YandexART 1.3 introduced: Better rendering and realistic images

Yandex added a new innovation in artificial intelligence technologies and announced YandexART 1.3, which can better interpret user commands and produce realistic images. This new version offers users the opportunity to turn their texts into art, thanks to its advanced hidden diffusion technology, and draws attention with the increase in image quality.

Improved technology and innovative approaches

YandexART 1.3 is equipped with the latent diffusion method, this technique makes it possible to produce high-quality graphics using less computing resources. The latent diffusion process begins with the development of an intermediate image representation, called a latent code, which contains the basic information of the image in compressed form. This code is converted into high-resolution images with the help of a neural network, and this process gives much faster and more effective results compared to gradual diffusion techniques.

The model's training dataset has been expanded to include over 850 million image-text pairs to enable it to better understand user requests. Yandex also added synthetic text, which are detailed image descriptions generated by the neural network, to this dataset. This allows YandexART to interpret text-based commands more accurately.

Additionally, two text encoders were integrated so that the model could better understand the details obtained from user prompts. Thanks to these encoders, YandexART can process text commands more effectively and convert them into machine-readable data.

Innovative areas of use in practice

YandexART 1.3 offers users the opportunity to work with various image formats. These formats include options such as 16:9, 4:3 or 3:4. This enables the images produced to be used on a wide range of media, from magazine covers to television screens.

Internal evaluations show that YandexART 1.3 performs significantly better than previous versions and other similar technologies. In tests, this new version outperformed Midjourney V5.2 by 57% and YandexART 1.2 by 63%.

The introduction of YandexART 1.3 further expands the boundaries of AI-based artistic expression and content creation. This technology, which allows users to create impressive and realistic works of art with just simple texts, ushers in a new era in the world of digital art and creativity. This innovative step by Yandex stands out as an indicator of how far technology can go when it meets art, by prioritizing user experience.