Research

Research behind Kandinsky models. Architectures, training methods,
and publications shared openly with the AI community.

10+Publications

3+Publication years

EMNLP · NAACLPeer-reviewed venues

All publications

KVAE: Family of Tokenizers for Multimodal Generative Models

IEEE Access2026

NABLA: Neighborhood Adaptive Block-Level Attention

Unlocking the Duality between Flow and Field Matching

Kandinsky 5.0: A Family of Foundation Models for Image and Video Generation

Time-Correlated Video Bridge Matching

One-Step Residual Shifting Diffusion for Image Super‑Resolution via Distillation

RusCode: Russian Cultural Code Benchmark for Text-to-Image Generation

VIVAT: Virtuous Improving VAE Training through Artifact Mitigation

IEEE Access2024

Improve your videos: Architectural improvements for text‑to‑video generation pipeline

Kandinsky 3: Text-to-image synthesis for multifunctional generative framework

Doklady Mathematics2024

CRAFT: Cultural Russian-oriented dataset adaptation for focused text‑to‑image generation

Kandinsky 3.0 technical report

Kandinsky: an improved text‑to‑image synthesis with image prior and latent diffusion

FusionFrames: efficient architectural aspects for text‑to‑video generation pipeline