Tuesday, 6 October 2026 · 6 min read
Kandinsky 6.0 Video generates synchronized 5-second videos with 44 kHz audio using dual-stream diffusion models
Kandinsky 6.0 Video introduces two diffusion models (3B and 29B parameters) that generate 5-second Full-HD videos with synchronized high-quality audio and lip-sync from text or images, using a dual-stream CrossDiT architecture and multi-stage training including reinforcement learning and distillation.
Based on the paper Kandinsky 6.0 Video: Foundation Models for Synchronized Video and Audio Generation by Team Kandinsky, Julia Agafonova, Bulat Akhmatov, Mikhail Aksyutin and 84 others
Challenges of synchronized video and audio generation
Generating videos with synchronized audio is more complex than generating either modality alone because it requires semantic, temporal, and emotional consistency across both modalities, including accurate lip-sync for speech.
Recent models have explored joint audio-video generation with various architectures and synchronization mechanisms, but open-source solutions often lag behind proprietary systems in clip duration, resolution, and synchronization accuracy.
Kandinsky 5.0 provided a strong video generation foundation, but extending it to synchronized audio-video generation requires new architectures, training strategies, and datasets to handle the audio modality and cross-modal alignment.
Sign in to read the whole paper
You have read the opening. The rest is free with a Bezen account, and no card is needed.
In the rest of this paper
- How the method works, step by step, with a flow chart
- Three key formulas from the paper, explained symbol by symbol
- Three figures from the paper, and how to read them
- How it was tested, what was found, and where it falls short