Google's Gemini to resume AI image generation of people after 6-month pause

Google's new image generation capabilities are powered by its latest image generation model Imagen 3

Google on August 28 said it will again allow users to generate images of people using its artificial intelligence (AI) chatbot Gemini, about six months after pausing the feature due to a controversy over producing historically inaccurate images.

The company said it will start rolling out an early access version of this capability in English language to Gemini Advanced subscribers as well as the business and enterprise subscribers of Google Workspace in the coming days. It plans to gradually expand the feature to more users and languages soon.

Gemini Advanced is the paid version of the Gemini AI chatbot, available to users as part of the recently launched Google One AI Premium Plan. This plan costs Rs 1,950 per month after an initial one-month free trial.

The new capabilities are powered by Google's latest image generation model Imagen 3 that was introduced at the tech giant's annual developer conference Google I/O 2024 in May.

Imagen 3 better understands natural language and the intent behind the user's prompt to produce "lifelike images with incredible" level of detail and far fewer distracting visual imperfections than prior models, Google said at the time.

One can create images in various styles including photorealistic landscapes, textured oil paintings or whimsical claymation scenes with "just a few words", Dave Citron, Google’s senior director of product management for Gemini Experiences said in a blogpost.

In February, Google had paused Gemini's ability to create images of people after facing severe criticism from consumers over inaccuracies in depictions of some historical images, including those of the US Founding Fathers. This was a few weeks after the tech giant had introduced image generation capabilities in Gemini.

Sundar Pichai, chief executive of Google parent firm Alphabet, also reportedly addressed these missteps in a memo to employees in February, calling them "completely unacceptable".

Shortly afterwards, Google DeepMind CEO Demis Hassabis had stated plans of relaunching the feature in the "next few weeks"

In a blogpost today, Citron said the company has made "significant progress" in providing a better user experience when generating images of people through Imagen 3.

"We’ve worked to make technical improvements to the product, as well as improved evaluation sets, red-teaming exercises and clear product principles" he said. The new model also comes with built-in safeguards and adheres to the company's product design principles, Citron said.

Users won't be able to generate photorealistic, identifiable individuals, depictions of minors or excessively gory, violent or sexual scenes, he said.

"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 said.

Rival players such as OpenAI's ChatGPT, Meta's Meta AI, and Microsoft's Copilot already allow consumers to create images of people on their respective chatbots.

Elon Musk's xAI also launched Grok 2 chatbot with image generation feature earlier this month. The company has partnered with Flux.1, a text-to-image AI model from Black Forest Labs, for the image generation capabilities. Black Forest Labs is a startup founded by former developers of the popular AI firm Stability.ai.

Apart from restarting image generation, Google will also start enabling people to create custom versions of Gemini chatbot, called Gems, for help in specific tasks. This could be a learning coach, career guide, gym buddy, coding partner or creative writing guide.

The feature, which was first introduced at Google I/O 2024, will be available to Gemini Advanced, Business and Enterprise subscribers in the coming days.

Rival OpenAI had also made its custom chatbot store, GPT Store, available to all users for free in May.