The model is built on the NEO-unify architecture and improves patchify layers, data quality, prompt enhancement, task formulation, and post-training. It supports native 4K generation, local and multi-reference editing, object and region control, and complex instructions involving counts, spatial relationships, styles, and layouts.
Released with public weights and an Apache 2.0 license, SenseNova-U1.5 can be run through its reference inference repository or tried in SenseNova Studio. It is useful for posters, infographics, brand assets, creative image generation, and controlled image editing.
