Case Study 2: The Format Bake-Off — One Product, Four Recording Choices, One Grade

Where Case Study 1 took apart a blockbuster's grade to prove that the look is built from what you capture, this one proves the same thing with your own hands and no budget at all. We are going to shoot a single object on a table four different ways — standard profile, Log, low bitrate, and (if we can) 8-bit versus 10-bit — grade them all identically, and watch which ones survive and which ones shatter. This is a shoot-along-and-edit: the deliberately un-glamorous afternoon that turns the invisible choices of this chapter into things you've seen fail with your own eyes. Follow along, and better, run your own version — it directly feeds the format decision your Production Checkpoint asks you to make for Project 1.

The brief

Before we shoot a real product job, we want to know — not guess — what our different recording formats actually buy and cost. So we set up a controlled little bake-off: one object, one light, one framing, held rock-steady, captured several ways, then graded the same way, so the only thing that varies is the format. Whatever differences we see are honestly caused by the recording choice and nothing else.

We choose the product on a table — one of the book's four anchor setups — for a specific reason: it is the setup that punishes format weakness most visibly. A product shoot usually wants a smooth, graded backdrop (a gentle gradient of color falling off behind the object) and a punchy, saturated look on the product itself. A smooth gradient is the single fastest way to reveal banding (§3.3), and a saturated color is where chroma subsampling (§3.3) and bit depth show their limits under a grade. If a format is going to fail, it will fail here, in the open, where we can see it — which makes this the perfect teaching setup even though your Project 1 is a talking-head. Everything we learn transfers straight to that talking-head's skin tones and soft-lit background.

The whole exercise lives on one sentence from the chapter: the recording format sets the ceiling on what post can do. We're going to measure that ceiling, four times, and pick the format that leaves the most room for the look we want.

📱 Phone-first note: you may not be able to shoot Log or 10-bit — that's fine. Do the comparison you can do (high-bitrate standard vs your phone's lowest-quality/most-compressed mode is genuinely revealing), and read the Log and 10-bit phases as the reason to reach for a manual app or a capable camera when a heavy grade is coming. The method — capture, grade identically, watch what breaks — is what matters, and it works at every tier.

Gear and settings

A format bake-off needs almost nothing but discipline.

⚙️ Settings Box: the format bake-off (what we'll compare).

Setting	Test values	What we're judging
Picture profile	Standard (Rec.709), then Log/flat	Ready-to-watch vs latitude to grade
Bit depth	8-bit, then 10-bit (if available)	Banding on the gradient backdrop under a grade
Chroma	4:2:0, then 4:2:2 (if available)	Edge/color cleanliness on the saturated product
Bitrate	Lowest mode, then highest mode	Macroblocking/smear vs clean detail
(Resolution / frame rate)	4K, 24 fps, held constant	Not tested today — Chapter 2 owns these
(Shutter)	1/50, held constant	Held so format is the only variable
(White balance)	Set correctly in camera, held constant	Isolate the format, not the color temp

The greyed rows are deliberately held constant. A valid test changes one thing at a time; today the variable is the recording format, so resolution, frame rate, shutter, light, and white balance all stay locked while we cycle the format.

The kit: a camera that can change picture profile and, ideally, bitrate and bit depth (a mirrorless camera is ideal; a phone with a manual video app does the standard-vs-bitrate half); something to hold it dead still (a tripod is important here — any drift between takes pollutes the comparison); one product with a saturated color and a smooth surface (a colored bottle, a ceramic mug, a piece of fruit — something with a hero color and a soft highlight); a plain backdrop that can hold a smooth gradient (a seamless paper, a painted wall, a curtain); and one controllable light. That's a complete bake-off kit, and most of it you already own.

The setup

We build a simple, classic product frame: the object slightly off-center on a clean surface, a smooth gradient falling off behind it, one soft key raking across it to bring out its shape and a gentle highlight, and a bounce to keep the shadow side from going dead. Two diagrams — the framing, then the room.

FIGURE CS2.1 — The product framing (16:9 frame, object on a third)

  ┌───────────────────────────────────────────┐
  │        smooth gradient backdrop             │   O = rule-of-thirds intersections
  │      (light → dark, top to bottom)          │       (the product sits near the
  │        O·······················O            │        left one; its hero highlight
  │        ·                        ·           │        catches the eye first)
  │        ·        ▟█▙  ← product   ·           │
  │        ·        ▜█▛  (hero color)·           │   The GRADIENT is the trap we've set:
  │        O·········│··············O            │   a smooth backdrop is where banding
  │                  │  soft reflection          │   shows first. If a format is weak,
  │        clean surface, gentle falloff         │   this background will reveal it.
  └───────────────────────────────────────────┘
   Framed for a smooth backdrop and a saturated hero color on purpose — this frame is
   engineered to expose format weakness under a grade. Composition is Chapter 6; today
   we just lock it and never touch it again.

FIGURE CS2.2 — The product setup (top-down view)

        [ soft KEY ] ▢
              \
               ↘   (rakes across the product to show shape + a hero highlight)
             ( S )  ← the product, on the table (the "subject" slot)
               ↗
        ○ white bounce (fills the shadow side so it isn't dead black)
              |
        [ backdrop ] : lit separately to fall off into a smooth gradient
              |
            [CAM]  on a tripod, locked; ~50mm-equiv; NOT moving between takes

   Legend (frozen): ( S ) = subject (here, the product)  ▢ = soft key  ○ = bounce/fill
   [CAM] = camera.  We change ONLY the camera's recording FORMAT between takes; the light,
   the product, the framing, and the white balance all stay exactly put.

The one discipline that makes the whole test valid: nothing moves but the format. Same product, same light, same framing, same white balance, same everything — we only reach into the menu and change the profile, the bitrate, or the bit depth. If the product drifts or the light shifts between takes, we can no longer trust that the format caused the difference we see in the grade. A sloppy bake-off teaches nothing; a clean one teaches the truth.

The shoot, format by format

We record ten to fifteen seconds in each format, label each clip immediately (this matters — flat Log clips look nearly identical to each other on a small preview and are miserable to sort later), and move to the next. We don't judge anything on the camera; the whole point is what happens in the grade.

Take 1 — standard profile, high bitrate: the ready baseline

We start with the honest, everyday choice: a standard (Rec.709) profile at the camera's highest bitrate, 8-bit if that's all we have.

FIGURE CS2.3 — Take 1, the standard-profile capture        [constructed teaching example]
  THE FRAME    The product on its third, saturated hero color popping, backdrop falling off smoothly.
  THE LIGHT    Soft key raking across the product; a clean highlight on its shoulder; bounce filling the
               shadow side. White balance set correctly, so the whites read neutral.
  THE SOUND    (Not our subject today — a product shot is silent.)
  THE EFFECT   It looks GOOD straight out of camera: contrasty, colorful, ready to hand over. The
               backdrop gradient is smooth; the product color is rich; nothing needs doing to watch it.
  THE LESSON   A standard profile at a healthy bitrate is a genuinely strong choice — it looks finished
               immediately. The question the grade will ask is: how far can we PUSH it before it breaks?

The read: this is the baseline every other take is measured against. Straight out of camera it's the best-looking of the bunch — which is exactly the seduction of a standard profile, and exactly why beginners never think about anything else until a grade breaks on them. We note that it looks great now, and reserve judgment for the grade.

Take 2 — Log (10-bit if we have it), high bitrate: the flat capture

Now the opposite philosophy: a Log (or the flattest) profile the camera offers, in 10-bit if available, at high bitrate, exposed a touch bright to protect the shadows (a preview of Chapter 5's §5.5).

FIGURE CS2.4 — Take 2, the Log capture        [constructed teaching example]
  THE FRAME    Identical framing, product, light, and distance — nothing changed but the profile.
  THE EFFECT   It looks WRONG: flat, grey, milky, desaturated. The hero color is muted; the backdrop is
               a dull, low-contrast haze; the whole thing looks like a mistake. On the camera, it's ugly.
  THE LESSON   This is Log doing its job — holding every scrap of range and color, committing to nothing,
               waiting for the grade. "Looks wrong out of camera" is the FEATURE, not the failure.

The read: if you didn't know better, you'd delete this take. It looks worse than everything else. But it is the take carrying the most information into the grade — the flat, uncommitted, information-rich capture that §3.4 promised. Whether that promise pays off is the entire question of the edit pass. We resist judging it by eye and label it carefully.

Take 3 — low bitrate, 8-bit: the "save card space" version

Now the trap we want to spring on ourselves: the camera's lowest-quality, most-compressed mode — the one a beginner picks "to save space."

FIGURE CS2.5 — Take 3, the low-bitrate capture        [constructed teaching example]
  THE FRAME    Same product, same light — but recorded at the stingiest bitrate the camera offers.
  THE EFFECT   On the camera it looks... fine. Almost indistinguishable from Take 1 at a glance. Maybe
               the smooth backdrop looks a hair less clean if you stare, but you'd never reject it here.
  THE LESSON   Low-bitrate damage HIDES on the camera and in casual viewing. It's a sleeper — the
               problems are latent, waiting for the grade to drag them into the light.

The read: this is the sneaky one, and the most important lesson of the day is that it looks fine right now. Nobody would catch the problem on set. That's precisely why over-compression sinks so many projects — the damage is invisible until you grade, by which point the shoot is long over. We note how normal it looks and set the trap for the edit.

Take 4 — 8-bit vs 10-bit, matched: the head-to-head

If our camera can do both, we shoot one more matched pair — the identical standard-profile shot in 8-bit and in 10-bit — so we can isolate bit depth alone in the grade. (If the camera is 8-bit only, we skip this and read the results below as the reason to want 10-bit for graded work.)

FIGURE CS2.6 — Take 4, the 8-bit vs 10-bit pair (what the grade will reveal)

  8-bit  (256 steps/channel):  gradient backdrop = ▓▓▒▒░░  → few rungs; a hard grade opens GAPS
  10-bit (1024 steps/channel): gradient backdrop = ▓▒░·▁    → many rungs; the same grade stays smooth

  On the camera, IDENTICAL. The difference is invisible until we stretch the gradient in the grade —
  then the 8-bit backdrop bands into stair-steps and the 10-bit one holds.

The read: the two clips are indistinguishable on the camera — as they must be, since the extra tonal steps only matter when stretched. We've now got four (or five) clips that look, on set, from "great" to "wrong" to "fine," and none of their real differences have shown up yet. Everything that matters is about to happen in the grade.

⚠️ Common Mistake: judging a format by how it looks on the camera. Every instinct on set screams that Take 1 and Take 3 are great and Take 2 (Log) is broken. Every one of those instincts is about to be proven irrelevant or backwards. The whole danger of format decisions is that they're invisible at the moment you make them — the flat Log looks worst and grades best; the low-bitrate clip looks fine and grades worst. If you judge format by the camera's screen, you will choose wrong every time. Judge it by the grade, which is the only place it reveals itself.

The edit pass

We bring all the clips into any free editor (DaVinci Resolve is the book's reference from Chapter 26 on) and do one thing: apply the same aggressive grade to every clip, and watch. Our target look is a real product look — deepen the backdrop into a rich, moody gradient, push the hero color to punch, add contrast, keep the highlight clean. Exactly the kind of grade a product job demands, and exactly the kind that finds every weakness.

First, we bring the Log clip back to normal. Take 2 gets a color-space transform (or the camera maker's LUT) to undo the flatness — the grey, milky image snaps into a normal-looking one, and now we grade from there. This is the extra step Log always costs, and the reason you never shoot Log for a job you won't grade.

Then we apply the identical grade to all clips and compare. Here's the timeline and what each clip does under the same push.

FIGURE CS2.7 — The grade, applied identically to every clip

  V1  [ T1 standard 8-bit ][ T2 Log 10-bit ][ T3 low-bitrate 8-bit ][ T4a 8-bit ][ T4b 10-bit ]
        same grade node ▸     same grade ▸       same grade ▸          same ▸       same ▸
      ──────────────────────────────────────────────────────────────────────────────────────
      RESULT under the identical aggressive grade:
        T1 standard 8-bit   → holds a MODERATE push; bands only if we shove the backdrop hard
        T2 Log 10-bit       → takes the grade beautifully; smooth backdrop, rich color, most room
        T3 low-bitrate 8-bit→ FALLS APART: backdrop bands, shadows block up, product edges smear
        T4a 8-bit           → backdrop bands under the hard push (few tonal steps)
        T4b 10-bit          → same push, backdrop stays SMOOTH (many tonal steps)

The results, take by take, are the whole payoff of the afternoon:

• Take 1 (standard, high bitrate, 8-bit) takes a moderate grade well — this is why it's a fine choice for light work. Push the backdrop gradient hard, though, and even this clean 8-bit clip starts to band. It has room, but not a lot. Bought: a great-looking, ready image and enough latitude for a modest grade. Cost: a low ceiling — it breaks if the grade gets ambitious.
• Take 2 (Log, 10-bit, high bitrate) is the star. The flat capture that looked broken on set now grades further than anything else — the backdrop deepens into a smooth, rich gradient with no banding, the hero color pushes hard and stays clean, the highlights and shadows both hold detail. Bought: maximum room to build the exact look we wanted. Cost: the extra grading step, the discipline to expose it right, and the need for 10-bit — a cost we happily paid because we had a grade to do.
• Take 3 (low bitrate, 8-bit) is the cautionary tale, and it's ugly. The smooth backdrop shatters into stair-stepped bands the moment we deepen it; the shadows block up into mud; the product's saturated edges smear and shimmer. Bought: small files. Cost: the entire grade. This footage cannot deliver the look, and nothing in post fixes it — the information was thrown away at capture.
• Take 4 (8-bit vs 10-bit) isolates the lesson cleanly: identical grade, and the 8-bit backdrop bands while the 10-bit one stays glassy-smooth. The only variable was bit depth, and it decided whether the gradient survived. This single pair is the most convincing argument in the chapter for 10-bit on any graded work.

✂️ In the Edit. Watch what just happened across five clips: the one that looked worst on set (Log) delivered the best grade, and the one that looked fine on set (low bitrate) delivered the worst. The edit is where the format's true value is finally spent — and by then the shoot is over and the choice is locked. Every "bought" and "cost" above was decided the instant we pressed record; the edit only revealed the bill. This is "you shoot for the edit" made literal: you are not choosing a format for how it looks on the camera, you are choosing it for how much room it gives the grade three weeks later. Choose for the edit you intend to have, because the edit can only ever spend what the capture saved.

What we'd lock for a real product job

The bake-off did its job. Our format decision, made by seeing rather than guessing:

• For a graded product job with a smooth backdrop: Log, 10-bit, high bitrate, exposed carefully — because the grade is heavy and the gradient is unforgiving, and Take 2 proved this format has the room. If the camera can't do 10-bit, a high-bitrate standard profile graded gently (Take 1's ceiling), or a plan to avoid big backdrop pushes.
• For 4:2:2 if keying or fine color work is involved: step to 4:2:2 — the saturated edges and any key need the color detail.
• Never the low-bitrate mode for anything graded — Take 3 settled that forever.
• The card-and-offload discipline unchanged: fast enough card for the high bitrate, offload and back up before wiping, confirm import and playback.

And the transfer to Project 1's talking-head is direct: a face's soft-lit cheek and the out-of-focus background behind it are exactly the smooth gradients that banded here. If you'll grade your talking-head toward a real look, everything Take 2 and Take 4 taught applies to that skin and that background. If you'll only correct it lightly, Take 1's standard profile is honest and fine — which is precisely what the Production Checkpoint concludes.

What we'd do differently

Even a clean test surfaces the next one:

• We eyeballed exposure across the profiles. Log wants a slightly brighter exposure than standard (Chapter 5), and matching brightness across the takes by eye muddied the comparison a touch. Next time we'd use the waveform (Chapter 5's tool) to set each take precisely, so only the format differs.
• We didn't test a 10-bit 4:2:0 middle rung. Many cameras offer 10-bit 4:2:0 — more tonal steps but still limited color detail. Testing it would map the ladder more finely between "8-bit 4:2:0" and "10-bit 4:2:2."
• We tested one grade. A gentle grade would have let even Take 3 look passable, hiding its ceiling. We chose an aggressive grade on purpose to expose the limits — but shooting a second, gentler grade would show how much the required push changes which format you need. A light correction needs far less than a heavy look.

None of these is a failure; each is the next test, surfaced by doing this one — the make-it, watch-it, improve-it loop the whole book runs on.

The bigger lesson

We spent an afternoon shooting an object that no one will ever watch, and produced no video anyone wants — and it was one of the most useful afternoons in the project, because we now know, in our hands and eyes, what each format costs and buys. The instinct is to trust the camera's screen: the standard clip looked great, the low-bitrate clip looked fine, the Log clip looked broken. Every one of those on-set judgments was either irrelevant or exactly backwards, and only the grade told the truth. That is the paradox at the heart of this chapter: the most consequential technical decision you make is the one you cannot see yourself making. A wrong frame rate you can spot on set (Chapter 2). A wrong format hides until it's too late to fix.

There's a deeper point, the same one Case Study 1 made from the top of the industry. Fury Road captured rich and flat so the grade could build a molten, mythic world. We captured rich and flat so the grade could build a moody product look. The budgets differ by a factor of a million; the strategy is identical — capture more information than the finished image shows, so the finish has room to become what you want. And the counter-lesson is identical too: reach up the flexibility ladder only as far as the look and your pipeline require. Take 2 was right for a graded product job and would be absurd overkill for a quick social clip of the same object. The skill isn't "always shoot Log" or "always shoot RAW." The skill is knowing, before you press record, what the edit will need — and this afternoon, for a few clips no one will watch, is how you learn to know it.

Discussion questions

1. The Log clip looked worst on the camera and graded best; the low-bitrate clip looked fine and graded worst. Why is judging a format by the camera's screen almost guaranteed to lead you astray?
2. The product setup was chosen because it punishes format weakness (smooth backdrop, saturated color). How would this bake-off's results differ on a busy, textured subject that hides banding — and what does that tell you about matching format to content?
3. Take 1 (standard) took a moderate grade but banded under a hard one. Describe two real jobs: one where Take 1's format is the right call, and one where it isn't.
4. The 8-bit vs 10-bit pair isolated a single variable. Why is changing one thing at a time the only way to trust what a test teaches you — and where else in Chapters 2–3 did you see this discipline?
5. Connect this bake-off to Case Study 1: what is the shared strategy between a blockbuster's grade and a no-budget product shot, and what is the shared warning?
6. Your Project 1 is a talking-head, not a product. Which of this case study's lessons transfer directly to a face and its soft background, and which don't?

Your turn

Run your own format bake-off — it directly informs the format choice your Production Checkpoint asks you to make:

1. Set up a controlled subject with a smooth gradient behind it (a colored object on a plain, softly-lit backdrop) and lock the camera on a tripod. Set white balance correctly and don't touch it.
2. Capture the same shot in every format you can: standard profile at high bitrate, standard at your lowest/most-compressed bitrate, and — if your camera offers them — Log, and 10-bit vs 8-bit. Label each clip the instant you shoot it.
3. In any editor, apply one aggressive grade to all of them: deepen the backdrop, push the color, add contrast — a real look, hard enough to hurt.
4. Watch them side by side, large. Rank them by how much grade they survived. Find the exact point where each one bands, blocks, or smears — that's its ceiling, measured.
5. Write your verdict for Project 1: which format gives your talking-head the room its intended grade needs, with a one-sentence reason tied to what you'll do in post — exactly as we concluded above.
6. Log one line in your Frame Log: the single most surprising thing a format did in the grade (probably that the flat one won).

It won't be pretty — it's a test, not a piece. But you'll walk into your real graded shoots having seen your formats break, which is the difference between a shooter who chooses a format on purpose and one who finds out the ceiling too late.

Key takeaways

• A format's true value shows only in the grade, never on the camera — the clip that looked worst on set (Log) graded best, and the one that looked fine (low bitrate) graded worst.
• Isolate one variable. Lock light, framing, white balance, and settings so the format is honestly the only thing that changed — the same discipline as Chapter 2's camera test.
• Standard profile: looks great immediately, takes a moderate grade, but has a low ceiling — it bands under an aggressive push. Right for light correction, wrong for a heavy look.
• Log (10-bit): looks "wrong" out of camera on purpose, and grades further than anything else — the right tool when you have a grade to do and the bit depth to support it.
• Low bitrate: hides its damage on set and shatters in the grade — banding, blocking, smearing that no post can fix. Never for graded work.
• Bit depth decides the gradient. The 8-bit vs 10-bit pair, identical but for depth, was the difference between a banded backdrop and a smooth one under the same grade.
• The strategy scales from a blockbuster to a phone: capture more information than the finished image shows, so the grade has room — but reach up the flexibility ladder only as far as the look and your pipeline require.
• It all transfers to your talking-head: a face's soft cheek and blurred background are the same smooth gradients that banded here — so choose Project 1's format for the grade you actually intend to do.