Updated: September 3, 2026 · Docs · Clothoff AI Editorial Team
What is HunyuanVideo?
Tencent's open-weight video model line, from the 13B release of December 2024 to HunyuanVideo-1.5 — downloadable weights, a custom Tencent license, no built-in filter.
HunyuanVideo is an open-weight video generation model developed by Tencent's Hunyuan team, first released with 13 billion parameters on Dec 3, 2024, extended with HunyuanVideo-I2V on Mar 6, 2025 and succeeded by the lighter HunyuanVideo-1.5 on Nov 20, 2025. Code and weights are published on GitHub and Hugging Face under a custom Tencent license, with no safety filter included.
| Version | Date | Developer | License / Access | Filters |
|---|---|---|---|---|
| HunyuanVideo (13B) | Dec 3, 2024 | Tencent Hunyuan | Open weights and code; custom Tencent license file in the repository | No safety filter stated in the model card |
| HunyuanVideo-I2V | Mar 6, 2025 | Tencent Hunyuan | Open weights; same license family | No built-in filter stated |
| HunyuanVideo-1.5 | Nov 20, 2025 | Tencent Hunyuan | Open weights and inference code; distilled 480p I2V and training code Dec 5, 2025 | No built-in filter stated |
The license file name and territorial terms were not restated on the README pages read; the repository license text governs and should be read before deployment.
Which HunyuanVideo versions exist?
Tencent released the original HunyuanVideo on Dec 3, 2024 as a 13B-parameter text-to-video model with open inference code and weights, one of the largest open video models at the time. HunyuanVideo-I2V, the image-to-video variant, followed on Mar 6, 2025, and an avatar-focused model later that spring.
HunyuanVideo-1.5 arrived on Nov 20, 2025 as a lighter model designed to run on consumer hardware; Tencent added distilled 480p image-to-video weights and the training and LoRA code on Dec 5, 2025. All versions are on GitHub and Hugging Face.
“We release the inference code and model weights of HunyuanVideo-1.5.”
— Tencent Hunyuan, HunyuanVideo-1.5 repository, accessed Sep 3, 2026
Does HunyuanVideo allow NSFW output?
The model cards describe no safety classifier, and the repositories ship none, so a locally run HunyuanVideo generates whatever it is prompted to generate. Filtering, if any, is added by whoever hosts the model, which is the pattern for every open-weight video model on this site.
The weights are governed by a custom Tencent license rather than Apache or MIT; its use restrictions and any territorial limits apply to every deployment, and the README pages we read on Sep 3, 2026 do not restate them. Separately from the license, intimate imagery of a real person without consent is unlawful in most jurisdictions.
How is HunyuanVideo used in tools we review?
No tool in our rankings names HunyuanVideo as a backbone. Its consumer-hardware footprint since version 1.5 and the availability of LoRA training code make it a plausible component for self-hosted services, so we track it alongside Wan and LTX-Video as the open-weight options such services can choose from.
For the tested services see the Deep Undress review and the video hub; none discloses a model family. Under how we rate the privacy axis, weighted 30 percent, asks about retention and deletion, which an unfiltered open backbone makes more important, not less.
What does the LoRA training code change?
Since Dec 5, 2025 Tencent has published training and LoRA code for HunyuanVideo-1.5, which lets anyone fine-tune the model on a small dataset, including a specific person's likeness. That is the technical step that turns a general video model into a targeted tool, and it is where consent becomes decisive.
Fine-tunes inherit the license terms, and a LoRA trained on a real person without permission has no lawful use for intimate content. See LoRA for how adapters work and image-to-video for the input mode the I2V weights implement.
Consent note: tools in our rankings are documented for use only with photos of yourself or consenting adults 18+. See Responsible AI; if an image of you was misused, report NCII.
Sources
- GitHub — Tencent-Hunyuan/HunyuanVideo-1.5 — release Nov 20, 2025; updates Dec 5, 2025, accessed Sep 3, 2026.
- Hugging Face — tencent/HunyuanVideo-1.5 — model card and weights, accessed Sep 3, 2026.
- GitHub — Tencent-Hunyuan/HunyuanVideo — 13B release Dec 3, 2024; I2V Mar 6, 2025, accessed Sep 3, 2026.
- Tencent Hunyuan — official site — developer overview, accessed Sep 3, 2026.
- HunyuanVideo — project page — project site named in the repository, accessed Sep 3, 2026.
Reference entry by the Clothoff AI Editorial Team; not legal advice. The access date marks the last verification.
Frequently asked questions about HunyuanVideo
Is HunyuanVideo open source?
The code and weights are downloadable from GitHub and Hugging Face, but the weights carry a custom Tencent license rather than an OSI-approved one. The license includes use restrictions that every deployment must follow, so HunyuanVideo is open weights rather than open source in the strict sense of the term.
When were the HunyuanVideo versions released?
Tencent's repositories date the original 13B HunyuanVideo to Dec 3, 2024, the image-to-video model HunyuanVideo-I2V to Mar 6, 2025 and HunyuanVideo-1.5 to Nov 20, 2025, with distilled 480p image-to-video weights and the training and LoRA code added on Dec 5, 2025, as read on Sep 3, 2026.
Does HunyuanVideo include a safety filter?
No. Neither the model cards nor the repositories describe an NSFW classifier or output filter, so any moderation comes from the operator hosting the model. This is typical of open-weight video models and is the reason our reviews focus on how a service handles uploads rather than which model it runs.
What hardware does HunyuanVideo-1.5 need?
Tencent positions HunyuanVideo-1.5 as a model that runs on consumer graphics cards, with distilled 480p image-to-video weights lowering the memory requirement further. The original 13B model needs far more video memory. Exact needs depend on resolution, clip length and the inference framework used.
Can HunyuanVideo animate a photo of a real person?
Image-to-video is supported through HunyuanVideo-I2V and the 1.5 weights, so technically yes. Legally, turning an identifiable person's photo into intimate video without documented consent is unlawful in the United States, the EU and elsewhere, and the Tencent license's use restrictions apply in addition.
Do the tools we review run on HunyuanVideo?
No reviewed tool names HunyuanVideo publicly. Because it can be self-hosted and fine-tuned, we consider it a plausible backbone for services that avoid moderated APIs, but our scores rest on tested behavior: upload handling, retention, deletion and moderation, not on the weights a vendor loads.