How On-Device AI Models Find Your Best Tinder Profile Photos

Summary

Tinder developed an on-device AI Photo Selector to recommend compelling profile photos directly from a user's camera roll without exposing private assets. To establish a reference face, the system extracts facial landmarks via Apple's Vision framework and normalizes the crop using affine transformations. To prevent funnel drop-offs from mandatory selfies, the architecture includes a fallback mechanism that detects a single recognizable face in existing profile photos. Processing workflows execute concurrently across thousands of local assets while downloading and initializing TensorFlow Lite models in parallel using Combine. A dedicated analytics queue synchronizes operation counts with DispatchGroup to ensure accurate metric capture before the global rollout.

Context

Manually sifting through camera rolls to select profile photos is overwhelming for users, often resulting in minimal effort, unengaging profiles, and negative impacts on engagement metrics. Additionally, requiring users to capture a live selfie to establish a reference face created user hesitation and significant funnel drop-offs.

Approach / What changed

Tinder built an on-device AI Photo Selector utilizing Apple's Vision framework, CoreML, TensorFlow Lite, and Combine. The system captures a reference face from a selfie or existing profile photos via BasePhotoInference, applies affine transformations for normalized face crops, concurrently scans device assets, and tracks privacy-safe performance metrics using a dedicated analytics queue.

Takeaways

• Normalized face cropping applies affine transformations and a least squares solution (Ax = B) onto a 112x112 grid to reliably align facial landmarks regardless of image angle.
• The BasePhotoInference fallback reduces funnel drop-offs by scanning existing profile photos for a single recognizable face instead of requiring a live selfie.
• Linearizing analytics through a dedicated DispatchQueue and DispatchGroup ensures synchronized and accurate metric reporting during concurrent on-device asset processing.

Related reading

• Grab · Jun 9, 2022

  How facial recognition technology keeps you safe

  Grab utilizes facial recognition technology across its platform for driver authentication, passenger verification, and digital electronic Know Your Customer (e-KYC) processes. The core pipeline consists of image preprocessing through face detection and alignment, anti-spoofing checks, feature extraction into high-dimensional vector embeddings, and downstream verification or search. To counter spoof attacks like screen replays, synthetic moiré patterns are generated and cropped face patches are used during training and inference to focus on local structures rather than global semantic noise. Face verification challenges involving shallow ID datasets and masked faces are resolved using semi-Siamese training architectures and masked data augmentations.

  Kai Feng Tee

• Tinder · Jan 9, 2023

  Building Obsidian, Tinder’s Design System

  As Tinder scaled over a decade, maintaining visual consistency across its client applications became unsustainable and increased engineering overhead. To address this, the UI Platform team developed Obsidian, a unified design system combining design tokens, reusable components, and documentation. Designers define base and context tokens in Figma using the Tokens Studio plugin, which are then transformed by Style Dictionary into native artifacts for iOS, Android, and Web. Web developers integrate these tokens using Atomizer to generate static stylesheets, while a Zeroheight documentation portal and a custom token lookup tool assist legacy migrations. The ongoing migration has streamlined collaboration between designers and engineers, resolved styling ambiguities, and improved color contrast accessibility across platforms.