The Deepmind logo is being displayed on a smartphone with the Google Gemini logo in the background in this photo illustration in Brussels, Belgium, on February 8, 2024.
In a blog post on Wednesday, Dave Citron, a senior director of product on Gemini, wrote that early access to Google’s new Imagen 3 generator will be available to Gemini Advanced, Business and Enterprise users starting in English in the coming days.
“We’ve worked to make technical improvements to the product, as well as improved evaluation sets, red-teaming exercises and clear product principles,” Citron wrote. Red-teaming refers to a practice companies use to test products for vulnerabilities.
Google paused its image generation feature earlier this year after saying it was offering “inaccuracies” in historical pictures. Embarrassing examples spread across social media.
One user asked the tool to generate an image of a German soldier in 1943, and it created a racially diverse set of soldiers wearing German military uniforms. Another query for a historical depiction of a medieval British king also generated a racially diverse set of images, including a woman ruler.
Citron said Imagen 3 doesn’t support photorealistic identifiable individuals, depictions of minors or excessively gory, violent or sexual scenes.
“Of course, as with any generative AI tool, not every image Gemini creates will be perfect, but we’ll continue to listen to feedback from early users as we keep improving,” Citron wrote. “We’ll gradually roll this out, aiming to bring it to more users and languages soon.”
