Google adds AI image generation to Search — but there’s a catch

Google‘s artificial intelligence (AI) image generator for the masses is here … sort of.

Hot on the heels of Adobe’s new Firefly 2 AI image generator, and OpenAI releasing its new DALL-E 3 image generator broadly within ChatGPT Plus (and offering it through investor/customer Microsoft’s Bing Image Creator), Google this week announced it, too, was putting a text-to-image generation AI feature directly into Google Search. However, only for those who opt into Search Generative Experience (SGE) in Google Search Labs, the company’s experimental beta testing service.

“We believe that generative AI in Search can help spark inspiration and help you get even more done. We look forward to continued testing and getting user feedback,” wrote Hema Budaraju, senior director of product management at Google Search in an official company blog post.

Not only that, but Google also unveiled a new feature that lets users compote Google Docs-style written drafts using AI in Search. These can then be exported to Gmail or Google Docs. Budaraju cited drafting a note to a home contractor as an example of how this could be useful.

Has potential, but access is limited for now

A Google spokesperson told VentureBeat that the new image creation capability was powered by Imagen, Google’s proprietary AI image generation foundation model. It has already been available since May through the Google Cloud AI development and deployment environment Vertex, but that requires a credit card to access (though initial usage is free).

Google also posted a cool and cute animated GIF of how its new search image generator AI in Search could be used to create several versions of a capybara wearing a chef’s hat, simply by a user typing “draw an image of a…” and the description of the desired image into the search bar.

Yet, even when VentureBeat opted into the Search Labs experience via the link provided in Google’s blog post on Friday, the image generation capability was not available to us.

Asked by VentureBeat about the lack of availability, a Google spokesperson noted via email: “We announced that these capabilities were just beginning to roll out starting yesterday (in English in the U.S.), so if you’re not getting them yet, please keep an eye out :).”

In fairness, it’s similar to what OpenAI has done with ChatGPT and its slow, two-week-long rollout of vision and audio capabilities.

Asked if and when image creation would be generally available in Google Search, outside of the Search Labs/SGE opt-in, Google’s spokesperson wrote: “SGE is an opt-in experiment available in Search Labs, which is a new program for people to test bold new products and ideas we’re exploring. Search Labs is available in the Google app (Android and iOS) and on Chrome desktop.”

So…no definite timeline on that.

Unfortunately, any confusion over who can access AI image generation in Google Search and how, is probably not a great start to that feature for Google, especially when there are numerous other competing AI image generators already on the market and easily accessible to all users.

Google still playing catchup when it comes to AI

Last month, Google updated its AI chatbot Bard — powered by its PaLM 2 foundation large language model (LLM) — to offer integrations with other Google Apps products, such as Gmail and Drive. However, VentureBeat’s test of the new integrations produced lackluster results. And a new report from Fortune magazine suggests Googlers themselves are unhappy with the product, and expressing internal doubts about its usefulness, with some even questioning what LLMs generally are useful for.

Bard also recently suffered an embarrassing faux pas when a sharp-eyed SEO consultant, Gagan Ghotra, discovered that Google Search was indexing Bard conversation links, and surfacing them in public search results. Even if the user had elected to share them with a single third party (Google has since updated this to correct the issue).

Meanwhile, OpenAI has reportedly passed $1.3 billion in annualized revenue, according to a report in The Information, driven largely by the success of its signature AI product, the LLM-powered ChatGPT, which offers a $20 per month individual subscription tier and enterprise plans, as well as a paid API.

And this week across X, people including this VentureBeat author posted numerous examples of how the company’s new GPT-4V or vision model — which powers the latest version of ChatGPT with vision and audio analysis capabilities — was delivering truly impressive, helpful results.

It even saved at least one user from getting a parking ticket — as good an argument in favor of the real-world usefulness of LLMs as this weary NYC driver has seen.

Chat GPT can now see and hear.

It can analyze videos, pictures, and audio and give you all the right answers to your questions.

I don’t have to get parking tickets anymore?? pic.twitter.com/aDM5J3xOih

— Orbit X. (@orbitxai) October 6, 2023

Ultimately, we at VentureBeat are still eagerly awaiting the public release of Google’s much-hyped new AI foundation model Gemini which aims to be “more capable” than ChatGPT.

Source