This paper introduces diffusion bridge implicit models (DBIMs), a novel approach for accelerating sampling in denoising diffusion bridge models (DDBMs) without additional training. DDBMs are powerful for interpolating between two arbitrary paired distributions but require computationally intensive sampling. DBIMs generalize DDBMs by defining non-Markovian diffusion bridges on discretized timesteps, resulting in generative processes ranging from stochastic to deterministic. DBIMs are significantly faster than DDBMs, achieving up to 25× speedup, and induce a novel ordinary differential equation (ODE) that inspires high-order numerical solvers. They maintain generation diversity by introducing booting noise in the initial sampling step, enabling faithful encoding, reconstruction, and semantic interpolation in image translation tasks. DBIMs are also connected to the probability flow ODE (PF-ODE) in DDBMs, offering a simpler and more efficient form. Experiments on image translation and restoration tasks show that DBIMs outperform DDBMs in both sample quality and efficiency, achieving state-of-the-art performance on high-resolution datasets. DBIMs are also extended to high-order numerical solvers, improving convergence and generation quality. Despite their advantages, DBIMs still lag behind GAN-based methods in one-step generation and face challenges in real-time applications. The paper also discusses the limitations and failure cases of DBIMs, highlighting their effectiveness in challenging inpainting scenarios. Overall, DBIMs offer a promising solution for efficient and high-quality sampling in diffusion models.This paper introduces diffusion bridge implicit models (DBIMs), a novel approach for accelerating sampling in denoising diffusion bridge models (DDBMs) without additional training. DDBMs are powerful for interpolating between two arbitrary paired distributions but require computationally intensive sampling. DBIMs generalize DDBMs by defining non-Markovian diffusion bridges on discretized timesteps, resulting in generative processes ranging from stochastic to deterministic. DBIMs are significantly faster than DDBMs, achieving up to 25× speedup, and induce a novel ordinary differential equation (ODE) that inspires high-order numerical solvers. They maintain generation diversity by introducing booting noise in the initial sampling step, enabling faithful encoding, reconstruction, and semantic interpolation in image translation tasks. DBIMs are also connected to the probability flow ODE (PF-ODE) in DDBMs, offering a simpler and more efficient form. Experiments on image translation and restoration tasks show that DBIMs outperform DDBMs in both sample quality and efficiency, achieving state-of-the-art performance on high-resolution datasets. DBIMs are also extended to high-order numerical solvers, improving convergence and generation quality. Despite their advantages, DBIMs still lag behind GAN-based methods in one-step generation and face challenges in real-time applications. The paper also discusses the limitations and failure cases of DBIMs, highlighting their effectiveness in challenging inpainting scenarios. Overall, DBIMs offer a promising solution for efficient and high-quality sampling in diffusion models.