Recently, diffusion models have been used to solve various inverse problems in an unsupervised manner with appropriate modifications to the sampling process. However, the current solvers, which recursively apply a reverse diffusion step followed by a measurement consistency step, often produce sub-optimal results. By studying the generative sampling path, here we show that current solvers throw the sample path off the data manifold, and hence the error accumulates. To address this, we propose an additional correction term inspired by the manifold constraint, which can be used synergistically with the previous solvers to make the iterations close to the manifold. The proposed manifold constraint is straightforward to implement within a few lines of code, yet boosts the performance by a surprisingly large margin. With extensive experiments, we show that our method is superior to the previous methods both theoretically and empirically, producing promising results in many applications such as image inpainting, colorization, and sparse-view computed tomography.Nevertheless, for certain problems (e.g. inpainting), currently used algorithms often produce unsatisfactory results when implemented naively (e.g. boundary artifacts, as shown in fig. 1 (b)). The authors in [26] showed that in order to produce high quality reconstructions, one needs to iterate back and forth between the noising and the denoising step at least > 10 times per iteration. These iterations are computationally demanding and should be avoided, considering that diffusion models are slow to sample from even without such iterations.Recently, another type of score-based approach, called Noise2Score [23], was proposed for image denoising without clean references. In contrast to the diffusion models, Noise2Score is computationally efficient as it is a deterministic single-step approach that does not involve any stochastic sampling. Unfortunately, the performance of Noise2Score is inferior to the diffusion models that rely Preprint. Under review.