Render Settings & Optimization

To understand every render setting, you have to understand noise. An offline engine estimates each pixel's colour by tracing many random light rays into the scene and averaging what they find. With few rays, that average is a rough guess and neighbouring pixels guess slightly differently - the image looks grainy, speckled, noisy. Each ray is a sample; adding more samples means a better average, so the noise falls and the image cleans up. That is the entire mechanism. Noise is not a bug or a broken setting; it is simply an estimate that has not yet had enough samples to settle.

This is where nearly all your render time goes: buying samples to beat down noise. And the relationship is unforgiving, because it follows the mathematics of averaging - to halve the remaining noise you must roughly quadruple the samples. Put differently, doubling your render time only cuts the noise by around thirty percent once you are past the easy early gains. The image that goes from 'unusable static' to 'basically clean' in the first two minutes might take another twenty to become 'mathematically perfect' - and your eye cannot tell the last two stages apart. This is the single most important fact in render optimization, and the figure shows its shape: a steep early drop, then a long, flat, expensive tail.

Some parts of a scene are noisier than others and soak up disproportionate samples: blurry (glossy) reflections, deep GI in dark interiors, small bright lights, glass and refraction, and depth-of-field blur. Recognising these noise sources is half of optimization - often a single glossy floor or an overly small, overly bright light is what makes an otherwise simple scene render for an hour. Modern engines use adaptive sampling to spend rays where the noise actually is, but you still help enormously by not creating unnecessary noise in the first place: sane light sizes, sane glossiness, GI no deeper than the eye needs. The mental model to carry is simple - render time is the currency, samples are what you buy with it, and noise is what you are buying your way out of. Everything else in this lesson is about spending that currency where it shows and refusing to spend it where it does not.

Because of that brutal tail, the smartest move in modern offline rendering is usually not to sample your way to a perfectly clean image. Instead you stop at 'nearly clean' - where the steep, cheap part of the curve has done its work - and hand the residual grain to a denoiser. A denoiser is an algorithm (increasingly an AI-trained one) that intelligently smooths out the remaining noise while trying to preserve real detail and edges. Both V-Ray and Corona ship capable denoisers, and using them well is one of the highest-leverage skills in the whole module.

The economics are dramatic. Chasing the last of the noise by brute sampling might take twenty extra minutes; a denoiser can clean that same residual in seconds. So the modern render recipe is: set a noise threshold for 'nearly clean', render to it, then denoise. This can cut a render from an hour to fifteen minutes for a result the eye reads as identical - and on animation, where every second of savings multiplies across hundreds of frames, it is transformative. The denoiser is the single biggest reason offline render times have become manageable for small practices and freelancers.

But a denoiser is a tool to tune, not a magic 'make clean' button, and it has failure modes worth respecting. Push it too hard - denoise an image that is still very noisy - and it cannot tell noise from detail, so fine textures, sharp edges and small highlights get smeared into a soft, waxy, plasticky look. The fix is to give it a reasonable starting point: render to moderate noise, not extreme noise, then denoise. Some engines also offer denoisers that run during the render and let you watch the trade live, and modes that denoise individual render elements so you keep post flexibility. The rule of thumb to internalise: the denoiser removes the last bit of noise cheaply; it cannot rescue an under-sampled render, and leaning on it too early costs you exactly the crisp detail that made you choose offline in the first place. Treat it as the final, cheap polish on an already-decent image, and it will save you hours; treat it as a substitute for sampling, and it will quietly rob your image of the realism you paid render time to get.

Resolution is the other big time multiplier, and it is the one people most often set wrong. Render time scales roughly with pixel count, so doubling both the width and height of an image quadruples the number of pixels - and the time. Rendering a web thumbnail at billboard resolution is one of the most common and most wasteful mistakes in archviz. The discipline is to match resolution to the delivery: a social post or a client-review WIP needs far fewer pixels than an A1 competition board or a printed billboard, and rendering the big size 'just in case' can multiply your day for no benefit.

This connects to the deeper principle that governs this entire lesson: match the quality to the job, not to your ego. A design-stage test render can be small, noisy and denoised hard - you only need to read the light and the composition. A client-review image needs to be clean and correctly sized for a screen. Only the final hero, destined for print or a marketing splash, earns full resolution and low noise. Deciding the target before you render - what is this image for, where will it be seen, how big - is what stops you from either under-delivering a blurry hero or, far more commonly, burning hours over-rendering something that will be viewed at postcard size.

Several practical moves flow from this. Use small, fast draft renders to set up light and materials, and only go big for the final. Use region / crop rendering to re-render just the corner you changed instead of the whole frame - invaluable when you are perfecting one material or one light. Keep a test resolution and a final resolution as named presets so you never fat-finger a huge draft. And remember that a denoised medium-resolution image often beats a noisy high-resolution one for the same time. Every one of these is the same idea wearing different clothes: spend pixels and samples where the job needs them, and nowhere else. The optimizer's question is never 'how good can I make this?' but 'what is the least this specific image can cost me and still do its job?' - and asking it on every render is what lets you deliver more images, faster, without your clients ever seeing the difference.

Put it together into a repeatable workflow and render times stop being a source of dread. Start rough and fast. Set a low-resolution draft and a loose noise threshold so you can iterate on light, materials and camera in seconds, not minutes - or better, use the engine's interactive/live mode (Corona's interactive render, V-Ray's IPR) so setup feels near real-time. Get the image right at draft quality before you spend a single expensive sample; there is no point rendering a clean version of a badly-lit scene.

Then optimize the scene, not just the settings. Track down the noise sources named earlier - an over-bright tiny light, an over-glossy floor, GI bounces set deeper than the eye needs - and tame them, because fixing the cause is always cheaper than sampling past the symptom. Hide or simplify geometry the camera never sees. Reuse and instance repeated objects. These scene-level moves often cut render time more than any sampling tweak. Next, set the target honestly: pick the final resolution from the delivery, and a noise threshold that will look clean after denoising rather than before.

Then render the final to that 'nearly clean' threshold and denoise to finish - checking the denoise did not smear detail. For anything you might revise, output render elements so the fixes happen in post, not in a re-render. And when the job is big - print resolution, or animation across hundreds of frames - reach for more hardware or a render farm rather than heroically waiting. The through-line of the entire module lands here: offline gives you the photoreal ceiling, but only optimization makes reaching it affordable. The best archviz artists are not the ones who render the most samples; they are the ones who reach exactly the quality the job needs and stop - because a good image delivered on time beats a perfect image that missed the deadline, every single time. Learn to feel where 'enough' is, build the draft-then-target-then-denoise habit, and the wall of settings resolves into one dial you know how to spend.