Privacy-First In-Browser Generative AI Web Apps
summary
Claims that browsers can run generative AI locally using WebNN with about 90% of native performance, and that models run on CPU, GPU or an NPU for better battery and speed. Demo shows an image classifier running on the NPU of a Copilot+ Surface (7th gen). Practical stack: ONNX models served to the browser via transformers.js or ONNX Runtime Web, cached with the Cache Storage API, model downloads range from tens of megabytes to a few gigabytes, with GPU/CPU fallbacks and dev‑flag support in Chrome/Edge.
AI-written summary · about
chapters (12)
- 0:00Introduction and demo
- 4:47Introducing myself and context
- 6:54Why in-browser AI is needed
- 11:13Web Neural Network standard
- 14:25AI ecosystem and hardware
- 20:10Benefits of WebNN standard
- 22:42Getting started with WebNN
- 27:17Frameworks and higher-level APIs
- 32:00Transformers.js for practical use
- 36:21Summary and key takeaways
- 39:11Built-in browser AI features
- 43:50Q&A and closing
related talks
32mPrivacy-first in-browser Generative AI web apps
46mThe Ultimate Guide to Browser-Based AI: Privacy-First, Offline-Ready and Future-Proof
21mUnlocking Client-Side AI in Your Web App
41mEscaping the Cloud: High-Performance AI in your Browser
16mThe Web is AI Ready—maximize your AI web development with WebNN
31mGenerative AI Directly in My Browser, Is It Possible? (spoiler, yes 😉)
18mWeb AI on next generation AI PCs
33mLocal Models: How to Run AI in the Browser