HunyuanVideo 1.5

HunyuanVideo 1.5 is Tencent's November 2025 open-source text-to-video and image-to-video model. At 8.3B parameters it's roughly 40% smaller than the original 13B HunyuanVideo, yet Tencent reports stronger visual quality and motion coherence thanks to a new attention mechanism (SSTA), glyph-aware text encoding, and a built-in super-resolution stage that upscales 480p/720p outputs to 1080p. The headline change is hardware: where the 13B needed ~40GB VRAM, 1.5 runs on consumer GPUs with a 14GB minimum via offloading. It has native ComfyUI and Diffusers support, ~4,500 GitHub stars, and a published technical report.

The honest picture: 'consumer GPU' still means a 24GB card for comfortable use, and 14GB-minimum setups need aggressive offloading that slows generation considerably. Default clips are short (~5 seconds at 24fps). The license is a custom Tencent open-source license rather than Apache/MIT, so commercial-use terms need a careful read. Closed models like Veo 3.1 and Sora 2 still lead on raw quality.

Clone github.com/Tencent-Hunyuan/HunyuanVideo-1.5, download weights from Hugging Face, or use the Diffusers HunyuanVideo15Pipeline / native ComfyUI support.