Quick answer: An AR filter is a real-time visual effect that uses a phone’s camera and a tracking algorithm to overlay digital elements onto a person’s face or the world around them. Snapchat calls its versions Lenses, TikTok calls its tool Effect House, and Instagram once ran on Meta’s Spark AR Studio before that tool shut down. The effect follows your face or environment frame by frame instead of sitting as a flat overlay, which is what separates it from a simple color filter.
What Is an AR Filter?
The word “filter” gets used loosely. A basic color filter just adjusts tone, brightness, or contrast across the whole frame and needs no tracking at all. An AR filter, or what Snapchat specifically calls a Lens, is different: it detects a face or a surface, builds a live model of it, and locks digital content to that model so the content moves as you move.
That distinction matters because the AR version is a small piece of software, not a static edit. It runs a tracking pipeline every frame the camera is open, which is why building one takes an actual development tool rather than a slider in a photo app.
Each major platform built its own version of that tool. Snapchat creators build in Lens Studio, TikTok creators build in Effect House, and Instagram creators built in Spark AR Studio until Meta retired it. The output looks similar to a viewer scrolling their feed, a face or a scene with something digital attached to it, but each platform’s filters only run inside that platform’s own app.
How Do AR Filters Work?
Face-based filters start by detecting a face, then mapping dozens of facial landmark points and building a 3D mesh that mimics the user’s expression as it changes. Lens Studio, Snap’s own AR development tool, follows those points in real time, closely enough to catch a blink, a smile, or a raised eyebrow and trigger effects off them.
Content then attaches to that mesh in one of a few ways: a flat, textured plane anchored to the head, a full 3D object positioned against a facial feature, or a texture drawn from the camera feed itself for effects centered on one part of the face.
World-facing filters work differently. Instead of tracking a face, they use the phone’s camera and motion sensors to map the surrounding space and anchor 3D objects to a fixed point in it, so a digital object appears to sit still on a table or a sidewalk as the camera moves around it. This is the same augmented reality principle used in wayfinding and shopping apps, just packaged for a single social post.
Both approaches lean on the phone doing all the work locally, in the few milliseconds between one camera frame and the next. Tracking has to hold up as a person turns their head, walks around a room, or steps into a shadow, and a filter that loses its lock tends to show it immediately, with digital content drifting off a face or floating free of the surface it was supposed to sit on.
Early face filters were mostly cosmetic: a pair of dog ears anchored to the top of the head, a flower crown resting above the eyes, a warped nose. Those effects only needed the face mesh and a handful of attachment points, which is why they became so common in the first wave of Snapchat Lenses.
Filters have gotten more layered since. A single lens can now combine face tracking with body tracking, background segmentation, and a triggered animation that fires on a specific expression, all running at once. Some go further and swap in a trained machine learning model to recognize something more specific than a generic face, which is where a tool like SnapML comes in.
Where Are AR Filters Used?
Snapchat introduced face-tracking Lenses in September 2015, letting users long-press a detected face to apply an effect. It added World Lenses in April 2017 for effects anchored in 3D space rather than on a face, then extended face tracking to full-body tracking in August 2020, following a person’s joints so an effect could react to how they moved rather than only to their face.
TikTok’s equivalent creator tool, Effect House, came out of beta in April 2022. Instagram and Facebook effects ran on Meta’s Spark AR Studio starting in 2019, though Meta shut that tool down in January 2025.
Brands have leaned on the format for advertising, too. Coca-Cola, Pepsi, and Taco Bell have all built sponsored AR lenses that turn a branded effect into something people choose to use and share on their own.
Discovery works differently from ordinary ads as a result. A lens usually reaches people because someone scanned a code, searched the in-app lens directory, or saw a friend post with it, which is why an effect that is genuinely fun to use travels further than one that simply buys placement.
AR Filters vs Deepfakes: What’s the Difference?
Both technologies alter a face on camera, but the intent and the mechanism differ. An AR filter overlays or reshapes elements on top of a face the tracking system can see is a face; it does not try to convince you the result is unaltered footage, and the effect usually disappears the moment tracking is lost.
A deepfake works the other way. It generates a synthetic likeness, often swapping one person’s face onto another’s, specifically aiming for the output to pass as genuine video. AR filters are built for playful or practical overlay; deepfakes are built to convincingly replace identity, which is why they raise a very different set of concerns.
Do AR Filter Creators Get Paid?
Snap pays lens creators through a rewards program, sharing money with creators whose lenses pull large audiences or run inside its paid subscription tier. The specific payout paths, eligibility rules, and rates have been reworked several times since Snap started paying for lenses, so the durable point is the arrangement itself: for some creators, building lenses is paid work rather than a hobby.
What the program will not pay for is telling. Effects judged to be little more than a color adjustment, a static image pasted over the frame, or a copy of someone else’s lens are excluded, which keeps the money pointed at work that actually uses the tracking pipeline.
What Comes Next for AR Filters?
The line between a filter and a full AR app keeps blurring. Tools like SnapML now let creators drop a custom machine learning model into a Lens, so an effect can recognize a specific object or scene instead of just a generic face or floor. As that gets easier to build, expect filters to keep moving from a fun face swap toward something closer to a lightweight AR app, still opened with the same long press that started the format back in 2015.