Open weightsFree
HunyuanVideo
Tencent
HunyuanVideo is a video generation model line from Tencent. The latest release is HunyuanVideo 1.5 (20 Nov 2025). This page tracks all 5 HunyuanVideo releases since 3 Dec 2024, with specs, licensing, and how to access each one. Official site →
HunyuanVideo 1.5released
Open weightsFreeCommunity APISelf-host
Lightweight open video model (T2V and I2V); runs on roughly 14GB VRAM
- License
- Tencent Hunyuan Community License
- Commercial use
- Allowed
- Availability
- Self-host, Community API
- Parameters
- 8.3B
HunyuanVideo-Avatarreleased
Open weightsFreeCommunity APISelf-host
Audio-driven multi-character human animation
- License
- Tencent Hunyuan Community License
- Commercial use
- Allowed
- Availability
- Self-host, Community API
HunyuanCustomreleased
Open weightsFreeCommunity APISelf-host
Multimodal-driven customized/subject-consistent video
- License
- Tencent Hunyuan Community License
- Commercial use
- Allowed
- Availability
- Self-host, Community API
HunyuanVideo-I2Vreleased
Open weightsFreeCommunity APISelf-host
Image-to-video built on HunyuanVideo
- License
- Tencent Hunyuan Community License
- Commercial use
- Allowed
- Availability
- Self-host, Community API
HunyuanVideoreleased
Open weightsFreeCommunity APISelf-host
Best open-source video model at launch; dual-stream transformer; competitive with closed models
- License
- Tencent Hunyuan Community License
- Commercial use
- Allowed
- Availability
- Self-host, Community API
- Parameters
- 13B
- Resolution
- 720p
FAQ
- When was HunyuanVideo 1.5 released?
- HunyuanVideo 1.5 was released on 20 Nov 2025.
- Is HunyuanVideo open source?
- Yes. HunyuanVideo 1.5 ships open weights under the Tencent Hunyuan Community License licence.
- Can HunyuanVideo be used commercially?
- Yes. Commercial use of HunyuanVideo 1.5 is allowed under the Tencent Hunyuan Community License licence.
- Where can I use HunyuanVideo?
- HunyuanVideo 1.5 is available via Self-host and Community API.
See HunyuanVideo in context on the video model timeline, or browse all model families.