Snap Develops Groundbreaking AI Text-to-Image Model for Mobile Devices
New technology will power high-resolution experiences for Snapchatters in the coming months
Snap has developed a groundbreaking AI model capable of generating high-resolution images for mobile devices in just seconds. This research model is the continuation of our long-term investment in cutting edge AI and ML technologies that enable some of today’s most advanced interactive developer and consumer experiences.
The new diffusion model is:
Compact & Fast: Our state-of-the-art model is relatively small, yet able to produce high resolution images in around 1.4 seconds on an iPhone 16 Pro Max.
Mobile-Friendly, Compute Efficient: It runs entirely on device significantly reducing computational cost compared to large server-reliant models.
Powered by Innovative Training Techniques: It’s able to produce stunning visual results by transferring rich representations from large-scale diffusion models to this smaller and more efficient model.
We are excited to bring this technology into production in the coming months to help power some of Snapchat’s AI features, such as AI Snaps, AI Bitmoji Backgrounds, and more. By implementing this in-house built technology, we’ll be able to offer our community fast, high-quality AI tools at a lower operating cost.
Snap has a long history of research excellence in model optimization and efficiency. We are inspired by the industry innovation that is making AI tools more efficient, affordable, and accessible and we look forward to continuing to contribute to the rapid pace of innovation, particularly for mobile first experiences.