Denoising in Corona produces noise-free images and significantly reduces render times. This is achieved by eliminating the need to wait for noise to refine during the rendering process.
The main purpose of denoising, along with adaptivity, is reducing the number of passes needed to get a noise-free image. The reported render time reductions are between 50 and 70%. Denoising also attempts to remove fireflies (single bright pixels) from the image.
There are 3 types of denoising in Corona:
- Corona Denoiser (called Corona High Quality in older versions) - for production rendering only
- NVIDIA GPU AI (fast preview, Optix) - for production and interactive rendering
- Intel CPU/GPU AI (speed-quality hybrid, OIDN) - for production and interactive rendering
You can use the Fireflies Filter on its own or alongside denoising in order to get rid of the bright spots (fireflies) in your rendered images.
Note: Starting from Corona 12, the new VFB 2.0 is the new default VFB in Corona, featuring an updated interface with improved user experience. For a complete overview, refer to the official documentation.
1. Corona Denoiser
This denoiser is a post-rendering effect which analyzes noise in the 3D space and smartly reduces it without producing splotches while maintaining sharp edges and texture details. It is not strictly a 2D post-process. The result can be interactively blended with the original non-denoised image directly in the VFB after the rendering is finished to achieve the desired level of denoising. This is especially useful if the denoised image appears to be over-processed (generally this should not happen but can sometimes be observed, for example in case of textures with very subtle patterns and rendering a small number of passes).
|
No denoising |
Corona Denoiser |
Note: Clicking "Cancel" in the Corona VFB will DISCARD denoising, and it will not be applied to the final rendering! To stop rendering and apply denoising, you need to use the "Stop" button.
The Corona Denoiser has 3 modes:
- High quality - full denoising
- Only remove fireflies - useful in scenes where fireflies are the main issue, much faster than full denoising
- Gather data for later - gathers denoising data for the standalone Corona Image Editor
There are two parameters controlling this denoising type:
- Denoise amount - this blends between the denoised and non-denoised image. The default value of 0,65 provides good balance between denoising and fine detail.
- Denoise radius - radius of the blurring filter. Increasing it may help reduce very strong noise. Decreasing can help prevent loss of fine details. The default value usually works best.
2. NVIDIA GPU AI Denoiser (fast preview, Optix)
This denoiser's main purpose, as the name suggests, is to provide fast, noise-free previews. It is applied almost in real-time during interactive or production rendering, and after the rendering is stopped. It requires a supported GPU to work (See: Hardware Requirements), which is detected during the installation.
|
No denoising |
NVIDIA AI Denoiser |
NVIDIA GPU AI Denoiser works in a similar way to the Corona Denoiser, however there are some substantial differences:
- It is preformed purely on the GPU (Corona Denoiser - on the CPU)
- It uses machine learning ("AI") to smartly blur noise, but also to produce detail (Corona Denoiser cannot add detail to the image)
- It can be used during interactive rendering too (Corona Denoiser can only be applied after the production rendering is finished)
- It works almost in real-time and is applied to the image progressively as it renders (Corona Denoiser takes time and is applied only once, after the production rendering is finished)
- It should be used mostly for interactive rendering, quick previews, and optionally for final stills
- NVIDIA GPU AI Denoiser, as opposed to the Corona Denoiser, does not feature any additional modes
The only setting available for this denoising type is Denoise Amount, which works exactly the same as for the Corona Denoiser – it blends the original non-denoised image with the denoised result, and this can be done interactively during rendering, or once rendering is finished.
3. Intel CPU / GPU AI (speed-quality hybrid)
This denoiser combines the features of the Corona Denoiser and the NVIDIA GPU AI Denoiser and uses machine learning to produce noise-free results. It is slower than the NVIDIA GPU AI Denoiser (seconds vs milliseconds), but it can be applied to interactive rendering too.
|
No denoising |
Intel AI Denoiser |
It can run on the CPU (production rendering only) or the GPU (production and interactive rendering). CPU denoising isn't limited to Intel CPUs only. Interactive rendering requires a supported NVIDIA GPU.
The only setting available for this denoising type is Denoise Amount, which works exactly the same as for the Corona Denoiser – it blends the original non-denoised image with the denoised result, and this can be done interactively during rendering, or once rendering is finished.
Denoising Limitations
There are some important things to bear in mind when using denoising:
- There are some important things to bear in mind when using denoising:
- Currently, none of the denoisers in Corona offers temporal consistency. This means that denoising generally shouldn't be used for animations as it may produce splotches, flickering, or other artifacts. In some cases, it may be possible to render animations with denoising, however this is dependent on scene type and base image quality.
- If a combination of AI denoising (NVIDIA or Intel) with high quality pixel filtering is detected, the pixel filtering is automatically changed to Tent. To learn more, see: High quality pixel filtering used with AI denoising or upscaling error messages
- Denoising can improve images which are noisy only to a certain degree. In case of extremely noisy images, denoising will not be able to help, and the result will be splotchy or full of artifacts.
- Denoising requires additional render elements, which are invisible to the user. These render elements can take a considerable amount of memory, for example:
3dsmax.exe RAM usage (Cinema 4D RAM usage would be similar) for an interior scene rendered in 1920x1080 resolution with 10 denoised render elements:
- No denoising – 6,5 GB
- Fast Preview Denoising – 7,3 GB (+GPU VRAM)
- Corona Denoising – 7,3 GB
Enabling Denoising
For Production rendering:
The mentioned 3 denoisers can be selected from
Render Settings > Corona > General settings > Denoising:
The Intel GPU AI or NVIDIA GPU (Optix) AI denoiser can be selected from:
Render Settings > Corona > Performance settings Tab > Interactive Rendering > Fast preview denoise during render:
Example Uses
1. Render in Higher Resolution, Denoise, Then Downscale
A simple and highly effective way to reduce overall render time is to render your image at a higher resolution than needed, apply denoising, and then downscale the result to the final output size using Corona's "Save 50%" option.
The workflow is straightforward: first, render the image in high resolution. Next, denoise it - ideally using the Intel denoiser for the best balance of speed and quality. Finally, downscale the denoised image to the desired resolution.
This approach often produces cleaner results with fewer noise artifacts, meaning you can achieve acceptable quality within fewer render passes. In many cases, this can reduce render times by several times compared to rendering directly at the target resolution.
Tutorial - Faster Corona Render Times:
2. Saving a CXR file and denoising it later
The "Gather data for later" option can be used to skip the actual denoising process but save the denoising data within the CXR file. Such file can be then denoised later using the Corona Image Editor or loaded into the Corona VFB and denoiser there. .
Since denoising is calculated later, you can use this approach to try different Denoise radius values without re-rendering (otherwise, the Denoise radius is baked in at render time).
For more information about CXR file saving for later use, see: How to save CXR files
Note:
- Using the "Gather data for later" option will allow you to denoise an image later in the Corona Image Editor but only using the Corona Denoiser.
- If you wish to denoise your image using the NVIDIA AI or Intel AI denoiser, choose either of those two, render your image, and save to CXR. You can then switch between these two denoising types in the Corona Image Editor.
3. Saving multiple versions of the rendering with different denoising amounts
Multiple "VirtualBeauty" passes can be used to save different versions of the same image at once with various amounts of denoising applied. This allows you to use masks (either generated from Corona or created manually) to blend between different denoising amounts in post, for example, if you wish to preserve flakes in car paint or fine bump mapping in carpets that might need a lower denoising level than other parts of the image. Individual denoising amount is specified in the properties of each VirtualBeauty pass regardless of the in-VFB setting:
Multiple VirtualBeauty elements visible on the Corona Multi-Pass list. The denoising value of each of the passes is different than the currently set "denoise amount" in Corona VFB. Click to enlarge.
4. Denoising other passes
Denoising can be applied to other passes too by enabling the "Apply denoising also to this render element" option in each pass's settings.
Denoising Sum-Up
- Corona Denoising
- Can be used for production rendering only (General settings tab).
- Is performed on the CPU. If your CPU can run Corona, it can also run this denoising type.
- Is applied after the production rendering is stopped.
- Is more precise, and so requires more time to be calculated.
- Maximum allowed resolution of the denoised image and the number of denoised render elements are dependent on the available system RAM.
- NVIDIA GPU AI Denoising
- Can be enabled for production rendering (General settings tab) and for interactive rendering (Performance settings tab).
- Is performed on the GPU, which must be a supported NVIDIA GPU (See: Hardware Requirements)
- Is applied during rendering, almost real-time, and also when the rendering is finished.
- Can be used for extremely quick previews.
- May be used for high quality stills (provided that the base image is of good enough quality and no denoising artifacts appear).
- Maximum allowed resolution of the denoised image and the number of denoised render elements are dependent on the available GPU RAM.
- Intel CPU / GPU AI Denoising
- Can be used for production rendering (General settings tab) and interactive rendering (the GPU mode only; Performance settings tab).
- Can be performed on either the CPU or the GPU. If your CPU can run Corona, it can also run this denoising type. Almost any NVIDIA GPU made in the past 10 years will run this denoiser too. Just make sure you have the latest drivers installed.
- Is applied during rendering, almost real-time, and also when the rendering is finished.
- Can be used for quick previews.
- Is much faster than Corona Denoiser, but slower than NVIDIA GPU AI denoising.
- May be used for high quality stills (provided that the base image is of good enough quality and no denoising artifacts appear).
- Maximum allowed resolution of the denoised image and the number of denoised render elements are dependent on the available system RAM or the GPU RAM depending on the selected mode.