Write it and Google will show it is the motto behind the tech giant’s new AI project – where written text is transformed into an auto-generated image.
Imagine a dog in a red sweater and sunglasses skating on a beach. Or better yet, write it down and Google’s new AI project Imagen turns it into a digital image.
In any case, that is the basic idea. The dog can just as easily be a panda, or a cat. The skateboard can be a bicycle, or a guitar. No matter what you write, Imagen can generate an image of it. But. There is always a but. The project is neither groundbreaking in its genre nor available to the public yet.
“New level”
Being able to describe something and then get a picture of it has existed as a technology for a while and so far the market leader has been DALL-E, according to tech site The Verge.
Google’s entry into the field, however, brings with it the company’s financial muscle and massive status in the tech world. According to Google itself, the AI project reaches “a new level of photorealism” and the technology “is preferred by people who can compare with other models, such as DALL-E”.
However, an imminent future where every thought can become a meaning that becomes a picture is not likely. Google itself writes that the project has some clear limitations in its current design. On the one hand, Imagen avoids things that are pornographic, and on the other hand, the data that Imagen uses is also known to contain a skewed worldview.
Distorted image
Slightly simplified, it can be said that Imagen so far uses very large amounts of public data, images that are linked to text descriptions, that have not been washed away from racist stereotypes or distorted gender roles. A text in the style of “a CEO with a hat at a party on a boat” would thus generate a white man as CEO. Humans in general, Google describes, are difficult for Imagen to generate at present – and those created are almost exclusively white.
No time aspect or end goal for the project has been set yet, Google only writes that they hope to make “more progress” in the future.