Google’s new text-to-image AI model delivers incredible results

Recently, artificial intelligence and machine learning, which are among the biggest research topics of computer scientists, are also research areas where large companies spend the most resources. One of the artificial intelligence teams of technology giant Google has managed to model an artificial intelligence with a very high ability to convert from text to images.

Google Imagen converts typed texts to images

Brain Team from Google Research teams announced an artificial intelligence model that converts texts to images. The AI, called the Imagen diffusion model, offers unprecedented photorealism and a deep understanding of language. Although Google’s Imagen model is similar to OpenAI’s DALL-E model, it renders images more realistically.

Researchers who want to measure Imagen’s capabilities against other artificial intelligence models, created a new test platform called DrawBench. With this test, more than 200 texts were presented to artificial intelligence models and they were asked to create images. These images were then rated by people. As a result of DrawBench, Google’s Imagen model came to the fore both in terms of visual and text comprehension.

Imagen will not be public

My Google Imagen model will not be available to the public at this time. The researchers noted that the databases they used to retrieve text data contained the negative aspects of internet language, and therefore the model could be prone to producing racist rhetoric, social stereotypes and pornographic content. Google stated that they will be able to offer a public structure on the Imagen platform in the future. Up to this point, you can try Imagen on its own site in a very limited sense.