Yandex introduces YandexART 1.3

Yandex has announced YandexART 1.3, an improved version of its neural network that transforms descriptions provided by users via text into ready-to-use images and animations.

According to the company's statement, YandexART 1.3 is equipped with technology developed for generating images through latent diffusion. Additionally, the image dataset used to train the model has been increased by 2.5 times. These improvements allow YandexART to better understand text prompts and create more realistic images in various formats.

The new version of YandexART is available for use in the "Shedevrum" application worldwide, including in Turkey.

The latent diffusion technique requires fewer computational resources and facilitates the creation of higher-quality graphics. The process begins by developing an intermediate image representation known as a latent code. This contains a compact description that includes essential information about the image in a compressed form.

The neural network then converts this code into a high-resolution image in a single step. It was stated that this is more effective than the multi-stage refinement of technical, gradual diffusion.

WHAT ARE THE NEW FEATURES?

The company added detailed image descriptions known as synthetic texts, which are generated by a neural network, to the training dataset to ensure the model better understands user prompts. The dataset in question has been expanded to include more than 850 million image-text pairs.

Furthermore, two text encoders have been included in the model to ensure YandexART takes more details from user prompts into account. These encoders allow YandexART to accurately interpret text commands and convert them into machine-readable data.

The updated YandexART allows users to create images in various formats, including 16:9, 4:3, or 3:4, making these images usable for magazine covers, televisions, and more media.

Internal evaluations reveal that YandexART 1.3 performed better than Midjourney V5.2 in 57 percent of trials and better than YandexART 1.2 in 63 percent of trials.