Google’s AI-powered search experience can now generate images

An illustration of the Google logo.
Illustration: The Verge

Google’s Search Generative Experience (SGE) will let you create images right from a text prompt starting Thursday, the company announced. Microsoft has offered the ability to make images from Bing Chat using OpenAI’s DALL-E model since March, so it’s not too surprising to see Google follow suit.

Here’s how it works. If you’re opted in to SGE through Google’s Search Labs program, you can just type your query into the Google search bar. After you do, SGE can create a few images based on your prompt that you can pick from. The tool is powered by the Imagen family of AI models, Google spokesperson Craig Ewer tells The Verge.

In a GIF, Google showed how the tool can be used to create images based on the prompt “draw an image of a capybara wearing a chefs hat and cooking breakfast.” As shown in the GIF, you can also refine the queries that Google ends up using to create each photo.

A GIF showing Google’s Search Generative Experience creating images based off the prompt “draw an image of a capybara wearing a chefs hat and cooking breakfast.” GIF: Google

You’ll be able to make Google’s AI-generated images from Google Images, too. As you’re browsing Google Images results, you might see a box that prompts you to generate images based on your query — those generated images will show up in a sidebar.

In a blog post, Google’s Hema Budaraju stresses that the company is trying to make sure it rolls out its image-generation tools responsibly. The tool isn’t supposed to be able to generate images that violate the company’s prohibited use policy for generative AI, Budaraju says. (Hopefully, that will prevent Google’s tool from being used to make images of popular cartoon and video game characters flying planes toward the twin towers, like Bing’s image generation tool was.)

Every image made with SGE will “have metadata labeling and embedded watermarking to indicate that it was created by AI,” Budaraju says. During “this initial stage,” Google won’t let you make images that “depict photorealistic faces or any prompts that include notable individuals,” Ewer says. (I asked if Google plans to allow those down the line, and Ewer says “it would be premature to say either way” and that the company will evolve its approach over time.) And Google is limiting image generation to users who are 18 or older, meaning teens who have access to SGE won’t be able to use it.

Google is also adding the ability to generate written drafts right from the search bar using SGE. Using your query, Google can generate a draft, and you can choose for the tools to make something that’s long, short, or has a casual tone. If you like the result, you can export that draft into Google Docs or Gmail.

I asked Ewer what models powered the draft writing feature, and was given this statement: “SGE is currently powered by a variety of LLMs. The models used in SGE are trained on both data from publicly available sources and on human supervision data.”

Since launching SGE in May, Google has steadily improved it with things like more video and better links. I’m still not a big fan of it — it’s still way too slow, and I’d rather just scroll through links — but these new image and draft creation tools do seem potentially useful.

Title: Google’s AI-powered search experience can now generate images
Source: The Verge – All Posts
Source URL:
Date: October 12, 2023 at 06:03PM
Feedly Board(s): Technologie