Google has demonstrated an artificial intelligence model that generates images based on the entered text. Her diffusion model Imagen, created by the Brain Team at Google Research, according to the developers, offers “an incredible degree of photorealism and a deep level of understanding of language.”
Examples of images created by Imagen.
This is not the first time we have seen similar AI models. DALL-E from OpenAI performed a similar function, turning text into visual images. Google’s version, however, creates more realistic images. The researchers conducted a comparative test by asking people to evaluate images generated by different AI.
“Users preferred Imagen to other models when comparing side–by-side, both in terms of sample quality and in terms of matching images to text,” Google said.
But despite all the successes, Google has also reported the problem of disturbing content generated by Imagen. This was due to the fact that not all data sets for AI were filtered, in addition, the LAION-400M data set was used, which includes a wide range of unacceptable content, including pornographic images, racist insults and harmful social stereotypes.
Imagen is not yet available to the average user, because the development of the model is not finished. The Brain Team still has a lot of work to do to create a framework and further configure the AI.