TBNAI
Kandinsky 6.0 Video: Foundation Models for Synchronized Video and Audio Generation
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 03/10/2026, 20:00:00 UTC
Briefing created: 17/10/2026, 00:14:25 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 06/10/2026, 04:18:25 UTC.
Briefings
Kandinsky 6.0 Video introduces a family of diffusion models for synchronized text-to-audio-video generation, featuring a 3B parameter Lite model and a 29B parameter Pro model. The Pro model significantly outperforms the 5.0 predecessor in human evaluation, maintaining competitive performance with leading audio-video models while utilizing a dual-stream CrossDiT architecture for temporal and semantic alignment. Both models generate 5-second clips with 44 kHz audio and lip-sync, with the Pro model's resolution boosted to Full-HD via a built-in super-resolution model.
Key points
- Kandinsky 6.0 Video Pro outperforms Kandinsky 5.0 Video Pro in side-by-side human evaluation, particularly in speech quality.
- The 29B parameter Pro model uses a dual-stream CrossDiT architecture to align video and audio streams through bidirectional cross-attention.
Context
This research builds on Kandinsky 5.0 Video capabilities, introducing a continuous pretraining strategy that trains audio streams from scratch on large-scale corpora before joint training on paired data. The models are released under MIT license with code, checkpoints, and diffusers integration for open research and deployment.
Limitations
Evidence covers only the 5-second video clips generated by the models. Human evaluation results are based on side-by-side comparisons rather than quantitative metrics. No specific dates for model training completion are provided, and the exact number of paired audio-video datasets used for joint training is not specified.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 17/10/2026, 00:14:25 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeihcvvq6lqjl6d3c25uv66cjkypfo7tq4bkpusf2fa7o32rsxng3cu