Google's AI project Gemini hit a snag earlier this year when its image-generating feature was found to create wild historical inaccuracies in the interest of diversity. Google says it's been working on that, and the new and improved people-generator will made available to some paid users soon.

It was an embarassment for Google, in this early-stage AI era, when its Gemini chatbot and image-generator was called out first by right-wing figures and then by the whole national press for its strange racial diversity tic. Namely, when asked to produce an image of America's founding fathers, it would insert random people of color into the group. And though there has never been a Black pope, Gemini wanted to make it so when told to produce an image of a pope.

This led to Google suspending the image-generating feature back in February, saying they would need to make some tweaks.

"Gemini's AI image generation does generate a wide range of people. And that's generally a good thing because people around the world use it. But it's missing the mark here," the company said in a statement at the time.

Now, as TechCrunch reports, Google’s AI research division DeepMind has been hard at work getting Imagen 3, the newest image-generation model, to stop creating these odd anachronisms. But as some fixes are put in place, only users with paid subscriptions to Gemini Advanced, Business or Enterprise will be able to access the people-generating feature.

The rest of Imagen 3's capabilities are being made available to all Gemini users this week.

"We’ve significantly reduced the potential for undesirable responses through extensive internal and external red-teaming testing, collaborating with independent experts to ensure ongoing improvement,” a company spokesperson tells TechCrunch. "Our focus has been on rigorously testing people generation before turning it back on."

Imagen 3 was reportedly trained on "a large dataset comprising images, text and associated annotations," the spokesperson said. And the model's training data was filtered for "safety” and “review[ed] … with consideration to fairness issues."

Google appeared to have overcorrected when it comes to diversity and racial sensitivity in creating the previous people-generation model — after a notorious controversy in 2015 when the Google Photos app was found to be labeling a photograph of two Black people as "gorillas."

In other Gemini news, the company announced the rollout this week of "Gems," which are customizable chatbots that can be focused on specific subject matter. A user can create their own Gem by typing in a set of specific instructions, and that Gem can be saved for future use — for instance, to ask fitness questions, or focus on a specific type of cooking.

"With Gems, you can create a team of experts to help you think through a challenging project, brainstorm ideas for an upcoming event, or write the perfect caption for a social media post. Your Gem can also remember a detailed set of instructions to help you save time on tedious, repetitive, or difficult tasks," the company explains.

The company says Gems are now rolling out "on desktop and mobile devices to Gemini Advanced, Gemini Business and Gemini Enterprise users in more than 150 countries," and are available in "most languages."

Previously: Google Suspends Gemini Image Module After Backlash Over Diverse Depictions of Founding Fathers, Nazis