September 13, 2026
SELECT, REVIEW, AND DELIVER AN IMAGE THAT CAN ACTUALLY BE USED

This is where Part XIII earns its promise: a generated draft becomes a publishable asset — or it does not ship. Everything before this lesson made options. This lesson makes decisions.
Dramatic is not usable
A dramatic image wins attention for three seconds. A usable image survives the next ten checks:
- Accuracy: does the product, person, label, or data match reality?
- Hands / faces / products: the classic failure zones — extra fingers, melted text, drifting loaf shapes.
- Brand consistency: does it match VISUAL-SYSTEM.md (57.4) side by side with its siblings?
- Copy space: is the reserved title/headline area actually empty and large enough?
- Crop safety: does the subject survive 16:9, 1:1, and 9:16 without decapitation?
- Legibility: any baked-in text readable at publish size? (Better: no baked-in text at all.)
- Accessibility: can the image's information reach everyone (alt text, contrast, no color-only meaning)?
- Rights / permission: do you have the right to publish every element — references, faces, brands, training-safe pipeline?
Run this list on the Classroom hero, the bakery loaf, and the research metaphor alike. The bakery's gorgeous loaf with the wrong scoring pattern fails. The hero with the title colliding into the subject fails. The metaphor that looks like a real screenshot of a real tool fails — it manufactures evidence.
The seven-row review table
Score each candidate pass/fail with a one-line note. One fail in rows 2, 5, or 6 usually blocks publication; the rest decide between revise, reshoot, or reject.
| # | Check | Pass? | Note |
|---|---|---|---|
| 1 | Creative fit — serves the one-sentence message and audience | | |
| 2 | Factual / product accuracy — nothing invented about the claim | | |
| 3 | Composition — subject, copy space, crop safety across placements | | |
| 4 | Technical defects — hands, faces, text errors, artifacts, resolution | | |
| 5 | Text errors — spelling, fake words, illegible labels | | |
| 6 | Risk / permission — faces, brands, likeness, source rights cleared | | |
| 7 | Crop / export readiness — master + variants at correct ratios and sizes | | |
Grade honestly. A "pass with revision" is a fail with homework: name the fix ("widen left copy space; regenerate background only") and re-review. Class 58 teaches repair workflows; this table tells you whether repair is warranted or the direction needs regenerating.
The right medium follows the claim and the trust needed
Not every job wants a generated image. Choose per claim:
- Generated image: best for illustration, metaphor, background plates, and concept exploration — where no one will mistake it for documentation. The research card-drawer metaphor, the Classroom geometric hero.
- Original photography: best when truth matters — the actual loaf, the actual founder, the actual shop. If the audience must trust "this is the product," photograph it.
- Licensed stock: best for fast, generic, non-claim context (a morning street, hands kneading — clearly illustrative, properly licensed). Never as fake evidence of your process.
- Screenshot: best for "this is what the tool actually shows" — captured, unedited except for crop/redaction, labeled as a capture with date/version.
- Diagram / chart: best for structured information — flows, comparisons, numbers. Data lives here, drawn from sources, never "beautified" into new values.
Combinations are normal: generated hero background plus real product photo composited in; screenshot plus diagram annotation. The rule never changes: a visual may illustrate a claim, but it must not make a testimonial, event, capability, or result appear to exist when it does not. When in doubt, caption the image's status plainly ("illustration," "product photo," "screenshot, March 2026").
Alt text is information, not SEO
Alt text answers: what information does this image contribute to someone who cannot see it? Write one concise sentence naming the content and function:
- Weak: "bakery image bread delicious IMG_0421" (keywords + filename, no information).
- Strong: "Illustration of a card drawer with one glowing card being pulled, representing targeted retrieval for the article's RAG explanation."
- Strong: "Photo of today's sourdough loaf on the bakery counter; headline text beside it reads 'Baked at 6 AM daily.'"
Name the subject, the relevant detail, and any text the sighted reader gets. Keep it under ~140 characters when possible. If the image is purely decorative (a background texture carrying no information), empty alt (alt="") plus the real content in surrounding text is the honest choice. Never stuff keywords; never paste the filename.
Exercise: ship a DELIVERY-PACK/
Take your winning candidate from 57.2's exploration, confirmed against 57.4's system, and build:
DELIVERY-PACK/
master.png (largest clean export, correct ratio per brief)
hero-16x9.png (desktop/OG crop)
feed-1x1.png (square crop, subject safe)
vertical-9x16.png (story/Reel crop IF the brief needs it)
alt-text.md (one entry per export)
caption-claim-check.md (claim, source/reference, status label)
review-checklist.md (the 7-row table, filled, dated, signed)
sources.md (brief, references, model/date/settings, permission notes)
Templates:
# alt-text.md
- master.png: [one sentence of contributed information]
- hero-16x9.png: [same, adjusted if crop changes meaning]
# caption-claim-check.md
- Published caption:
- Claim the image supports:
- Evidence/source:
- Status label: (illustration / product photo / screenshot [date] / diagram from [source])
- What this image does NOT claim:
Finish line: a DELIVERY-PACK/ with final export, crop variants, alt text, caption/claim check, filled review table, and source note — everything a publisher needs without asking you a question.
Verify fast: hand the pack to a collaborator and ask them to publish cold. If they hesitate on ratio, alt text, or "is this real?", the pack is incomplete. Common failure: shipping the dramatic favorite instead of the usable winner. When the two differ, ship the usable one and keep the dramatic one in explorations.
Check your understanding
1. Name three differences between a dramatic image and a usable one. 2. Which review-table rows block publication outright, and why? 3. A research article needs a chart, a product launch needs a loaf photo, a lesson needs a hero. Assign each a medium and defend the trust logic. 4. Write alt text for your delivery image in one sentence. What did you leave out, and why?
Next
Class 57 closes with a publishable image and a repeatable path: brief → direction → engine choice → system → review → delivery. Class 58 opens the neighboring skill: improving an existing photograph — relight, repair, reframe — without changing what it truthfully shows.
ARTICLE DISCUSSION
JOIN THE
CONVERSATION.
Got a question, a take, or a better way to do this? Log in and leave a comment.
