The Yandex team has announced a prototype of its own neural network for creating images based on user text prompts, launching a public beta test.
The "Shedevrum" application is available on Android and iOS mobile devices.
The beta version of "Shedevrum" 0.1.4 can be downloaded for free from the Google Play and App Store app stores. Descriptions can be written in Russian and English.
After installing the application, users can view images and like them. To create pictures based on your own requests, you need to submit an application and wait for it to be approved. In the future, this feature will be available to all registered users.
The neural network creates images using a cascading diffusion method: first it generates images corresponding to the request, and then gradually increases their resolution and saturates them with details.
For beta testing, the first version of the generative model was trained on 240 million examples of images with text descriptions. Training continues on a dataset of 500 million examples, and the quality of images will gradually improve in future updates.
Now on home
