HunyuanVideo 1.5 🧠 Model Open source

What it is

HunyuanVideo is Tencent's open-weight video generation family. The current release, HunyuanVideo 1.5, came out in November 2025. It is an 8.3-billion-parameter model built to run on consumer GPUs, and Tencent followed it with training code, LoRA scripts and step-distilled variants in December 2025. We found no newer HunyuanVideo release by September 2026.

Key features

• Text-to-video and image-to-video at 480p and 720p
• 1080p through a built-in super-resolution stage
• About 5 seconds by default (121 frames)
• Step-distilled 480p image-to-video model that renders in 8-12 steps
• Runs on about 14 GB of VRAM with offloading. Training and LoRA code are public

Where you can use it

Download from GitHub and Hugging Face (tencent/HunyuanVideo-1.5) and run it locally. ComfyUI supports HunyuanVideo natively, and Krea lists an older Hunyuan model as legacy. Hosted options are mostly third-party inference sites.

Pricing and rights

The model is free to download under the Tencent Hunyuan Community License. That is a custom licence, not Apache or MIT. Check its territory and usage terms before commercial use. The model card does not mention native audio. You pay only for your own compute.

Who it is for

ComfyUI users and developers who want a smaller open video model to fine-tune, and who can live without sound.

Verdict

HunyuanVideo 1.5 is a good open base model for local work and LoRA training. It has not been updated in 2026, has no native audio, and its default clips are short next to Wan and LTX.