel coding (JSCC), particularly under low-latency constraints. By leveraging powerful learned image priors, diffusion-based generative decoders can achieve strong perceptual quality under limited channel budgets. However, they normally have high decoding latency due to iterative stochastic denoising. To overcome this limitation and enable low-latency decoding, we propose a flow-matching (FM)-based generative decoder under a new land-then-transport (LTT) paradigm, which tightly integrates the physical wireless channel into a continuous-time probability flow. We first construct a Gaussian smoothing path for AWGN channels whose noise schedule monotonically indexes the effective noise levels, and derive a closed-form analytical teacher velocity field along this path. A deep neural-network based student vector field is then trained via conditional flow matching (CFM), yielding a deterministic, channel-aware ordinary differential equation (ODE) decoder with complexity linear in the number of ODE steps; at inference time, it only requires an estimate of the effective noise variance to set the ODE initialization time. We further show that Rayleigh fading and MIMO channels can be converted, via linear MMSE equalization and singular-value-domain processing, into AWGN-equivalent channels with calibrated effective starting times (the time t⋆ on the Gaussian path whose noise level matches the effective channel noise). Thus the same probability path and trained velocity field of AWGN decoders can be reused for Rayleigh and MIMO channels without retraining. For a fixed number of complex channel uses per image, experiments on MNIST, Fashion-MNIST, and DIV2K over AWGN, Rayleigh, and MIMO channels demonstrate that the proposed decoder consistently outperforms JPEG2000 +LDPC, DeepJSCC, and diffusion-based baselines, while achieving a favorable perceptual visual quality with as few as a small number of ODE steps. The results show that the proposed LTT framework provides a deterministic, physically interpretable, and computation-efficient solution for generative wireless image decoding for various channels.