The paper introduces the Sub-Path Linear Approximation Model (SPLAM), a novel approach to accelerate diffusion models while maintaining high-quality image generation. SPLAM leverages the consistency property of diffusion models, treating the Probability Flow (PF) ODE trajectory as a series of sub-paths and using Sub-Path Linear (SL) ODEs to estimate errors progressively along each sub-path. This method reduces cumulative errors and enhances image generation quality. The paper also presents an efficient distillation method to incorporate pre-trained latent diffusion models, such as Stable Diffusion. Extensive experiments on datasets like LAION, MS COCO 2014, and MS COCO 2017 demonstrate that SPLAM achieves state-of-the-art performance in few-step generation tasks, with significantly reduced training time compared to existing acceleration methods.The paper introduces the Sub-Path Linear Approximation Model (SPLAM), a novel approach to accelerate diffusion models while maintaining high-quality image generation. SPLAM leverages the consistency property of diffusion models, treating the Probability Flow (PF) ODE trajectory as a series of sub-paths and using Sub-Path Linear (SL) ODEs to estimate errors progressively along each sub-path. This method reduces cumulative errors and enhances image generation quality. The paper also presents an efficient distillation method to incorporate pre-trained latent diffusion models, such as Stable Diffusion. Extensive experiments on datasets like LAION, MS COCO 2014, and MS COCO 2017 demonstrate that SPLAM achieves state-of-the-art performance in few-step generation tasks, with significantly reduced training time compared to existing acceleration methods.