Path tracing has been successfully utilized in modern animated films to produce photorealistic images. However, with path tracing being a Monte Carlo method, a large amount of light paths must be sampled to reduce image noise to acceptable levels. Although a wide variety of acceleration and importance sampling techniques have been developed to reduce the cost of individual samples and reduce the number of required samples, path tracing remains a very computationally expensive approach. Monte Carlo error estimation and reconstruction approaches the problem from a signal processing perspective by reducing the sampling budget and instead applying image denoising filters to lower noise levels. Although these image denoising filters can be successfully applied to post-process images into near ground truth results, estimating the right parameters to use remains an open problem. We propose that limitations of automatic parameter estimation can be overcome by taking advantage of the ground truth image in the mind of artists. To that end we introduce a workflow and associated user interface to let artists locally tune and apply filters to noisy Monte Carlo images. The interface lets users explore the high dimensional parameter spaces of filters through an image navigation paradigm, which means that users pick the right filter and settings by choosing the end result they want. We organize a case study to evaluate the efficacy of interactive filter tuning and to compare the efficiency of slider-based parameter choice with the image navigation paradigm. With the results from the case study we show that our workflow allows even novice users with no prior image processing experience to quickly select appropriate filters. Users are able to produce higher quality results using our local parameter painting approach compared to global optimization in the same time span. Users also prefer the image navigation interface for filter choice and tuning to sliders, but we are only able to quantify the benefits on images that require filters with a high dimensional parameter space.