15-18: The 2D World - A Guide to Photorealism

Reading: Eran Dinur, The Complete Guide to Photorealism for Visual Effects, Visualization and Games (Focal Press, 2022), Part 4, chapters 15-18, pages 171-209. This page condenses four chapters. Unlike Part 3, most of this transfers directly to After Effects work, so read it properly.

The two rules Dinur opens with

2D looks limiting next to 3D. No lighting, no shading, no texturing, no modeling, transformations restricted to two axes. But that ignores hundreds of years of painting in which artists conveyed depth, light, texture, atmosphere and scale with nothing but a brush and a canvas. They did it through mastery of color and a deep understanding of three-dimensional space, and the digital process is essentially the same.

So he gives his students two pieces of advice, in this order:

Mastery of color is the key to successful compositing and matte painting.
Never think of compositing as stacking layers - as "A over B" or "layer 5 over layer 4". Every image or piece of footage represents a three-dimensional world, and every 2D element added to it should be approached as a three-dimensional entity placed within a three-dimensional space.

His summary: the key to photoreal 2D is to never think of it as 2D. The seemingly unattainable third dimension is accounted for through the thoughtful use of scale, perspective, atmosphere and depth of field.

The benefits are real. 2D is much faster than 3D rendering and shows results in real time, which is why final lighting and color adjustments on CG renders, and lens characteristics like depth of field, are usually left to compositing. And the biggest advantage is working with photographs and footage: the detail, the imperfections, the wear and tear, the subtle shading, the warmth of a physical lens are all already there. The challenge is combining different photographic elements - Dinur's phrase is combining puzzle pieces from multiple unrelated puzzles into one coherent image.

↑ Contents

Chapter 15: Integrating 2D Elements

Color matching

Making an element sit in a new environment is one of the most crucial skills for a compositor, and one of the hardest to master. Composite the same tree over five different backgrounds and the leaves stay green - but a different green each time.

This is where the six-layer approach from Chapter 3 pays off. The green of the leaves exists as an absolute color only in a virtual world with no external influences. In reality it is already affected by the lighting and by the camera's own color interpretation. Placed in a new environment, it has to be adjusted for that environment's lighting, atmosphere and camera color interpretation. The five color operations - gain, gamma, lift, offset, saturation - are the tools.

There is no single methodology, and it is more personal preference than science. Some start by matching black and white points, others by brightness or hue. Dinur's own order: big brush strokes first - gain and gamma for general brightness, contrast and overall tint - then finer adjustments for blacks and highlights, then specific ranges like shadows, midtones and highlights, then a hue correct tool if specific hues still need work. He relies on his eyes more than on sampling, because color matching is context-sensitive and relative rather than absolute.

The exercise he sets his students is worth doing yourself: take a photo and a color-corrected version of it, and adjust the original to match the second as closely as possible. It can be an easy drill or a mind-twisting puzzle. Mechanical, yes - but so is a pianist practising scales.

Edges: the whole problem in one section

CG elements come with a built-in alpha. Every 2D element has to be cut out of its original background first. If all edges were sharp and well defined this would be simple, but they hardly ever are - edges carry sub-pixel detail like hair, fur, fabric and tiny irregularities, or are fuzzy and semi-transparent from translucency, defocus or motion blur.

And extraction is only half the problem. Any edge that is not 100% opaque carries some of the original background with it. That is obvious trouble when the original background is busy, but even a clean green screen causes it: if the luminance of the old background does not match the new one, you get dark or bright edges around the element, and the composite dies.

Which produces the central dilemma: preserve edge detail and softness and risk ugly bright or dark outlines, or tighten the extraction and end up with unnaturally sharp, harsh edges. Both damage photorealism - either route produces a comp that looks Photoshopped. The goal is to keep the edge detail and its original softness while making sure the semi-transparent edges blend seamlessly with the new background.

His methods:

Match background luminosity at the sourcing stage. If you know what the element is going into, solve it before you shoot. Shooting a green screen element for a night shot, he asks the cinematographer to take the screen lighting down a notch or two - still bright enough to key, but dim enough to avoid unnatural bright edges over a dark background. Same principle when picking stock: a tree extracted from a bright sky and composited into a dark forest will show bright edges, so look for tree elements shot against a similarly dark background.
Spill suppression. Semi-transparent edges carry not just the background's luminosity but its hue - in this case an extremely unnatural green or blue. Spill is needed not only at edges but wherever the subject catches bounce or specular from the screen. Nuke's IBK algorithm uses a clean version of the screen as a color roadmap for subtracting spill, which is more accurate because a real green screen is never uniform. Careful, though: heavy despill on grainy footage creates black holes and noise where green pixels turn black, in which case degrain first and re-grain after.
Edge and core extractions. Junior compositors try to pull one perfect matte. It is far more effective to make two: a hard-edged core matte that ensures no holes in the body and face, and a softer extraction for the edges that preserves hair, fabric, and motion-blurred or defocused detail.
Edge reconstruction. Heavily motion-blurred or defocused areas are so transparent and wispy that they contain too much background to extract. Often the best approach is to not extract them at all but to recreate them: pull a tighter matte along the last solid pixels, then place a dilated, blurred copy of the element underneath itself to replace the missing soft edges, using directional or vector blur for the streaking. The result carries none of the original background.
Matte-less extractions. This sounds odd but the logic is sound: compositing an element over a background is really just changing the color of the background pixels - the matte only defines where those changes happen, not any physical thickness. For thin wispy detail like hair, fur or grass, it is often better to modulate the background color with the foreground directly, using an additive or multiply operation. For dark hair over a bright green screen, isolate the channel with the most contrast, brighten the non-hair pixels to 1 and clamp, then multiply the background by that image: dark hair darkens the background while white areas do nothing. For bright blond hair over a darker blue screen, darken non-hair areas to zero and add instead. Usually complementary to standard keying, and often substantial in bringing back very fine detail.
Edge blur and edge color. A slight blur to extracted edges reduces noise, improves blending and adds a photographic touch, since lenses soften boundaries. Most effective when very subtle - 1 to 3 pixels is usually enough, higher only if the element is out of focus. Reference the footage for how soft real edges are at comparable focus. Since edge blurring also affects grain, use the edge-detect matte to add grain back only where it went missing.
Light wrap. When a subject is in front of a light source or a bright area like a window, some light bleeds over its edges. Frequently misused as a remedy for bad edges. In most cases objects display a fairly solid silhouette even when backlit, and light bleeding is limited to extremely bright areas. Exaggerated light wrap gives unnaturally fuzzy, ghost-like edges. Use it sparingly and always in direct relation to areas of strong luminance behind the subject.

↑ Contents

Chapter 16: Integrating CG Elements

Render passes

Dinur frames it well: imagine what photographers could do if they could separate a photo into direct and indirect light, highlights, shadows, transparency, refraction and depth, or by individual light. Not possible photographically; routine in CG. Render passes - also called AOVs, render elements or render channels - are integral to VFX because they allow adjustment at comp without going back to a slow 3D render.

They split into lighting passes and utility passes.

Lighting passes. The complete render is the beauty pass. It can be broken into diffuse, specular, transmission, subsurface scattering and so on, and further into direct and indirect versions of each: a direct diffuse pass has only illumination from light sources, indirect diffuse only illumination from reflected light; direct specular shows reflections of light sources including sky and environment domes, indirect specular shows reflections between objects. Dinur notes the separation is completely artificial - in the real world there is no difference between reflecting the sky and reflecting a nearby tree - but it is extremely helpful.

Two approaches in comp: start from the separate passes and re-add them, which works because light accumulates, so an additive recombination reproduces the beauty exactly - like an audio mixing console, giving full control over every component in the mix. Or, for minor tweaks, start with the beauty and add or reduce only specific components.

His caution, and it echoes the whole book: changing the balance between lighting components may break the PBR energy conservation. Not necessarily bad - the goal is a convincing image - but worth keeping in mind, and he admits to often tweaking components ad nauseam only to return to the beauty at the end. Sometimes the well-balanced PBR result works best unaltered.

Some clever uses:

Increasing indirect diffuse produces a poor man's subsurface scattering, reducing contrast and softening features - good for taking the edge off elements that feel too harsh and CG.
Boosting indirect specular is technically incorrect but makes a surface feel more metallic, by enhancing the illusion of increased reflections at grazing angle.
Indirect components are the noisiest, so when re-rendering at higher samples is not an option, separate the indirect passes, denoise or degrain them, and plug them back in.
Per-pass color correction is extremely effective, because a blanket grade hurts photorealism. The classic case is changing diffuse color without affecting specular color - since specular reflections in dielectrics are not tinted, keeping the specular unaffected while grading the diffuse preserves the correct material.

Utility passes carry no visual content; their pixel data is information for manipulating or diagnosing the render.

Shadow matte and ambient occlusion. These bridge the gap between 3D and 2D. To composite a CG chair onto a photo of a room, you need contact shadows under its legs, and possibly cast shadows on the floor. The CG artist models the interacting elements of the footage - the floor, nearby objects - as accurately as possible and adds them to the scene as shadow casting/receiving objects only, invisible in the beauty but active in the AO and shadow matte. AO is perfect for contact and proximity shadows because it is completely independent of the actual lights: multiply the background by the AO pass and the contact areas darken while unaffected areas stay untouched. A shadow matte pass, derived from the actual lighting, then supplies the correct softness and intensity of each defined shadow, used as a mask for a global shadow grade.
Point Position and Facing Normal passes. These are 3D-to-2D translators. The Position pass (P pass) stores the location of every surface point in real 3D space in the RGB channels as XYZ, and needs 16- or 32-bit float such as EXR. The Normal pass (N pass) stores facing direction in RGB with a range of -1 to 1 - a floor is 0,1,0 and a right-hand wall is -1,0,0, relative to world coordinates rather than the camera. The Position pass creates a dense point cloud of the scene inside the compositing application without importing geometry, which is how a compositor knows exactly where a character's feet land in 3D when placing 2D dust elements. They can also project textures onto rendered geometry at comp, or drive relighting tools. A cheap trick: use the Y channel of the Normal pass as a mask to brighten only up-facing areas.
Matte and ID passes. Colored mattes only work in pure red, green or blue, since mixed colors cannot be isolated accurately - so one matte pass holds up to three separate mattes. Cryptomatte assigns a unique ID to every object and material and lets the compositor isolate any combination by point and click, eliminating multiple matte passes.
Z depth. Serves two essential purposes: accurate 2D depth of field, and depth-based atmospheric effects. Its limitations matter. Most engines output a non-normalized depth pass representing any distance to infinity, which is generally good but awkward as an atmospheric mask, often forcing the compositor to push the black and white points. Transmission and refraction cannot be represented - all transparent surfaces read as fully opaque, so an out-of-focus element can suddenly snap into focus when it moves behind glass. Volumetric elements like smoke and fog are similarly problematic, since a depth pass cannot show multiple layers of semi-transparent pixels. And reflected elements are not represented, so you cannot use depth to focus on a figure reflected in a mirror.
Deep compositing solves much of that: deepEXR stores an array of values per pixel at varying depths, including surfaces hidden behind foreground objects. Layering order stops mattering entirely, because elements are automatically arranged at the correct depth - true 3D depth-based layering, especially useful for particles and volumetric smoke. The catch is that deep data can only come from 3D rendering, not live footage, and the files are substantially larger.

Improving CG in comp - the sweetening tips

CG is purely mathematical and therefore inherently perfect, and that perfection stands in the way of photorealism. Part 3 addressed it in modeling, texturing and shading. These are the 2D remedies, and they are simple enough to use tonight:

Render resolution. The obvious assumption is that render resolution should match output. Often it should not. CG renders are almost always too sharp compared with photographed footage and need softening to sit in a comp. Rendering at 80% or 75% and resizing back to 100% with a cubic or quadratic filter produces the same effect as softening, and cuts render times and file sizes as a bonus. (The exception is a scene with many thin straight lines, like suspension bridge wires, where full resolution avoids aliasing.)
Hazing effect. Nothing to do with atmospheric depth. Blur the CG element a little - 10 to 40 pixels depending on resolution - and merge that blurred version back over itself at an opacity just high enough to slightly soften and smear the render. A very simple, very effective way to take the edge off CG, as long as the mix stays low and it does not become a cheesy dreamy-photo look.
Highlights bloom. The same procedure but with a max operation instead of over or screen, so only the brightest areas get hazed. Mimics real highlight bloom. Works best subtle - manufacturers work to minimize lens scattering, so overdone blooms look wrong.
Albedo pass and contrast. A recurring problem: the render looks terrific alone but does not sit in the comp, and contrast is usually the culprit. Renders often look right over clean black and too punchy against the target footage. Adjusting contrast with lift and gamma is tricky because it is hard to affect lighting-related color without also flattening the texture color. The diffuse albedo pass contains all the texture color information minus lighting and shading, so blending the beauty with a bit of the albedo pass reduces light-shadow contrast without flattening any texture detail.

↑ Contents

Chapter 17: Lighting in 2D

Relighting with color

In 2D there are no lights to place, no surfaces set to react in specific ways, no automatic shadows or reflections - and, most importantly, lighting is already present in the image and often has to be undone before it can be modified. That is challenging and sometimes impossible.

So, exactly as a traditional painter does, the only way to create or change lighting in 2D is through the selective use of color: coloring the areas of light and shadow, with the paintbrush replaced by animated roto shapes when the element is moving.

The pre-baked lighting in the source is the defining factor in how well it can be relit. An element shot in direct sunlight requires removing the baked light and shadows first. Even the highest-quality imagery may not have the dynamic range to grade highlights down and lift shadows that far, and extreme grading gives clamped flat areas or ugly artifacts. Best is to source material whose lighting already matches the target. Failing that, start with overcast or flatly lit material - exactly the same rule as texturing and 3D scanning.

Reflections

Up to now specularity has been one thing, whether it appears as a small highlight or a mirror image. In 2D the distinction matters. Our perception is not especially sensitive to the top of the color range and we are lenient about misplaced or inaccurate highlights, so generic highlights can be painted in fairly easily - a few scattered daubs is often all it takes, as countless traditional paintings show. The catch with film and video is motion: the effect of specularity is lost if the painted spots do not move in tandem with the object, light and camera. Without proper movement, static highlights read as texture rather than specular reflection.

Clear reflections on windows, mirrors, water or polished metal are much harder, mainly because what is visible in the reflection is not necessarily seen by the camera. The reflected image represents the point of view of the mirror, which can be entirely different from the camera's.

Two scenarios he works through:

A matte painting of a mountain lake. It seems obvious to grab the background above the shore and flip it. But the camera's line of sight is roughly horizontal while the lake is seeing the environment vertically from the bottom up. That 90-degree difference in field of view barely affects distant elements, but anything close to the water edge looks different - a deer viewed face-on by the camera is seen from underneath by the reflection. Hard to recreate in 2D without hand-painting or 3D. For most purposes the simple flip works fine; the discrepancy shows only in close-ups with distinct focus on elements near the surface.
An over-the-shoulder shot of an actor at a window with a green screen behind it. A proper sense of window glass can only come from reflections - but the camera is shooting the actor's back while the glass reflects their face. A traditional painter simply paints it; a photoreal composite needs real footage from the reflection's point of view. Options: put physical glass in the set window and luma-key the reflection back over the added background, though the glass often reflects the camera, crew or an unbuilt part of the set; or shoot a separate witness camera take from the window toward the actor's face, which works only if the actor is not moving much. His point is broader: not every reflection can be created from scratch in post. Sometimes the material must be shot on the day.

Two rules for faking reflections:

Apply the Fresnel effect. Reflection should be stronger at shallow viewing angles and fade as the angle steepens. In the lake example a consistently strong reflection looks unnatural - it should fade toward the camera, which in 2D terms means fading from top to bottom. The same applies to any horizontal reflective surface such as a shiny floor or table, and to curved surfaces where reflection is stronger toward the sides.
Break them up correctly. As surface irregularity increases, the reflected image loses definition, and at a certain point only the highest-intensity areas - light sources, the sky - remain as broken highlights. For small ripples, modulate or distort the reflection with animated undulating noise whose scale and intensity match the ripples and whose perspective matches too, getting smaller with distance. When waves are stronger, blur the reflected image before modulating it, so only the strongest illumination differences like the sun show up in the broken reflection, while finer detail like clouds does not.

Creating shadows in 2D

Shadow mattes and AO only work for CG elements lit and rendered in 3D. For 2D elements the angle, perspective, falloff and softness must all be approximated manually, as in traditional art. The first step is always to study the shadows already in the footage and estimate the location and character of the light sources, even if they are not in frame. Easy in an outdoor daytime scene, since you only need the sun's position; much harder at night or indoors with several sources.

Shape, angle and stretching. The element's alpha channel is the starting shape, then transformed, scaled, rotated and skewed for correct perspective and stretching. Easy on a still, much harder when the element moves. For a walking actor, the shadow must always stick to the feet where they touch the ground. It is more important that the shadow sticks to the contact areas than that every small detail of its shape is right - we are far more susceptible to floating shadows than to missing detail. For sunlight, gauge the sun's position on an imaginary half dome and trace a straight line from it through the subject at its contact point; angle and length should stay consistent across the scene regardless of where elements are. For any other light, the angle changes with the subject's position relative to the source, so animate the line with its origin tracking the light and its pivot tracking the subject.
Softness and falloff. It is usually not necessary to consider every factor from Chapter 8 - matching the overall softness and falloff of existing shadows in the footage is enough. If the footage gives no reference, consider the scale of the lights: large lights give soft shadows, small focused ones give sharp shadows with a narrow penumbra. Blur the shadow matte using local masks. When the light comes at a steep angle the penumbra is fan-shaped, and the falloff happens gradually along the sides as well as the length.
Breaking up shadows. The hardest part is matching the shadow's shape and deformation to the underlying environment. Easy on flat even ground, hard on an irregular surface. On relatively flat but irregular ground - dirt, pebbles, grass, water - breaking up the shadow matte with simple noise is often quite effective, though not an accurate representation. It is not a solution for large irregularities like stairs, or where the shadow falls over distinct objects; there the roto shape has to be animated to reflect the deformations.
Shadow color - the mistake to avoid. Junior compositors overlay the shadow as if it were a separate element, which forces a consistent color when the shadow's color actually depends on the color of the surface. Color-correcting the footage through a matte is the preferable method, because it preserves the underlying hue. But simply gaining down may push the blacks too low, so it is usually necessary to gain down while slightly lifting the blacks and reducing saturation. Tint the shadow toward the hue of the remaining lights - in daylight that means the sky color, at night or indoors the combined color of whatever is left, which is rarely as cool as a daytime exterior. Keep the tinting subtle: the slight bluishness of shadowed areas on a clear day is nowhere near as saturated as the sky itself. And since our eyes are sensitive to black mismatches, always check the blacks in the shadows you create against existing shadows in the footage.
Contact and proximity shadows in 2D are created by dilating and blurring the element's alpha, or by drawing simple soft mask shapes. Closer means darker and sharper, so the intensity and softness must be animated to match the action - a walking actor's ground contact shadow is darker and sharper when the foot is down, softer and lighter as it lifts. Compositors focus on the obvious contact areas and often neglect the subtler proximity occlusion, but a very soft, very subtle darkening of a wider area underneath a walking actor genuinely helps integration - an almost imperceptible effect that goes a long way.

Atmospheric depth

Photorealism relies strongly on accurate depth, and imagery falls apart when aerial perspective is off or inconsistent. In VFX it is typically added at comp even when fog or smoke were rendered in 3D, because the final adjustment needs to happen interactively in the context of the finished image. For CG, the Z depth pass is the base. For 2D elements, artists must rely on their eye, or manually slice depths by painting masks or rotoscoping.

Referencing the footage. Assess the element's distance from camera based on how deep it sits in the shot, then find areas of the footage at the same depth to use as reference. Adding a building to a photo of a city? Existing buildings at a similar depth tell you the amount of hazing. Look at the blacks first - not just how lifted they are, but their specific RGB balance, and temporarily raise the viewing gamma to see into them. Matching the almost imperceptible color detail in the low range is the key step in matching aerial perspective.

When the footage gives no depth clues - elements placed deeper than anything present, or a fully CG scene - the sky is the best guide. Lifting distant elements toward the color of the sky is the basic way to simulate aerial perspective. But since humidity, dust and aerosols congregate in the lower atmosphere, and most distant elements sit near the horizon, push toward the horizon color rather than the zenith. For very large distant elements like mountains it is not enough to push blacks and midtones; atmospheric components like low-lying cloud or fog may partly or completely obstruct distant elements, and their interaction has to be shown - fog blanketing the foothills and dissipating higher up, leaving the peaks clearer than the base. Treat the sky as a true atmosphere - a three-dimensional volume that engulfs the scene and interacts with it - rather than a 2D background sitting behind it.

And the technical rule that makes it work: while lifting the low end for atmospheric depth, do not brighten the higher midtones and highlights. Clean air leaves highlights intact and haze may dim them, but highlights never get brighter because of the atmosphere. So lift or gamma are the right operations for depth grading. An effective technique: grab the sky from the footage, blur it enough to erase any discernible detail, and subtly screen it over the composited elements. It must be a screen merge (A+B-AB), not an additive one (A+B), because screen protects the top range and prevents highlights from brightening while still lifting the low end and lower midtones.

↑ Contents

Chapter 18: Lens and Camera Effects

Dinur's framing: the significance of adding lens and camera characteristics at the compositing stage cannot be overstated. This is not the icing on the cake - this is the essence of photorealism. Even the most realistic CG render feels artificial without the appropriate touch of defocus, distortion, flares and grain. Lens effects are the final glue binding all the elements in a matte painting or composite, because they strengthen the illusion that everything in the image was physically captured by a camera.

Chapter 10 described the phenomena. This chapter is about applying them.

Defocus. 3D depth of field is the most accurate and avoids Z-depth's problems, but it is baked in and cannot be adjusted later, and it needs higher sampling and slower renders. 2D defocus is far more flexible - focal point, depth of field and defocus amount all tweakable in real time - and is an efficient alternative as long as you know the Z depth limitations, particularly transparent and volumetric objects. Compositing defocus tools usually let you set the number of polygons to match the aperture blades, adjust their curvature, add highlights bloom with a variable threshold, gamma shift, and change the defocus aspect ratio for anamorphic. In a fully 2D workflow there is no depth pass, so convincing depth of field needs a good eye and a clear notion of where each element sits in 3D space; focus gradation can be somewhat simulated with soft masks. When integrating into live action, focus level and bokeh type should be matched to areas at a comparable depth.
Chromatic aberration is easily replicated by slightly offsetting one or two of the RGB channels - 1 to 4 pixels is usually enough for typical fringing, more for stylised effect. Since transverse aberration is stronger toward the frame edges, simulate it by scaling the selected RGB channels from the center of the frame rather than offsetting them uniformly.
Lens distortion is indispensable for camera tracking, and some tools offer elaborate models mimicking real spherical and anamorphic lenses as well as rectilinear and fish-eye projections. Beyond the technical role, it adds a great deal of realism to fully CG games and static renders, as long as it stays within a plausible range that matches the context.
Lens flares serve a triple purpose: visual drama, a sense of a physical lens, and tying elements together, since a flare happens over everything in the image. The characteristics from Chapter 10 have to be implemented in context - movement and intensity fluctuations tied to the movement of the relevant light sources, and random disturbances and imperfections added to avoid an overly clean look. Two routes: procedural generation gives accurate motion and full control over shape and color, but is too clean and perfect out of the box and needs degrading and varying through noise or texture modulation - the extra effort is the difference between a tacky effect and believable visuals. Real footage carries all the photoreal detail and imperfection built in, but matching the motion of a full flare to a specific shot is difficult, so stock footage of single isolated flare components is usually the most useful, because you can combine and animate them in context.
Motion blur. 3D motion blur is baked in like defocus, so it is often more practical to render a motion vector pass alongside the beauty and apply blur in comp, which shortens render times and allows fine-tuning against the footage. The gotcha: applying motion blur to a complete render produces bad edges and artifacts where moving and static objects overlap, or where objects move at different speeds and directions - so render in separate layers. For 2D elements, motion blur is limited to two axes, but Z-axis motion blur is generally less noticeable, and 2D vector information can be generated by tracking the element. Where tracking fails, fake it with a directional blur, keyframing amount and angle to match. Extracted and rotoscoped edges often need added motion blur to match the original footage.
Grain is added last, on top of all the components, and its contribution is as crucial as it is subtle. Matching and adding grain often requires removing previous grain from different elements first, to prevent mismatched, doubled or frozen grain. Degraining is a balancing act between removing as much grain as possible and preserving sharpness and detail. Grain size and softness should be matched and adjusted separately for each RGB channel, and both dark and bright areas examined. Grain in the blacks is particularly important - artificially added grain does not show well in the blacks, yet in both digital and film the grain is usually stronger there, so most grain tools offer a separate control for it. Reference is the surrounding shots in VFX, or the desired camera or film stock in visualization and games.

↑ Contents

Terms from Part 4

Spill suppression - removing the screen's hue from semi-transparent edges and bounce areas.
Core and edge mattes - two separate extractions, hard for the body and soft for the edges.
Edge reconstruction - recreating soft edges rather than extracting them.
Matte-less extraction - modulating the background directly with add or multiply, for fine detail.
Light wrap - light bleeding over an edge from a bright background. Frequently misused.
Render pass / AOV - a separated component of a render.
Beauty pass - the complete rendered image.
Position (P) and Normal (N) passes - 3D-to-2D translators storing XYZ location and facing direction in RGB.
Cryptomatte - automatic per-object and per-material ID mattes.
Z depth - per-pixel distance from camera. Cannot represent transparency, volumes or reflections.
Deep compositing - multiple depth samples per pixel, making layer order irrelevant.
Witness camera - a second camera shot from the reflection's point of view.
Motion vector pass - per-pixel motion data for applying motion blur in comp.
Degraining / denoising - removing grain before re-matching it.

↑ Contents

Check yourself

What are Dinur's two pieces of advice to compositing students, and why is the second one the important one?
Why does a semi-transparent edge cause trouble even over a clean green screen?
Why pull two mattes instead of one?
When is light wrap appropriate, and when is it being misused?
What can you do with an albedo pass that lift and gamma cannot?
Why is ambient occlusion the right pass for a contact shadow?
Name three things a Z depth pass cannot represent.
Why is flipping the background a technically wrong but usually acceptable lake reflection?
What is wrong with overlaying a gray shape to make a shadow, and what should you do instead?
When adding atmospheric depth, which part of the tonal range must not move, and which merge operation protects it?
Why should chromatic aberration be scaled from the center rather than offset uniformly?
Where is grain strongest, and why does that make it hard to add convincingly?

↑ Contents