Google has released the latest version of Imagen 3, its AI text-to-image generator, to users in the US, as spotted by VentureBeat. The tool, which you can access on Google’s AI Test Kitchen, is supposed to generate images with “better detail, richer lighting, and fewer distracting artifacts” compared to Google’s previous models.
Google first announced its updated Imagen 3 tool during I/O in May, but it only appears to have made the tool generally available through its Vertex AI platform within the past several days. Some Reddit users started experimenting with Imagen 3 last week, and Google published a research paper on the tool on Tuesday.
Like other AI image generators, Imagen 3 can create detailed images based on your prompt. You can also edit the image by highlighting a certain part and describing what you want to change.
:format(webp)/cdn.vox-cdn.com/uploads/chorus_asset/file/25574101/imagen_3_sonic.png)
Image: Emma Roth / Imagen 3
There seem to be some guardrails in place, as the tool will decline to generate images of public figures, like Taylor Swift, and also won’t produce images of weapons. And while it will stop short of generating named copyrighted characters, you can get around this pretty easily by describing the character you want to create.
I got the tool to generate images that look very much like Sonic the Hedgehog and Mario, while my colleague was able to create characters resembling Mickey Mouse. I also found that it will generate logos belonging to companies like Apple, Macy’s, Hershey’s, and even Google, which you can see in the image at the top of this article.
Despite these somewhat flexible guardrails, Imagen 3 still stands in stark contrast to Grok, the AI image generator that lives on Elon Musk’s X platform. Grok has been used to generate all sorts of wild content, including images with drugs, violence, and public figures doing questionable things.
But Google’s AI tools have run into some issues as well. Earlier this year, Google stopped letting people generate images with its Gemini AI chatbot after users found that it created historically inaccurate images.
