⚡️ Kandinsky 6.0 Video vs Competitors: Benchmark Analysis
The flagship of the new open-source Kandinsky 6.0 Video lineup — the Pro model with 29 billion parameters — shows significant improvements over its predecessor and competitors.
📊 Benchmark results:
vs Kandinsky 5.0:
• Clear advantage across all criteria
• Improved visual quality
• Better prompt adherence
• Enhanced camera movement control
vs LTX 2.5 (open-source):
• Preferred in visual quality and speech quality
vs Veo 3.1 Fast (closed-source):
• Superior in image animation tasks
• Better visual quality
• Improved source image adherence
• More natural camera movement
🏗 Technical architecture:
• Dual-stream architecture (video + audio)
• Audio pretrained on 40M tracks; joint training on 7M audio-video segments
• 44 kHz audio output
• Full HD resolution support
• Lip-sync technology
• Lite version (3B) runs on consumer GPUs (16–32 GB VRAM)
🔧 Deployment:
Open-source under MIT license. Available through:
• Diffusers
• FastVideo
• ComfyUI
• SGLang
• vLLM-omni
Anton Frolov, Sber: "We're releasing it free of charge and without restrictions."
🔗 Hugging Face
Post #2432
284

- ❤ 1