Denoising Techniques For Blender Cycles

Noise in Blender Cycles renders manifests as random speckles across surfaces and backgrounds. It originates from the path tracing algorithm Cycles uses to simulate light behavior. Each ray traced samples only a part of the total lighting information. More samples are required to approximate the full global illumination in a scene.

Insufficient samples lead to variation between pixels, appearing as graininess or noise. Lower sample counts increase noise. Complex lighting interactions requiring more ray bounces also produce noisier renders. The noise is particularly noticeable in darker areas with subtle lighting.

What Causes Noise in Cycles Renders

The core reasons for noise in Cycles renders are:

• Low sample count – Tracing fewer light paths per pixel leaves gaps in lighting information
• Complex lighting – Indirect, caustic, subsurface scattering effects need more samples
• Small light sources – Concentrated emissive objects increase noise
• Glossy/rough surfaces – Glossy BSDFs require more ray bounces to resolve
• Particle systems – Lots of small geometries make noise management tougher

In essence, anything requiring more ray bounces while having insufficient samples generates noise. Effects like motion blur and depth of field also play a role with their extended sampling.

Noise Reduction Settings in Cycles

Cycles provides built-in sampling settings under Render Properties > Sampling to reduce noise at the cost of render time.

• Samples – Primary control raising total paths per pixel traced
• Preview Samples – Lower samples for quick preview renders
• Viewport Samples – Samples used for final render viewport preview
• Noisy Image Preview – How quickly the render starts showing in the viewport
• Max Bounces – Limits ray recursion for diffuse/glossy/transmission rays
• Min Light Bounces – Early ray termination for quick preview renders
• Diffuse Bounces/Glossy Bounces/Transmission Bounces/Volume Bounces – Individual bounce limits per ray type

Using higher samples and bounce values reduces noise at the cost of longer render times. There are also various integrator settings providing noise reduction tweaks.

Using Denoising Nodes

While increasing samples is the main way to remove noise, it quickly becomes time-prohibitive. AI-based GPU/CPU denoising provides efficient noise reduction after the initial low sample render.

The Blender Compositor includes dedicated denoising nodes. These filter renders using algorithms recognizing noise patterns while preserving detail. This allows clean results with up to 5x lower sample counts.

Denoising Node Types

• OpenImageDenoise – Best overall quality with GPU/CPU support
• Optix AI Denoiser – Nvidia RTX GPU acceleration with AI training
• Intel Open Image Denoise – Alternative CPU-based denoiser

OpenImageDenoise works on all systems providing great versatility. Optix further boosts performance for RTX cards. The Intel denoiser focuses on fast CPU denoising.

Example Node Setup for Denoising

Adding denoising nodes is straightforward – they are appended after the Render Layers node carrying the beauty/emission passes.

A typical non-destructive node setup would be:

1. Render Layers node
2. Viewer node connected to Render Layers Direct output
3. Denoising node connected to Render Layers Image output
4. Composite node for corrections connected to Denoising node
5. File Output node to save final image

This allows comparing the original render and denoised versions. Further grading can enhance the denoised image before output.

Denoising Controls and Tuning

The available denoising tuning differs across OpenImageDenoise, Optix, IntelOIDN. But common controls include:

• Input passes – Beauty, Albedo, Normal
• Samples – Affects denoising quality and timing
• Prefilter spots/pixels – Removes hot pixels before processing
• Prefilter size – Spot/pixel filter kernel radius
• Denoise RGB/A – Toggles color/alpha channel denoising
• HDR/Square norm – Handling of image signal ranges

Tuning revolves around balancing render sample count, denoiser sample count, and prefiltering. Feature tuning helps retain fine textures, gradients, and small details.

Best Practices for Reducing Noise

Achieving low-noise Cycles renders boils down to these guidelines:

1. Use portal lights to introduce indirect lighting gradually
2. Control max ray bounces to limit oversampling in simple areas
3. Cull backfaces and use simplified collision proxies for intricate assets
4. Bake lighting for static meshes and reuse across frames
5. Use Irradiance Volumes/Cubemaps to cache indirect lighting on static objects
6. Denoise renders post-processing rather than pushing sample counts
7. Use Optix AI denoiser for RTX GPU renders, OpenImageDenoise otherwise

This methodology focuses sampling power on key areas while saving time in simpler zones. Targeted caching and culling complement selective focus.

Optimizing Render Settings for Lower Noise

Aside from samples and denoising, optimizing other render settings helps combat noise:

• Clamp Indirect – Reduces over-brightening of indirect bounces
• Light Falloff – Controls attenuation curve of light objects
• Caustics – Lowers clamp for caustic photons going through glass, water
• Filter Glossy – Blurs bright samples over nearby pixels
• Anisotropic Samples – Better sampling of anisotropic shaders
• Ambient Occlusion – Controls noise vs accuracy of occlusion

Effects like motion blur, depth of field, volumes also require tailored sampling for clean results without needing overall samples.

Useful Add-ons and Extensions for Denoising

While native Blender nodes provide capable denoising, dedicated add-ons enhance the workflow:

• D-Noise – Auto-generates complex denoising node groups with fine tuning
• Compositor Log Viewer – Displays render passes as colors for debugging
• Speed Denoise – One-click batch denoising of multiple renders
• Intel Open Image Denoise – Custom node bringing Intel’s latest denoising algo

The D-Noise and IntelOIDN add-ons in particular provide extra functionality missing from the stock denoisers. Overall, nodes remain effective for production use.

FAQs on Cycles Denoising

Why does denoisingsmoothen some surface details?

The filtering that strips noise can also smoothen fine textures without enough sampling. Increasing sample counts retains more high-frequency detail.

Can multiple denoising passes be layered?

Yes, chaining multiple denoisers lets artists use their various strengths. Common multi-pass strategies go from basic to fine-tuned for quality.

Which GPUs support AI/Optix denoising in Cycles?

Optix AI acceleration needs RTX cards from the 2000/3000/4000 series. Older generations only get basic Optix without AI denoising.

Is it better to render with GPU or CPU for denoising?

GPU rendering allows leveraging graphics card horsepower for denoising computation. This makes for significantly faster results.

Can animated sequences be denoised?

Yes, video clips can be batch processed through denoising nodes or scripts to filter noise across every frame.