Google’s new and improved Imagen 3 AI image generator is insanely realistic

Google has quietly released its upgraded text-to-image generator Imagen 3. Available for US users, it promises a significant leap in image quality and detail. And judging from the examples… I’m scared of how good it is.

First teased at Google I/O in May, Imagen 3 is now accessible through the company’s AI Test Kitchen. According to Google, the new model boasts “even better detail, richer lighting and fewer distracting artifacts” than the company’s previous models.

Like its competitors, Imagen 3 transforms text prompts into images. You can also edit existing images with precision, thanks to the tool’s ability to understand and modify specific elements. While Google has implemented safeguards to prevent the generation of harmful or misleading content, such as images of public figures or weapons, the model’s creativity is still impressive.

“We’ve designed Imagen 3 to generate high-quality images in a wide range of formats and styles, from photorealistic landscapes to richly textured oil paintings or whimsical claymation scenes,” Google explained.

Possible concerns

It’s worth noting that Imagen 3’s approach to content moderation differs from that of other platforms. You can’t just generate an image of Trump praying, for example. However, the tool can generate images resembling copyrighted characters and famous logos if you know what prompts to use. Folks at The Verge created an image of “a cartoonish blue hedgehog running in a field,” and Imagen 3 returned an image of Sonic the Hedgehog. While Google is taking a more cautious approach than its competitors, it’s obviously still possible to bypass the restrictions.

When it comes to image realism, here’s an example I found particularly interesting – and concerning since I’m a crocheter. There has been a surge of AI-generated crochet images followed by fake patterns. Pinterest is filled with them, and worse – scammers have been selling these fake patterns on Etsy. But it’s still easy to spot AI-generated crochet (at least it is for me). However, with Imagen 3, it seems to me that it’s becoming incredibly realistic, opening doors to even more people getting scammed. And this is just crochet…

The future of Imagen 3

Despite possible concerns, Google has ambitious plans for Imagen 3. The company intends to release multiple versions of the model, each optimized for specific tasks. “We’ll be expanding Imagen 3’s availability across Google products, like the Gemini app and web experience, Workspace, Ads and more,” Google writes. Additionally, features like inpainting and outpainting, popular in Imagen 2, are slated for integration into Imagen 3.

Since I don’t live in the US, the tool is still unavailable for me to test personally. But if you’re our US readers, you can try it out now for free, and feel free to let me know what you think.

[via The Verge; lead image generated with Google Bard]