A video generation framework for Diffusion Transformer models that creates videos at flexible resolutions and from images, with training support for baseline and LoRA models.
2.3K198PythonApache-2.0Updated 2 days ago
Tencent-Hunyuan/HunyuanVideo
An open-source 13B-parameter video generation foundation model from Tencent that creates videos from text prompts.
Open-source alternative to:Runway
Luma Dream Machine
- Stars
- 12,646
- Forks
- 1,358
- Open issues
- 187
- Last commit
- 14 days ago
What is it?
HunyuanVideo is an open-source video generation model from Tencent, released with PyTorch model definitions, pre-trained weights, and inference code. It generates videos from text prompts and is built on a diffusion transformer that operates in a latent space compressed by a causal 3D VAE.
The model uses a multimodal large language model as its text encoder rather than the CLIP and T5 encoders common in earlier text-to-video systems. It also includes an optional prompt rewrite model, based on Hunyuan-Large, with Normal and Master modes that adjust prompts for intent or visual quality.
The repository covers single-GPU inference with a command line tool or a Gradio web server, multi-GPU parallel inference through xDiT, and FP8 quantized weights to reduce GPU memory use. The default 720p, 129-frame setting needs about 60GB of GPU memory and is tested on Linux with NVIDIA GPUs.
Key features
- 01Generates videos from text prompts using a diffusion transformer architecture.
- 02Uses a decoder-only multimodal LLM as the text encoder.
- 03Provides a causal 3D VAE that compresses video into a compact latent space.
- 04Supports multi-GPU parallel inference via xDiT and USP.
- 05Offers FP8 quantized weights that save about 10GB of GPU memory.
- 06Includes a Gradio web demo and a command line sampling script.
Use cases
- Researchers experimenting with open-source text-to-video generation models.
- Developers who want to self-host video generation on their own NVIDIA GPUs.
- Teams building video generation features that need fine-tuning or custom pipelines.
- Creators testing text-to-video output locally with a Gradio interface.
Quick start
- 1
Clone the repository and follow the Linux installation guide in the README
- 2
Download the pretrained model weights as described in the README
- 3
Run single-GPU inference from the command line as described in the README
Things to know
- The 13B-parameter model needs a large GPU, and the README points to FP8 weights to reduce memory use.
- The metadata does not name a clear license, so check the license terms before commercial use.
In collections
Topics
FAQ
- What is HunyuanVideo?
- An open-source 13B-parameter video generation foundation model from Tencent that creates videos from text prompts. HunyuanVideo is an open-source video generation model from Tencent, released with PyTorch model definitions, pre-trained weights, and inference code. It generates videos from text prompts and is built on a diffusion transformer that operates in a latent space compressed by a causal 3D VAE.
- Is HunyuanVideo free to use? What license does it have?
- HunyuanVideo is open source under the Other license. Check the license terms before using it in commercial products.
- Is HunyuanVideo still maintained?
- Actively maintained. The last commit to HunyuanVideo was on Oct 8, 2026.
- What can I use HunyuanVideo for?
- Researchers experimenting with open-source text-to-video generation models. Developers who want to self-host video generation on their own NVIDIA GPUs. Teams building video generation features that need fine-tuning or custom pipelines. Creators testing text-to-video output locally with a Gradio interface.
- How do I install HunyuanVideo?
- Clone the repository and follow the Linux installation guide in the README Download the pretrained model weights as described in the README Run single-GPU inference from the command line as described in the README
- Where can I get HunyuanVideo?
- The source code and installation instructions are on GitHub: https://github.com/Tencent-Hunyuan/HunyuanVideo
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
Open-source large-scale video generative models for text-to-video, image-to-video, speech-to-video and character animation.
AI & Machine Learning#video-generation#text-to-video#image-to-video#diffusion-model#mixture-of-experts
18.1K2.4KPythonApache-2.0Updated last month
A PyTorch library of pretrained diffusion models, noise schedulers and building blocks for image, video and audio generation.
34.8K7.4KPythonApache-2.0Updated 6 hours ago
Open-source codebase for efficient high-resolution image and video generation using linear diffusion transformers, with training and inference pipelines.
9.6K769PythonApache-2.0Updated 2 days ago
A node-graph GUI, API and backend for running diffusion models locally to generate images, video, audio and 3D content.
AI & Machine Learning#stable-diffusion#diffusion-models#node-editor#image-generation#video-generation
142K16.7KPythonGPL-3.0Updated 7 hours ago
A benchmark suite and evaluation toolkit that scores text-to-video and image-to-video generative models across many quality and semantic dimensions.
1.8K135PythonApache-2.0Updated 2 months ago