The battle of the AI art generators has been heating up as big tech giants enter the ring. While the space is currently dominated by DALL-E 2, Stable Diffusion and Midjourney, Microsoft, Meta and Google have all announced text-to-image (or text-to-video) tools on the way. Now Google's just made its offering public – or at least part of it.
Google first gave us a glimpse of Imagen, its AI image generator, back in May with the publication of a proof-of-concept paper and the launch of website showcasing some of the results of the tool. It's just released limited features of the tool to the public – and they're bizarrely specific (for more on how AI art generators work, see our piece on how to use DALL-E 2).
A product created in Google's AI Test Kitchen, Imagen is a powerful AI text-to-image art generator that can turn written prompts into convincing images. The results shared so far suggest that it can hold its own against the best AI art generators, which have generated a lot of controversy due to their increasingly convincing results and concerns about copyright, abuse and the future of artists' jobs.
Despite the controversy, the genie truly seems to be out of the bottle with major tech heavyweights entering the scene. Microsoft Designer looks set to offer what sounds like a form of AI clip art while Meta's Make-a-Video promises to offer a text-to-video generator. Google has beaten them to a public release with the launch of Imagen (opens in new tab) in beta. but it's holding back on most of the tool's power.
We won't yet be given the ability to use text prompts like "a cute corgi living in a house made out of sushi" or a dragon fruit wearing a karate belt in the snow (animals and fruit are rather popular subjects for AI imagery, see the results above). All we can create are fantasy buildings and Pixar-style monsters.
For now, we're getting access to two functions from Google Imagen: ‘City Dreamer’ and ‘Wobbler’. The first allows users to describe a city to model SimCity-like isometric designs. Say you want a city made of cookies, for example. City Dreamer will suggest designs for plots and buildings. Wobbler, meanwhile, allows users to create and dress a small monster. They can choose the material to make it from, for example, felt or marzipan, dress it up and then make it dance by prodding it.
Such a specific tool, with a very specific aesthetic, offers nowhere near the creative options offered by existing AI image generators. But that doesn't mean Google doesn't have them. Given the controversy around AI art, it seems Google is playing it safer than other platforms have done. It says that the beta release is intended to collect community feedback so that it can fine-tune the tool and correct any problems before the full release.
The other catch is that, as with many such tools, there is a waiting list. You can apply via the AI Test Kitchen app, available on Google Play (opens in new tab) and the iOS App Store (opens in new tab).
What is Google Imagen?
Google Imagen is the search and tech giant's AI-powered text-to-image generator. Google revealed the tool in May in a research paper and a website that shows capabilities that looks very similar to those of tools such as DALL-E 2. It's now made two limited parts of the platform available to the public in beta via its AI Test Kitchen.