Aug 29, 2024 16:00:00

Google to reopen AI image generation feature for Gemini paid users, after pausing it in February 2024 following criticism over racial depictions

Google has announced that it will introduce the image generation model '

Imagen 3 ' to the image generation function of the multimodal AI ' Gemini ' on August 28, 2024. At the same time, it has been announced that the image generation function of people, which had been suspended due to issues with racial depictions, is expected to be resumed.

Google Gemini updates: Custom Gems and improved image generation with Imagen 3
https://blog.google/products/gemini/google-gemini-update-august-2024/

Google to Let Some Users Generate Images of People After Scandal - Bloomberg

https://www.bloomberg.com/news/articles/2024-08-28/google-to-let-some-users-generate-images-of-people-after-scandal

Google AI reintroduces human image generation after historical accuracy outcry | Ars Technica
https://arstechnica.com/ai/2024/08/months-after-controversy-google-ai-can-generate-images-of-humans-again/

Google takes aim at custom GPTs with the launch of Gemini Gems
https://www.androidpolice.com/embargo-aug-28-1200-et-google-geminis-new-gems-let-you-create-your-own-personal-ai-expert-for-any-topic/

In an announcement on August 28, 2024, Google announced, 'We will upgrade Gemini's creative image generation capabilities and introduce our latest image generation model, Imagen 3, to Gemini in the next few days and make it available to users in all languages.' Imagen 3, released on August 16, 2024, features improved prompt comprehension compared to the previous Imagen 2 , as well as significantly improved text rendering capabilities. Imagen 3 is summarized in detail in the following article.

Google has released 'Imagen 3', a high-quality image generation model that can also render characters, so I tried using it - GIGAZINE

Google says about Imagen 3, 'Imagen 3 sets a new standard by generating images with just a few prompts. By chatting to Gemini, you can generate a variety of images, including photorealistic landscapes and highly textured oil paintings.'

In addition, Google has announced that it will resume generating images of people for paid Gemini Advanced, Business, and Enterprise users in the coming days. Regarding Gemini image generation, Google was criticized in February 2024 for 'inaccuracies in depicting historical images,' and apologized, saying that 'attempts to generate a wide range of results led to misguided results.'

Google apologizes for Gemini generating 'racially diverse Nazis' - GIGAZINE

Google later explained that 'as we tuned Gemini to portray a range of people, we failed to account for cases where it clearly shouldn't have done so,' and 'over time, Gemini became much more cautious than we intended and began refusing to answer certain prompts altogether,' andreported that it had turned off the person image generation feature to make significant improvements, including extensive testing.

Google said, 'We have been working on technical improvements to our product, improving our evaluation set, red teaming, and creating clear product principles ,' and announced that it will resume the person image generation function. Google said, 'With Imagen 3, we have made great progress by improving the user experience when generating images of people. However, we do not support the generation of realistic and identifiable individuals, depictions of minors, or excessively cruel, violent, or sexual scenes. Of course, as with other generative AI, not all images generated by Gemini are perfect, but we will continue to listen to user feedback and continue to improve. We aim to make the person image generation function available to more users and languages.'

In addition, Google announced that it has released a new feature called 'Gem' for Gemini Advanced, Business, and Enterprise users that allows them to customize Gemini and create their own chatbots. Gem is a feature announced at ' Google I/O ' held in May 2024, and allows users to create their own chat AI, similar to ' GPTs ' provided by OpenAI's ChatGPT.

Google says, 'Gem can remember a detailed set of instructions, saving you time on tedious, repetitive or difficult tasks.' Gems also offers examples such as an 'idea pro' that provides inspiration for birthday presents and a 'career advisor' that can help you hone your skills and develop a plan to achieve your career goals.