LogInGoogle introduced the neural network that creates images from a description called Imagen. The AI system is said to deliver "an unprecedented degree of photorealism".
You might have heard about DALL-E – an AI system by Open AI that can generate realistic images and art using a description in natural language. Not long ago, the team released an updated version of the tool, DALL-E 2, which is capable of creating concepts, attributes, and styles that were never seen before and/or impossible to see in real life and, thanks to the improvements, can make realistic edits to existing images from a natural language caption.
Hot on the heels, Google introduced its own neural network that creates images from a description called Imagen. The tool was developed by Google Brain, a deep learning artificial intelligence research team under the umbrella of Google AI.
According to the developers, the new AI system is capable of delivering "an unprecedented degree of photorealism" and "a deep level of language understanding".
The neural network recognizes text using large transformer language models and generates the first image of 64x64 pixels based on the received data. It further utilizes text-conditional super-resolution diffusion models to gradually increase the resolution to 256x256 and then to 1024x1024 pixels, while adding more details to the image.
The company invited a group of testers to compare images created with Imagen, DALL-E 2, VQ-GAN+CLIP, and Latent Diffusion Models. The experiment showed that people most often preferred images generated by the Google neural network: both in terms of fidelity and quality of the resulting image and image-text alignment.
Here are some examples of how Imagen works:
A dog looking curiously in the mirror, seeing a cat.
An art gallery displaying Monet paintings. The art gallery is flooded. Robots are going around the art gallery using paddle boards.
A cute corgi lives in a house made out of sushi.
A strawberry mug filled with white sesame seeds. The mug is floating in a dark chocolate sea.
The company offered to try out the neural network in a limited tech demo on Imagen's website. So far, users cannot enter queries themselves but are only able to choose select words from those offered.
It is currently unclear when Google will offer to use Imagen to the fullest extent. The project is still in the experimental phase.
You can learn more about Imagen here. Also, don't forget to join our new Reddit page, our new Telegram channel, follow us on Instagram and Twitter, where we are sharing breakdowns, the latest news, awesome artworks, and more.