The improved version of Google's image generation engine, Imagen 3, will once again enable users to create human figures, after the feature was removed following complaints about the creation of racist and false images, including depictions of black Nazis. According to the company, the new engine has built-in safeguards to prevent the creation of false, violent, or sexual images.
The upgrade will be powered by Imagen 3, the latest version of Google’s AI text-to-image generator. The tech giant quietly launched Imagen 3 through the AI Test Kitchen category, which allows experimental testing of new features. Now, the engine will be integrated into Gemini, the AI engine, and will be capable of creating anything from photorealistic landscapes to oil paintings using just a few words.
Google suspended Gemini's ability to create images of people in February after users discovered that it had generated false or inaccurate historical images. According to Google, the upgraded model comes with built-in safeguards and, according to the company, performs better compared to other available image generation models.
The site will not allow users to create photorealistic images of public figures, content involving minors, or violent and sexual imagery. Google plans to expand the ability to create AI-generated people to more users and languages soon. However, the Imagen feature is different from Google’s new Remagine, which allows AI elements to be integrated into photos taken with the new line of Google Pixel 9 phones.
It is worth recalling that in February of this year, Meta, the owner of Facebook, Instagram, and Threads, announced that it is promoting the development of tools to detect AI-generated images, in an effort to combat deepfakes—false images created online.
10 billion AI images a year
According to estimates, since 2022, nearly 20 billion AI-generated images have been uploaded to the internet, including fake images of public or private individuals that were uploaded without their consent, or misleading political information intended to distort the truth.
For some time, Meta and other social media giants have known that they would have no choice but to take action. Last year, the UK passed the Online Safety Bill, making it a crime to upload fake images of a person without their consent.
US lawmakers have already stated in the past that they failed in their duty to protect users online and that only legislation will compel social networks to act to prevent the spread of fake news. Meta's move is expected to lead other companies to build trust and control standards over the information published.
Meta admitted that it is still not possible to detect all AI-generated content, and that some will try to bypass tagging technology. However, they said they intend to continue looking for ways to monitor some of the content uploaded and call on users to share information themselves about content generated using AI, so that the company can add a tag label to it.
In recent months, AI-generated images have become so sophisticated that it is sometimes difficult to tell they are not real. For example, last January, fake images of pop star Taylor Swift were uploaded to social networks, which were reportedly created using AI.
In the UK, a slideshow of eight images showing Princes William and Harry at King Charles' coronation was circulated on Facebook, with more than 78,000 likes. One of the images appeared to show an emotional embrace between William and Harry after reports of a rift between the brothers. None of the eight images were real.
Another fake image showed former US President Donald Trump, also created using AI, after he was charged with 13 counts of alleged election fraud.