The Year in AI Papers
The year/Independent research

Paper 2511.14993

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

Vladimir Arkhipkin, Vladimir Korviakov, Nikolai Gerasimenko, Denis Parkhomenko, Viacheslav Vasilev, Alexey Letunovskiy, Nikolai Vaulin, Maria Kovaleva, Ivan Kirillov, Lev Novitskiy, Denis Koposov, Nikita Kiselev, et al.

Published
Nov 2025
Research lab
Independent
Citations
9
GitHub
805 stars

01 In brief

Summary

Kandinsky 5.0 is a family of foundation models for high-resolution image and video synthesis, comprising three model line-ups: Image Lite (6B parameters), Video Lite (2B), and Video Pro (19B).

The models are built on a unified latent diffusion architecture with a CrossDiT backbone and trained using flow matching.

Key innovations include the NABLA sparse attention mechanism, which reduces training and inference time by 2.7× for long or high-resolution videos, and a multi-stage training pipeline incorporating pre-training, supervised fine-tuning (SFT), distillation, and RL-based post-training.

The data pipeline curates over 500 million images and 250 million video scenes, with specialized datasets for image editing and Russian cultural content.

Human evaluations show Kandinsky 5.0 Video Lite outperforms Sora and Wan models in visual quality and motion dynamics, while Video Pro excels in visual quality and motion but lags in prompt following compared to Veo 3.

The models are open-sourced under the MIT license, with code and checkpoints available on GitHub and Hugging Face.

The report also discusses limitations, including text alignment and temporal consistency for complex dynamics, and outlines future work toward a unified foundation model.

02 From the paper

Abstract

This report introduces Kandinsky 5.0, a family of state-of-the-art foundation models for high-resolution image and 10-second video synthesis. The framework comprises three core line-up of models: Kandinsky 5.0 Image Lite - a line-up of 6B parameter image generation models, Kandinsky 5.0 Video Lite - a fast and lightweight 2B parameter text-to-video and image-to-video models, and Kandinsky 5.0 Video Pro - 19B parameter models that achieves superior video generation quality. We provide a comprehensive review of the data curation lifecycle - including collection, processing, filtering and clustering - for the multi-stage training pipeline that involves extensive pre-training and incorporates quality-enhancement techniques such as self-supervised fine-tuning (SFT) and reinforcement learning (RL)-based post-training. We also present novel architectural, training, and inference optimizations that enable Kandinsky 5.0 to achieve high generation speeds and state-of-the-art performance across various tasks, as demonstrated by human evaluation. As a large-scale, publicly available generative framework, Kandinsky 5.0 leverages the full potential of its pre-training and subsequent stages to be adapted for a wide range of generative applications. We hope that this report, together with the release of our open-source code and training checkpoints, will substantially advance the development and accessibility of high-quality generative models for the research community.