Google has announced plans to reintroduce its Gemini artificial intelligence tool for generating images of people, following a temporary suspension earlier this year due to inaccuracies in the images it produced. The tool will initially be available to Gemini Advanced, Business, and Enterprise users in English within the coming days.

In a blog post on Wednesday, Dave Citron, Senior Director of Product at Gemini, explained that the new version of the image generation tool, known as Imagen 3, has undergone significant technical improvements. Citron noted that Google has enhanced its evaluation processes, conducted rigorous “red-teaming” exercises to identify vulnerabilities, and established clear product principles to guide the tool’s use. Red-teaming is a practice used by companies to stress-test products for potential flaws or weaknesses.

Google initially paused the AI image generation feature in February after users reported several inaccuracies in the images it generated, particularly in historical contexts. Some users shared examples on social media, including an image of a German soldier from 1943 that showed a racially diverse group of soldiers in German military uniforms, and a depiction of a medieval British king that included a woman ruler. These errors led to the tool being temporarily withdrawn.

The updated Imagen 3 tool will not support the creation of photorealistic images of identifiable individuals, depictions of minors, or content that is excessively gory, violent, or sexual. Citron emphasized that while not every image generated by the tool will be perfect, Google is committed to listening to user feedback and making ongoing improvements.

Google plans to gradually roll out the enhanced tool to more users and languages in the near future, continuing its efforts to refine the capabilities and accuracy of its AI-driven image generation technology.