Home Creators Posts Import Register Favorites Logout

Content

Houdini’s new Copernicus toolset includes a very useful node for denoising renders, Denoise AI. Let’s have a look at how it works!

Denoise AI can be found inside any Copernicus (COP) network. You can add a network to wherever you’re currently working (in /obj or /stage, for example), or visit the dedicated /img compositing context.

Dive inside your newly-created COP network, add a File node and import a single image or sequence. Finally, connect a Denoise AI node.

Denoise AI lets you choose between two denoising algorithms: Intel’s Open Image Denoise (OIDN) or Nvidia’s OptiX. OIDN comes with Houdini, but OptiX is only available if you have an Nvidia GPU.

Both denoisers can make use of normal (N) and albedo AOVs, plus OptiX can take advantage of a motionvector AOV.

Here’s a simple setup to denoise an image sequence with OIDN. Wire up the inputs as shown and enable the Auxiliary Planes Are Clean and Prefilter Auxiliary Planes options. Turning these two on produces the best results, but it also slows processing down a bit.

Note that you can add outputs for all the AOVs contained in the image by clicking Add AOVs from File.

Setting up OptiX takes a few more steps. You’ll want to connect the motionvector AOV as well as the Previous Frame input. There won’t be any “previous frame” on the first frame of an animation, nor will there be a motion vector. Side FX recommends the following approach:

On the first frame, create a black (0, 0, 0) motion vector Constant and refer to the first, noisy frame in the sequence for the prevframe input. Then, on the second frame, bring in the motionvector AOV and the previous denoised frame.

This can all be automated with Switch nodes and an expression to control the switch activation. In this case we’re using the $F>1 expression to tell the switches to enable input 0 on frame 1 and input 1 on frame 1 and beyond. This means our Constant and the first, noisy frame are used on frame 1, then the correct AOV and previous denoised frame are used thereafter. We use a Time Shift node to load one frame back.

Whether using OIDN or OptiX, you’ll want to connect a ROP Image Output to save the denoised images to disk. Just update the file name, frame range, etc. as desired and click Render to Disk.

That’s all there is to it!

Final Thoughts

This is an easy way to denoise renders using one of two high-quality, production-tested algorithms. I find that even renders without the extra AOVs have the potential to clean up nicely. Results can be hit or miss depending on the image content and noise level, so give everything a try to see what performs best for you!

Files

Comments

No comments found for this post.