HunyuanVideo

Why we included this project

Tencent's HunyuanVideo is an open-weight text-to-video model that ships the PyTorch definitions, pre-trained weights, and sampling code needed to generate short clips from a prompt. If you want to run a large video diffusion model on your own hardware rather than through an API, this repo is the practical place to start: it documents the two-stage diffusion-transformer design and includes the exact weights and inference pipeline to reproduce the paper's results. Because it's one of the bigger open-weight generators, it also works well as a reference when you're studying how modern DiT-based video models are put together or want to fine-tune the base model. The surrounding pieces, including a Hugging Face Diffusers pipeline, a prompt-rewrite model, and the newer 1.5 release with efficiency gains, let you move from research code to a more packaged setup when needed.