Google Gemini is getting a range of new features, including access to its impressive Imagen 3 artificial intelligence image generator and a new custom chatbot feature called Gems.
Imagen 3 was first announced at Google I/O earlier this year and was previously only available through the ImageFX experiment. It includes improved realism and better text rendering on images.
Gems were first announced in May. They work a little like GPTs in ChatGPT in that you can get the AI to act a certain way or play a certain character. They can be tagged into a chat instance on Gemini web.
Both of the new features will only be available to people paying $19 per month for a Gemini Advanced subscription through Google One or Gemini for Workspace, but access will be expanded over time.
Personalization with Gemini Gems
Google Gems are a clever idea. They allow you to quickly create a custom version of Gemini tailored to a specific purpose. For example, you could create one aimed at a specific goal such as training for a marathon, where it would focus on fitness, health and nutrition.
“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,” the company explained in a statement. Adding: “Your Gem can also remember a detailed set of instructions to help you save time on tedious, repetitive or difficult tasks.”
This includes giving it details of a class you’re taking, topics you want to study or even your own writing style, where it can edit your work to be grammatically accurate without losing your unique flow.
Gems will work in most languages and across 150 countries but will initially only be available to Gemini Advanced subscribers or those with a business or enterprise account.
Better images with Imagen 3
I have been testing Google DeepMind’s impressive Imagen 3 image generation model for a month or so as it is available in the Google Labs service ImageFX. It is finally coming to the Gemini chatbot.
Our design principles are clear: From start to finish, you remain in control of the creative process. If the initial image you get doesn’t meet your expectations, simply tell Gemini what you’d like to change and it’ll give you a new image.
The new model will be available to Gemini Advanced subscribers in English but will eventually be available to all users in multiple languages around the world. It comes with safeguards against things like inappropriate content and copyright. It isn’t clear how much better this will be than Imagen 2 in Gemini which got Google into trouble for being overly sensitive on certain topics.
All images generated using Imagen 3 in Gemini will also be watermarked with SynthID so that if they are used anywhere can be flagged as AI-generated. They will also only be made in a square format.
Google says Gemini will also now be able to generate images of people again, but not photorealistic depictions of identifiable individuals, minors or anything sexual.
According to Google: “Our design principles are clear: From start to finish, you remain in control of the creative process. If the initial image you get doesn’t meet your expectations, simply tell Gemini what you’d like to change and it’ll give you a new image.”
This is one of the most powerful reasons for building an AI image generator into a chatbot. The AI chat model acts as the go-between, refining prompts and sending better instructions to the image generator than a human could communicate.
