The repository hosts UniPic, a unified multimodal series featuring three distinct modeling paradigms. UniPic-3 achieves state-of-the-art multi-image editing with support for 1-6 input images and 8-step inference via CM and DMD distillation. UniPic-2 leverages diffusion post-training for text-to-image generation and fine-grained image editing. UniPic-1 provides a 1.5B parameter unified autoregressive model for joint visual understanding and generation tasks.