The new version of the neural network is already used in Masterpiece, and will later appear in other Yandex services
The Yandex team presented a new version of the diffusion neural network Yandex AI Rendering Technology (YandexART), which creates images and animations in response to user text requests.
YandexART 1.3. switched to new technology for generating images — latent diffusion. In addition, the dataset on which the model was trained was increased by 2.5 times. Thanks to this, YandexART better understands text queries and creates even more realistic images in different formats.
The press service explained:
Latent diffusion technology consumes less computing resources and allows you to create more realistic graphics. It forms an intermediate representation of the picture in the form of a latent code — a compact description containing basic information about the image in a compressed form. The neural network then expands the code into a full high-resolution image in one step. This approach is more effective than multi-stage image refinement in cascade diffusion.
In addition, the YandexART update will give users the ability to create images in different formats such as 16:9, 4:3 or 3:4.
