People imagery must consider representation and diversity

What it is

Representative people imagery examines the range of people, assigned roles, and composition across photography, illustration, avatars, tutorials, and campaigns. It prevents a product from making one appearance the unmarked “ordinary user” or repeatedly placing some people in assisted, low-power, or decorative roles. Diversity does not mean mechanically fitting every identity into every image, and appearance cannot reliably reveal identity. Localization delivery studies the actual target audience, audits coverage and roles, and supplies authorized assets that fit the specific context.

Why it happens

Image libraries amplify collection bias through search tags, default featured sets, and reused templates. Even when no individual picture is conspicuously harmful, the product as a whole can repeatedly omit people across occupation, age, skin tone, body, disability presentation, family structure, and clothing. Counting who appears also misses role bias: who is consistently the expert, decision-maker, caregiver, recipient, or risk subject. Maintainable governance stores visible attributes, narrative role, setting, provenance, license, and self-described metadata separately. Coverage analysis operates on the asset portfolio weighted by exposure, and target-market reviewers examine composition and context; software does not guess identity from faces.

Studying it

Weight an asset audit by high-exposure touchpoints and record who appears, which role they hold, whether they appear alone or in interaction, and who remains after cropping. Use reliable metadata and necessary human review for classification; do not infer hidden identities from images. In target-market sessions, let participants browse a complete journey and identify which scenes seem to address people like them, which feel like decorative collage, and which role assignments feel implausible. Probe the composition rather than asking only whether the set is “diverse enough.” When comparing replacements, measure task comprehension, credibility, brand fit, and perceived representation, analyzing by relevant lived experience and product context instead of treating a country as a homogeneous sample.

Where it stops holding

Population proportion is not a quota for every page, and a count cannot prove respectful or accurate portrayal. Some identities are not visible, and false annotation creates additional harm. Real-person assets also require appropriate consent, withdrawal, sensitive-attribute handling, and protection of minors. A service designed for a particular population can foreground that audience without making it the sole image of an entire market. Visual representation and inclusive language affect one another but are different workflows: this entry audits image selection, composition, roles, and asset metadata rather than replacing copy review.

Applying it

• Register photos, illustrations, avatars, tutorials, and video with provenance, license and consent scope, self-description or reliable metadata, visible attributes, narrative role, setting, crop variants, market review state, and expiry. Prohibit automated face inference from filling identity fields.
• Audit the complete journey with exposure weighting for home, onboarding, help, and paid media, crossing “who appears” with “in which role.” Track absence, role confinement, tokenistic insertion, and exclusion by cropping as separate issues.
• Define an authentic situation and role before sourcing or producing replacements. Invite relevant communities and target-market reviewers to examine composition, clothing, assistive devices, and relationships. Use replaceable resource keys rather than baking people imagery into shared components.
• Generate each market's complete visual journey before release and review representation, task clarity, authorization, and alternative text together. Block further publication when review expires or consent is withdrawn, and fall back to an approved non-person or neutral asset until a replacement is approved. Replace all derived crops, caches, and distributed artifacts through a versioned asset update.

Related

• Same group: S3.02.1 Gesture icons carry cross-cultural offense risk · S3.02.2 Animal, religious, and bodily imagery can be sensitive · S3.02.3 Physical-object metaphors may lack a local referent
• Adjacent: C3.34.3 Hand imagery in gesture tutorials should represent varied skin tones and hand shapes · T1.08.1 Avoid default assumptions about ability, gender, or age
• Search terms: inclusive imagery · representation audit · diversity in visual assets