HunyuanVideo
Tencent-Hunyuan/HunyuanVideoReviewed by AI
Checked by RepoPicks Radar on
What is HunyuanVideo?
HunyuanVideo is Tencent's open-weights text-to-video model with the code to run it. You type a description and it produces a video of about five seconds, up to 720p. It runs from the command line or from a simple web page it starts locally (Gradio). The hardware bar is high: the authors list 45 GB of video memory for 544x960 and 60 GB for 720p, test only on Linux with an NVIDIA card, and the model files are downloaded separately from Hugging Face.
HunyuanVideo is free and can replace Runway, Kling, and Pika.
How easy is it to install HunyuanVideo?
D · terminal, several tools or a specific graphics card · Linux
Linux is the only system the authors test: you create a conda environment, install a specific PyTorch build for your CUDA version, install the requirements, optionally build flash-attention, download the model files and then run a Python script (a Docker image is offered as an alternative). An NVIDIA card is required and the stated peak memory is 45-60 GB.
Coming soon: install in 1 click with the RepoPicks app.
HunyuanVideo on GitHub
- Stars
- 12,754on Nov 10, 2026
- Downloads
- —all releases
- Growth
- +30stars in 6 days
Stars over time
Facts
- Runs on
- Linux
- You need to install
- Python, NVIDIA graphics card, FFmpeg
- To build from source
- Python, Graphics card, FFmpeg
- Disk space
- 1,010 MB–4.5 GB · estimate
- Accounts or keys
- None
- Cost
- Free
- License
- NOASSERTION (Other)
- Can replace
- Runway, Kling, Pika, LTX Studio, Veo 3
- Last change on GitHub
- Oct 8, 2026
HunyuanVideo risk signals (5)
- AttentionDependencies with known vulnerabilities
Pinned dependencies with known vulnerabilities (OSV): torch: GHSA-63cw-57p8-fm3p, GHSA-887c-mr87-cxwp, GHSA-qfhq-4f3w-5fph; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74; accelerate: GHSA-4j2p-28q2-5m79, PYSEC-2026-3804; tqdm: GHSA-g7vv-2v7x-gj9p, PYSEC-2026-1976; gradio: GHSA-39mp-8hj3-5c49, GHSA-5cpq-9538-jm2j, GHSA-6655-8ph2-63j3
automatic check
- Attentionlicense excludes eu uk south korea
LICENSE.txt, first line: "THIS LICENSE AGREEMENT DOES NOT APPLY IN THE EUROPEAN UNION, UNITED KINGDOM AND SOUTH KOREA AND IS EXPRESSLY LIMITED TO THE TERRITORY", and section 5(c): using or displaying the works or their output outside that territory "is unlicensed and unauthorized".
AI review
- Infolicense extra commercial conditions
The Tencent Hunyuan Community License is not a standard open-source licence: section 4 requires a separate licence from Tencent above 100 million monthly active users, and section 5(b) forbids using the outputs to improve any other AI model.
AI review
- Infoneeds datacenter class gpu
README Requirements table: 60 GB of GPU peak memory for 720px1280px129f and 45 GB for 544px960px129f; "The model is tested on a single 80G GPU". FP8 weights are said to save about 10 GB.
AI review
- Infodocker command runs privileged
The README's Docker commands use --privileged and --security-opt=seccomp=unconfined, which give the container far-reaching access to the host machine.
AI review
Evidence is shown in its original language.
What Radar could not check about HunyuanVideo
- How many gigabytes the model files take once downloaded from Hugging Face, since the download instructions live in a separate file
- Whether it can be made to run on a normal home graphics card with FP8 weights and CPU offload, and how slow that is
- Whether the install works on Windows or macOS, which the authors do not test
- How long one 5-second video takes on a single GPU outside the authors' multi-GPU measurements