Articoli

Have you ever seen a puppy in a nest emerging from a cracked egg? What about a photo that’s overlooking a steampunk city with airships? Or a picture of two robots having a romantic evening at the movies? These might sound far-fetched, but a novel type of machine learning technology called text-to-image generation makes them possible. These models can generate high-quality, photorealistic images from a simple text prompt.

Within Google Research, our scientists and engineers have been exploring text-to-image generation using a variety of AI techniques. After a lot of testing we recently announced two new text-to-image models — Imagen and Parti. Both have the ability to generate photorealistic images but use different approaches. We want to share a little more about how these models work and their potential.

How text-to-image models work

With text-to-image models, people provide a text description and the models produce images matching the description as closely as possible. This can be something as simple as “an apple” or “a cat sitting on a couch” to more complex details, interactions and descriptive indicators like “a cute sloth holding a small treasure chest. A bright golden glow is coming from the chest.”

In the past few years, ML models have been trained on large image datasets with corresponding textual descriptions, resulting in higher quality images and a broader range of descriptions. This has sparked major breakthroughs in this area, including Open AI’s DALL-E 2.

How Imagen and Parti work

Imagen and Parti build on previous models. Transformer models are able to process words in relationship to one another in a sentence. They are foundational to how we represent text in our text-to-image models. Both models also use a new technique that helps generate images that more closely match the text description. While Imagen and Parti use similar technology, they pursue different, but complementary strategies.

Imagen is a Diffusion model, which learns to convert a pattern of random dots to images. These images first start as low resolution and then progressively increase in resolution. Recently, Diffusion models have seen success in both image and audio tasks like enhancing image resolution, recoloring black and white photos, editing regions of an image, uncropping images, and text-to-speech synthesis.

Parti’s approach first converts a collection of images into a sequence of code entries, similar to puzzle pieces. A given text prompt is then translated into these code entries and a new image is created. This approach takes advantage of existing research and infrastructure for large language models such as PaLM and is critical for handling long, complex text prompts and producing high-quality images.

These models have many limitations. For example, neither can reliably produce specific counts of objects (e.g. “ten apples”), nor place them correctly based on specific spatial descriptions (e.g. “a red sphere to the left of a blue block with a yellow triangle on it”). Also, as prompts become more complex, the models begin to falter, either missing details or introducing details that were not provided in the prompt. These behaviors are a result of several shortcomings, including lack of explicit training material, limited data representation, and lack of 3D awareness. We hope to address these gaps through broader representations and more effective integration into the text-to-image generation process.

Taking a responsible approach to Imagen and Parti

Text-to-image models are exciting tools for inspiration and creativity. They also come with risks related to disinformation, bias and safety. We’re having discussions around Responsible AI practices and the necessary steps to safely pursue this technology. As an initial step, we’re using easily identifiable watermarks to ensure people can always recognize an Imagen- or Parti-generated image. We’re also conducting experiments to better understand biases of the models, like how they represent people and cultures, while exploring possible mitigations. The Imagen and Parti papers provide extensive discussion of these issues.

What’s next for text-to-image models at Google

We will push on new ideas that combine the best of both models, and expand to related tasks such as adding the ability to interactively generate and edit images through text. We’re also continuing to conduct in-depth comparisons and evaluations to align with our Responsible AI Principles. Our goal is to bring user experiences based on these models to the world in a safe, responsible way that will inspire creativity.

A good night’s sleep can help you feel rested and recharged, while also playing a critical role in your overall health and well-being. Since the introduction of Fitbit’s sleep features in 2009, sleep tracking has been incredibly popular – making information previously only available through a sleep lab accessible to users via their wrist. To date, we’ve analyzed 22 billion hours of sleep data, equivalent to the lifespan of over 5,000 tortoises. (Fun facts: a tortoise lives for up to 500 years and there are 60,000 giant tortoises in the world.)

We know our users are motivated to gather even more insight into their sleep, which is why today we’re introducing Sleep Profile. This new Fitbit Premium feature offers a new longitudinal analysis of your sleep patterns. It also makes your sleep data even easier to interpret with fun animal characters, so you can take steps to improve your sleep quality and, in turn, your overall health.

What sleeper type are you?

These critical data points gathered from your sleep analysis inform your designated Sleep Animal, providing a fun way to understand your sleeping style. Our extensive research and testing found six animal archetypes, each chosen for their distinct trends that correlate each animal’s behavior with common user sleeping behaviors and patterns.

The following graphics highlight top traits and habits of each sleeper type, and how yours can impact your day-to-day:

How does it work?

Wear your Fitbit device to sleep for at least 14 nights per calendar month to receive an assessment. The more you wear the device to sleep, the more precise the evaluation. Sleep can vary significantly from night to night, so it’s helpful to analyze your sleep data over a longer period from your own sleep environment to uncover more insights into your habits and how you can improve.

As a Premium member you will see your Sleep Profile, including your sleep animal and monthly sleep analysis, on the first day of every month. Animals can change from month to month, and data collected throughout the month will inform the next animal you receive the following month. There is no “ideal” animal – each one can be used to better understand your sleep patterns and how you can improve your sleep quality.

Based on your sleeper type’s behavior, you will see where your sleep patterns compare to others like you – whether it’s setting a more consistent sleep schedule or creating a bedtime routine like trying the mindfulness content found within the Fitbit app to help you fall asleep faster. Getting quality sleep can play a critical role in your overall health and well-being, with poor sleep being correlated to increased potential health risks including diabetes, cardiovascular problems, obesity, poor cognitive functioning and more. Premium uplevels your Fitbit experience by helping you better understand your sleep patterns, your readiness to exercise and how your body responds to stressors, all with additional guidance and insights to help you achieve your health, fitness and wellness goals.

Sleep Profile is rolling out in the Fitbit app to Premium users with Sense, Versa 3, Versa 2, Charge 5, Luxe or Inspire 2 devices upon launch. Users will receive their first profile during the week of July 4, followed by monthly profiles delivered in the Fitbit app on the 1st of each month. (Note: Sleep Profiles aren’t intended for medical purposes and users should consult their health professionals for questions about their health.)

A couple of years ago, my partner and I moved to Kentucky, not knowing anyone in the state. Before moving, I extensively researched local websites and online magazines, trying to understand the neighborhoods and get a sense of the community. Somewhere along the way I found Lussi Brown Coffee Bar, a local business run by by a queer woman. Not knowing whether I’d get a response, I eagerly reached out through email, asking questions to get a sense of the community. To my delight, the owner, Sarah Brown (she/they), quickly responded and provided an overview of the rich history of the LGBTQ+ community in Lexington. And of course, she shared recommendations of some of their favorite LGBTQ+ owned businesses in the state too!

As we moved into town, Sarah and their girlfriend welcomed us with open arms, very much making the community immediately feel like home. And our physical home brought that same love, too. Unintentionally, we rented an apartment on a short street filled with LGBTQ+ folks from their 20s through their 70s — in fact, our neighbors called it Kentucky’s “Barbary Lane,” a nod to the tight knit, beloved street of LGBTQ+ folks in Armistead Maupin’s novel “Tales of the City.”

With that same spirit, we want to make it easier for others to find LGBTQ+ owned businesses in their own community. Starting today, merchants in the U.S. with a verified Business Profile on Google can add an LGBTQ+ owned attribute to their profile, making it easier for customers to find and support them through Search and Maps. This new offering joins the Black-owned, Latino-owned, veteran-owned and women-owned business attributes we already offer, and is yet another way people can support diverse businesses.

As we celebrate Pride, it’s important to remember visibility and representation are critical, all year round. A flag in the window of a small business has the power to bring queer folks together, to celebrate our joy, honor our history, and fight for our diverse community. It’s our hope that this attribute will allow business owners to celebrate their identity and community with the world.