Chapter 33 — Quiz

Twenty-five questions to check your grasp of generative imaging, the editing/generating line, disclosure, copyright, and where the photographer's value now lives. Mix of multiple choice, true/false-with- justification, short answer, and "read the image" seeing questions. Write your answers before opening the key. For the justification and short-answer items, a correct conclusion with a wrong reason earns half credit — in this chapter, the reasoning is the point.

Multiple choice

1. The most fundamental difference between a camera and a generative image model is that: - (a) the camera is more expensive - (b) the camera measures light that was really there; the model predicts pixels that were not - (c) the model is always lower resolution - (d) the camera cannot be fooled

2. A diffusion model creates an image by: - (a) photographing a scene through a digital lens - (b) copying the nearest matching image from its training set - (c) starting from noise and removing it step by step, steered by your prompt, until an image emerges - (d) scanning the internet live for a real photo

3. Generative models reliably struggle with all of the following EXCEPT: - (a) the correct number of fingers on a hand - (b) legible real text within the image - (c) plausible-looking generic surface texture - (d) physically consistent shadows and reflections

4. Which tool invents the LEAST new content? - (a) generative fill - (b) AI upscaling - (c) AI masking/selection - (d) generative extend

5. On the capture-to-conjured spectrum, the pipeline first begins genuinely inventing content at: - (a) AI denoise - (b) AI lens correction - (c) AI masking - (d) AI upscaling

6. The Chapter 29 test that still governs AI editing says manipulation is any edit that: - (a) took more than ten minutes - (b) used a generative tool - (c) changes the record of what was actually there - (d) was done on a phone

7. "Generative fill" specifically means: - (a) increasing the file's resolution - (b) selecting an area and having the model replace it with newly generated content - (c) removing noise from a real capture - (d) correcting lens distortion

8. The two layers of honest AI disclosure are: - (a) a watermark and a copyright symbol - (b) plain-language labeling (for humans) and content credentials/provenance (verifiable, embedded) - (c) a hashtag and a caption - (d) the file name and the folder name

9. Which statement about copyright and AI output is most accurate as a general shape? - (a) AI images are always owned by the person who typed the prompt - (b) photographs you capture are clearly yours, while purely-prompted images may be protectable by no one - (c) AI images can never be copyrighted under any circumstances - (d) the model's company always owns the output

10. As plausible AI images become infinite and nearly free, the economic value of a true, witnessed photograph: - (a) collapses to zero - (b) is unaffected - (c) rises, because the scarce thing is now what the flood cannot make - (d) depends only on the camera used

11. Which of these is NOT something a generative model can structurally do? - (a) produce a plausible generic landscape quickly - (b) average many photographers' styles - (c) be present at a real, singular moment and witness it - (d) generate infinite variations

12. "The AI generated it" as a defense when a generation depicts a recognizable real person: - (a) fully removes your liability - (b) has protected no one — you chose the prompt and published the result; right of publicity still applies - (c) only matters for celebrities - (d) transfers responsibility to the model's company

True / False — and say why

For each, mark True or False and give a one-sentence justification. A right label with a wrong reason is only half right.

13. "Because the image looks completely photorealistic, it must be a record of something real."

14. AI masking/selection is controversial because it changes the identity of your pixels.

15. Removing a sensor-dust spot from the sky and removing a real person from the sky are ethically the same operation because both might use a fill tool.

16. A single #ai hashtag buried at the end of a forty-tag caption is adequate disclosure for a news photograph.

17. Keeping your unaltered RAW files helps establish human authorship and that a real scene was captured.

18. Disclosure that appears only in a visible caption (and not in the alt text) fully informs every viewer of an image's AI origin.

Short answer

19. In one or two sentences, explain why "did I use AI?" is a weaker test for the editing/generating line than "did the source pixels come from the scene?"

20. Name three of the five capacities from §33.6 that a generative model structurally cannot have, and say in a phrase why each is not a temporary limitation.

21. State the three independent copyright questions about AI imaging, and confirm in one line that an answer to one does not settle the others.

22. Why must the editing/generating line "live in you as a decided rule" rather than being judged from the finished image?

Read the image (seeing questions)

23. You are handed an image presented as a captured wildlife photograph of an owl on a branch at dusk. Name three specific visual "tells" — drawn from how generative models fail — that would make you suspect generative fill or full generation, and say what a genuine capture would show instead for each.

24. A real-estate listing photo shows a house with a clean, empty blue sky and a perfectly tidy yard. You learn the original capture had a prominent power line crossing the sky and a neighbor's trash bin visible at the edge, both removed with generative fill. Using the chapter's tests, decide whether this is enhancement or manipulation, and explain why the genre matters to your answer.

25. Below are two finishes of the same dawn-beach capture, in which a red bucket sat near the waterline: - Finish A: denoised, shadows lifted, warmth graded, horizon straightened, cropped tighter. The bucket remains. - Finish B: all of the above, plus the bucket generative-filled away and replaced with invented sand.

A viewer cannot tell them apart by eye. (i) Which finish is still a photograph in the chapter's sense, and why? (ii) What does the fact that they're visually indistinguishable prove about where the editing/ generating line must live? (iii) For each finish, write the honest one-line disclosure you would attach.

Answer Key

1. (b) — A camera measures real light; a model predicts pixels with no scene behind them. (§33.1)

2. (c) — Diffusion = reverse the noise, steered by the prompt, until an image condenses. (§33.1)

3. (c) — Generic plausible texture is exactly what models are good at; specifics (fingers, text, shadows) are what they fail. (§33.1)

4. (c) — AI masking invents nothing; it only understands your real frame. (§33.2)

5. (d) — Upscaling invents plausible fine detail; everything before it refines the real signal. (§33.2)

6. (c) — Manipulation = changing the record of what was there, regardless of tool or effort. (§33.3, Ch.29)

7. (b) — Select an area, model replaces it with new generated content. (§33.2)

8. (b) — A human-readable label and verifiable embedded provenance; use both. (§33.4)

9. (b) — Captured photos are clearly yours; pure prompts may lack the human authorship copyright requires. (§33.5)

10. (c) — Scarcity shifts to the true/witnessed image the flood can't produce. (§33.6)

11. (c) — Witnessing a real singular moment is precisely what a non-present, non-record model cannot do. (§33.6)

12. (b) — You chose and published; right of publicity/consent attach regardless of tool. (§33.5, Ch.32)

13. False — Photorealism is the model's default output, not evidence; looking real and being real are fully decoupled. (§33.1)

14. False — Masking is uncontroversial because it changes nothing about the pixels' identity; it only understands the frame so you can adjust your own real pixels. (§33.2)

15. False — The Chapter 29 test turns on whether the removed thing was incidental to the photo's claim: the dust was never in the world (enhancement); the person was genuinely there (manipulation). (§33.3, Ch.29)

16. False — Technically disclosed ≠ disclosed; the test is whether a normal viewer in the normal way would understand, and a buried hashtag fails it — especially at news stakes. (§33.4)

17. True — The unaltered RAW is your strongest evidence of human authorship and of what the scene actually contained. (§33.5)

18. False — Disclosure that reaches only sighted viewers has a gap; the AI origin must also appear in the alt text so non-sighted viewers receive it. (§33.4)

19. Because tools keep changing and blurring, but the question of whether an image is a record of light or a fabrication of plausible pixels is stable; anchoring ethics to the image's relationship to reality survives the technology, while "did I use AI?" does not (AI denoise is honest; a clone-stamp manipulation predates AI). (§33.3)

20. Any three of: witness (it isn't there / isn't a record), presence to a real moment (the moment is real and gone — only a present person catches it), relationship/consent (a human transaction with a real subject), voice (the residue of a real particular life), responsibility (only a person can owe and honor duties). None is temporary because each follows from what a generative model is. (§33.6)

21. (1) Output ownership — is the generated image mine to own? (2) Training-data legality — was it lawful to train on scraped images? (3) Output infringement — can a generation infringe a work or a real person? Each is independent; e.g., the output could be unownable yet still infringe. (§33.5)

22. Because competent generation is invisible in the result — an edited and a generated finish can be indistinguishable by eye — so a rule based on appearance cannot work; the rule must be a decision you make about the image's source and honesty before your finger is on the tool. (§33.3)

23. (Seeing — model answer.) Tells: (i) the owl's eye catchlights don't match the scene's actual light direction/sources — a genuine capture's catchlights point at the real light (Chapter 5); (ii) feather or bark texture that repeats or "swirls" unnaturally — real plumage/bark has irregular, non-tiling detail; (iii) physically impossible geometry where talons meet the branch, or a shadow that doesn't agree with the highlights — a real capture has one consistent shadow logic. Other valid tells: mushy or invented "text" on any sign/band, inconsistent depth of field, a background that doesn't resolve into a real place. (§33.1)

24. (Seeing — model answer.) This is manipulation, because a listing photo makes a claim about reality — "this is what you would see from this house" — and removing a real power line and a real bin falsifies that claim, materially, for a buyer making a decision. The genre is decisive: the same removal of an incidental stray object from a piece of personal landscape art could be defensible enhancement (Chapter 29), but here the image's entire purpose is to represent the property truthfully, so the standard is strict and the removal misleads. (§33.3, §33.5, Ch.29)

25. (Seeing — model answer.) (i) Finish A is still a photograph: every pixel's source is the light that actually reached the sensor; it's the best version of what was there. Finish B asserts an empty beach that wasn't empty — pixels exist that no light made — so it has crossed into generating. (ii) Their visual indistinguishability proves the line cannot be judged from the result; it must live in the photographer as a decided rule about the image's source and honesty. (iii) Finish A: "Real capture; denoised, graded, cropped — no elements added or removed." Finish B: "Real capture with object removed via generative fill (bucket replaced with generated sand)." (§33.3, §33.4)

Topics to review, by question

• Q1–3, 13, 23: What models do/don't do; plausibility vs. evidence; failure modes — §33.1.
• Q4–5, 14: The AI editing pipeline; sorting tools by invention — §33.2.
• Q6–7, 15, 19, 22, 24, 25: The editing/generating line; the enhancement/manipulation test; removal — §33.3 (with Chapter 29).
• Q8, 16, 18, 25(iii): Disclosure on two layers; provenance/content credentials; matching detail to stakes — §33.4.
• Q9, 12, 17, 21: Copyright's three questions; ownership; training data; infringement and real people — §33.5 (with Chapters 32, 35).
• Q10–11, 20: Where the photographer's value is irreplaceable; the economics of scarcity — §33.6.