Cryogenic carbon capture (CCC) can preferentially desublimate $\text {CO}_2$ out of the flue gas. A widespread application of CCC requires a comprehensive understanding of $\text {CO}_2$ desublimation properties. This is, however, highly challenging due to the multiphysics behind it. This study proposes a lattice Boltzmann (LB) model to study $\text {CO}_2$ desublimation on a cooled cylinder surface during CCC. In two-dimensional (2-D) simulations, various $\text {CO}_2$ desublimation and capture behaviours are produced in response to different operation conditions, namely, gas velocity (Péclet number $\textit {Pe}$) and cylinder temperature (subcooling degree $\Delta T_{sub}$). As $\textit {Pe}$ increases or $\Delta T_{sub}$ decreases, the desublimation rate gradually becomes insufficient compared with the $\text {CO}_2$ supply via convection/diffusion. Correspondingly, the desublimated solid $\text {CO}_2$ layer (SCL) transforms from a loose (i.e. cluster-like, dendritic or incomplete) structure to a dense one. Four desublimation regimes are thus classified as diffusion-controlled, joint-controlled, convection-controlled and desublimation-controlled regimes. The joint-controlled regime shows quantitatively a desirable $\text {CO}_2$ capture performance: fast desublimation rate, high capture capacity, and full cylinder utilization. Regime distributions are summarized on a $\textit {Pe}$–$\Delta T_{sub}$ space to determine operation parameters for the joint-controlled regime. Moreover, three-dimensional simulations demonstrate four similar desublimation regimes, verifying the reliability of 2-D results. Under regimes with loose SCLs, however, the desublimation process shows an improved $\text {CO}_2$ capture performance in three dimensions. This is attributed to the enhanced availability of gas–solid interface and flow paths. This work develops a reliable LB model to study $\text {CO}_2$ desublimation, which can facilitate applications of CCC for mitigating climate change.