Google presents Imagen, an artificial intelligence capable of creating images from a simple description. The technology combines language understanding with photorealistic imagery design. The proposed result is truly stunning. And the artificial intelligence is even able to reproduce known designs and use drawing or painting styles.
Advances in machine learning and artificial intelligence have been enormous in recent years. And it seems obvious that it will manage to evolve until it catches up with science fiction novels. We knew that artificial intelligence is capable of responding to natural language prompts. Virtual assistants like Assistant, Siri or Alexa confirm this to us every day. We also know that artificial intelligence is able to associate words with images. If you search for “dog” or “beach” in Google Photos, the engine offers you relevant results.
Read also – Deepfake: this AI will make film dubbing much more believable
We also knew that an AI is able to create more or less realistic images. We are not only talking about Deepfake, although it could possibly be related to it. We’re just talking about recreating a scene from… nothing. But, imagine for a moment that we combine these three technologies: language understanding, meaning interpretation and visual production. This gives Imagen, the new technology developed by google brainthe firm’s team dedicated to artificial intelligence.
Imagen interprets a sentence and transforms it into an image
Imagen is a simply stunning AI: it creates an image from a description. You can discover at the top of this article a montage of four images provided on the official Imagen website. The description of the bottom left one is simple: “ a dog looks at an ice cream with curiosity. And he sees a cat there “. These two simple sentences were enough. For the visual at the bottom left: a giant cobra sits in a field. It is made of corn grain. At Phonandroid, we were all amazed at the realism of these images.
By going to the official website, you will even have the possibility of “testing” Imagen. Indeed, a multiple-choice text is proposed. You select the parts that interest you and it offers you a suitable image. And, even more surprisingly, the system offers a choice of a photorealistic montage or an image imitating styles of famous artistslike Rembrandt.
On this official page, you will also find a text on the limitations and the societal impact that such technology could have. Car, Imagen could be powerful enough to create deepfakes of public figures, without the victim being able to do much about it. Google confirms that Imagen blocks racist, pornographic or degrading images. But these are not the only limits that the firm imposes on its system, Google also evoking the tendency of its AI to create images with rather peanut-type humans. Which is not very inclusive.