Case Study 23.2 — Shoot-Along: A Backlit Window Portrait, Phone-Only, Fighting (and Using) the Computation
"You don't take a photograph, you make it." — attributed to Ansel Adams
This is a different kind of case study from the first. Where Case Study 23.1 analyzed a landmark genre, this one is a production — a complete, from-scratch shoot you can follow step by step, in a location you can reach on foot, with nothing but a phone. The brief is deliberately a trap: a portrait against a bright window, the single most common scene where a phone's computation both rescues you (HDR) and betrays you (portrait mode). We will run it phase by phase, showing the thinking, the Described Photograph of each result, and exactly what we adjusted and why. The goal is not a perfect portrait. It is to feel, in your own hands, the difference between accepting what the phone computes and directing it.
The brief and the constraints
The brief: Make one keeper-quality environmental portrait of a person seated beside a bright window — their face well-lit and detailed, the bright view outside still holding some detail (not a blown-white rectangle), and a sense of the room around them. Then make a second, tighter version with subject-background separation using portrait mode — and make it convincing.
The constraints: - Phone only. No dedicated camera, no flash, no reflector you had to buy. (We'll improvise fill from what's in the room — Chapter 11's principle, no gear.) - One location: the kitchen window — our recurring source of soft, controllable light. - Natural light only, mid-morning, so the window is bright but not direct-sun harsh. - One willing subject, who can hold a pose for a few seconds at a time and is comfortable being directed. - Time: about forty-five minutes, the same humble budget that made the glowing egg in Chapter 1.
This is the classic high-contrast portrait problem: the window is many stops brighter than the shaded face. Expose for the face and the window blows out; expose for the window and the face goes to mud. It is exactly the scene HDR was built for — and exactly the scene where portrait mode's depth map will struggle, because the subject is close to a bright, detailed background. We will use the first and fight the second.
Gear and starting settings
The whole kit fits in one hand. Here is where we start, knowing we'll adjust everything.
⚙️ Settings Box — Backlit window portrait, phone-only (a starting point, not a recipe)
Control Starting point Why / adjust when Mode Photo (auto), HDR ON for the wide shot The scene's range exceeds one exposure; let the merge work Capture RAW + JPEG if available JPEG for the share; RAW to override the pipeline later (§23.6) Expose for The bright window (tap it, drag down if needed) Blown highlights can't be recovered; HDR will lift the face Focus Tap the near eye Sharpness goes where the meaning is (Chapter 4) White balance Auto to start; lock once it looks right Mixed window-cool + room-warm light can drift frame to frame Stability Both elbows braced; hold through the HDR pause The merge needs aligned frames (§23.1) For the tight shot Portrait mode, moderate blur, subject stepped forward Give the depth map real distance and don't max the blur (§23.4) Adjust from here as the room and the light dictate. The box is where you start the thinking, not where it ends.
The setup: a lighting map made of a window and a wall
You do not need lights. You need to place the person in the light you already have — the whole lesson of
Part III, applied with a phone. Here is the top-down map. (Frozen symbol legend, stated once: ( S ) =
subject; [CAM] = camera/phone; ☀ = the window as our key light; ○ = an improvised reflector — a white
wall, a tablecloth, a sheet of paper.)
FIGURE CS23.2 — Backlit window portrait setup (top-down view), phone-only
[ BRIGHT WINDOW ☀ ]
│ (the brightest thing in the scene; our key + our problem)
│
( S ) ───────────► turned ~45° toward the window
↙ ↘
☀ window light ○ improvised fill
rakes across the (a white wall / sheet / tablecloth
near side of the camera-right, bouncing window light
face from camera-left back into the shadow side)
│
[ CAM ]
│
(phone at the subject's EYE LEVEL, braced)
Key idea: the window both LIGHTS the face (turn the subject toward it) and
THREATENS to blow out (it's behind/beside them). HDR's job is to hold both.
The two moves that make this work, before a single frame: turn the subject partly toward the window so its soft light rakes across the near side of the face (Chapter 5's direction lesson — get the light off the front), and put something bright on the shadow side to bounce a little fill into the dark cheek (Chapter 11's fill principle, achieved with a white wall, a sheet, or a piece of foam board — no purchase). Now the face has modeling — bright near cheek, gently shadowed far cheek — and the window is positioned as both our light and our challenge.
Phase 1 — The honest baseline: see the problem
Before we let the phone rescue anything, we shoot the problem honestly, so we know what the scene really gives a single sensor frame. HDR off. We shoot it twice — once exposed for the face, once for the window — exactly the "pick your battle" failure the whole chapter is about.
FIGURE CS23.3 — "The baseline failures" [constructed teaching example]
VERSION A — exposed for the FACE (HDR off): The face is bright and correct, modeled nicely by the window
light. But the window itself is a featureless white rectangle — the view outside, the frame,
the sky, all gone, clipped to pure white. The eye is yanked to that glaring white block instead
of the person. A failed photograph, but an honest one.
VERSION B — exposed for the WINDOW (HDR off): The view outside is lovely — you can see the street, the
sky, the detail. But the face has collapsed into a dim, muddy silhouette, the features barely
readable, the shadow side near-black. Also a failed photograph. The scene is simply bigger than
one frame can hold.
THE LESSON The scene spans more dynamic range than a single exposure can contain. Neither version is
usable — and seeing *both* failures is what makes you trust (and judge) the HDR merge that
follows. You can't appreciate the rescue until you've seen the drowning.
This is the most important phase and the one people skip. By shooting the honest failures first, you have calibrated your eye: you now know what the scene really looks like to the sensor, so when HDR hands you a miracle, you can tell whether it's a faithful miracle or an overcooked one.
Phase 2 — Let HDR work: the wide environmental frame
Now HDR on. We tap the bright window to set exposure (protecting the highlights — blown highlights have no data to recover, §23.2), and trust the merge to lift the face. We brace both elbows and hold through the HDR pause — the capture is several frames, and jerking away early would leave it half-merged (§23.1's common mistake). We tap the near eye for focus.
FIGURE CS23.4 — "The window portrait, HDR merge" [constructed teaching example]
THE FRAME The subject fills the right two-thirds, seated, turned ~45° toward a tall window on the left
that fills the left third. Through the glass: a soft, detailed view — rooftops, a pale sky,
not blown out. The room around the subject reads warm and lived-in. A mug on the sill catches
a small highlight.
THE LIGHT Soft window light raking across the near cheek (bright), the far cheek gently shadowed and
lifted by the bounced fill — three-dimensional modeling, the opposite of flat. The window
itself is bright but HELD: detail survives in the sky and frame. Warm room, cool window: a
pleasant temperature contrast.
THE MOMENT A calm, present beat — the subject looking softly toward the light, an unposed expression.
(Held still enough for the multi-frame HDR capture to resolve cleanly.)
THE CHOICES HDR on; exposed for the window; focus on the near eye; phone at eye level, braced; held
through the capture. Framed to include enough room to read "environmental portrait."
THE EFFECT The eye lands on the lit face first (brightest + sharpest), travels to the soft window light,
registers the held detail outside, and rests. Both the face AND the window have detail — the
impossible-in-one-frame scene, now a single natural-looking photograph.
THE LESSON HDR's merge recovered the full range that defeated both baseline frames. This is computation in
its home territory, serving the photograph rather than imposing on it. We exposed for the
highlights, let the merge lift the shadows, and got what the eye actually saw.
What we adjusted, and why. Our first HDR attempt looked slightly flat — the tone mapping had opened the shadows a touch too far, sanding away the modeling that made the face three-dimensional, drifting toward the grey "HDR look" (§23.2). The fix was twofold: where the phone offered "natural" vs. "rich" HDR, we chose natural; and we re-exposed a hair brighter on the window tap, which let the merge keep a little more contrast on the face. The lesson: HDR is not a single thing you turn on — its tone mapping has a taste, and your job is to nudge it back toward what your eye saw when it overreaches.
💡 Why It Works: Exposing for the highlights and letting HDR lift the shadows is the counterintuitive heart of this scene. Beginners expose for the face (the thing they care about) and then the window is already blown beyond recovery — there's no data left in pure white to merge back. By exposing for the window, we keep highlight data the merge can use, and we exploit the fact that shadows almost always hold recoverable detail. Protect what can't be recovered (highlights); rescue what can (shadows). That single rule turns the hardest common portrait scene from a failure into a keeper.
Phase 3 — The tight shot: fighting portrait mode into submission
Now the second deliverable — a tighter, more "professional"-looking frame with the background softened so the subject pops. This is portrait mode, and this is where the same scene that HDR rescued will try to betray us, because the subject is close to a bright, detailed window — a hard case for the depth map (§23.4).
We do not just point and shoot. We set the scene up to be easy for the depth map first, then shoot:
- Step the subject forward, away from the window. A metre or two of real distance between subject and background is the single biggest thing we can do — it gives the depth map a clean, easy edge to read (§23.4's common-mistake fix). A real lens would want the same separation.
- Choose moderate blur, not maximum. We set the simulated aperture to a middle strength. Maxing it would exaggerate every segmentation error — the hair halo, the cutout edge (§23.4).
- Mind the hard edges. Our subject has some flyaway hair and is wearing glasses — two of portrait mode's four nightmares. We can't remove them, so we'll shoot and inspect, ready to reshoot or to abandon the fake blur if it mangles them.
FIGURE CS23.5 — "The tight portrait, synthetic bokeh" [constructed teaching example]
THE FRAME A head-and-shoulders crop, subject filling most of the frame, the window and room now a soft,
blurred warm-and-cool wash behind. The face is sharp; the background dissolved.
THE LIGHT Same soft window key on the near cheek, shadow side lifted by fill. The blurred background
renders the bright window as a gentle glow rather than a hard rectangle — the blur actually
*helps* the highlight problem.
THE MOMENT A direct, quiet gaze toward the camera — a more formal portrait beat than the environmental
frame.
THE CHOICES Portrait mode; subject stepped ~2 m forward of the window; MODERATE simulated blur; focus and
exposure on the near eye; held steady. Inspected at 100% immediately after capture.
THE EFFECT The subject pops convincingly from the softened background — the "shot on a real camera" look a
phone lens can't produce optically (§23.4). At a glance, it works.
THE LESSON Synthetic bokeh is believable when you give it an easy scene: real subject-background distance
and moderate blur. We engineered the conditions for the depth map to succeed *before* pressing
the button — computation rewards a scene that's already easy to read.
What we adjusted, and why — the honest part. The first portrait-mode frame, shot before stepping the subject forward, was a mess: a clear halo along the flyaway hair, and the top edge of the glasses had a weird partial-blur where the depth map couldn't decide if the lens was near or far (§23.4's artifacts). Stepping the subject two metres forward fixed most of the halo — the bigger real depth gap gave the algorithm an easy edge. The glasses still showed a faint seam, so we made two more choices: we turned the subject's head a few degrees so the glasses caught the window light cleanly (less for the depth map to misread), and we dialed the blur down one more notch. The remaining tiny seam we judged acceptable. The key production lesson: when portrait mode fails, the answer is rarely "max the blur" or "give up" — it's to change the physical scene (distance, angle, light) so the depth map has an easier job, then use a moderate blur that hides the seams instead of exaggerating them.
⚠️ Common Mistake (caught live): On one frame we let the subject hold a coffee mug out toward the camera for a casual feel. Portrait mode blurred the mug — it read the outstretched hand as a different distance from the face and decided it was "background," even though it was the closest thing in the frame. The depth map isn't "near = sharp"; it's "subject = sharp," and it had decided the subject was the face, not the hand. The fix was to bring the mug back in line with the body, or shoot that frame in normal mode without the fake blur. Anything reaching toward or away from the camera at a different distance than the detected subject is a portrait-mode trap.
Phase 4 — The edit pass: undoing what the pipeline overdid
A phone JPEG arrives already heavily processed (§23.5). Our keeper frames look good but show two telltale pipeline fingerprints we want to walk back, working in a mobile editor (the full mobile workflow is Chapter 25 — here we just make the corrective moves):
- The skin looked slightly waxy — the noise reduction and any face-smoothing had sanded a little too much texture, drifting toward plastic (§23.5's "watercolor"). Because we shot RAW, we opened the RAW version and dialed noise reduction down, recovering the honest skin texture (pores, fine lines) that makes a face look like a person rather than a mannequin.
- The window had drifted a touch too blue from the auto white balance fighting the mixed light. We nudged the temperature slightly warmer for a more natural, inviting feel — a decision the auto pipeline made for us that we simply re-decided (§23.6).
The corrections were small, but they are the whole point of the chapter: the processed JPEG is the phone's opinion, and a thinking photographer keeps the parts that serve the image and overrides the parts that don't. Had we needed accurate color or maximal skin character (a documentary portrait, say), we'd have leaned harder on the RAW and trusted the pipeline even less.
🔗 Connection: Notice how this shoot quietly reused half the book: Chapter 5's light direction (turn the subject toward the window), Chapter 11's fill principle (a white wall as a no-cost reflector), Chapter 4's focus-on-the-eye and the optical reason a phone can't blur a background, and Chapter 3's dynamic-range thinking behind the HDR. Computational photography doesn't replace the fundamentals — it sits on top of them. A photographer who knows where the light should fall and what the frame is about will direct the computation; one who doesn't will be directed by it.
Discussion questions
- We shot the "honest baseline" failures (Phase 1) before letting HDR rescue the scene. Why is that step worth the time, rather than just shooting HDR from the start?
- In Phase 3, stepping the subject two metres forward dramatically improved the synthetic bokeh. Explain why in terms of the depth map — what made the algorithm's job easier?
- Portrait mode blurred the outstretched coffee mug (Phase 3's caught mistake). Explain why, given how the depth map decides what to keep sharp. How is "subject = sharp" different from "near = sharp"?
- We exposed for the bright window, not the face, in Phase 2. Defend that choice using what highlights and shadows can and can't recover.
- In the edit pass we dialed noise reduction down and re-warmed the white balance — undoing things the phone had done automatically. When is overriding the pipeline like this essential, and when is it over-fussy? Where's the line?
Your turn
Run this exact shoot in your own kitchen-window location, with your own willing subject, in about forty-five minutes. Follow the four phases deliberately:
- Baseline: HDR off, shoot the two honest failures (exposed for face, exposed for window). See the dynamic-range problem.
- HDR wide: HDR on, expose for the window, brace, hold through the capture, judge the tone mapping (nudge toward "natural" if it looks flat or haloed). Make the environmental frame.
- Portrait tight: Step the subject forward, moderate blur, mind the hair/glasses/hands. Inspect at 100% and fix the scene (distance, angle) rather than maxing the blur. Make the tight frame.
- Edit: If you shot RAW, walk back any waxy skin (noise reduction down) and any color drift (white balance). Keep the parts of the pipeline that helped; override the parts that lied.
Keep your best frame from Phase 2 or 3 — it's a strong candidate for the chapter's Portfolio Checkpoint (a tricky scene rescued by computation). And write the two curation lines: what the computation added (the range HDR held; the blur portrait mode faked) and what it faked or got wrong (a hair seam; a too-flat tone map; a waxy patch of skin you had to recover). You will have a real portrait and a working understanding of how to make a computational camera do what you want — which is the entire skill this chapter teaches.
Key takeaways
- The backlit window portrait is the canonical scene where a phone's computation both rescues you (HDR holds the face-and-window range one frame can't) and betrays you (portrait mode's depth map struggles with a subject close to a bright, detailed background).
- Shoot the honest baseline first (HDR off, exposed for face, then for window) to see the dynamic-range problem — you can't judge the rescue until you've seen the drowning.
- Expose for the highlights, let HDR lift the shadows. Blown highlights can't be recovered; shadows usually can. Prefer "natural" tone mapping and nudge it back when the merge goes flat or haloed.
- Make portrait mode's job easy before pressing the button: step the subject forward for real background distance, use moderate blur (maximal exaggerates every seam), and mind the four traps — hair, glasses, reaching hands/objects, cluttered close backgrounds. When it fails, fix the scene, not the blur slider.
- The processed JPEG is the phone's opinion; shoot RAW so you can override it — walk back waxy skin (noise reduction down) and color drift (white balance) in the edit. Keep what the pipeline got right; re-decide what it got wrong.
- Computational photography sits on top of the fundamentals — light direction, fill, focus, dynamic range. Know those, and you direct the computation; don't, and it directs you.