Spatial transcriptomics (ST) has transformed our understanding of cellular interactions and tissue architecture, but integrating ST data across technologies remains challenging due to differences in gene panels, data sparsity, and technical variability. We introduce Lloki, a novel framework for integrating imaging-based ST data from diverse technologies without requiring shared gene panels. Lloki decomposes ST integration into two distinct alignment tasks: feature alignment to address gene panel differences and batch integration to reduce technical variability. For feature alignment, we match ST gene expression sparsity to a single-cell reference by propagating features between spatially proximal cells with similar expression. This modified gene expression is then processed by a single-cell foundation model to obtain unified features. For batch alignment, we integrate feature-aligned data using a conditional autoencoder trained with a novel loss function that preserves biological variation while aligning batches. Evaluations across technologies demonstrate that Lloki outperforms existing batch integration methods and significantly enhances downstream tasks, including physical slice alignment and spatial gene program identification.