Chapter 33 — Exercises

This chapter is unusual: some of its most important exercises are done with a camera, but several are done with your judgment — sorting tools, drawing lines, writing disclosures, taking positions. That is exactly right, because the skill this chapter builds is not a technique but a stance, and a stance is built by arguing with concrete cases until you know where you stand. Difficulty is marked ⭐ (warm-up), ⭐⭐ (core), ⭐⭐⭐ (stretch). Do the starred and odd-numbered ones at minimum; model answers and critiques for those are in the book's Answers to Selected Exercises. Keep what you shoot — some of it feeds the Portfolio.

A standing rule for this chapter: when an exercise has you use a generative tool, you are doing it to understand it, not to deceive anyone. Label every constructed or AI-altered image as such, every time.

Part A — Sort the tools (Read the Frame / concept drills)

The first skill of this chapter is triage: given any AI feature, knowing instantly how much it invents and therefore which side of the line it sits on. These drills build that reflex. They are quick, they need no camera, and they are the foundation for every judgment that follows — you cannot decide whether a use is honest until you can say precisely what the tool does to the record.

Exercise 33.1 ⭐ — Place them on the spectrum. Without looking back at Figure 33.2, write these five tools in order from least inventive to most inventive: generative fill, AI denoise, AI upscale, AI masking/selection, AI lens-correction. Then write one sentence for each saying what it does to the record of your real capture.

Exercise 33.2 ⭐⭐ — Editing or generating? For each operation below, label it Editing (refines what you captured) or Generating (invents new content), and give your one-line reason: (a) AI-denoise a noisy ISO 12800 night frame; (b) generative-fill a flock of birds into an empty sky; (c) AI-mask the sky to darken it; (d) extend the left edge of the frame with invented scenery to change the crop; (e) modest 1.5× upscale of a landscape for a 13×19 print; (f) replace a dull overcast sky with a generated dramatic sunset.

Exercise 33.3 ⭐⭐ — The two-second crossing. List three generative operations that live one tap away from a harmless edit in a phone camera/gallery app (e.g., "magic eraser" sits next to crop). For each, name the harmless neighbor and explain why the location of the button tells you nothing about what it does to the record.

Exercise 33.4 ⭐⭐⭐ — Predict the failure. Knowing only that a diffusion model optimizes for plausibility, not truth, predict five specific kinds of detail it will tend to get wrong in a complex generated street scene. For each, explain why the mechanism causes it (don't just name it). Then, if you have access to any generator, test one prediction and report what you found.

Part B — See the line (Read the Frame)

Now you put the triage to work on real images and real decisions. The hard truth these exercises drive home is the one from Figure 33.4: the editing/generating line is usually invisible in the result, so you must learn to locate it by reasoning about an image's source and claim, not its appearance. Do these slowly; the goal is to internalize a rule you can apply at the speed of a finger hovering over a tool.

Exercise 33.5 ⭐ — Write the two-finish pair. Take any real photo you own that has a small distraction in it. Write two Described Photographs (six fields each) of the same capture: VERSION A finished by editing only (the distraction remains; you've improved tone/crop/grade), and VERSION B finished by generative-removing the distraction. The fields should be identical except THE CHOICES and THE EFFECT. The point is to feel how invisible the difference is in the result.

Exercise 33.6 ⭐⭐ — Resolve the removals. For each described removal, decide enhancement or manipulation using the Chapter 29 test, and justify in one sentence: (a) removing a sensor-dust spot from a clear sky in a fine-art print; (b) removing a "Closed" sign from the door of a restaurant in a press photo about the restaurant reopening; (c) removing a stranger who wandered into the far edge of your personal landscape; (d) removing a wedding ring from a portrait delivered to a client; (e) removing a power line from a real-estate listing photo of a house's view.

Exercise 33.7 ⭐⭐ — Read the fingerprints. You are handed an image claimed to be a captured wildlife photograph. Describe five visual "tells" that would make you suspect generative fill or full generation (think about what §33.1 said the models get wrong). For each tell, say what a genuine capture would show instead.

Exercise 33.8 ⭐⭐⭐ — The honest re-shoot. Find one of your own photos that you are tempted to "fix" with generative fill (a distracting object, an empty sky you wish were dramatic). Instead of fixing it in software, design the re-shoot that would solve the problem with light, timing, vantage, and framing (Chapters 1, 5, 9). Write the shot plan. Then, if you can, go execute it and compare the honest capture to what a generative fix would have faked.

Part C — Disclosure and provenance (Settings/Workflow drills)

If the line is invisible in the result, then honesty depends on telling people — which makes disclosure a practical craft, not a moral afterthought. These drills train you to write disclosures that actually inform (not merely satisfy a rule), to use both layers, and to match the detail of your label to the stakes of the image. A disclosure no one will see is not a disclosure; a disclosure scaled wrong for its context either under-informs or buries the signal. Get both right here.

Exercise 33.9 ⭐ — Match the disclosure to the stakes. Write the appropriate plain-language disclosure label for each use: (a) a personal Instagram art post that is a full AI generation; (b) a news photo that was only AI-denoised and cropped; (c) a magazine illustration that composites a real photo with generated elements; (d) an ad featuring a generated human face. One sentence each; match the detail to the stakes.

Exercise 33.10 ⭐⭐ — Find the burial. Here is a described disclosure practice: "I put #ai as the last of forty hashtags, three line-breaks below the caption, and I let the upload strip the file's metadata." Explain, in terms of the chapter's test, why this is technically disclosed but not disclosed. Then rewrite it into an adequate disclosure for a news context and for a personal-art context.

Exercise 33.11 ⭐⭐ — Two layers, on purpose. For one real AI-assisted image of your own (e.g., a denoised, masked, graded capture), write both layers of disclosure: (1) the plain-language sentence a human would read under the image, and (2) a plain-English description of what the embedded content credentials should record about how it was made. Explain why neither layer alone is sufficient.

Exercise 33.12 ⭐⭐⭐ — Alt text with disclosure. Write alt text (an image description for a blind or low-vision viewer, Chapters 1 and 34) for a heavily AI-edited image of yours that also discloses the AI origin within the alt text itself. Explain why disclosure that reaches only sighted viewers is incomplete, and how doing both at once sharpens your own seeing.

Part D — Use the tools to understand them (Shoot This)

You cannot have an informed stance on tools you have never touched. These exercises put the AI tools in your hands so that you understand their edges — where denoise stops restoring and starts smearing, where upscale stops extending and starts fabricating, what an honest composite looks like when it is presented honestly. Use everything here as a teaching demonstration, fully labeled; the point is comprehension, never deception.

Exercise 33.13 ⭐ — Denoise honestly. Make one genuinely noisy real photograph (shoot at your highest ISO in dim light, Chapter 3). Run AI denoise on a copy. Compare the two at 100%. Note what it cleaned and what it smeared or invented (a freckle, a star, fine fabric texture). Keep both, labeled, and write one line on where denoise stopped being restoration.

Exercise 33.14 ⭐⭐ — Upscale to the breaking point. Take a small image and upscale it progressively — 2×, 4×, 8× — with AI upscaling. At each step, find and describe the invented detail (eyelashes, bricks, gibberish text). Identify the magnification where it crosses from "plausible restoration" to "fabrication," and explain how you knew.

Exercise 33.15 ⭐⭐ — The labeled composite. Make one deliberately AI-composited image — a real photograph of yours with an openly generated element added — presented honestly as an AI composite, with a plain-language label baked into the caption and (if your tools allow) content credentials embedded. The exercise is not the composite; it is the honest presentation of it.

Exercise 33.16 ⭐⭐⭐ — The comparison piece. Build the teaching pair from Figure 33.4 with your own capture: one real photo finished two ways — editing (content untouched) and generating (one element added or removed) — displayed side by side and captioned to teach a viewer exactly where the line is. This can go straight into your Portfolio as the chapter's disclosed comparison piece.

Part E — Where you still matter (Shoot This)

This is the heart of the chapter turned into assignments: go make images whose value lives precisely where the machine cannot reach — witness, presence, relationship, voice. Each of these sends you out to produce something true and specific and yours, and then to articulate what about it a generation could never supply. Keep your best frames; several are strong candidates for the Portfolio.

Exercise 33.17 ⭐ — Photograph a witness. Make one photograph whose value is witness — something true and specific that is happening now, dated to today, in one of your recurring locations (the busy intersection, the market). It need not be dramatic; it must be real and yours. Caption it with the date and place.

Exercise 33.18 ⭐⭐ — Could a model have made this? Shoot a small set (20 frames) in real light of a real subject, keep your best 3, and for each ask the chapter's question: could a generative model have produced this? Rank them by how confidently the answer is "no — because it really happened and only I would have framed it this way." Write why your top frame resists the machine.

Exercise 33.19 ⭐⭐ — Relationship and consent. Photograph a real person you know, in real light, at a real moment, with their informed consent (Chapter 32). Afterward, write two or three sentences on what the relationship contributed to the frame that no generated face could carry. This is the irreplaceable column of Figure 33.7 made concrete.

Exercise 33.20 ⭐⭐⭐ — The voice test. Lay out five of your own keepers from across this book. Identify the recurring choices — your light, your distances, your moments — that make them recognizably yours (Chapter 39 previews this). Write a paragraph arguing why this particular voice is something a model that averages ten thousand photographers cannot reproduce.

Part F — Position and policy (Reflection / Portfolio)

A working photographer in this decade is expected to have a position, not just opinions — a stance specific enough that clients, editors, and viewers can rely on it. These exercises build that position in stages, from a single line you will never cross to a full written stance and an enforceable contest policy. Resist vagueness at every step: a stance is only worth the specifics it commits to.

Exercise 33.21 ⭐ — One line you will never cross. Write the single AI operation you will never perform, and the reason, in one sentence. (You will expand this into a full stance in Exercise 33.24.)

Exercise 33.22 ⭐⭐ — The three copyright questions. In your own words, state the three independent copyright questions from §33.5 and answer each as best you can for your own country, today — flagging clearly where the answer is unsettled. Note one practical step (e.g., keeping RAWs) that protects you regardless of how the law resolves.

Exercise 33.23 ⭐⭐⭐ — The contest rulebook. You are asked to write the AI policy for a local photography contest. In about 200 words, draft rules that: define what counts as a "photograph" for the contest, state what AI processing is allowed (e.g., denoise, masking) and forbidden (e.g., generative add/remove), require disclosure, and say how disputes about provenance will be handled. Make it specific enough to actually enforce.

Exercise 33.24 ⭐⭐ — Write your stance (Portfolio). Complete the Portfolio Checkpoint: write your 150–300-word stated stance on AI in your work. It must name which tools you use, where on the capture-to-conjured spectrum your line sits, what you will never do, and how you disclose. Vague language is a failed answer; commit to specifics a viewer could hold you to.

Part G — Interleaved (mixing prior chapters)

The chapter does not stand alone — it completes threads from across the book. These exercises deliberately braid Chapter 33 with earlier skills: reading light (Chapter 5) becomes your forensic tool; file discipline (Chapter 30) becomes provenance; RAW development (Chapter 26) and critique (Chapter 31) frame honest use. Interleaving like this is how durable understanding forms — and how you'll actually work, since no real shoot uses one chapter at a time.

Exercise 33.25 ⭐⭐ — Light tells the truth. Take a generated or heavily AI-filled image (yours or a clearly-labeled public example) and analyze its light with everything you learned in Chapter 5: is the direction consistent across all objects? Do the shadows agree on one or more real sources? Do catchlights match? Write where the light betrays that no single real scene was ever lit. (This is your strongest forensic tool.)

Exercise 33.26 ⭐⭐ — Provenance meets backup. Connect §33.5 to Chapter 30. Describe how your file-management and 3-2-1 backup workflow should treat RAW originals as evidence of authorship and provenance. What file would you produce, and in what form, if someone challenged whether one of your images was a real capture?

Exercise 33.27 ⭐⭐⭐ — The full honest pipeline. Take one real capture from RAW (Chapter 26) to a finished, shared image, using AI tools only on the editing side of the line (denoise, masking for dodge/burn per Chapter 28, modest upscale), and produce: the finished image, a one-line plain disclosure, and a short written log of every step. The deliverable proves you can use AI honestly end to end.

Exercise 33.28 ⭐⭐ — Critique with the new vocabulary. Return to the critique method of Chapter 31 (describe → analyze → interpret → judge) and apply it to a generated image and a captured image of the same subject. Where does the critique vocabulary work identically, and where does the question "what is this an image of?" change everything?

Stretch challenge

Exercise 33.29 ⭐⭐⭐ — A week of the real. For seven days, make one photograph each day whose value is something a model cannot produce — witness, presence, relationship, or voice — and one Light Log entry. At week's end, lay the seven out and write a page on what they have in common that no generator could fake. This is the chapter's thesis proven in your own work, and likely the seed of your Portfolio piece.

Model answers and critiques for the starred and odd-numbered exercises appear in the book's "Answers to Selected Exercises." Wherever an exercise asks you to take a position, there is no single correct answer — but there are answers that are specific and defensible and answers that hide. Aim to be held to yours.