Key Points

• Quick previews: even complex scenes will be (noisily) rendered in a few seconds, allowing you to quickly notice and correct errors. No more waiting half an hour to discover you had the wrong lighting setup.

• Typically, a simple render will take 10x as long to produce “clean” output with Falcon as with Cheetah.

• Falcon renders generally look much better than Cheetah renders. (You thought Cheetah renders looked fine?)

• Falcon renders have “noise” in much the same way digital photographs do. The more “samples” you render the less noise there will be.

• Falcon is more realistic than the old “Cheetah” renderer. Special faux-realistic tricks that the Cheetah renderer uses such as radiosity, ambient occlusion, depth of field, and blurred reflection and transparency, are computed automatically and at little or no extra cost (in setup or rendering time). Certain costly material features, such as reflection and transparency blur, likewise do not impact rendering time.

• Falcon just works. Very little finessing of lighting and rendering setups will be required — e.g. shadow blotches, issues with radiosity and transparent objects, lighting artifacts in corners will not be a problem and don’t require fake lighting setups or render tags.

• Falcon renders motion blur. Again, this has little or no impact on rendering time.

Falcon is a “Unidirectional Path Tracer”

The original Cheetah 3D renderer — now referred to as “Cheetah” — was a Ray Tracer, which means that each pixel in the final image is rendered by tracing back from the observer through the pixel into the scene, and then guesstimating how brightly lit each thing you bump into is.

AI draft: generated from the Cheetah 3D 8.1.1 manual; not yet reviewed by the author.

Falcon is a “unidirectional path tracer”. In simple terms, this means it’s an “unbiased” or “physically-based” renderer that works by tracing random paths of light. For each pixel it follows a ray from the camera into the scene, and every time the ray hits something it makes a random (“stochastic”) decision about where the light goes next — e.g. if a surface has a 50% chance of reflecting light, the path tracer flips a coin and follows one of the two possible paths — while checking how much light reaches that point from the lights in the scene. Each such path is one “sample”. A single sample is a terrible guess, but if you average enough of them everything evens out — which is why the image starts very noisy, and smooths out as the “law of large numbers” evens out the effect of random decisions.

The quality of a rendered image is referred to in terms of “samples per pixel” (Falcon’s Samples setting is the number of paths traced per pixel). If a rendered image is 1000 x 1000 pixels (one million pixels) then 100 samples per pixel means that 100,000,000 paths contributed to the image. Set Samples to 0 and Falcon will keep going until you stop the render yourself.

How far each path is followed is limited by Falcon’s Ray depth (bounces) settings — Total, Diffuse, Glossy, and Specular bounces, plus Transparency depth. More bounces give a more accurate result, but take longer and need more samples to clean up the noise.

End of AI draft.