HunyuanVideo (Tencent)

HunyuanVideo is Tencent's family of open-weight video generation models, part of its wider Hunyuan model line. The original HunyuanVideo, released on December 3, 2024, had more than 13 billion parameters and was the largest open video model at the time. Its successor, HunyuanVideo 1.5 (November 20, 2025), cut that to 8.3 billion parameters and runs on a single consumer GPU. Alongside Alibaba's Wan, it's one of the two Chinese model families that released serious open weights for text-to-video. Its license sets it apart: it doesn't apply in the European Union, the United Kingdom or South Korea.

Models

HunyuanVideo 1.5 is a diffusion transformer paired with a 3D causal VAE that compresses video 16 times spatially and 4 times in time. It uses a selective and sliding-tile attention scheme to cut the cost of attention over long clips. It generates 121 frames by default. Tencent says it needs a minimum of 14 GB of GPU memory with model offloading, which brings it within reach of a single consumer card. Tencent has since added distilled variants for faster generation and, in December 2025, training code and LoRA fine-tuning. As of October 2026, no 2026 successor appears on Tencent's Hugging Face or GitHub pages.

The License

HunyuanVideo uses the Tencent Hunyuan Community License. Its "Territory" is "the worldwide territory, excluding the territory of the European Union, United Kingdom and South Korea", and the license says it "does not apply" there, so a company or user in those places has no license to use the model at all. A second clause requires any licensee whose products had more than 100 million monthly active users in the month before a Hunyuan release to request a license from Tencent, which it may grant "in its sole discretion". The license gives no reason for the carve-outs. The excluded markets include the EU, where the EU AI Act's transparency duties for generated media have applied since August 2026.

How It Compares

The license, more than quality, usually decides between HunyuanVideo and its open rivals. Wan 2.2 is Apache-2.0 and usable anywhere. Lightricks' LTX-2.5 is free under its community license for companies with less than $10 million in revenue. MiniMax's H3 weights go further than Tencent's and exclude the United States as well. On hardware, HunyuanVideo 1.5 is the lighter model to run, and it's a credible choice where its license applies. None of these open models leads the 2026 rankings: the top of Artificial Analysis's boards belongs to closed, audio-native models such as Wan 3.0, and no HunyuanVideo model was listed there as of October 20, 2026. For the detailed comparison, see Wan vs HunyuanVideo; for the wider field, generative video.