HunyuanVideo
Tencent's open text-to-video foundation model with inference code and weights, needing a large NVIDIA GPU.
Code and weights for generating video from text prompts, with FP8, multi-GPU and ComfyUI routes.
HunyuanVideo is a large video-generation model from Tencent, released with PyTorch model definitions, pretrained weights and inference code. It suits researchers and technically capable studios with serious GPU hardware. Its licence excludes the European Union, the United Kingdom and South Korea, so check where you are before you start.
What it is
You clone the repository, download the weights and generate video from a text prompt using the command line or a Gradio web demo. The repo also covers prompt rewriting, FP8 quantised weights, multi-GPU parallel inference through xDiT, and Diffusers and ComfyUI integrations. Later models such as HunyuanVideo-1.5 live in separate repositories.
Who it's for
It suits ML engineers, researchers and production teams outside the excluded regions who own or rent high-memory NVIDIA GPUs. It does not suit anyone in the EU, UK or South Korea, where the licence says it does not apply, nor laptop users, nor people wanting a ready-made video editor.
Key features
Text-to-video generation
Generates video from text prompts at settings up to 720px by 1280px and 129 frames.
Prompt Rewrite model
A companion model adapts prompts to the format the video model was trained on.
FP8 quantised weights
Released FP8 weights cut GPU memory use.
Multi-GPU inference
Parallel inference code powered by xDiT speeds up generation across several GPUs.
Gradio web demo
Run a local Gradio server for point-and-click prompting.
Ecosystem integrations
Works with Diffusers and ComfyUI, with community GGUF and speed-up projects listed.
Up and running in 3 steps
Our honest verdict
What's great
- Weights, inference code and a technical report are all published, so you can run generation on your own hardware.
- The README lists community routes for lower memory and faster sampling, including FP8, GGUF quantisation and caching methods.
- Diffusers and ComfyUI support means you can slot it into existing generative workflows.
- The licence states that Tencent claims no rights in the outputs you generate.
Where it falls short
- It needs an NVIDIA GPU with CUDA: the README gives 60GB peak memory for 720p and 45GB for 544p, and recommends an 80GB card, so this is not a desktop or laptop tool.
- The licence does not apply in the European Union, the United Kingdom or South Korea, and any use or output use outside its Territory is unlicensed.
- It is not an open-source licence: you may not use the model or its outputs to improve any other AI model, and a 100 million monthly-active-user threshold requires a separate licence from Tencent.
- The README states Linux as the supported operating system, and setup involves conda, CUDA-specific PyTorch builds and a separate weights download.
Pricing & license
The code and weights are free to download under Tencent's community licence, and the repository shows no paid plan. The README links a hosted Playground at https://video.hunyuan.tencent.com whose pricing is not shown there. Self-hosting costs your GPU hardware or rental.
Open remote AI-training roles
876 open remote AI-training roles, updated November 15, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Can I use HunyuanVideo in the EU or UK?
No. The licence states that it does not apply in the European Union, United Kingdom and South Korea and limits use to a Territory that excludes them.
What GPU do I need?
An NVIDIA GPU with CUDA. The README lists 60GB peak memory for 720px by 1280px at 129 frames and 45GB for 544px by 960px, and recommends 80GB.
Is the output free to use commercially?
The licence says Tencent claims no rights in outputs and you are responsible for them, but it restricts use to the Territory and bars using outputs to improve other AI models. Read the full agreement before commercial use.
