Google DeepMind's Generative AI Adding Text-to-Image and Draft Creators

Friday - October 13 2023, 14:45 UTC - 1 year ago

Google has unveiled its AI-generated Search Generative Engine (SGE) which gives users the ability to create images through textual prompts and create AI-drafted documents such as emails, research articles or essays. The AI-created documents and images will be watermarked and labeled to avoid false attributions. Google is also soon launching a tool ‘About this image’ which will give users insights about the origin and spread of an image from their search.

Google announced Thursday that it’s adding another generative AI feature to its Search engine. Google is giving users the ability to create images using textual prompts. For now, it’s only available to users in the United States who have opted for Google’s experimental Search Generative Engine (SGE), which integrates generative AI into the world’s most visited website.

The company unveiled SGE at the I/O 2023 and will be rolled out to other countries soon, said Google.

A user could type in "hands holding flowers with a view of mountains in the background," the search engine would spew up to four images in the results. Because Google doesn’t want users to claim these images as human-generated, every image generated through SGE will have metadata labeling and embedded watermarking to indicate that AI created it.

In late August this year, Google unveiled the AI watermarking technology called SynthID, reported Interesting Engineering. SynthID is an AI tool developed by Google DeepMind that can watermark and recognize AI-generated images. The tool is available only to users of Google’s text-to-image AI generator ‘Imagen.’ .

"That’s why we’re building safeguards into this experience and blocking the creation of images that run counter to our prohibited use policy for generative AI, including harmful or misleading content," said the company blog.

To fight misinformation and establish the credibility of images on Google, the company mentioned in yesterday’s blog that they will soon launch a tool called ‘About this image.’ It will reveal information about an image, like when that image and similar images were indexed by Google, where it may have first appeared, or where else it’s been seen online (like on news, social, or fact-checking sites).

The other genAI announcement .

SGE has also been integrated with AI capabilities for drafting content in response to user queries. As part of the experiment, a user can ask for a written draft as a jumping-off point to a research article, an email, or an essay. The user can tell SGE to shorten the draft or change the tone to be more casual.

For example, suppose a user is dissatisfied with a product delivered via an online shopping website and wants to send an email to return the product. The user could type into the search engine: "Write a draft to notify an online shopping company that the product they sent is inferior and ask the procedure to return it." .

From the looks of it, the draft creator could synthesize information from various sources and help a user get a summarized draft of what they’re looking for instead of scrolling through the search engine for hours. It’s similar to another generative AI feature Google unveiled in May, which allowed the search engine to provide extra context in search results.