What Toronto's Diverse Professional Teams Are Learning About AI Headshots

The concern about AI headshots is usually framed as a quality problem. The plasticky skin. The symmetrical face that looks rendered rather than lit. The photo that colleagues recognize as AI immediately. Those concerns are real, and Toronto companies have been hearing them since 2024. The more substantive issue is identity — and the research on that is no longer ambiguous.

AI headshot generators do not just produce portraits that look slightly artificial. They systematically alter the appearance of people of color — lightening skin tones, modifying facial features, smoothing hair textures, and in some cases removing religious markers without asking permission. For organizations with stated commitments to diversity, equity, and inclusion, that pattern is a contradiction built into the product itself.

The evidence on this has moved from anecdote to academic literature.

A 2024 peer-reviewed study in JAMA Network Open generated 1,000 AI physician headshots across five popular platforms. Across those outputs, 82% depicted White physicians, against 63% of the actual U.S. physician workforce. Three of the five platforms produced zero images of Latino physicians (Capturely, citing JAMA Network Open 2024). An Asian American MIT student uploaded a selfie requesting a more "professional" result. The AI returned an image with lightened skin and blue eyes. When the incident was reported, the platform's founder acknowledged that the models "aren't smart enough" to avoid this outcome (CNW Media, 2025).

A 2026 controlled study examining AI portrait editing across more than 5,000 images found that 62 to 71% of edited outputs showed lighter skin tones than the source photo. Black and Indian subjects experienced skin lightening at a rate of 72 to 75%, compared to 44% for White subjects. The researchers describe this consistent pattern across all three models tested as a "default to White" prior — a systematic drift in the algorithm's output even when edit instructions said nothing about skin tone (Seochan et al., January 2026).

In March 2026, a J.S.D. candidate at Berkeley Law tested more than 25 AI headshot generators as part of an anti-discrimination research project. Every one removed her hijab from the generated images. None asked. None offered a choice. The hijab disappeared from each output without explanation or opt-out (Complete AI Training, March 2026).

These findings frame the decision professional services firms in Toronto face when evaluating AI headshot tools for a team directory update, a website refresh, or a new LinkedIn rollout for recent hires.

What a firm publishes on its team page signals who works there, what the organization values, and who a candidate or client expects to meet in the room. When that signal is generated by a model with a documented bias toward Eurocentric appearance standards, the gap between the published image and the actual person creates a trust problem alongside the visual one. A client who meets the real employee after seeing an AI-brightened headshot absorbs that gap even without naming it; a candidate scanning a team page before deciding whether to apply is running the same calculation.

The counterargument is cost. AI headshot platforms charge anywhere from $15 to $99 per person depending on the tier, and they eliminate the logistics of scheduling a team in one place. That trade-off is defensible for a small team with uniform headshot needs; for a firm of 30 with a public DEI commitment and a range of skin tones and religious expressions, the documented bias changes the calculation.

A full team day at Omilia Visuals in Toronto covers up to 30 people and starts at $3,200 — approximately $107 per person — with on-set art direction, same-day delivery, and photographs that reflect how each person actually looks (corporate headshot packages). The per-seat cost is comparable to premium AI platforms; the identity distortion is not a category that platforms have priced in.

The skin-lightening bias the research confirms is not limited to fringe applications. It appears across multiple well-known platforms, in controlled academic benchmarks, and in documented accounts from users who uploaded accurate photos and received back images of someone who did not look like them. A poll of 1,600 people in 2026 found that 38% described AI-generated headshots as "soulless" (Metroplex Headshots, 2026). Across the professional photography industry, 2026 is the year the backlash arrived (Blue Bend Photography, 2026). Enterprise firms reversing course cite one reason: the technology produces inaccurate representations of their people.

For a firm whose team page is meant to represent an actual diverse workforce, choosing a headshot tool that alters the appearance of your people is a brand decision, and the research makes clear what that decision communicates about whose appearance the firm treats as the default.

Before the next headshots update, it is worth asking: does the tool you use reflect the people on your team, or does it reflect an algorithm's assumptions about what professional looks like?

Omilia Visuals photographs corporate teams across Toronto, with experience across a wide range of skin tones, backgrounds, and cultural expressions. If you are booking headshots for a diverse team, see how corporate sessions work at Omilia.