If it is challenging enough to have a layperson figure out what is meant by trending technology such as ‘neuronal network technology’ in AI (Artificial Intelligence), try a simpler idea: how about let us just get rid of that grainy ‘noise’ in our rendering. We demand photo-realism. This and this alone had baffled many software engineers for all time.
Besides, developing art-directed technology that keeps our user artists and visualizers at the edge of not tipping over the frustration of wanting more speed, less quality versus more quality, less speed? There is the demand for greater flexibility: how to achieve picture perfect with simple and fast controls yet, not take away that creative versatility that personalized every work of art rendered?
Cebas Visual Technology has been in the business of creative visual effects for more than a quarter of a century now – sometimes eureka! – at times groping in the dark for that unique piece of software engineering. Now, following last September’s release of the new finalRender trueHybrid™, cebas will deliver with its upcoming subscription drop an enhanced finalRender featuring an AI-Denoiser based on NVIDIA’s OptiX 5.0 platform. This AI- trained denoising solution, tested and certified by NVIDIA on tens of thousands of 3D scenes offers realtime denoise functionality with outstanding quality.
NVIDIA has spent a lot of resources and time to develop this new technology that is aimed at helping other software developers in tackling one of the pressing issues when developing a modern rendering system. Physically accurate unbiased renderers, like finalRender, have to deal with noise while rendering an image. True unbiased renderers are prone to more noise than old school biased renderers, one reason being the methods used to follow light rays in unbiased path-tracing.
We could explore noise in general and the reason for it in an endless wall of technical descriptions and white papers. It can easily fill a whole book if not several books! Do not worry we are not going to do it here.
NVIDIA did the heavy lifting already for the developers and delivered a remarkable tool for us to integrate and enhance on – which we did. finalRender Subscription Drop1 will incorporate this new AI-Denoiser technology fully integrated into finalRender. Early testing already shows promising speed gains by reducing the amount of render passes (samples) needed to get a clean image. Depending on the scene, we could see speedup factors ranging from 2 to 5 times or even 10 times faster in some situations!
New finalRender AI-Denoiser incorporating NVIDIA OptiX 5.0 Update Coming Soon for Subscribers from cebas Visual Technology Inc. on Vimeo.
About this Image
The PC workplace image was rendered with 100 samples, only. This was done to illustrate the power and strength of the AI-Denoiser, integrated in finalRender. The zoomed in keyboard section (pixel blow up) should resemble a flat shiny surface illuminated by an Area light. Without the AI-Denoiser, noise is clearly visible in the image. The same picture with AI-Denoising active, shows the noise fully removed. Other remarkable areas to watch out for are the soft shadows of the cup and the pen.
As seen in this example image, AI-Denoising is pretty impressive and it does not really eat up render time at all. In consecutive passes, the AI-Denoiser is processing its results in the millisecond range! Our integration of the AI-Denoiser into finalRender is fully transparent and without any complexity at all for the artist! Following our philosophy – one button is enough – you either turn it on or off. No complex features or controls needed, and all the denoising magic is done by the Artificial Intelligence.
So far our testing did not reveal any major headaches besides the known issues or better, the technical challenges when using transparent objects (glass) or extreme contrast values. Purely artificial image setups (no noise at all, all black areas) will also startle our integrated Artificial Intelligence, which is expected somehow. You can’t tell an AI to look for noise and create photoreal images by not supplying it the minimum – a real image.
Scene #2: fR-Denoiser, Motion Blur, Depth-of-Field for detailed surfaces working in sync simultaneously.
Scene #3 : fR-Denoiser, Depth Of Field - metal surface