Meet the team that turns your ideas into work.

Discover EightBots
EIGHTBOTS
Contact salesSign up
Kandinsky LabKandinsky 3

Kandinsky 3 Flash

Flash is a trained distillation of Kandinsky 3. It is also used as a refiner in the 3.1 release.

A trained fast model also serves as a refiner

Training used an adversarial diffusion-distillation approach in latent space. The repository provides a separate Flash weight file; generation and refinement are uses of that file.ai-forever/Kandinsky3.1 — model card and files

Specifications & access

Kandinsky 3 Flash specifications
DetailValue and conditions
AccessDownloadable
Model identifiersKandinsky FlashKandinsky 3.0 RefinerKandinsky 3.1 Flashai-forever/Kandinsky3.1/weights/kandinsky3_flash.ptai-forever/Kandinsky3.1
Weight licenseApache 2.0ai-forever/Kandinsky3.1 — model card and files
SamplingFour steps for image generationKandinsky 3.0 technical report, revised June 2024

Related models

  • An image diffusion model with a larger text encoder and U-Net, trained with attention to Russian cultural subjects.

    Launch date not confirmedDownloadable