This is the largest model for generating images from a text description in Russian. -v-prilozhenii-na-smartfone-97fd090.jpg” alt=””Sberbank” introduced the Kandinsky neural network. You can try it in the application on your smartphone” />
Sberbank officially introduced the Kandinsky neural network. This is an improved version of the multimodal ruDALL-E, which generates images by description and was introduced last year.
The developers noted that they were able to significantly improve the quality of this model by retraining it on 179 million images with text descriptions. Kandinsky can generate images with arbitrary aspect ratio, and can also use a new way to increase the resolution.
Images are created in three stages:
- First, one neural network (Kandinsky) generates a given the number of images according to the description.
- Then the second (ruCLIP Large) selects the most successful and most relevant to the given text description of the work.
- And after the third one increases the size of the pictures.
A distinctive advantage of the Kandinsky model compared to previous versions is a higher degree of detail in the generated images.
Anyone can try out the Kandinsky neural network. It is available in the Salyut mobile app and on Sberbank smart devices upon request “Turn on the artist”.
Kandinsky will ask you to say a description, and after a few minutes he will return with nine options for images. It remains only to choose one of them to increase the size.
Salute! Smart DevicesDialogPrice: Free e32f8b3.png” alt=””Sberbank” introduced the Kandinsky neural network. You can try it in the app on your smartphone” />
Salut – Virtual AssistantsSberDevices LLCPrice: Free
< img class="aligncenter" src="/wp-content/uploads/2022/06/sber-predstavil-nejroset-kandinsky-ejo-mozhno-oprobovat-v-prilozhenii-na-smartfone-6f6fc21.png" alt="« "Sber" introduced the Kandinsky neural network. You can try it in the application on your smartphone" />