When Oakland band the Westones uploaded a photo to run a Facebook ad, Meta's AI suggested "improvements" that replaced every member of the racially diverse group with white AI-generated figures. It's not an isolated case.

The company says advertisers control the experience. Many of its AI features are on by default and require per-ad manual opt-out.

When Chris Hoog, saxophonist and bandleader for Oakland funk-soul group the Westones, tried to run a Facebook ad for an upcoming show this month, he uploaded a live photograph from the High Sierra Music Festival, shot by local photographer Evan Lanam. What Meta's AI ad interface served up in return, as Hoog documented in a July 14 Instagram reel, was a menu of unsolicited "improvements" aimed at higher click-through rates — and every AI-generated person in those alternatives appeared to be white. The Westones are a racially diverse group.

The suggestions escalated as Hoog scrolled. Early options tweaked lighting and contrast. Deeper in, Meta's AI had replaced Hoog and trumpeter Brendan Liu with two different men gripping AI-generated wind instruments at anatomically implausible angles. One option substituted the entire seven-piece band with a one-armed woman in a sundress standing on astroturf, holding what looked like a Slurpee cup of wine. The final option: a giant disembodied forearm protruding from Liu's torso, gripping a liquor bottle with dummy-text label copy, while Liu continued playing the trumpet.

"It's clear that they're testing their capabilities around brand placement and furthering the capitalist angle," Hoog told Coyote Media, which first reported the story on July 17, "but to do that without any kind of acknowledgement, just hoping that I'm going to click 'next' and not even realize what we've done, is just ridiculous." Meta did not respond to Coyote Media's request for comment.

The case is not isolated. Brie Read, CEO of Snag Tights, told Glossy that Meta's AI altered her brand's ads to show models with three legs and, in multiple instances, replaced Black models with white women — despite having manually opted out of AI features. Ads consultant Jessica Gleim told Business Insider the same tools generated inappropriate and bizarre imagery for a female-founded pajama brand she advises. Business Insider reported in July 2026 that eight separate advertisers and agency executives described Meta's AI ad tools as consistently producing inaccurate and strange creative output.

Meta spokesperson Simone Levien told Marketing Brew: "Advertisers can opt out of AI creative testing at any time... advertisers are always in control of their experience." The design runs the other direction: Meta's AI ad features are auto-enabled by default, require per-ad manual opt-out, and — per the company's own transparency policy published in February 2025 — only trigger AI-disclosure labels for "significant edits" or content featuring "photorealistic humans." Minor alterations carry no disclosure.

The racial homogenization is not new to Meta's systems, and not unacknowledged internally. Meta's own "Building Generative AI Responsibly" guide states that training-data imbalances "may manifest in potentially biased outputs, such as disparities in gender or skin tone representation." A 2024 study on arXiv found Meta's budget optimization tool directed roughly 64% of ad spend toward images featuring light-skinned models. The company settled a U.S. Department of Justice housing discrimination lawsuit in 2022 — agreeing to build a "Variance Reduction System" to reduce racial disparities in ad delivery — and in 2023 the Lawyers' Committee for Civil Rights Under Law filed a separate civil rights complaint alleging Meta's AI steered for-profit college ads disproportionately to Black users.

The Hoog ad eventually ran with the original Lanam photo intact. Two days later, he noticed Meta had auto-attached AI-generated background music. The Westones' music wasn't in Meta's library, so there was no swap option; he found a "disable" toggle buried in advanced settings.

Meta has not addressed the image-alteration incidents reported this month, has not disclosed what guardrails exist on its ad AI's ability to alter racial presentation in uploaded photos, and has not clarified whether the Variance Reduction System from the 2022 DOJ settlement — which addressed ad delivery targeting — applies to the AI tools that modify creative assets before an ad is submitted.