HunyuanVideo 1.5

HunyuanVideo 1.5, released November 20, 2025, is Tencent's open-source video generation model: an 8.3B-parameter Diffusion Transformer with a 3D causal VAE and a custom Selective and Sliding Tile Attention variant that cuts 10-second 720p synthesis time nearly 2x versus FlashAttention-3. A two-stage pipeline generates 5–10 second clips at 480p–720p, then a dedicated super-resolution network upscales to 1080p. Crucially, it fits in 14GB of VRAM, making it one of the few credible open video models runnable on consumer hardware.

The honest picture: it is a mid-tier 2025 open model, not a closed-leader beater — motion stability and fine aesthetic quality trail Seedance, Veo 3, and Sora-class systems. There is no native audio (Tencent points to the separate HunyuanVideo-Foley line), and the license is Tencent's Hunyuan community license rather than Apache/MIT, so check commercial terms. For privacy-sensitive or offline video work and fine-tuning experiments, though, it is currently the most practical open option.