AR and social media meet in the phone camera. On Snapchat and TikTok, augmented reality means Lenses and effects that add 3D effects, objects and transformations to the live camera view. Meta offered third-party AR effects through its Spark platform until it shut Spark down in January 2025. Brands use these effects for campaigns and product try-on. Sephora ran a Ramadan shoppable Lens on Snapchat in the Middle East, and the NFL and sponsors such as Pepsi and Uber Eats launched Super Bowl 2025 filters across Instagram and Snapchat.
Every effect has to run on the user’s own phone, inside a budget the platform sets, and the campaign results cited here are reported by Snap and an agency. Both points shape what a brand should expect.
What does AR mean in Snapchat, Instagram and TikTok?
Snapchat describes Lenses as AR experiences that add 3D effects, objects, characters and transformations to a Snap. Face Lenses change how the person in front of the camera looks, and World Lenses place effects and characters into the surroundings, such as a dancing Bitmoji on a desk. In mobile AR generally, try-on uses the front camera, while AR of the surroundings uses the rear camera with on-device SLAM or VIO tracking.
TikTok’s equivalent is Effect House, which had been adopted in close beta by more than 450 creators. According to TikTok, their effects had appeared in 1.5 billion videos with over 600 billion views. Its documentation covers effect types such as Segmentation, Face Mask, Head Tracker, Face Stretch and 3D Face.
Meta Spark launched in 2017, and Meta said more than 600,000 creators from over 190 countries had built effects with it. After 14 January 2025, effects built by brands and creators on Spark were no longer available.
The scale figures for Snapchat are Snap’s own. Snap says over 250 million Snapchatters engage with Lenses every day, and that Lenses are played with more than 6 billion times a day on average. The general definition of what augmented reality is and where it is applied beyond social apps is covered separately.
How does a face lens run on a phone, and what does it cost?
A face lens is a real-time vision and graphics pipeline running on the phone, with face models built for mobile GPUs. The pieces are well documented, and they run in this order:
- Find the face. Google’s BlazeFace detector (Bazarevsky et al., 2019) runs at 200 to 1,000+ FPS on flagship devices. It was built to hand a facial region of interest to downstream models for keypoints, geometry, expression or segmentation.
- Track landmarks or a mesh. Google researchers (Kartynnik et al., 2019) inferred a 468-vertex 3D face mesh from a single camera at 100 to 1,000+ FPS on mobile GPUs, depending on device and model variant. Snap’s Lens Studio exposes 93 tracked face landmarks and a Face Mesh that mimics the user’s expression in real time.
- Track the world (World Lenses only). Markerless AR estimates the device’s pose and the scene geometry with visual-inertial odometry or SLAM, so graphics stay aligned with the room.
- Render and composite. The effect is drawn over the camera image, with depth cues such as occlusion and shadows. Try-on also needs lighting that matches the scene, or the product looks pasted on.
The detection and mesh numbers can mislead. BlazeFace’s speed was measured on flagship devices, for detection alone. Stacked features and less powerful phones cut the frame rate further, as both platforms warn. Lens Studio tells creators to target 30 FPS and stay above 15 FPS on most devices, and Effect House supports at most 30 FPS. Stable registration and low motion-to-photon latency are key usability factors for AR.
Measured battery and memory cost of try-on
The phone-cost measurement comes from a research paper on lighting estimation for front-facing mobile AR. The authors streamed the front and rear cameras together on an iPhone 11 Pro to capture better lighting for virtual try-on. In a 10-minute session, dual-camera streaming used on average 5% of the battery, 2.5 times the cost of the front camera alone. Cutting rear-camera capture to 3 minutes brought it down to 3%. Memory rose from 62.7 MB to 78 MB, a 24.4% increase (78 / 62.7 ≈ 1.244). Their proposed fix is a timer that switches the rear camera on at intervals, shortening the interval when the scene changes, plus a trigger from the inertial sensor when the phone moves significantly.
Platform guidance points the same way. Snap notes that using less RAM means less memory bandwidth, which reduces heating and battery use and can raise frame rate. TikTok warns that an effect with a high memory increment is more likely to crash the app.
Performance budgets the lens platforms publish
| Constraint | Snapchat (Lens Studio) | TikTok (Effect House) |
|---|---|---|
| Frame rate | Target 30 FPS, above 15 FPS on most devices | Highest supported 30 FPS |
| Activation time | Under 650 ms on Snap’s benchmark device to run as an ad | Not stated in these sources |
| Package size | Under 8 MB | Not stated in these sources |
| 3D geometry | Under 100,000 triangles in total recommended (the guide also gives a 60,000-triangle figure) | Not stated in these sources |
| Textures | None above 2048 × 2048 | Not stated in these sources |
| Expensive features | PBR and Uber materials raise activation time; Unlit preferred for simple elements | Generative effects and Body Avatar Drive very high impact; segmentation high |
| Device reach | Low RAM use keeps lenses stable on lower-end devices | Many effects fail on Android phones without WebGL support |
Lens Activation Time is the time a Lens takes to render on screen after it has loaded on the device. On one of its own Instagram projects, the agency Superside kept the filter under 4 MB, using a low-poly style and choosing objects carefully, to fit Instagram’s size limits.
Which brands have used AR on social media, and what did it do for them?
Sephora and Snapchat ran a multi-format campaign that combined AR Lenses with influencer content and product education, including a Ramadan shoppable Lens in the Middle East. For the campaign as a whole, Snap’s case study reports a 12-point rise in brand awareness, a 14-point rise in ad awareness, a 5.6% increase in add-to-cart in Saudi Arabia and a 9-point lift in purchase intent. Snap presents the result as Lenses starting to affect the lower funnel rather than only awareness.
The NFL’s Super Bowl 2025 filters let fans wear team face paint, catch footballs in AR and pose with AR versions of star players. Superside reports that the filters drew massive participation but gives no figures. An earlier platform-level example is Snapchat’s Sky Filters, announced in September 2017, which restyled photos of the sky with options such as starry night and rainbow.
Read these results for what they are. Snap published the Sephora figures and an agency blog the NFL account, so they show what is possible rather than what a typical campaign achieves. More augmented reality marketing examples and what they look like in practice are collected on that page, including campaigns with and without a commerce path.
How does AR virtual try-on help shoppers choose?
Snap’s Catalog-Powered Shopping Lenses show how the mechanics work. The Lens pulls product name and price from a brand’s product catalogue into on-screen Product Cards, and the shopper swipes through them to try on several products and click to buy, all in one Lens. Snap says Snapchatters who experience a shoppable AR Lens are 2.4 times more likely to purchase than those who do not, and that Lenses drive a 14% higher share of incremental sales contribution than video. The 2.4 times figure compares Snapchatters who experienced a shoppable Lens with those who did not. The comparison therefore cannot separate the Lens’s effect from self-selection.
Researchers analysed data from an international cosmetics retailer to test AR’s effect on online sales. They found that AR use in the retailer’s mobile app was associated with higher sales for less popular brands, products with narrower appeal and more expensive products. The effect was stronger for customers new to the online channel or the product category. The authors conclude that AR is most effective when uncertainty about the product is high, because it builds purchase confidence. That study covers a retailer’s own app, not a social platform. The mechanism (reducing doubt before purchase) plausibly carries over to social try-on, but that is an inference.
Try-on also raises the engineering bar. Front-facing try-on of glasses or hats needs lighting that matches the user’s real environment, and the front camera alone sees little of that environment, which is why the researchers above turned to the rear camera at a battery cost. Serving many products and faces at scale brings its own problems, covered in our piece on the cold-start engineering problem in AR beauty try-on at scale, with the wider category picture in augmented reality in the beauty and cosmetics industry.
Advantages and disadvantages of AR in social media marketing
What brands gain
Two advantages stand out: reach and reduced purchase uncertainty. Over 250 million Snapchatters already engage with Lenses every day, and the cosmetics-retailer study finds AR most effective where product uncertainty is high. Snap’s Camera Kit also lets developers bring Snap AR into their own apps, so an effect is not confined to the Snapchat feed.
Treat headline uplift numbers with care. A 2026 compilation by Amra & Elma states that brands using AR face filters achieved up to a 131% conversion lift over traditional digital ads. The compilation names no study behind the 131% figure, so it is a marketing claim, not a lift any given campaign should expect.
What brands risk
- Platform shutdown. Meta’s closure of Spark on 14 January 2025 made effects built by brands and creators unavailable.
- Biometric privacy. Snap reached a $35 million settlement in an Illinois class action alleging that Lenses collected users’ biometric data without consent under the state’s BIPA law, as reported by Engadget. Snap’s position is that Lenses can recognise an eye or a nose as part of a face but cannot identify whose face it is. Any AR that analyses the camera feed in real time raises these questions.
- Device fragmentation. An effect that passes QA may still fail on less powerful phones, and many Effect House effects cannot run on Android devices without WebGL support.
- User discomfort with beauty filters. A Consumer Reports survey cited by TechCrunch in 2022 found that 59% of Americans find beauty filters troubling.
Building effects that outlive the platform
An effect now has to fit a phone in the middle of the range and survive the platform it was built for. Activation time, memory increment and stacked features such as segmentation decide whether a lens reaches the audience, not the tracking model’s speed on a flagship. One hedge against platform risk is a brand’s own app, since Snap’s Camera Kit brings Snap AR into developers’ own apps.
