Lesson 58.1 gave you the boundary: improve the rendering, never rewrite the facts. This lesson applies it to the jobs students actually bring to class: a product photo that needs to sell, a headshot that needs to impress, a deck visual that needs to look clean, and a document scan that needs to read clearly.

The workflows differ by placement, but the principle never changes. A beautiful incorrect asset is not a win. It is a liability.

Product images: accuracy first, beauty second

Product geometry, labels, colors, sizes, ingredients, features, and customer claims require special care. The buyer cannot touch the object. The photograph is the promise.

Safe product workflow, in order:

1. Lock the truth. Photograph the real product in neutral light. Record true color values, label text, dimensions, and any visible flaws. This reference decides every later edit. 2. Clean globally. Correct white balance against a neutral reference, lift exposure, recover shadows, reduce noise. Check color against the physical object on a calibrated view, not just on a phone screen. 3. Clean locally. Remove dust, stray reflections, and background clutter with tight masks. Never touch label lettering, certification marks, ingredient lists, or size indicators. If text is illegible, reshoot — do not let an upscaler invent sharper letters. 4. Handle the background deliberately. Background removal, neutral replacement, relighting, and crop expansion are acceptable when the background is not part of the claim. A clean white, soft gray, or brand-color sweep is honest. A marble kitchen, gym shelf, or mountain trail implies a context the product never earned — use it only for clearly illustrative lifestyle layouts, and keep one accurate plain-background master. 5. Expand with restraint. Outpainting a little extra background for a banner crop is fine. Extending the product itself, duplicating items, or inventing packaging details is not. 6. Generate around the asset, not over it. The durable pattern for lifestyle scenes: keep the accurate product cutout untouched on its own layer, then generate the supporting scene behind or beside it. If the generated scene casts impossible shadows or reflections, fix the scene — never warp the product to match.

Special cases: apparel must keep true color and drape; food must keep actual portion and ingredients; cosmetics and supplements must keep label claims verbatim; refurbished or used goods must keep visible wear disclosed. When in doubt, ship the slightly less glamorous accurate image. Returns and chargebacks cost more than a reshoot.

One workflow, five placements

Decks deserve a warning: five product images should look like one campaign, not five unrelated experiments. Reuse one background rule, one lighting direction, and one crop logic. Documents deserve a stricter one: a scanned contract, receipt, or ID must never be generatively "cleaned" into different words, numbers, or signatures.

Profile portraits: flattering and credible

A good profile photo helps someone look confident, approachable, and put-together — on a dating app, a professional bio, a speaker page, or a social profile. AI can clean up distracting backgrounds, fix a weak crop or bad light, build a stronger setting around a genuine portrait, and help the person test which image feels most like their best self. That is a real use case, not a footnote.

The failure mode is equally real: a random glamour shot with no relationship to the person's life. Over-smoothed skin, reshaped jawlines, borrowed wardrobes, and fantasy settings create matches, interviews, and introductions that collapse on first meeting. The goal is your best self, still unmistakably you.

Write a short profile-image brief before generating anything:

• The real person. Age range, features that must stay exact (face shape, eyes, smile, hair, glasses, visible marks), and what friends would recognize instantly.
• The feeling. Dating: warm, relaxed, genuine. Professional: competent, approachable, current. Speaker: authoritative and energetic. Pick one primary feeling, not three.
• A fitting setting and activity. Choose something from the person's real life: a bright café, a workshop, a trailhead, a clean studio backdrop. Never a yacht, jet, or conference stage they have never stood on.
• Wardrobe and color. What they actually wear when at their best, in colors that suit them and meet platform norms.
• The platform crop. Square for avatars and bio cards; vertical for dating and social profiles. Compose for the crop — eyes on the upper third, no amputated hands, no text under interface overlays.
• Stay-them rules. No face reshaping, no age reduction, no body reshaping, no invented hairstyles, no removed glasses or marks. Light, background, and crop may improve. Identity may not.

Then test honestly: show the edit to someone who knows you and ask, "Is this me on a good day?" If they hesitate, you went too far.

Practical exercise: two placements, one person

Prepare a genuine personal portrait — your own or a volunteer's with permission — for two placements:

1. Square profile/card (LinkedIn, bio page, speaker card): head and shoulders, clean background, confident calm expression, copy-safe margin. 2. Vertical social/dating image (4:5 or 9:16 safe): slightly wider scene, real-life setting, warm approachable feeling.

Deliver each export plus a one-page edit record: source photo reference, intended first impression in one sentence, every edit listed (background cleanup, relight, crop, retouch), and the details that must stay consistent with the person. Keep the source file. If a platform, employer, or date asks what changed, you can answer precisely.

Finish line: two strong, platform-ready profile images plus a transparent edit record — flattering, credible, and defensible.

Common failures: smoothing skin until texture vanishes (recovery: dial retouching back until pores survive at 100 percent); expanding into a fake luxury background (recovery: return to a real, modest setting); shipping only one crop and letting the platform auto-crop the forehead (recovery: compose and export per placement from the approved master).

Check your understanding

• Why must label text be reshot rather than regenerated when it is illegible?
• What is the "generate around the asset" pattern, and which layer must never be warped?
• What six elements does a profile-image brief require, and what do stay-them rules forbid?
• Why do you export square and vertical versions separately instead of relying on one generation?

In the next lesson you will make this work durable: a naming, folder, and review system so every approved master, variant, and export survives contact with collaborators, campaigns, and channel crops.