Photography Art Direction

Photography & Art Direction

Imagery is the fastest read on a page. A viewer judges "is this real, made by people who care?" from a hero photo before reading a word. Generic stock and default AI output are the loudest slop signals there are — louder than a font, because a bad image is content, not chrome. This skill makes imagery a deliberate, stated, authored choice, exactly as the doctrine Mission demands ("the moat is looking human-made", doctrine/design-doctrine.md §0).

Use When

• Selecting or commissioning photography for a brand, site, hero, deck, report cover, or app.
• Defining a photo treatment system — colour grade, grain, contrast, duotone, crop ratios, focal-point rules — so every image looks like it came from one art director.
• Killing stock-photo cliché: smiling handshakes, headset call-centre agents, "diverse team laughing at salad," lens-flare cityscapes, isolated-on-white product shots with no story.
• Writing an art-direction brief / mood board before a shoot, a stock pull, or a buy.
• Directing an AI image generator for brand imagery — the prompt, the negative constraints, and the accept/reject gate that keeps the output off the slop tells.
• Auditing existing imagery and deciding remake / re-grade / re-crop / replace.

Do Not Use When

• The asset is a vector illustration, spot graphic, or hero illustration → use the illustration skill in this group (sibling).
• The asset is an icon / icon set → iconography-system-design (this group).
• It is data visualisation or a chart → group 12 (data-visualization, dashboard-and-data-product-design).
• You are running the generic visual slop audit across a whole product (UI + features) → the cross-cutting visual-product-slop-audit (group 00/04). This skill owns imagery direction; that one owns the product-wide reject gate.

Required Inputs

• The brand: voice/positioning, audience, the one feeling the imagery must carry.
• Where the image lives: hero / inline / background / report cover / OG card / avatar — and its rendered size and aspect ratios.
• Budget & sourcing path: original shoot, commissioned, premium/editorial stock, free stock, or AI-generated (and whether AI output is acceptable for this surface at all).
• Technical envelope: max file weight, format (AVIF/WebP), dark/light backgrounds, text-overlay zones. (For LCP weight budgets see doctrine/references/web-performance-budgets-2026.md.)

Workflow

1. State the photographic intent first (the anti-slop non-negotiable #1, doctrine/design-doctrine.md §2). One sentence: subject world + light + distance + mood. e.g. "Real clinicians mid-task, available daylight, close and slightly off-centre, calm not triumphant." If you cannot say it, you are not ready to pick an image.
2. Choose the sourcing tier against budget and surface, in this order of preference: original shoot → commissioned/editorial → premium stock used non-obviously (crop/grade it so it stops looking like stock) → AI-generated only where permitted → free stock (last resort, high cliché risk). See references/anti-stock-direction.md §Sourcing ladder.
3. Define the treatment system — the look that unifies every image: colour grade, black point/contrast curve, grain, optional duotone/monotone, and the crop/focal-point/aspect rules. Specify it concretely (values, ratios), per references/photo-treatment-system.md. This is what makes a stock or AI image read as yours.
4. Direct the crop & focal point — most "stock look" dies in the crop. Push the subject off the thirds, crop into the action, leave intentional negative space for text. Define safe text zones and the art-directed ratios (references/photo-treatment-system.md §Cropping).
5. If sourcing AI: write the prompt + the negative-constraint list + the accept/reject gate keyed to the slop tells (doctrine/references/ai-slop-taxonomy.md §Visual tells) — waxy skin, melted backgrounds, gibberish signage, extra fingers, warped logos, mismatched lighting. AI imagery is admissible only when it can pass that gate AND a human re-grades/crops it into the treatment system. Per doctrine §2 sourcing-authority rule, an AI tool's own aesthetic defaults are never the authority for the look — they are what we art-direct away from. See references/anti-stock-direction.md §AI-image direction.
6. Run the reject gate on every candidate (stock or AI): cliché test, slop-tell test, treatment-fit test, performance-weight test. Anything that fails is re-graded, re-cropped, or replaced — never shipped "because it's close."
7. Accessibility & weight: every content image needs meaningful alt text (decorative → empty alt), sufficient contrast for any overlaid text (doctrine/references/wcag-2.2-criteria.md), and a weight inside the LCP budget (doctrine/references/web-performance-budgets-2026.md).

Decision Rules

Capability Contract

Read and search are required for the brief, surfaces, assets, licences, and consent. Editing is allowed only for authorised treatment or selection. Network sourcing, purchase, publication, and representation of real people require separate authority and recorded provenance.

Degraded Mode

If required evidence or tooling is unavailable, use the scoped fallback below and mark the result unverified. Without image assets, deliver the brief, shot list, crop map, and treatment recipe. Without licence, release, or provenance evidence, use placeholders and block embedding or publication.

Anti-Patterns

• The slop cliché set: smiling handshake, headset agent, team-laughing-at-salad, isolated-on-white with no story, generic-gradient-city, fake-candid-with-perfect-teeth.
• Stock used raw — straight off the library, uncropped, ungraded, recognisable. The image is fine; shipping it unauthored is the failure.
• No treatment system — ten images from ten sources, ten different grades, reads as a scrapbook, not a brand.
• AI image shipped un-vetted — waxy skin / 6 fingers / melted background / warped logo (the ai-slop-taxonomy.md tells). This is the highest public-backlash risk (taxonomy §2, "shovelware assets").
• Centre-punched crop with the subject staring down the lens — the default everything-looks- the-same composition.
• Decorative AI sparkle/gradient as "imagery" with no subject and no benefit — slop by § product-tells.

Outputs

• A one-page art-direction brief / board: intent sentence, sourcing tier, treatment spec (grade/grain/crop/ratios), reference frames, do/don't, and (if AI) the prompt + negative list
  • accept/reject gate.
• A reusable treatment recipe (concrete values) that any later image can be run through.
• A reject log for audited candidates.

Examples

• examples/art-direction-board.md — a complete, real art-direction brief for a sample brand (a community health-tech product, "BrightSoma Clinics"): intent, sourcing ladder decision, full treatment recipe, crop rules, and an AI-image direction block with negative constraints and the accept/reject gate. Reusable as a template — not lorem.

References

• references/photo-treatment-system.md — the concrete treatment + cropping/focal-point system (grade, grain, duotone, ratios, safe text zones, weight).
• references/anti-stock-direction.md — the cliché kill-list, the sourcing ladder, how to make stock un-stock, and the AI-image direction handoff (prompt + negatives + gate).
• doctrine/design-doctrine.md — Mission §0 ("looks human-made") and Anti-Slop Charter §2 (state-the-choice; sourcing-authority asymmetry rule).
• doctrine/references/ai-slop-taxonomy.md — the visual tells (the reject-gate checklist) and the "shovelware assets" public-backlash risk.
• doctrine/references/wcag-2.2-criteria.md — alt text & text-over-image contrast.
• doctrine/references/web-performance-budgets-2026.md — image weight inside the LCP budget.