Optimizing Blender’S Denoiser Settings For Your Specific Scene
What is Denoising and Why it’s Useful
Denoising refers to algorithms and techniques used to remove random noise from rendered images while preserving important scene details. Noise is inherent in photorealistic rendering with techniques like path tracing due to the stochastic nature of light transport simulations. Denoising removes this noise to produce clean final renders.
Using denoising can speed up render times by allowing more noisy renders from lower sample counts. Denoising the noisier images produces final images free of random noise in much less time than simply increasing samples to reduce noise. This makes denoising incredibly useful for optimizing render times.
Denoiser Settings Overview
The Blender denoiser includes several settings across two main categories – sampling and filtering settings – that control different aspects of the denoising process. Optimizing these settings is key to maximizing quality and efficiency for a given scene.
Sampling
Sample Count
The sample count determines how much Monte Carlo noise will be present in renders before denoising. Higher samples means less inherent noise but longer render times. Finding the optimal balance reduces noise while keeping renders fast.
Prefilter
Prefilter controls neighborhood scaling when transforming samples into a feature vector used by the denoiser. Higher values smooth out noise more but can cause detail loss.
Normal
The normal setting determines how much surface normals are taken into account during denoising instead of just color. Higher values preserve edges better but are slower to compute.
Variance
Variance controls how much sample variance at each pixel is included in the feature vector. This helps discriminate between noise and detail. Higher values improve edge discrimination but slow down computation.
Filtering
Radius
Filter radius controls the size of the window each pixel looks at to filter out noise. Larger radii work better for smoothing out noise but can cause some loss of fine detail.
Strength
Strength determines how much filtering is applied once features are extracted. Higher strength removes more noise but starts to degrade edges, texture details, and small features.
Optimizing for Different Scenes
Balancing tradeoffs between quality and speed with the denoiser ultimately depends greatly on the type of scene and specific noise behavior. What works for a still life or product render may not translate to landscapes or characters.
Indoor Scenes
Interior renders involve many complex lighting interactions with glossy and diffuse surfaces. Indirect bounces display higher amounts of raw noise compared to direct illumination.
Increasing Sample Count
Using a moderately high sample count reduces dark splotches from indirect bounces while keeping render times reasonable compared to extremely high values.
Lowering Filter Radius
A small filter radius avoids blurring intricate details on objects like furnishings while sufficiently removing noise in flat, continuous surfaces like walls.
Outdoor Scenes
Exteriors involve direct sun lighting, broad area lighting sources like skies, and complex global illumination across terrain and vegetation. This can introduce banding artifacts and graininess.
Increasing Normal/Variance
Higher normal and variance samples help distinguish between noise and hard shadows or light/dark foliage edges to prevent detail loss and gradient banding issues.
Raising Filter Strength
Moderate filter strength levels smooth out grains in broad gradient areas representing terrain and skies in large exterior spaces better than smaller radii.
Characters
Organic models involve many curves, changing depths, fine featured surface textures, and thin protruding elements. High frequency texture details and depth complexity introduce difficult to remove noise.
Higher Prefilter
Using higher prefilter settings better retains difficult geometrical texture details on skin, hair, and clothing materials while smoothing noise in flatter regions.
Moderate Filter Radius
The radius shouldn’t over-blur high-resolution normal, displacement, and procedurally generated textures but still sufficiently filter noise on large surface expanses.
Putting it All Together
Optimizing denoising ultimately takes trial and error testing. However, these general guidelines improve baseline settings.
Start with Defaults
Begin by assessing noise levels and render times at default sampling and filter settings to understand standard behavior for a scene.
Tweak Based on Noise and Render Times
First adjust sampling settings targeting remaining excessive noise areas until render times become impractical. Then modify filter settings if additional smoothing is still required in areas.
Find Optimal Balance for Your Needs
Compare settings combinations empirically to select options providing the best compromise between noise reduction, detail preservation, and render speed for final frames based on strict quality requirements or looming deadlines.
