Google is turning its image generator function back on for some Gemini AI users, after temporarily suspending the tool over concerns it could not reliably depict White people.
The Alphabet-owned tech company said in a blog post on Wednesday that the latest generation of its text-to-image tool, Imagen 3, will soon be available to users who pay for Gemini Advanced, Gemini Business and Gemini Enterprise subscriptions.
Over the coming days, we’ll also start to roll out the generation of images of people, with an early access version for our Gemini Advanced, Business, and Enterprise users, starting in English. We’ve worked to make technical improvements to the product, as well as improved evaluation sets, red-teaming exercises and clear product principles,” the company said.
Users of the tool can write text prompts that Imagen 3 turns into visual images. In its blogpost Wednesday, Google provided a sample prompt and corresponding image.
“An animated image of a tiny dragon hatching from an egg in a sunlit meadow, surrounded by curious glowing butterflies. Vibrant colors, detailed scales,” the science fiction-themed prompt read.
In another example, Google instructed the tool to generate an image of a “ball gown made of paper napkins in an elegant showroom.”
Google added that Imagen 3 has new built-in safeguards so that it adheres to the company’s design principles, which include limiting Gemini from generating “instructions for self-harm,” or in the case of Imagen 3, creating “pornography, or excessively gory images.”
Google came under fire when it first rolled out its text-to-image tool and user prompts to generate historical figures, such as German soldiers during World War ll, and popes, all of whom have historically been White and male, went awry. Some of Gemini’s images portrayed Nazi soldiers as Black and Asian and popes as female.
At the time, CEO Sundar Pichai called its AI app problems “unacceptable.”
Google assured users that this time around, it has “made significant progress in providing a better user experience when generating images of people.”
“We don’t support the generation of photorealistic, identifiable individuals, depictions of minors or excessively gory, violent or sexual scenes,” the company said in a statement. But it included a caveat, given that the AI-based tool is still novel and in development.
“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. We’ll gradually roll this out, aiming to bring it to more users and languages soon.”
On Thursday, the image-generation function was not available to all Gemini Advanced subscribers.
Prompted to depict a human being, the tool responded, “Image generation of people is coming soon to Gemini Advanced.”
