EndoSparse is a framework for real-time sparse view synthesis of endoscopic scenes using Gaussian Splatting. The method leverages prior knowledge from multiple foundation models during the reconstruction process to improve geometric and appearance quality under sparse-view conditions. Experimental results show that EndoSparse achieves superior results in terms of accurate geometry, realistic appearance, and rendering efficiency compared to state-of-the-art methods. The framework incorporates geometric prior knowledge from Depth-Anything and image appearance priors from Stable Diffusion to guide the optimization of the 3D Gaussian Splatting process. EndoSparse is designed to handle sparse observations, making it suitable for real-world clinical scenarios where only a limited number of views are available. The method uses diffusion priors to regularize the synthesized results and ensures that the rendered depth maps are consistent with depth estimation predictions. The overall training objective combines multiple loss terms, including RGB reconstruction, diffusion prior, and geometric prior, to optimize the 3D Gaussian Splatting model. EndoSparse demonstrates significant improvements in geometric and visual accuracy, even with sparse observations, and is capable of real-time rendering. The framework is robust to sparse-view limitations and has the potential for practical deployment in clinical settings.EndoSparse is a framework for real-time sparse view synthesis of endoscopic scenes using Gaussian Splatting. The method leverages prior knowledge from multiple foundation models during the reconstruction process to improve geometric and appearance quality under sparse-view conditions. Experimental results show that EndoSparse achieves superior results in terms of accurate geometry, realistic appearance, and rendering efficiency compared to state-of-the-art methods. The framework incorporates geometric prior knowledge from Depth-Anything and image appearance priors from Stable Diffusion to guide the optimization of the 3D Gaussian Splatting process. EndoSparse is designed to handle sparse observations, making it suitable for real-world clinical scenarios where only a limited number of views are available. The method uses diffusion priors to regularize the synthesized results and ensures that the rendered depth maps are consistent with depth estimation predictions. The overall training objective combines multiple loss terms, including RGB reconstruction, diffusion prior, and geometric prior, to optimize the 3D Gaussian Splatting model. EndoSparse demonstrates significant improvements in geometric and visual accuracy, even with sparse observations, and is capable of real-time rendering. The framework is robust to sparse-view limitations and has the potential for practical deployment in clinical settings.