Google Unveils Imagen 3, Its Latest AI-Powered Text-to-Image Generator

Google’s Imagen, introduced in 2022, has seen annual upgrades. Following Imagen 2, the latest iteration, Imagen 3, was unveiled at this year’s Google I/O developer conference. Leading the presentation was Demis Hassabis, head of Google DeepMind.

We’re introducing Imagen 3: our highest quality text-to-image generation model yet. 🎨

It produces visuals with incredible detail, realistic lighting and fewer distracting artifacts.

From quick sketches to very high-res imagery, here’s a look at what it can create. 👀 #GoogleIO pic.twitter.com/XMrQYGeSiO
— Google DeepMind (@GoogleDeepMind) May 14, 2024

For those unfamiliar, Imagen is Google’s proprietary AI model for text-to-image generation. It enables users to generate AI-created images from textual descriptions. Demis highlighted:

“…this is our best model yet for generating text, which has been a challenge for image generation models. In side-by-side comparisons, independent evaluators preferred Imagen 3 over other popular image generation models.”
~ Demis Hassabis, head of Google DeepMind

Additionally, Demis compared the new AI media generator, Imagen 3, with its predecessor, Imagen 2. He noted that Imagen 3 is notably more “creative and detailed,” stating:

“Plus, this is our best model yet for generating text, which has been a challenge for image generation models.”
~ Demis Hassabis, head of Google DeepMind

While a new AI image generator may seem appealing, it inevitably raises concerns about deep fakes. Google seems to have considered this, as they have announced that Imagen 3 will utilize Google’s SynthID. This feature applies an invisible watermark to generated media, allowing users to distinguish authentic images from potentially manipulated ones.

Google also touts Imagen 3 as “photorealistic,” “incredibly detailed,” and with “fewer distracting artifacts,” using typical marketing jargon. However, as usual, Google has not disclosed details about the training data used for Imagen 3 or its other AI tools.

Imagen 3 is accessible through the ImageFX tool on Google’s AI Test Kitchen. However, users must sign up before testing it. To sign up, visit the ImageFX tool page and select Settings at the bottom.

In the tool, Imagen 2 is the default setting. However, for “Best Quality,” users can opt for Imagen 3 by requesting access. This involves filling out a quick Google Form and waiting for approval. Developers and enterprise customers will also gain access to Imagen 3 “soon” through Vertex AI.

What are your thoughts on Google’s latest version of Imagen? Share them in the comments below!