HunyuanVideo

Name: HunyuanVideo
Author: HunyuanVideo

Free plan LinuxSelf-hosted

What HunyuanVideo is

HunyuanVideo is an open-source video foundation model with PyTorch definitions, pretrained weights, and inference code for generating video from text or images. It supports text-to-video and image-to-video workflows, with listed maximum resolution of 1080p and commercial use permitted. The model has over 13 billion parameters and uses a decoder-only multimodal language model as its text encoder. Its unified architecture combines dual-stream and single-stream Transformer designs with full attention. Generation configurations include 540p and 720p at 129 frames. FP8 weights are available; the repository says they reduce GPU memory use by about 10GB. HunyuanVideo requires an NVIDIA CUDA GPU, with stated peaks of 60GB for 720p and 45GB for 544p generation. Linux is the tested operating system, with Docker images and an installation guide. Diffusers, xDiT parallel inference, and community ComfyUI tools are listed integrations.

Who it suits

It suits developers and creators who can run Linux-based, NVIDIA CUDA workflows and want open-source video generation. Commercial projects are permitted.

What it does well

Open-source and free
Supports text-to-video and image-to-video
Commercial use is permitted
Provides pretrained weights and inference code
FP8 weights reduce stated memory needs

What to know first

Requires an NVIDIA CUDA GPU
Linux is the tested operating system
No API access
No AI avatar generation

Limits

The stated peak GPU memory requirement is 60GB for 720p 129-frame generation and 45GB for 544p 129-frame generation. Generation configurations are provided at 540p and 720p with 129 frames.

Getting started

Open the HunyuanVideo GitHub repository
Follow the Linux installation guide or use the provided Docker images
Install on a system with an NVIDIA GPU and CUDA
Download the inference code and pretrained weights
Choose standard or FP8 weights based on available GPU memory
Run text-to-video or image-to-video generation

Verdict

HunyuanVideo offers open-source generation with text and image inputs, commercial use, and community integrations. Its substantial GPU requirements and Linux-focused setup may limit suitable environments.

Questions

Is HunyuanVideo free?

Yes. It uses a free pricing model and is open source.

Which operating systems and hardware are supported?

Linux is the tested operating system. HunyuanVideo requires an NVIDIA GPU with CUDA.

Can it generate video from images?

Yes. It supports both image-to-video and text-to-video generation.

What resolution can it produce?

The listed maximum resolution is 1080p, with repository configurations at 540p and 720p using 129 frames.

Can businesses use HunyuanVideo?

Yes. Commercial use is permitted.

Does it offer an API or AI avatars?

No. API access and AI avatar generation are not provided.

Facts from HunyuanVideo's own pages

AI avatars: No · github.com, 19 Sept 2026
API access: No · github.com, 19 Sept 2026
Architecture: The model uses a unified image and video generative architecture with a hybrid dual-stream to single-stream Transformer design and full attention. · github.com, 11 Oct 2026
Commercial use: Yes · github.com, 19 Sept 2026
Community tools: The project lists community contributions including native ComfyUI support and a ComfyUI wrapper. · github.com, 11 Oct 2026
Downloads: The project provides inference code and pretrained model weights, with FP8 weights also available to reduce GPU memory use by about 10GB. · github.com, 11 Oct 2026
Founded: 1998 · github.com, 28 Sept 2026
Generation: The repository supports text-prompt video generation at 540p and 720p, with 129-frame configurations. · github.com, 11 Oct 2026
Hardware requirements: The model requires an NVIDIA GPU with CUDA; the stated peak GPU memory requirements are 60GB for 720p 129-frame generation and 45GB for 544p 129-frame generation. · github.com, 11 Oct 2026
Headquarters: Shenzhen, China · github.com, 28 Sept 2026
Image-to-video: The project links a separate HunyuanVideo image-to-video model with inference code and checkpoints. · github.com, 11 Oct 2026
Integrations: The repository says HunyuanVideo is integrated into Diffusers and offers parallel inference code powered by xDiT. · github.com, 11 Oct 2026
License: The repository links a license file, but the opened page does not state license terms in its displayed content. · github.com, 11 Oct 2026
Lower-memory option: FP8 quantized weights save about 10GB of GPU memory, according to the repository. · github.com, 11 Oct 2026
Maker: Tencent says it was founded in 1998 and is headquartered in Shenzhen, China. · tencent.com, 11 Oct 2026
Maximum resolution: 1080p · github.com, 19 Sept 2026
Model size: The model has over 13 billion parameters. · github.com, 11 Oct 2026
Operating system: Linux is the tested operating system, and the repository provides a Linux installation guide and Docker images. · github.com, 11 Oct 2026
Product: HunyuanVideo is an open-source video foundation model; the repository provides PyTorch model definitions, pretrained weights, and inference and sampling code. · github.com, 11 Oct 2026
Prompt rewriting: Its prompt rewrite model offers Normal and Master modes; the Master mode adds detail about composition, lighting, and camera movement and may lose some semantic details. · github.com, 11 Oct 2026
Purpose: HunyuanVideo is an open-source video foundation model with code and pretrained weights for video generation. · github.com, 11 Oct 2026
Support: The repository invites users to join its WeChat and Discord communities and to share projects using HunyuanVideo. · github.com, 11 Oct 2026
Text encoder: The model uses a decoder-only multimodal large language model as its text encoder. · github.com, 11 Oct 2026
Text-to-video: The released model generates videos from text prompts, with supported output settings up to 720p and 129 frames. · github.com, 11 Oct 2026