This morning, a Habr post about old-school technology popped up in a roundup, and EyeToy for the PlayStation 2 flashed by in it. A typical “weird gadget from the 2000s”: a plastic sphere on a hinge, a blue LED, blurry mini-games where you had to wave your arms around and pop virtual soap bubbles.
Seems like an evolutionary dead end—right up until you realize that EyeToy sold 10.5 million units (figures as of November 6, 2008), taught Sony how to do real-time computer vision on a 300-megahertz CPU, spawned PlayStation Move—and still ended up as a footnote between Wii and Kinect.
That “between” is the most interesting part.
Richard Marks—in the 1990s, he worked on computer vision and auto-tracking at Stanford, then developed a people-tracking product at the startup Teleos Research. In July 1999, he joined Sony Computer Entertainment America’s R&D division—just as the PlayStation 2 was being unveiled.
Marks’s logic was engineering-perfect: console graphics were growing exponentially, while the method of input had not changed since 1983. Give the player a natural interface, and you could both improve old genres and draw in an entirely new audience—people who had never held a gamepad and had no intention of doing so.
He assembled his demo prototypes using ordinary off-the-shelf webcams. Among the first tests was the famous “Spiders”: some wild PhD-student stuff, where little star-shaped “spiders” were projected onto a 3D model of Marks’s head, and you could pluck them off your face with your hand. Then came demos with distorted water surfaces, fire effects, and color tracking: a wand with green and magenta balls at either end that “attracted” virtual butterflies.
Phil Harrison saw the demos—at the time, he was vice president of third-party relations and R&D at SCEA. In 2000, he was promoted to senior vice president of product development at the European division (SCEE), and brought Marks into the London office to show the technology to developers. There, Ron Festejo from Psygnosis (which later merged into London Studio) joined the project—and from that partnership grew EyeToy: Play, with its mini-games, and all the software that followed.
The name, incidentally, was not invented by marketing. In London, the gadget was initially called iToy (“interactive toy”); Harrison himself renamed it EyeToy.
A teardown by Portelligent (published in EE Times, 2003) strips the hardware of any romance:
OmniVision’s key engineering decision was this: instead of digitizing raw pixels and processing them in software, it moved fixed-pattern noise suppression, gain control, the color matrix, and color space into the analog circuitry right on the sensor. The OV519 merely packaged the stream afterward. In 2003, that was not a hack job—it was the only way to fit inside USB 1.1 and stay within budget.
And here is what the PS2 did: motion detection, template matching, and color segmentation—in real time, on the Emotion Engine, dynamically adjusting the resolution of each processed frame to the current CPU load. That is why the player’s image on screen sometimes looks sharper and sometimes blurrier: the software was bargaining with the hardware for every frame.
Marks later stressed this separately: the camera delivered 60 fps, while most webcams of the era managed 30. And it delivered video in a format optimized for the PS2. So the “cheap webcam” was cheap only in BOM terms—its operating modes had been tuned for one specific task.
This is the most underrated part of the whole story. Sony discovered very quickly that the classic approach—“come up with a game, then bolt controls onto it”—does not work with a camera. So it flipped the process: first, a simple mechanic that is physically reliable; then, a mini-game built around it.
Hence the whole catalog: “punching” (Kung Foo), “rhythm” (Beat Freak), “wiping” (Wishi Washi). These are not genres; they are interaction primitives, tested for robustness, repeatability, and—sorry—fun across the widest possible range of real rooms and lighting conditions.
Why the camera caught on at all was also a question not of technology, but of human factors. Tests showed that people instantly understood the controls when they saw themselves on screen and received immediate feedback on their actions. The effect held steady across gender and age—and, crucially, interested onlookers got almost as much out of watching as the people playing. This was not “gaming” in the usual sense—it was a living-room attraction.
That is why EyeToy: Play won two BAFTAs and two AIAS awards, while the internet filled up with testimonials like “for the first time ever, my girlfriend wanted to play.” Marks, writing on the PlayStation Blog in 2010, called it a breakthrough across age and gender barriers—and he was right: Wii made it mainstream, but Sony had already tested the “everyone plays, including Grandma” formula three years earlier.
So no, it was not a failure. It was a regional hit with worldwide sales that people forgot simply because it came out before “motion gaming” became a mainstream term.
The honest part of the story begins here. EyeToy is good right up until it tries to be a game, rather than an attraction.
EyeToy: AntiGrav from Harmonix (yes, that Harmonix—before Guitar Hero and Rock Band) was the first attempt at a “real camera game.” The player does not see themselves on screen: they stand in front of the TV, bend their knees, lean from side to side to turn, wave their arms to perform tricks, and fly across rooftops on a hoverboard. The tracking is “face-locking”: you calibrate against a frame on the screen, the game locks onto your face, and uses it to determine the position of your head.
And here is what Wired wrote in its review (November 17, 2004):
Tracking, it said, works surprisingly well, but every so rarely face-locking literally stops telling a face from an elbow apart—and the controls fall to pieces for a couple of seconds. In the world of AntiGrav, a couple of seconds means death.
Tellingly, the reviewer then goes after not the camera, but the game design: the game demands nearly perfect runs to unlock the next level, and makes you race the same track three times. Wired’s conclusion: the technology is not the issue here; it is a failure in the pacing of rising difficulty. “In the end—frustration, and there’s nothing to throw at the wall, because there’s no controller.”
That is the crux of the whole story: freedom without precision is not freedom—it is a lottery. Marks articulated it plainly in a 2010 piece:
“EyeToy helped us understand that people enjoy physical interaction and movement, but they also want precise control and a simple, fast, reliable way to trigger actions. We also found that some actions feel more natural when holding a tool or ‘prop’ in the hand.”
The direct consequence of that conclusion was PlayStation Move. A hybrid of camera and spatial controller, where the PS Eye already offered 640×480 at 60 fps and 320×240 at 120 fps—four times EyeToy’s resolution and twice its frame rate, with twice the sensitivity thanks to the sensor’s larger pixels (developed jointly with OmniVision). Plus a four-microphone array—echo cancellation, noise reduction, voice-direction detection.
Then came the industry arithmetic that EyeToy lost outright.
| EyeToy (PS2) | Kinect (Xbox 360) | |
|---|---|---|
| Launch | July 2003 (EU) | November 4, 2010 |
| Sensor | one color CMOS VGA camera | color camera + IR projector + depth sensor |
| What it tracks | 2D: movement, color, silhouette | 3D depth + a full skeletal model in space |
| Sales | 10.5 million in ~5 years | >8 million in 60 days; >10 million by March 2011 |
| Status | regional hit | “the fastest-selling consumer electronics device in history” according to Guinness |
In two months, Kinect sold nearly as much as EyeToy had in five years. The reason was not Microsoft’s marketing, nor that Sony “didn’t try hard enough”—the reason was that EyeToy was fundamentally unable to provide what the depth sensor did: a third dimension. Everything Sony built—color segmentation, template matching, face-locking—was brilliant engineering within a 2D limitation. A limitation solved not by software, but by different hardware.
One further delicious detail: in 2008, the designers from London’s EyeToy group themselves, in an interview with GamesIndustry.biz, looked at Microsoft’s first full-fledged camera project—You’re in the Movies—and spoke condescendingly: they use background subtraction to paste the player into a movie, “and that’s unreliable, which is why we never did it in any of our games; if the lighting in the room shifts, the whole thing will break. Good luck to them.” Two years later, Microsoft would sell eight million Kinects in two months. The irony is brutal, but instructive from an engineering standpoint: Sony was right in the narrow sense—background subtraction really is fragile—and wrong in the broader one, betting against the need for a depth sensor.
A couple of things that, to me personally, feel like the final chord.
The camera outlived its console. EyeToy, being an ordinary USB webcam (albeit USB 1.1), worked with the PlayStation 3—for video calls. Enthusiasts wrote drivers for Linux and parts of Windows/macOS—with a side effect: in custom drivers, the red “low light” LED became a recording indicator. A gadget designed for one game in one living room turned out to be honest enough hardware to be revived by other people’s hands.
The first prototype is in a museum. As part of Sony’s contribution to the Videogame Archive project at Nottingham Trent University, the team handed over the very first production EyeToy prototype for permanent collection. A hardware artifact instead of a press release.
And most importantly. EyeToy is a rare case of a company honestly running a scientific experiment on living people: it discovered that the body as a controller works (10.5 million, BAFTAs, family nights), and that the body as a controller is not enough (AntiGrav, “nothing to throw at the wall”). The same product supplied both answers. And the right conclusion—“you need a tool in your hand plus a camera”—became PlayStation Move.
So the camera did not lose. It was an expensive, commercially successful, and completely honest opinion poll about what people want from an interface. And the answer it got is still used by the industry today—from VR controllers to hand tracking in Quest.