Help:Kandinsky 12b
Kandinsky 12b is one of the newer models (June of 2022) for generating images based in Generational Adversarial Networks (GANs). It was developed by Mr. Shonenkov, from Russia.
The dynamics for the final user are simple: Starting from a text prompt, the neural network iterates an image (or in this case, a number of images in parallel) and displays the final result. Thus, with normal parameters, 16 square images are generated; nominally of the same theme (at least, sharing the prompt). At the time of writing, these images are 256x256 pixels each.
It is one of ruDALLE's models. Kandinsky 12b seems to focus on the "surrealist style" and uses more parameters than the original model.
It is trained so that the text prompts are in the Russian language. The English prompts are translated by the bot, but the translation is imperfect and the results of generating in English are not ideal.
Kandinsky 12b is currently on an experimental stage. There are no payment models currently on public offer at the time of writing.
Usage on Shonenkov AI's Discord channel is free, but you can only place one prompt every some hours (this changes with waves of demand and degree of model optimization). Prompts take several hours to generate, since they are queued with those of all other users. Generation of prompts is sequential, in a First-In-First-Out (FIFO) waiting sequence.
The images generated by Kandinsky 12b can be used for your own projects, just by giving credit to the model and "Shonenkov AI" (the software engineer responsible).
You can join the Telegram channel at the following link: shonenkovAI
