Google is reinstating a feature that lets users create AI-generated images of people after suspending the capability earlier this year amid criticism it produced misleading and historically inaccurate depictions.
In the days to come, the ability to create images of people from within the company’s Gemini suite of AI models will return to Gemini Advanced, Business and Enterprise customers, the company announced in a blog post on Wednesday. Like other image-generation tools such as Midjourney, Stable Diffusion and OpenAI’s Dall-E, Google’s Imagen 3 lets users turn typed or spoken text prompts into visual representations almost instantly.
“We’ve worked to make technical improvements to the product, as well as improved evaluation sets, red-teaming exercises and clear product principles,” reads the post from Dave Citron, senior director of product management for Gemini Experiences. Red teaming means simulating harmful behavior to spot and correct product vulnerabilities.
Two weeks after Google launched Gemini in February, the company paused some image generation features when critics, most notably tech entrepreneur Elon Musk, who’s working on rival AI products, accused the tool of being “woke” for producing images of people that didn’t align with the historical ethnic and gender realities of the times they represented. One controversial photo, for example, depicted Black men and women wearing World War II-era German military uniforms, while another showed a female pope.
Such images quickly went viral, and Jack Krawczyk, a Google AI product lead who on social media often welcomed user input on the company’s offering, reduced his public online presence after being harassed over the issue.
Sundar Pichai, CEO of Alphabet and its Google subsidiary, sent an internal memo to employees saying the image issues had “have offended our users and shown bias.” He added that “to be clear, that’s completely unacceptable and we got it wrong.” Google’s stock value fell as the company rushed to respond to the controversy.
Humans Created By Imagen, The Sequel
Now, almost seven months later, the capability to turn prompts into images of people will is back, at least for some users, starting in English.
“With Imagen 3, we’ve made significant progress in providing a better user experience when generating images of people,” the blog reads.
The post includes several examples of images generated by Imagen, but none of them involve humans. Instead, one shows an image produced by the cute, non-controversial prompt “tiny dragon hatching from an egg in a sunlit meadow, surrounded by curious glowing butterflies.” Another shows a photorealistic image of a mountain vista.
People, Yes. Minors Or Sexual Images, No
The tool, however, “will not support” the generation of photorealistic, identifiable people, the post indicated. Nor will it support depictions of minors or excessively gory, violent or sexual images. A Google spokesman clarified that “will not support” means the tool won’t create such images when prompted to do so.
“Of course, as with any generative AI tool, not every image Gemini creates will be perfect,” Wednesday’s blog post said, “but we’ll continue to listen to feedback from early users as we keep improving.”
