REI ran an Instagram ad of a two-handlebar bike after Meta's AI quietly altered the photo

Tombstone icon

In June 2026, outdoor co-op REI ran an Instagram ad for a Van Rysel road bike in which the bicycle had two sets of handlebars, misrouted chains, mismatched brakes, and gibberish frame text, atop a human figure that looked stitched together from several people. The original was a real photo from a Van Rysel shoot featuring cyclist Amity Rockwell. REI said Meta had auto-enrolled it in an AI "personalization" tool that altered the vendor image without its consent; Meta countered that its Advantage+ creative AI is off by default and that advertisers must review outputs before running them. The mangled ad stayed live for roughly a week before REI pulled it, unenrolled, and apologized.

Incident Details

Severity:Facepalm
Company:REI
Perpetrator:Marketing / Ad-platform AI
Incident Date:
Blast Radius:AI-altered ad showing a physically impossible bike and a real athlete's altered likeness ran ~1 week on Instagram; viral backlash; brand-values embarrassment for an environmentally branded co-op; vendor and athlete blindsided

Bikes are not a hard object to photograph. They have two wheels, one set of handlebars, a chain that goes where the chain goes, and brakes that humans have been arranging the same way for a century. So when outdoor co-op REI ran an Instagram ad for a Van Rysel road bike that had grown a second set of handlebars out of the saddle, the internet did what it does: zoomed in, lost its mind, and started counting the other things that were wrong.

A bike that could not exist

The ad showed a woman standing beside a Van Rysel road bike in a park. The more people looked, the worse it got. As cataloged by Business Insider, The Seattle Times, Inc., and a detailed reconstruction at PPC Land, the bike sported an extra pair of handlebars sprouting from behind the seat, chains routed through impossible geometry, brakes mounted in the wrong places, a missing crank arm, and frame lettering that dissolved into gibberish. The human in the shot didn't escape either; observers described a figure that looked assembled from several different people, with a face that wasn't quite anyone's.

This was not a bike REI sells, because it's not a bike that can be built. The picture was a real photograph run through a generative AI model that "improved" it into nonsense. Reddit threads picked it apart impossibility by impossibility, and the phrase "AI slop" got attached to a brand that has spent decades cultivating the opposite image.

The original image, it turned out, was entirely fine. Van Rysel North America confirmed to Business Insider that the source photo came from a professional shoot featuring cyclist Amity Rockwell, and that "any later alterations were not made by Van Rysel." Rockwell herself seemed as confused as everyone else, writing on Instagram that "this was an official shoot. That I got hired for. So why are they AI deep frying the images? To alter a product they're supposedly selling? And my face along with it lol? I'm so lost." A real athlete's face, altered by a machine, pushed out in a national ad, and nobody asked her.

"Meta did it" versus "you ran it"

REI's explanation pointed squarely at its ad platform. In statements to Business Insider and The Seattle Times, a spokesperson said: "Meta auto-enrolled us in an AI personalization tool that produced an inaccurate and inappropriate alteration of a vendor-provided image in some of our ads," adding, drily, that "while a two-handled bike might be interesting, it is not something you will find in our assortment." The co-op said it had unenrolled from the tool and apologized, noting that "product accuracy and our vendor relationships matter."

Meta did not take the blame quietly. A Meta spokesperson told The Seattle Times that the relevant suite, Advantage+ Creative, which houses the generative image feature, is off by default, and that turning it on, plus applying any AI-generated variation to a live campaign, "requires a deliberate action" by the advertiser. Meta's own terms for these tools, quoted by Business Insider, warn that AI ad outputs may be "inaccurate, incomplete, misleading, offensive, and/or inappropriate," and put the responsibility for reviewing them on the advertiser.

So which is it? The honest answer is that both things can be true and both look bad. PPC Land's timeline has the ad running from roughly June 15 and REI issuing its acknowledgment to Fast Company on June 22, after the backlash crested over the weekend. REI is not the first advertiser to report this exact surprise; Business Insider has previously documented other brands finding Meta settings like "test new creative features" and "automatic adjustments" flipped to on, generating bizarre creative they never approved. A default that quietly opts brands into having their images rewritten by AI is, charitably, a setting designed to be missed. That's a real platform problem, and Meta owns it.

But a generated image still has to be chosen and run. Meta's position is that a human at the advertiser had to apply the AI variation and push it live, which means somebody in the ad pipeline either reviewed a two-handlebar bike and shrugged, or reviewed nothing at all and let an automated system serve whatever it produced. For a week. The platform built the trap; the advertiser walked into it without checking the floor.

Why a co-op got roasted harder than most

Plenty of brands run ugly ads. REI got a sharper kicking because of who REI claims to be. It's a member-owned cooperative that has made environmental stewardship a core part of its identity, the company famous for closing on Black Friday and telling people to go outside. Generative AI is, fairly or not, now associated in the public mind with heavy energy and water use, so the optics of that specific company outsourcing a product photo to an image model were always going to land badly. "So much for caring about the environment," ran one widely upvoted comment quoted by Business Insider.

There's also the vendor-relationship angle, which is quieter but arguably worse. Van Rysel supplied an accurate photo of its own bike, shot with an athlete it hired. The ad pipeline then deformed the product and the person and broadcast the result under REI's name. If you're a brand partner, watching your gear get "AI deep fried" into a physically impossible mutant in someone else's ad is not a confidence builder.

The accountability gap is the actual story

The two-handlebar bike is funny. The mechanism behind it is the part worth keeping. This is what happens when image generation gets wired into an ad system that auto-applies "enhancements," and the chain of custody for what actually reaches the public gets blurry enough that, when it breaks, everyone can point somewhere else. Meta points to its terms and its default toggles. REI points to Meta's auto-enrollment. The vendor points out it sent a clean photo. The athlete points out nobody asked her. And the ad still ran.

The fix isn't exotic. Somewhere between "vendor delivers an accurate photo" and "ad goes live to millions," a human being needs to look at the final creative and confirm it depicts a real product and a real person accurately. That used to be automatic, because making an ad took effort and somebody saw it before it shipped. Bolt a generative model into the middle that silently rewrites the image after the humans have signed off, default it to on, and you've removed the one checkpoint that catches a bike with two sets of handlebars.

REI unenrolled and apologized, which is the right move and also a tell: the safest setting for "let an AI quietly redraw our products" turned out to be off. Advertisers running Meta campaigns can disable Advantage+ creative enhancements in Ads Manager and audit their live creative, ideally before the next impossible bicycle shows up in somebody's feed. The tools will keep offering to improve your photos. The bikes would prefer you didn't let them.

Discussion