Chapter 35: Compositing and Visual Effects

Overview

Here is a fact that should reframe this entire chapter before you read a word of technique: most of the visual effects you have ever seen, you have never seen. Not the spaceships and the dragons — those you noticed. The ones that did their job are the ones you walked right past. The clean sky behind a talking head that was actually a grey parking structure. The phone screen showing a flawless app that, on set, was a green rectangle. The wide establishing shot of a city street with the film crew's own vans painted out of the reflections. The two twins in a drama who were one actor. Effects done well are invisible by definition, because the instant you notice an effect, part of your brain stops believing the picture — and a video that the audience stops believing is a video that has failed.

That is the whole ethic of this chapter, and it is the same ethic that has run through the entire book, now pointed at the most tempting toys in post-production. Motivate every choice. A composite, a key, a stock element, or a screen replacement is worth doing only when it serves the story the piece is telling — and it is worth doing well only so that the audience feels the result without ever catching the method. The beginner reaches for effects to show off. The professional reaches for them to solve a problem the story has, and then spends most of the effort hiding that an effect was used at all.

The other idea to install now is the one this book has repeated in every room: the finish begins on set. A clean composite is not rescued at the timeline; it is lit and shot to be composited, and then finished in post. A green-screen key that fails did most of its failing on the day, under bad light, with a wrinkled cloth and the subject standing too close. Get the shoot right and the key is easy. Get the shoot wrong and no plugin on earth will fully save it. So this chapter, like the ones on light and sound before it, is really two chapters braided together: how to shoot an effect, and how to finish it.

By the end, you will be able to light and shoot a clean chroma key, pull and clean a matte, cut a subject out by hand when there is no screen, drop a licensed stock element or a replacement screen into a shot so it belongs there — and, just as importantly, decide when not to use an effect at all. And because compositing is the craft of convincingly faking what a camera saw, this chapter ends where that power has to end: with consent, honesty, and the ethics of synthetic media, which are not an afterthought here but the point.

In this chapter you will learn to:

• Explain compositing and the alpha channel — how a foreground, its matte, and a background stack into one seamless image.
• Light and shoot a green screen for a clean chroma key, treating the screen and the subject as two separate lighting jobs.
• Perform keying: pull a matte, despill the edges, and clean the matte without destroying the hair.
• Use a mask and roto (rotoscoping) to isolate and repair a shot when nothing green was ever there.
• Composite stock elements and replace screens so they match the plate — and understand where VFX helps a story and where it wrecks it.
• Apply the ethics of synthetic media and deepfakes: consent, no deceptive impersonation, and disclosure.

Learning Paths

This is an advanced, optional-in-parts chapter — but the judgment in it is for everyone. Most beginners never need a green screen; every maker needs to know when an effect is a good idea, and every maker needs the ethics in §35.6.

• 📱 Phone-first: §35.1 (what a composite is) and §35.6 (ethics) are your core, and both are free. A cheap green cloth, even light, and a free keyer will get you a real key — §35.2 and §35.3 show you how without buying anything. Skip the deep matte mechanics until you actually need them.
• 🎥 Creator: §35.5 (stock elements and screen replacements) is the highest-return section for a channel — sky replacements, screen inserts, and clean licensed overlays are everyday tools. §35.6 keeps you out of trouble as AI tools land in your editor.
• 💼 Pro-track: all of it. Clients ask for keys, screen replacements, and clean-ups constantly, and the money is in making them invisible and licensed. §35.2–35.4 are your production and finish craft; §35.6 is your liability and your reputation.
• 🎓 Student: §35.1 gives you the vocabulary (composite, alpha, matte, key) and §35.6 gives you the framework every media-ethics discussion now turns on. Case Study 1 is your analysis anchor; Case Study 2 is the shoot to reproduce.

35.1 What compositing is; layers and the alpha

Start with the most ordinary version of the thing, because the exotic version only scales it up. You have a shot of a person talking to camera in a plain room, and you have a second image — a clean, branded background, or a city skyline, or a softly-lit studio that does not exist. You want the person to appear to be in that second image. The act of combining them so the seam disappears is the whole subject of this chapter.

Compositing is the craft of combining two or more image layers into a single shot so that they read as one photographed image. The key words are layers and read as one. You are stacking pictures — a foreground over a background, often with several elements in between — and your entire job is to make the stack look like a single thing a camera saw, not a collage. Every technique in this chapter, from a green-screen key to painting out a stray light stand, is a way of stacking layers and hiding the joins.

Sitting above compositing is the broader word. VFX — visual effects — is any imagery in the final frame that was created or altered outside the live moment of filming: composites, computer-generated imagery, set extensions, screen replacements, removals and clean-ups, matte paintings, particle elements. The useful distinction to hold is between visual effects and practical (or "special") effects. A practical effect happens in front of the camera on the day — real rain from a rig, a real spark, a real puff of stage smoke. A visual effect is added after, in post. This book is about production and finishing, so we live in the VFX side; but the best VFX supervisors will tell you the same thing this book keeps saying — do it for real when you can, and reserve the effect for when reality can't be filmed, controlled, or afforded.

Now the single most important technical idea in the chapter, the one that makes compositing possible at all: the alpha channel. A normal color image has three channels — red, green, and blue — that together say what color each pixel is. A composited layer carries a fourth channel, alpha, that says how transparent each pixel is. Alpha is not a color; it is an opacity map, and it is easiest to understand when you look at it on its own, rendered as a black-and-white image. That black-and-white image is called the matte (some traditions call it the "mask" or "key" — we will keep matte for the transparency image and use mask for a shape you draw, in §35.4). White in the matte means fully opaque: keep this pixel. Black means fully transparent: drop it, and let whatever is on the layer below show through. And the greys in between — this is where all the craft lives — mean partial transparency: a pixel that is half this layer and half the one beneath. Greys are how you render the soft, feathery, semi-transparent things that betray a bad composite when they are wrong: the fine hair at the edge of a head, the motion blur on a moving arm, the smoke, the glass, the soft edge of anything real.

FIGURE 35.1 — A composite is layers stacked through their alpha (the matte decides what shows)

  WHAT YOU SEE (final)          =   FOREGROUND       through its ALPHA      OVER    BACKGROUND
  ┌────────────────────┐            ┌────────────┐       ┌────────────┐             ┌────────────┐
  │  subject in the    │            │  subject   │       │ WHITE=keep │             │  new room  │
  │  new room, edges   │    =       │  as shot   │   ×   │ (opaque)   │      +      │  (plate,   │
  │  clean, sitting in │            │  on green  │       │ BLACK=drop │             │  stock, or │
  │  the room's light  │            │  (R,G,B)   │       │ GREY=edge  │             │  a graphic)│
  └────────────────────┘            └────────────┘       └────────────┘             └────────────┘
      the illusion                    layer 2               its matte                 layer 1
                                      (on top)          (the alpha channel)           (underneath)

  THE MATTE (alpha) read as black-and-white:
     ██████████████    WHITE  = fully opaque   → the foreground shows 100%
     ··············    BLACK  = fully clear    → the background shows through
     ▒▒ edge ▒▒        GREY   = partial        → hair, motion blur, glass, smoke (soft edges)

  Every layer carries its own alpha. Compositing = stacking layers and letting each matte decide what is
  kept, dropped, or blended. The black and the white are easy; the GREY edges are where the seam hides.

Read the figure as a recipe you will run over and over: take a foreground, generate a matte that says which parts of it to keep, and lay it over a background. When people say a composite "looks fake," they are almost never reacting to the black or the white — those are trivial. They are reacting to the greys: a hard, cut-out edge where soft hair should be; a green halo the matte left behind; an edge that buzzes and chatters frame to frame because the matte can't decide. Master the greys and you master compositing. Everything in §35.2 and §35.3 is, at bottom, a fight to get those edge greys right — and most of that fight is won on set.

One more principle before we shoot anything, because it governs every choice downstream. A composite has to agree with itself. The foreground and the background must share the same light (direction, quality, color, contrast — everything you learned in Chapters 11 and 12), the same lens feel (focal length and depth of field, Chapter 4), the same camera height and angle, the same motion (a locked-off foreground over a moving background will never sit right), and, after everything, the same grade (Chapter 32) and the same grain. A subject lit hard and cool, dropped over a background lit soft and warm, will look wrong even if the matte is flawless — because the human eye reads light before it reads edges. This is why compositing is a Part VII skill sitting on top of everything in Parts I–III: you cannot fake a shot into a world whose light you never learned to see.

🚪 Threshold Concept: the best visual effect is the one you never notice. For your whole life, "special effects" meant the loud stuff — the explosions and the monsters. The truth of the working craft is the opposite. The overwhelming majority of VFX shots in professional film and video are invisible: a removed sign, a replaced sky, a cleaned-up reflection, a joined-together take, a screen filled in. Once you accept that the goal is disappearance, not spectacle, your judgment changes permanently. You stop asking "what cool effect can I add?" and start asking "what problem does the story have that only an effect can solve, and how do I make sure no one can tell I solved it that way?" That question is the entire professional posture toward effects, and it is free.

💡 Why It Works: alpha turns a picture into a cut-out you can stack. The reason a fourth channel changes everything is that opacity, stored per-pixel, lets a flat rectangle behave like an irregular, soft-edged cut-out without actually being cut. The computer doesn't "know" where the subject ends; the matte tells it, pixel by pixel, and the greys let the boundary be soft instead of a scissor-cut. That is why generating a good matte is 90% of compositing: the matte is the shape of the illusion. Get it right and the stack looks photographed; get it wrong at the greys and it looks like paper glued on glass.

🔄 Check Your Eye. 1. In one sentence each, what do the R/G/B channels of an image tell you, and what does the alpha channel tell you? 2. In a matte, what do white, black, and grey each mean? 3. A composite has a technically perfect edge but still "looks fake." Name two non-edge things that are probably mismatched between the foreground and background.

Check yourself

1. R/G/B say what color each pixel is; alpha says how transparent (opaque vs. see-through) each pixel is.
2. White = fully opaque (keep the foreground); black = fully transparent (show the background); grey = partial (a soft or semi-transparent edge — hair, motion blur, glass).
3. Any two of: light (direction, quality, color, contrast), lens/depth of field, camera height/angle, motion, grade/color, or grain. The eye reads light before edges, so a lighting mismatch reads as "fake" even with a clean matte.

35.2 Green screen: lighting and shooting the key

If you want a subject you can drop into any background, the oldest and most reliable trick is to film them in front of a color the software can find and delete. That color is your green screen (or blue screen): a solid, evenly-colored backdrop, placed behind the subject, chosen and lit so that in post you can identify its color and make it transparent. The technique of keying on a color is chroma key — "chroma" meaning color — and the two terms travel together: you shoot a green screen in order to chroma key it out.

Why green? Three real reasons, and they are worth knowing because they explain when to break the rule. First, green is far from human skin tones — skin lives in the red-orange range, with very little green in it, so keying out green is unlikely to also key out your subject's face. Second, digital sensors see green best: the common Bayer color filter over a sensor has twice as many green photosites as red or blue, so the green channel is the sharpest, least-noisy channel you record — which means a green backdrop gives the cleanest edges to key against. Third, green needs less light than blue to reach a good, saturated value. Blue screens still have their place: you switch to blue when your subject has to be green (foliage, a green product, green wardrobe), or when you want the subtler spill that blue tends to throw. But green is the default, and for a beginner it is almost always the right call.

Now the sentence that this whole chapter is built on: a clean key is lit and shot, not fixed in post. The most common misconception about green screen is that it is a post-production effect. It is a production discipline that gets finished in post. And the discipline comes down to seeing that you have not one lighting problem but two, and that you must solve them separately.

Problem one: light the screen flat and even. The keyer in post works by selecting a color. If your screen is a single, uniform green value across its whole surface, the keyer has one easy color to find. But a green cloth lit by one lamp is not one value — it is bright where the lamp hits, dark in the corners, and darker still wherever a wrinkle throws a tiny shadow. Now the "green" you have to key ranges from near-black to near-white, and the keyer either misses the dark parts (leaving green patches) or grabs too wide a range (eating your subject's edges). So you light the screen as its own job: two soft sources, one from each side at roughly 45 degrees, equal power, aimed to wash the cloth as evenly as you can, with no hot center and no dark corners. And you get the cloth taut — stretched, clamped, or steamed flat — because every wrinkle is a shadow and every shadow is a second green value you did not want. A painted, matte, wrinkle-free green wall is the gold standard precisely because it is perfectly flat and even; a bedsheet safety-pinned to a curtain rail is the hard-mode version of the same idea.

Problem two: light the subject for the world they will land in. This is the half beginners forget, and it is where "motivate every choice" comes back with full force. The subject is not being lit for the green screen; the subject is being lit for the background plate — the image you are going to composite them into. If the plate is a warm, window-lit room (say, the café from earlier chapters, with its soft daylight key from Chapter 13), then the subject must be lit warm and soft from the matching direction, or they will look pasted onto a place they are obviously not standing in. So you light the subject with the full grammar of Chapter 11 — a key from the plate's key direction, fill to the plate's contrast ratio, a rim to separate them — and you keep that light off the screen, and the screen's light off them.

That last clause is the crux, and it has a name: spill. Spill is green light bouncing off the screen and landing on your subject — a green fringe on the hair, a green sheen on the shoulders, a green kick under the chin. Spill is the number-one enemy of a clean key, because it means the exact color you are trying to delete is now on your subject, so deleting it eats into them. You fight spill three ways, all on set: put distance between subject and screen (six to ten feet if the room allows), so the bounced green falls off before it reaches them; add a rim or back light (Chapter 11) to overpower the green edge with a clean, motivated one; and, if you can, flag the screen's light so it doesn't wrap around onto the subject. Distance also solves a second on-set problem for free: a subject standing close to the screen casts a shadow onto it, wrecking your flat, even value. Move them forward and the shadow falls on the floor, out of the key.

FIGURE 35.2 — Green-screen setup: light the SCREEN and the SUBJECT as two separate jobs (top-down)

        [========= GREEN SCREEN — taut, matte, wrinkle-free =========]
           ↑                                                    ↑
           │   even wash, both sides, equal power                │
        ▢ ─┘   (screen lights aimed ONLY at the screen)          └─ ▢
      (screen L)                                                 (screen R)

                     ←──── 6–10 ft of separation ────→
                        (distance kills spill AND the subject's shadow)

                               ▣  rim / back light
                                ╲   (a clean edge that overpowers
                                 ▼   the green spill; separation)
                              ( S )  ← subject lit for the PLATE, not the screen
                            ↗        ↖
                       ▢ key           ○ fill
                    (from the plate's   (to the plate's
                     key direction)      contrast ratio, Ch.11)
                               │
                             [CAM]  match lens, height, and depth of field to the background plate

  Two lighting problems, solved apart: (1) the SCREEN — flat, even, one green value, easy to key;
  (2) the SUBJECT — lit to match the world they'll be dropped into, with a rim to fight spill.
  DISTANCE is the free tool that keeps the two apart. Never light the subject with the screen's light.

Walk the figure and you can see the whole shoot. The two screen lights (▢) do nothing but wash the cloth evenly. The subject stands well forward of it. Their key, fill, and rim (▢ ○ ▣) are aimed at them, matched to the plate, and kept off the screen. The camera matches the plate's lens and height. Nothing here is expensive; the entire method is organization of light in space, and a phone shooter with two cheap LED panels, a clamp light for a rim, and a steamed green cloth can execute every line of it.

🎒 Gear Note: the cheapest key that works. You do not need a studio. A green screen can be a few dollars of matte green fabric (get the wrinkles out — steam it, stretch it, or clamp it taut), or a can of chroma-green matte paint on a spare wall, which is the best value in all of VFX. For light, two soft sources on the screen and any motivated light on your subject will do — window light (Chapter 13) can even be your subject key if the plate matches it. The two things money cannot replace are distance (put the subject well forward of the screen) and evenness (a flat, wrinkle-free, evenly-lit screen). Phone-first path: shoot in the flattest, highest-quality mode your phone offers (see the Tech box in §35.3 on why the recording format matters), get your separation, and key it in DaVinci Resolve, which is free. The discipline is identical from a phone to a soundstage; only the ceiling changes.

⚙️ Settings Box: shooting a green-screen key (a starting point to adapt). Match the plate first; these are defaults, not laws.

text SCREEN matte green, TAUT and wrinkle-free; subject 6–10 ft in front SCREEN LIGHT two soft sources, ~45° each side, equal power, flat even wash, no hotspots SUBJECT LIGHT key + fill + RIM, matched to the background plate's direction/quality/color (Ch.11–13) EXPOSURE screen a clean mid green — DON'T overexpose it; check it reads even on the waveform (Ch.5) SHUTTER 180° rule (Ch.2): motion blur must MATCH the plate — 1/50 s at 24 fps LENS / DoF match the plate's focal length and depth of field (Ch.4); a locked-off camera is easiest WHITE BALANCE set it correctly and LOCK it (Ch.12) — don't let auto WB drift the green mid-shot CODEC the least-compressed, highest chroma option you have — 4:2:2 or better keys far cleaner than 4:2:0 (Ch.3); shoot the best quality your camera offers for anything you'll key

The two settings beginners get wrong: they overexpose the screen (a blown, uneven screen keys worse, not better), and they shoot a heavily-compressed format that turns the green edges to mush. Expose the screen to a clean, even mid-green, and capture the best quality you can.

⚠️ Common Mistake: subject too close to the screen. It is the error that causes more bad keys than everything else combined. Pressed up against the cloth, the subject (a) is bathed in green spill that the key then eats out of their edges, and (b) casts a shadow onto the screen that destroys its even value. The fix is free and physical: move them forward. Six to ten feet of separation removes most spill and drops the shadow onto the floor, out of the key. If your room is too small for that, add a rim light and flag the screen's spill — but distance first, always. You cannot despill your way out of a subject who was standing in a green room.

🔄 Check Your Eye. 1. Why is green the default keying color? Give two of the three reasons. 2. Green screen is often called a post effect. Correct that: what are the two lighting jobs you must solve on set? 3. What is spill, and name two on-set ways to reduce it.

Check yourself

1. Any two of: green is far from skin tones (so keying it won't grab faces); sensors resolve green best (twice as many green photosites in a Bayer array → sharpest, cleanest channel); green needs less light than blue to reach a good value.
2. (1) Light the screen flat and even — one uniform green value — and (2) light the subject separately, matched to the background plate they'll be composited into.
3. Spill = green light bouncing off the screen onto the subject, contaminating their edges. Reduce it with distance (subject well forward of the screen), a rim/back light that overpowers the green edge, and flagging the screen's light off the subject.

35.3 Keying and cleaning the matte

Now the footage is on the timeline and it is time to make the green disappear. Keying is the post-production process of generating a matte from your footage — most often by selecting the chroma-key color and turning it transparent — and then refining that matte until the subject's edges are clean enough to sit over a new background. Keying is not one click; it is a short, ordered sequence of moves, and doing them in order is the difference between a clean composite and a green-fringed cut-out.

The good news is that you already understand the core tool. Back in Chapter 32 you used a qualifier to isolate a color and grade only it — click the skin, and the software selects every pixel in that hue-saturation-luma range. A keyer is exactly the same machinery pointed at a different goal: instead of grading the selected color, you make it transparent. Pulling a key is qualifying the green and turning the selection into a matte. If you can build a qualifier, you can pull a key.

Work it in this order:

1. Pull the key. Point the keyer at the green and let it generate a first matte. In DaVinci Resolve, the free tool this book anchors to, this is the "3D Keyer" or the dedicated "Delta Keyer" node — you sample the green with a few strokes and it builds the initial transparency (other editors call their version a "Chroma Key," "Keyer," or "Ultra Key" effect; Appendix E has the translations). Do not judge it yet in color. Switch the viewer to show the matte in black-and-white — this is the single most important habit in keying, because the matte is the truth and the color view flatters you.

2. Read the matte. In black-and-white, you want three things: the background is pure black (fully transparent), the subject is pure white (fully opaque), and only the genuine soft edges — hair, motion blur — are grey. If the "black" background is actually dark grey with texture, the screen wasn't even and some green will show through as a haze. If the "white" subject has grey holes in it (a green shirt, a reflective forehead), the key grabbed part of your subject. If the edges are a fat band of grey, the key is soft and floaty. You diagnose all of this in black-and-white before you ever look at the composite in color.

3. Despill. This is the step that separates a real key from a first-day key. Despill is removing the green light that contaminated the subject's edges — that green rim on the hair and shoulders you fought on set but never fully beat. Despill replaces the green fringe with a plausible edge color (usually pulled from the subject or neutralized), so the edge doesn't glow radioactive against the new background. Almost every keyer has a despill or spill-suppression control; use it. A composite where the subject has a faint green halo screams "green screen" louder than any other single tell, and despill is the cure.

4. Clean the matte. Now you refine the black and white without wrecking the grey. Three gentle moves: clip the black (push the almost-black background to fully transparent so no haze remains), clip the white (push the almost-white subject to fully opaque so they don't go semi-transparent and ghost the background through their chest), and shrink and soften the edge a hair — pulling the matte in by a pixel or two hides any last green fringe, and a slight blur on the matte edge lets the subject sit into the background instead of sitting on top of it like a sticker. The discipline here is gentleness: every one of these controls, pushed too far, eats the hair. If the fine strands at the top of the head turn into a hard helmet, you clipped too hard. Back off until the hair breathes again.

5. Garbage matte, then composite. Before the keyer even runs, it helps to crop out everything that was never green — the edge of the cloth, a light stand, the C-stand at the frame edge — with a rough drawn shape called a garbage matte (it is just a mask, §35.4). That way the keyer only has to deal with actual green. Finally, lay the keyed foreground over the background plate, and now judge it in color, against the real background, because edges that looked fine over black can fringe against a bright sky.

FIGURE 35.3 — Pulling and cleaning a key (signal flows left → right; here on Resolve's Fusion page)

  [ MediaIn:  ]     [  KEYER  ]        [ DESPILL ]        [ MATTE       ]      [ MERGE  ]      [ result  ]
  [ green-    ]  →  [ sample   ]   →   [ remove the ]  →  [ CLEAN-UP:   ]  →   [ FG over  ]  → [ over the]
  [ screen FG ]     [ green →   ]      [ green that ]     [ clip black, ]      [ BG plate ]    [ new     ]
  [           ]     [ make matte]      [ tints edges]     [ shrink,     ]      [          ]    [ world   ]
                                                          [ soften edge ]           ▲
                     ▲ garbage matte (a mask, §35.4) crops out stands and          │  [ MediaIn: BG plate ]
                       the cloth edge BEFORE the keyer, so it only sees green.      └──( the plate: room,
                                                                                        stock, or a graphic )

  CHECK THE MATTE IN BLACK-AND-WHITE at every step:
     background must be PURE BLACK       subject must be PURE WHITE        only hair/edges GREY
     green fringe on the edges?  →  despill harder.
     chewed-up, helmet hair?     →  you clipped/shrank the matte too far — back off.
     buzzing, chattering edge?   →  the screen wasn't even; fix on set next time, soften slightly now.

The figure is also a warning: notice how much of it points back to the shoot. A buzzing, chattering matte — edges that shimmer frame to frame — almost always means the screen was lit unevenly or the format was over-compressed. You can soften it a little in post, but the real fix was on set, which is the chapter's refrain. The cleaner the shoot, the shorter this node chain and the more invisible the result.

One optional finishing touch worth naming because it sells the shot: a light wrap, letting a thin band of the background's light and color bleed onto the very edge of the foreground, the way light really would wrap around a person standing in that environment. It is subtle and easy to overdo, but a touch of it is often the last few percent that makes a composite read as photographed rather than assembled. And after everything, the composite still has to be graded together (Chapter 32): the foreground and background pushed to one shared look, one contrast, one color temperature, and — if the plate is grainy or noisy — a little matching grain added over the whole thing, because a clean subject over a grainy plate is a dead giveaway.

🔬 The Tech: why the recording format decides how well you can key. (Optional — skip it and every practical step still works.) Back in Chapter 3 you met chroma subsampling: to save data, most cameras record full brightness detail but throw away color detail, storing it at reduced resolution. In 4:2:0 (what phones and most consumer cameras shoot), color is recorded at roughly a quarter resolution — which means the green-to-subject boundary, a purely color edge, arrives blocky and smeared. Key that and the edges are coarse and hard to clean. In 4:2:2 the color is at half resolution, and the same edge keys noticeably cleaner; 4:4:4 keeps full color and keys cleanest of all. Bit depth matters too: more tonal steps (10-bit over 8-bit) mean smoother mattes with less banding. This is the deepest reason "the key starts on set": if you know a shot will be keyed, capture the least-compressed, highest-chroma format your camera offers. You cannot add color detail in post that the codec discarded on the day.

🔗 Connection: a keyer is a qualifier with a different job. The chroma keyer in §35.3 and the color qualifier you used to protect skin in Chapter 32 (§32.4) are the same underlying tool: both isolate a range of hue, saturation, and luma. The qualifier feeds that selection to a grade; the keyer feeds it to transparency. If you got comfortable qualifying skin on the vectorscope's flesh-tone line, you already have the instinct for pulling a key — you are selecting a color precisely and refining the selection's edges. Same machinery, one aimed at color, one aimed at alpha.

🔄 Check Your Eye. 1. Why should you judge a key in black-and-white (the matte view) before you judge it in color? 2. What is despill, and why is it the step that most separates a clean key from an amateur one? 3. Your keyed subject's hair has turned into a hard "helmet" with no fine strands. Which cleanup move did you overdo, and what do you do?

Check yourself

1. The matte is the truth of your transparency: in black-and-white you can see whether the background is pure black, the subject pure white, and only the real edges grey. The color view hides haze and semi-transparency that the matte reveals.
2. Despill removes the green light contaminating the subject's edges (the green rim/halo). It's decisive because a green fringe is the single loudest "green screen!" tell, and no amount of matte cleanup fixes edge color — only despill does.
3. You clipped/shrank/eroded the matte too far (or the edge blur too hard), eating the fine hair. Back the cleanup controls off until the strands read as strands again; accept a hair of softness rather than a helmet.

35.4 Masking, rotoscoping, and clean-ups

Green screen is a luxury you plan for. Most of the compositing you will actually do in a working life happens on footage that has no green in it at all — a shot you already have, where you need to isolate or repair part of the frame. For that you draw shapes by hand.

A mask is a shape you draw to isolate a region of the frame — to include it or exclude it. You have already used a cousin of this: the power window in Chapter 32 was a mask used to limit a grade. Here the mask does the compositing job — it says "this region, and not that one." A garbage matte is a mask. A shape drawn around a sign so you can paint it out is a mask. When the thing you want to isolate holds still, a single drawn mask is all you need.

But things move, and a mask drawn on frame one is wrong by frame two. Roto — short for rotoscoping — is drawing a mask and then animating it, frame by frame, so it follows a moving subject or object that cannot be keyed. It is the manual version of a matte: where a chroma key generates transparency automatically from a color, roto generates it by hand from a shape you keep adjusting as the subject moves. You set the shape, step forward a few frames, adjust it to the subject's new position with keyframes (the same keyframes you met in Chapter 34), and repeat — for the length of the shot. It is exactly as laborious as it sounds, and that labor is the entire reason green screen exists: a green screen is a way of buying yourself out of roto. When you can plan and light a key, you do; when you're handed footage that was never shot for it, you roto.

Where all of this pays off invisibly is in clean-ups — the removals and repairs that make up the silent majority of professional VFX work:

• Rig and object removal: a boom mic that dipped into the top of frame, a light stand reflected in a window, a crew member caught in a car door, a modern road sign in a period piece, a brand logo you don't have the rights to show, a blemish or a wire.
• Paint-outs and patches: covering the unwanted thing with clean pixels — cloned from another part of the frame, or borrowed from a clean plate (an extra shot of the empty location with the offending object gone — the single most useful thing you can grab on set for any removal).
• Tracking the patch: if the camera moves, the patch has to move with it. Motion tracking attaches your patch (or a replacement element) to a point or a plane in the shot so it sticks frame to frame instead of floating. A patch that doesn't track is the classic tell of a rushed removal.

The craft in a clean-up is the same craft as in a key: hide the seam. A patch has to match the grain, the noise, the motion blur, and the grade of everything around it, or it sits on the surface of the shot like a smudge. The most common failure is a patch that is too clean — a smooth, noiseless rectangle over grainy footage — which is why matching grain is the last step of nearly every removal.

🎞️ Read This Sequence. Here is the chapter's anchor composite rendered as a Described Shot — the branded talking-head you'll build for Project 3. Read it as a target: an effect that has disappeared into a place.

FIGURE 35.4 — "The talking head that isn't where it looks"   [constructed teaching example — Project 3]
  THE FRAME    Medium shot of a spokesperson to camera, framed right-of-center on the thirds, a clean,
               softly-lit branded studio behind them with the product on a plinth. It reads as one room.
  THE MOVE     Locked off. (Static is the easy mode for compositing — nothing has to track.) The stillness
               also suits a calm, trustworthy product pitch.
  THE LIGHT    Soft key camera-left, gentle fill, a clean rim separating hair from the background — and,
               crucially, the SAME soft, slightly warm light is painted into the background plate, so the
               person and the "room" share one light. The rim was shot on set to beat green spill.
  THE SOUND    Clean lav dialogue (Ch.14), a low room-tone bed and soft music under it (Ch.33). Sound sells
               the space as much as the picture: a believable room needs a believable ambience.
  THE CUT      Cuts to a screen-replacement insert of the product's app (see §35.5), then back. The composite
               has to hold at the cut — the light and grade must match the inserts around it.
  THE EFFECT   The viewer never thinks "green screen." They think "nice studio." The effect has disappeared
               into a place, which is the entire goal — an invisible composite serving a motivated brief.
  THE LESSON   A composite succeeds when foreground and background share one light, one lens, one grade, and
               one grain — and when it's motivated (the brief wanted a clean branded space we didn't have).

Notice the lesson threaded through the figure: the composite is motivated (the branded brief needed a clean studio the budget didn't have), locked off (so nothing has to be tracked), and matched (one light, one grade). That is the anchor for your own Project 3 work — a spokesperson or product composited into a branded environment that reads as real because you shot the foreground to belong there.

✂️ In the Edit: shoot the clean plate, or pay for it in roto. Here is "shoot for the edit" pointed straight at VFX. If anything will be removed or replaced — a sign, a screen, a rig that keeps creeping into frame — grab a clean plate on the day: lock the camera, clear the frame of the offending object (or shoot a beat before it arrives), and roll a few seconds of the empty background. In post, that clean plate is a free, perfect patch that already matches the light, grain, and grade, because it is the same shot. Forget it, and you're rotoscoping and cloning the fix by hand, frame by frame, for hours. Ten seconds of clean plate on set is the cheapest VFX budget line there is — and the one beginners always skip. (Media note: keyed and composited shots are heavy; keep them organized and use proxies, Chapter 27, so your timeline stays alive.)

🔄 Check Your Eye. 1. What's the difference between a mask and roto (rotoscoping)? 2. Why does green screen exist, framed in terms of roto? 3. What is a clean plate, and why is grabbing one on set worth ten seconds of your shoot day?

Check yourself

1. A mask is a drawn shape isolating a region; roto is animating that mask frame-by-frame (with keyframes) to follow a moving subject that can't be keyed. A mask is a shape; roto is a shape over time.
2. Green screen is a way to avoid roto: it generates transparency automatically from a color, so you don't have to hand-draw and animate a matte around a moving subject. You green-screen the shots you can plan; you roto the ones you can't.
3. A clean plate is an extra shot of the empty location with the object-to-be-removed gone. It's a perfect patch that already matches the shot's light, grain, and grade — turning an hours-long paint-and-track job into a simple cover.

35.5 Stock elements and screen replacements

You do not have to film everything you composite. A huge amount of professional VFX is assembled from stock elements — pre-made, licensed image and footage assets you drop into a shot instead of shooting them yourself: skies and backgrounds, particle elements (smoke, fire, rain, snow, dust, sparks), atmospheric overlays (haze, god rays, light leaks, lens flares), textures, and clip footage. Need rain the day was too dry to have? A rain element. Need a dramatic sky behind a flat grey one? A sky plate. Need a warm flare across a lens for a transition? A flare element. These exist as licensed assets, and using them well is a core finishing skill.

The mechanical trick that makes many elements easy is the blend mode. Elements like fire, sparks, flares, and light leaks are usually delivered on a black background, and you composite them not with a matte but with a mode called screen (or "add"): screen mode makes black transparent and lets only the bright stuff through, so the fire or flare lays over your shot without a rectangle around it. (Dark elements use "multiply," the opposite.) You do not need the math; you need the habit — light-on-black elements go on a screen layer, and the black vanishes. It is the fastest, most reliable composite there is, which is why light leaks and flares are the first VFX most creators ever use.

Then there is the most useful invisible effect in commercial and product work: the screen replacement — replacing what appears on a phone, laptop, TV, or monitor in the shot with a clean image you control. It is everywhere, and for good reason. Real screens filmed by a real camera misbehave: they flicker and band because the screen's refresh rate fights the camera's shutter (the rolling shutter you met in Chapter 2), they throw moiré shimmer, they're often too dim or blown out, they catch glare — and, half the time, you didn't even have the final app or graphic ready on the day you shot. So you shoot the device with a clean, trackable screen (a solid color, or corner markers) and replace it in post.

The method is a corner-pin (or planar) track: you track the four corners of the screen as they move through the shot, pin your replacement image to those four points, and it locks to the screen frame by frame — tilting and sliding exactly as the real screen does. Then you sell it: add the screen's glow spilling onto the hand and face, a soft reflection across the glass, a hint of glare, and a touch of motion blur if the camera moves — because a real screen is not a flat, matte sticker; it is a light source with reflections. Skip those and the replacement reads as pasted-on. Add them and no one will ever know the phone was showing a green card on set.

FIGURE 35.5 — Screen replacement: track the corners, replace, then sell it with light

  THE PLATE (as shot)              THE TRACK                    THE COMPOSITE (finished)
  ┌────────────────────┐           ┌────────────────────┐       ┌────────────────────┐
  │  hand holds phone; │           │   +──────+  ← four  │       │  phone shows the   │
  │  screen: solid     │           │   │ track │   corner│       │  brand's real UI,  │
  │  color or markers, │    →      │   │ points│   pins  │   →   │  GLOW on the hand, │
  │  flicker / glare   │           │   +──────+   follow │       │  a soft reflection │
  │                    │           │   the moving screen │       │  + a hint of glare │
  └────────────────────┘           └────────────────────┘       └────────────────────┘
   why replace: rolling-shutter     corner-pin / planar track     add BACK the light the
   flicker & banding (Ch.2),        locks the new layer to the    screen would throw: glow,
   moiré, glare, dim screen —       four corners frame by frame,  reflection, glare, slight
   OR you lacked the final UI       so it tilts and slides with   motion blur → it reads as a
   on the shoot day                 the real device               real, glowing screen

Two disciplines make stock and screen work look professional instead of pasted. First, match the plate: a crisp 4K stock fire over grainy phone footage, a sky with the light coming from the wrong side, a screen replacement with no glow — each is a mismatch the eye catches instantly. Grade the element to the shot (Chapter 32), match its grain and motion blur, match its perspective and light direction. An unmatched element is worse than no element. Second, and non-negotiable, license it. Stock elements are somebody's copyrighted work. "I found it online" is not a license — a line this book's own license page states plainly, and one that has ended careers and cost lawsuits. Use reputable royalty-free or properly licensed sources, read what the license actually permits (some bar commercial use; some require credit), and keep your license records with the project. We cover licensing, releases, and copyright in full in Chapter 38; for now, treat every stock element like the borrowed property it is.

⚠️ Common Mistake: the element that doesn't belong to the shot. The tell of amateur VFX is almost never the effect itself — it's the mismatch. A sky replacement lit from the left over a scene lit from the right. A stock explosion that's sharper and cleaner than the grainy footage around it. A replaced screen with no glow, no reflection, and razor edges, sitting on the phone like a decal. The fix is always the same: make the element share the shot's light direction, grade, grain, motion blur, and perspective. Composite the element, then ask a fresh pair of eyes "is anything added here?" — if they can point to it, it isn't matched yet.

🎬 On Set → in post: replace a screen, invisibly. Shoot a short clip of someone using a phone or laptop, but put a solid, evenly-lit color card (or corner markers) where the screen content goes, and keep the shot fairly steady. In post: corner-pin track the four corners, pin in a replacement image (a brand's UI, a message, a video), then sell it — add a soft screen glow onto the hand/face, a faint reflection, and a hint of glare. Constraint: the replacement must track the screen with no sliding or floating, and it must throw light onto the scene. Self-review: show it to someone and ask, "what's on the phone?" — if they answer the content without noticing the effect, you've done a screen replacement right. This is a strong, motivated option for your Project 3.

🔄 Check Your Eye. 1. Why do you composite fire, sparks, and light-leak elements with a "screen" blend mode instead of a matte? 2. Give two reasons a real on-set screen (a phone or monitor) is often replaced in post. 3. Name three things you add back to a replaced screen to make it read as real, and one non-negotiable rule about any stock element you use.

Check yourself

1. Those elements come on a black background; "screen" (or "add") mode makes black transparent and lets only the bright pixels through, so the element lays over the shot with no rectangle — no matte needed.
2. Any two of: rolling-shutter flicker/banding (refresh vs. shutter), moiré, too dim/blown out, glare, or you didn't have the final UI/graphic on the shoot day.
3. Add back: screen glow onto the hand/face, a soft reflection, a hint of glare, and slight motion blur if the camera moves (any three). The rule: it must be licensed — "found it online" is not a license; use royalty-free/properly licensed sources and keep the records (Chapter 38).

35.6 Consent, deepfakes, and the ethics of synthetic media

Everything in this chapter is a technique for convincingly showing an audience something that did not happen in front of the camera. That is not a dark art; it is the job — a screen replacement, a sky, a clean studio behind a spokesperson are all benign, useful fictions the audience is happy to accept. But the same tools that place a person in a room they were never in can place words in their mouth they never said, and that is where compositing stops being a craft question and becomes an ethical one. This section is not optional, and it is not a footnote. In an era when the tools are getting cheaper and more convincing every month, it may be the most important thing in the chapter.

Two terms first, defined honestly. Synthetic media is imagery, video, or audio that is generated or substantially altered by software — increasingly by AI — rather than captured from reality: a face swapped onto another body, a voice cloned from recordings, a person made to appear to say or do something they never did, or an entirely invented "person" who never existed. A deepfake is the sharpest case: synthetic media that replaces or synthesizes a real, identifiable person's likeness or voice so convincingly that it can be mistaken for genuine footage of them. The technology is neither good nor evil on its own. A consented digital double, an actor de-aged with their agreement, a performer who licenses their voice for dubbing, a synthetic voice that gives speech back to someone who lost theirs — these are legitimate, even beautiful uses. The ethics do not live in the tool. They live in three questions you must answer before you make it.

Question one — consent: may you use this person at all? Never use a real person's likeness or voice without their informed consent for the specific use. "Informed" and "specific" both matter: someone who agreed to be interviewed did not agree to have their face composited into a scenario they'd object to, and someone who let you record a voiceover did not agree to have that voice cloned to say new things. The hardest line, and the one this book draws in bold: never make a real person appear to say or do something they did not. That is the seam where synthetic media becomes fraud, defamation, non-consensual intimate imagery, political disinformation, and cruelty — and it is a line that special care surrounds when the person is a minor, is vulnerable, or is dead and cannot consent at all. If you cannot get genuine, documented consent for the specific thing you want to make, you do not make it. There is no clever workaround; consent is the gate.

Question two — impersonation: is this deceptive? Consent settles whether you may; it does not settle whether the result deceives. Even with a signed release, synthetic media used to impersonate — to make an audience believe a real person genuinely said or did the thing — is a deception, and the book's license says it plainly: synthetic media must never be used to impersonate a real person without their consent, and where it could mislead, it must be disclosed. This is the same duty you met as the ethics of nonfiction back in Chapter 21: documentary and journalistic work carries an absolute obligation not to fabricate or deceptively edit. Compositing does not suspend that obligation; it supercharges it, because your tools are now good enough to fabricate seamlessly.

Question three — disclosure: must you tell the audience? When you use synthetic or heavily-manipulated media in a way a reasonable viewer could mistake for real, unaltered footage — a resurrected performer, a cloned voiceover, an AI presenter, a manipulated image in a factual piece — you disclose it. Disclosure is how you keep the audience's trust while still using the tool: label it, credit it, make the fiction knowable. In fiction the audience already knows it's fiction, so a monster or a set extension needs no disclaimer. In anything that presents itself as real, disclosure is not politeness; it is the difference between an effect and a lie.

FIGURE 35.6 — The synthetic-media decision: three questions before you fake a person

  Are you using a REAL person's likeness or voice?
        │
        ├── NO (a fully invented or generic element) ──► proceed; but still DISCLOSE if it could
        │                                                be mistaken for documentary/real footage
        │
        └── YES
             │
             ▼
        Do you have their INFORMED CONSENT for THIS specific use?
             │
             ├── NO ──►  ✋ STOP. Do not make a real person appear to say or do anything they didn't.
             │           (Fraud, defamation, non-consensual imagery, political disinformation, and
             │            "putting words in their mouth" all begin exactly here.)
             │
             └── YES
                  │
                  ▼
             Could the audience MISTAKE it for real / unaltered?
                  │
                  ├── YES ──► DISCLOSE it. Label the synthetic media. In documentary or
                  │            journalism this duty is absolute (Chapter 21).
                  │
                  └── NO (obvious fiction the audience knows is fiction) ──► proceed
   CONSENT gates whether you MAY. DISCLOSURE gates whether you must TELL. Get both — and get them in writing.

The legal landscape around all of this is real, active, and moving fast — rights of publicity and likeness, an emerging patchwork of deepfake and synthetic-media laws, protections against non-consensual intimate imagery, platform rules requiring disclosure of AI content, and election-integrity laws in many places. But laws vary enormously by jurisdiction and are changing as you read this, so this book will not hand you a specific statute to rely on, and nothing here is legal advice. What travels everywhere is the ethic, not the citation: consent, no deceptive impersonation, disclosure. Get releases for likeness and voice (Chapter 38 covers releases in full), document your consent, and when you're unsure whether something crosses a line, treat the person as you would want a stranger with these tools to treat you. We return to AI's larger role in the craft — and the working videographer's job in an AI world — in Chapter 40.

♿ Accessibility & Inclusion: the tool can include or exploit — you choose which. The same synthetic-media technology cuts both ways, and responsible practice uses it on the right side. On the inclusive side: a synthetic voice can give narration to a creator who has lost their speech (with their consent); AI-assisted captioning and audio description widen who can follow your work; a consented digital double can let a performer work who otherwise couldn't. On the harmful side sit the impersonations §35.6 forbids — and a hazard that is easy to forget in effects work: flashing and strobing. Muzzle flashes, glitch effects, hard light-leak cuts, and rapid composited flashes are a genuine photosensitivity risk (seizures), the same duty you met with motion graphics in Chapter 34 — keep flashing gentle and infrequent, and if a piece contains intense flashing, warn viewers. And whenever your work blends the real and the synthetic, disclosure is an accessibility act too: every member of your audience has a right to know what they're actually looking at. Effects that exclude, deceive, or endanger part of your audience aren't finished work — they're a failure of the craft's responsibility.

🔗 Connection: "motivate every choice" becomes "own every choice." The throughline of this whole book — that every light, move, cut, and grade should have a reason you can say out loud — reaches its sharpest point here. With synthetic media, you must be able to say your reasons and stand behind them ethically. In a world where any image can be faked, the trustworthy maker is the one who can account for what's real, who consented, and what was disclosed. Your consent records and your honesty are, increasingly, your professional reputation. The license at the front of this book (and Chapters 21, 38, and 40) all point at the same standard: the craft and the responsibility are inseparable.

🔄 Check Your Eye. 1. Define synthetic media and a deepfake, and state the one line this book draws in bold. 2. Consent, impersonation, disclosure: which question governs whether you may make something, and which governs whether you must tell the audience? 3. Give one genuinely legitimate use of synthetic media and one that is never acceptable.

Check yourself

1. Synthetic media = imagery/video/audio generated or substantially altered by software (often AI) rather than captured; a deepfake = synthetic media convincingly replacing a real, identifiable person's likeness/voice. The bold line: never make a real person appear to say or do something they did not.
2. Consent governs whether you may (no consent → stop). Disclosure governs whether you must tell (if it could be mistaken for real, disclose it).
3. Legitimate (any): a consented de-aging or digital double, a licensed voice for dubbing, a synthetic voice restoring someone's speech with consent, disclosed AI narration. Never acceptable: impersonating a real person without consent — fraud, defamation, non-consensual imagery, political disinformation.

Production Checkpoint

Your task: add one composited element to Project 3 — done cleanly, and motivated by the story. Pick the one that best serves your 5-minute branded piece:

1. A green-screen key. Shoot a spokesperson (or yourself, or a product) on a green screen — screen lit flat and even, subject well forward and lit to match a background plate — and composite them into a clean branded environment. Follow §35.2 on the shoot and §35.3 on the key. This directly advances the talking-head/product setups you've built all book.
2. A screen replacement. Replace a phone, laptop, or monitor screen in the piece with the brand's real UI or message: shoot a trackable screen, corner-pin track it, and sell it with glow, reflection, and a hint of glare (§35.5). Often the most motivated choice for a branded piece — the product frequently is the on-screen content.
3. A clean graphic composite or clean-up. Composite a licensed stock element that the story needs (a sky, an atmosphere, a light element), or paint out something that shouldn't be there (a rival's logo, a stray rig, a blemish) using a clean plate.

Whichever you choose: (a) write the one sentence of why the story needs it before you build it — if you can't, don't add it; (b) make it invisible — match the light, the grade (Chapter 32), the grain, and the motion; and (c) if any real person's likeness or voice is synthesized or manipulated, apply §35.6 — get consent, don't impersonate, disclose if it could mislead. Confirm every stock element is licensed (Chapter 38).

Why this matters: this is the final layer of the finish, and it is the purest test of the book's discipline. A composite you can justify in one sentence and that no one can spot is craft; an effect you added because it looked cool, or that anyone can point to, is decoration — and decoration is how finishing goes wrong. Pull this off and Project 3 is not just edited, colored, mixed, and titled; it is finished, ready for the delivery and export you'll do in Chapter 36.

Summary

• Compositing is combining two or more image layers into one seamless shot. VFX is any imagery created or altered outside the live moment of filming; the best VFX is invisible, and it is motivated by the story.
• The alpha channel is a per-pixel transparency map; rendered black-and-white it's the matte. White = opaque (keep), black = transparent (drop), grey = partial (soft edges — hair, motion blur, glass). The greys are where a composite succeeds or fails.
• A composite must agree with itself: foreground and background sharing one light, lens/DoF, camera height, motion, grade, and grain. The eye reads light before edges.
• Green screen / chroma key — the whole discipline in one line: a clean key is lit and shot, not fixed in post. Two separate jobs:

• Green is default (far from skin; sensors resolve green best; needs less light than blue). Shoot the least-compressed, highest-chroma format you can — 4:2:2 keys far better than 4:2:0 (Chapter 3).
• Keying order: pull the key → read the matte in black-and-white → despill the edges → clean the matte gently (clip black, clip white, shrink/soften — without eating hair) → garbage matte → composite and grade together. A keyer is a Chapter 32 qualifier aimed at transparency.
• Mask = a drawn shape isolating a region. Roto (rotoscoping) = animating that mask frame-by-frame to follow motion when you can't key. Green screen exists to avoid roto. Clean-ups (rig removal, paint-outs, patches) are the invisible majority of VFX — grab a clean plate on set to make them trivial.
• Stock elements (skies, particles, flares, light leaks) drop in via blend modes — light-on-black elements use "screen" so the black vanishes. Screen replacement (corner-pin track + glow/reflection/glare) fixes flickering, moiré, dim, or unfinished on-set screens. Always match the plate and always license it.
• Synthetic media / deepfake ethics — the non-negotiable core of the chapter:

Spaced Review

The finish is a stack, and effects sit on top of the color and motion work you just did. Pull these back up from memory.

1. (Chapter 32) A keyer and a color qualifier are the same underlying tool. Explain what that tool does, and what each of the two feeds its selection into.
2. (Chapter 32) After compositing a subject over a new plate, you grade the shot. Why must the foreground and background be graded together to one look — and what does protecting skin on the flesh-tone line have to do with a keyed subject?
3. (Chapter 34) Rotoscoping animates a mask over time using keyframes. In your own words, what is a keyframe and what does easing it change about the motion — and why does that matter for an animated matte?
4. (Chapter 34) Both motion graphics and composited effects (flashes, glitches, light leaks) carry a photosensitivity duty. What's the risk, and what's the practical rule?
5. (Chapters 11 & 3) Two on-set decisions make or break a key: how you light it and how you capture it. Name the lighting rule (from three-point lighting) and the capture rule (from codecs) that most determine whether a key will be clean.

What's Next

Project 3 is now finished in every layer that matters: cut, colored, graded, mixed, titled, and — with a clean, motivated composite — given whatever the story needed that the camera couldn't capture. It looks and sounds like professional work because it is professional work. But a finished timeline is not yet a video anyone can watch: it lives inside your editor, at full quality, in a format no platform wants. Chapter 36 is where you get it out the door — the export dialog demystified, delivery codecs and bitrates, the exact specs each platform expects, the high-quality master you keep versus the deliverables you send, and the captions that make it reachable by everyone. The finish is done; now we deliver it. Project 3 locks at delivery — and Chapter 36 is where you lock it.