Generate AI in the Browser with Chrome AI - Raymond Camden

Add powerful, on-device generative AI to your web app—no server required. Learn how with the new Chrome AI APIs.

#1about 1 minute

Introduction to generative AI in the browser

The speaker sets expectations for the talk, assuming basic familiarity with generative AI and JavaScript, and provides links to the presentation materials.

#2about 5 minutes

Understanding the fundamentals of Chrome AI

Chrome AI integrates the Gemini Nano model directly into the browser for task-focused operations, requiring progressive enhancement and a one-time model download.

#3about 3 minutes

Exploring the features of the Chrome AI APIs

The APIs are available for both browsers and extensions, supporting features like streaming, sessions for conversational context, and multimodal input with images and audio.

#4about 3 minutes

Using the translator and language detector APIs

The translator API converts text between languages, while the language detector API identifies the language of a given text and provides a confidence score.

#5about 2 minutes

How to use the summarizer API for text

The summarizer API can generate different styles of summaries, such as key points or headlines, but may sometimes include external context not present in the original text.

#6about 3 minutes

Generating and correcting text with built-in APIs

The writer and rewriter APIs generate or transform text based on tone and length, while the new proofreader API identifies spelling and grammar errors.

#7about 1 minute

Leveraging the flexible general purpose prompt API

The prompt API offers a flexible, session-based interface for general-purpose tasks, supporting system instructions, structured output, and multimodal inputs like images and audio.

#8about 7 minutes

A three-step guide to implementing Chrome AI

Implementing any Chrome AI feature involves checking if the API exists, verifying its availability, and then creating an instance while handling the model download progress.

#9about 4 minutes

Live demos of the translator and summarizer APIs

A demonstration shows the translator API converting English to Mandarin and the summarizer API condensing the Gettysburg Address, highlighting its speed and options.

#10about 4 minutes

Demonstrating the rewriter and prompt APIs

The rewriter API is used to make text more casual and shorter, while the prompt API analyzes the content of images to generate detailed descriptions.

#11about 2 minutes

Enhancing image analysis with geolocation data

Combining the prompt API with EXIF geolocation data from an image allows the model to generate significantly more context-aware and accurate descriptions.

#12about 2 minutes

Final resources and where to learn more

The presentation concludes with links to official documentation, online playgrounds for testing, and information on joining the early preview program for updates.