I’ve been trying out Google’s new AI image generator Bard. Here’s how it turned out
As it approaches its first birthday, Google’s conversational AI tool Bard is getting another upgrade.
Early reviews found Bard lagging behind competing chatbots, in part because its responses were less human-like. Since then, Google has integrated its own AI model Gemini, which has helped Bard expand its training beyond text to video, audio and images, and is now trying to make further strides in the utility space where generative AI remains a hot ticket .
In a blog post on Thursday, Jack Krawczyk, Bard’s head of product, said the tool now lets users create images for free.
When someone types a prompt, such as “Create an image of a hot air balloon flying over mountains at sunset,” Bard creates what Google describes as “custom, large-scale visuals to help bring your idea to life.”
However, it takes a while – about 13 seconds to be exact.
Although most of the queries we tried produced relevant images and/or responses, they did not have a 100% accuracy rate.
For example, when I asked Bard to create an image for a news story about the latest updates to the tool, he refused to create the requested image. When I asked him to simply create an image about the Google Bard, he created this blonde cyborg:
Over the past year, the market has been filled with chatbots like OpenAI’s ChatGPT, Microsoft’s Bing AI, Anthropic’s Claude — and yes, Google’s Bard — as big tech companies seek to stake their claim in the next wave of search. These chatbots access massive data sets and use large linguistic models to provide textual, and now image or even video, responses to consumer queries. It is a rapidly developing field and is already becoming closer to human conversation. However, although bots may provide answers confidently, they are not always accurate – and remain vulnerable to abuse.
Google’s post noted that Bard includes a distinction between visuals created with Bard and original human artwork, and embeds watermarks in the pixels of the generated images. To test this, I asked him to create an image of Botticelli’s The Birth of Venus. I showed an exact copy, but dirtier. Those faces! Those hands! However, there is an option to report a legal issue and give each photo a like or dislike.
In the wake of Taylor Swift’s deepfake, Google said it seeks to limit “violent, offensive or sexually explicit content” and applies filters to avoid creating images of specific people. In fact, she refused to create a photo of Super Bowl quarterbacks Patrick Mahomes and Brock Purdy on a picnic or one of Beyoncé at the bank.
“We will continue to invest in new technologies to improve the safety and privacy protection of our models,” Krawczyk wrote.
When I asked Bard to create a photo of Lisa Lacy at work, Bard said she didn’t have enough information about that person to help. However, he was able to create an image of the more general query of a journalist at work – with not one but two sandwiches on his desk.
She refused to create an image of a man throwing a coin from the Hoover Dam because “throwing objects into the Hoover Dam is prohibited.” (I offered to create a scenic or historical photography photo instead.)
Bard was happy to create images of historical moments such as the signing of the Declaration of Independence.
In addition to adding the image builder, Google is expanding the availability of Gemini Pro in Bard from English to more than 40 languages. This includes a double verification feature, which allows users to verify Bard’s responses through web content.
This will help expand the tool to more than 230 countries and regions, according to the post.
Google first added Gemini Pro to Bard in December 2023, to give it “more advanced understanding, reasoning, summarization, and coding capabilities.”
Editors’ Note: CNET uses an artificial intelligence engine to help create some stories. For more see This post.