HunyuanVideo

HunyuanVideo is Tencent's open-weights video foundation model, released December 2024 with over 13 billion parameters — the largest open-source video model at launch and the first major open release to ship both inference code and weights. It focuses on text-to-video generation (image-to-video arrived with the later 1.5), producing 5-second 720p clips at 24fps with emphasis on cinematic quality, motion dynamics, and text-video alignment.

Benchmarks (all provider-reported — Tencent-run professional human evaluation, not independent):

• Motion quality: 64.5% vs Runway Gen-3's 48.3% (1,500+ prompts, 60 professional evaluators — provider-reported)
• Claimed to outperform Runway Gen-3, Luma 1.6, and three top Chinese video models on visual quality, motion diversity, text-video alignment, and stability (provider-reported human eval)

No standardized automated benchmark (e.g. VBench) scores were published by the provider or found independently; the evidence base is Tencent's own human-evaluation program described in the technical report (arXiv 2412.03603).

Note: HunyuanVideo-1.5 (Nov 2025, tencent/HunyuanVideo-1.5) unifies text-to-video and image-to-video and is Tencent's current line; this entry covers the original 1.0 release.

Professional human evaluation (provider-reported): 64.5% motion quality vs Runway Gen-3 48.3% (1,500+ prompts, 60 evaluators). No independent or standardized automated benchmark scores found.