Google rolled out a series of AI updates, including a new text-to-image tool called ImageFX, as well as the ability to generate images in the AI chatbot Bard for free.
Users can ask Bard to generate photos using Imagen 2, Google’s most advanced text-to-image model, developed by the company’s AI division, DeepMind, and released in December. Google said Bard is now powered by its Gemini Pro AI language model.
While Bard is only getting the ability to generate images now, the feature has been available to users of the GPT-4-powered competitor ChatGPT Plus for months. ChatGPT uses Dall-E 3 to create images.
But unlike ChatGPT Plus, which requires a monthly subscription of $20, people in 230 countries and territories can currently use Bard for free in more than 40 languages.
Also read: Google’s New AI Feature Lets You Circle Anything on Your Phone To Search for it
Bootstrapping safety in Bard
In a blog post on Feb, 1, Google claimed that Imagen, the model that’s been embedded into Bard, delivers its highest-quality AI-created images yet, rendering tough tasks like human faces and hands realistically.
Bard’s image generation capabilities come as AI is increasingly being used to create deepfake porn material targeting celebrities like Taylor Swift. In Australia, lawmaker Georgie Purcell criticized a news channel after it used AI to make her boobs look bigger on TV.
Google said it made “significant investments” in Imagen 2 training data safety. It added that Bard’s image capability was “designed with responsibility in mind.”
The Mountain Valley, California-based tech company says it put guardrails in its AI image generators to “limit problematic outputs like violent, offensive, or sexually explicit content.” It also applied filters to cut “the risk of generating images of named individuals.”
Google said it conducts “extensive adversarial testing to identify and mitigate potential harmful content. In addition, images generated in Bard will be infused with SythID, a digital watermark created by Google, which indicates that pictures were made using AI when they appear in Search or Chrome.
The watermarks cannot be seen by the human eye or heard by the human ear (in the case of music made using Google AI tools), but they can be used for identification, the company says.
Introducing #ImageFX ✨ a new image generation tool powered by Imagen 2, Google's most advanced text-to-image model. We’re excited to add this to our suite of generative AI tools at Labs, and to help bring your most creative ideas to life. Try it out: https://t.co/e07Zz3Pdx4 pic.twitter.com/MVXEEt9c2b
— labs.google (@labsdotgoogle) February 1, 2024
More Google AI features
Alongside the debut of Bard’s image capability, Google revealed ImageFX, a new image-generation tool that allows people to create pictures using only text the same way they would with, say, Dall-E 3.
According to the company, the AI image generator features a prompt interface that includes “expressive chips,” which supposedly improves creativity by allowing people to experiment with “adjacent dimensions of their creations and ideas.”
Users can access ImageFX via Google’s experimental platform for new ideas and products, Labs. This is the same platform where people can play around with things such as the Search Generative Experience. Like Bard, the tool uses Imagen 2 to generate images.
Google is also embedding Imagen 2 across its offerings, including ads, Duet AI in Workspace, SGE, and Google Cloud’s Vertex AI.
The company said it made improvements to MusicFX, its music generator, claiming it can now generate music faster with higher-quality audio. Generated songs can now last up to 70 seconds.
Updates were also made to TextFX, a generative AI tool that aims to improve navigation and the overall user experience.
and then