Tencent-Hunyuan /HunyuanVideo-1.5

About HunyuanVideo-1.5

HunyuanVideo-1.5 is Tencent's open video model for text-to-video and image-to-video. It runs on consumer NVIDIA GPUs with 14 GB of memory and offloading turned on. You get 480p and 720p clips, plus a super-resolution step that reaches 1080p. Linux only, and some weights are still coming soon.

Best for: Developers and creators who want an open video model on their own GPU for short-drama shots.

• Generates 480p and 720p video from text or a single image
• Runs on consumer NVIDIA GPUs with 14 GB memory and offloading
• Upscales results to 1080p with a few-step super-resolution model
• Speeds up 480p image-to-video with a step-distilled model (8 or 12 steps)
• Cuts render time with cache options (deepcache, teacache, taylorcache) and fp8
• Trains and fine-tunes with LoRA, FSDP and the Muon optimizer

1. 1
   Clone the repo: the command below then cd HunyuanVideo-1.5

   git clone https://github.com/Tencent-Hunyuan/HunyuanVideo-1.5.git

   Copy
2. 2
   Install dependencies: the command below and the next command

   pip install -r requirements.txt

   Copy

   pip install -i https://mirrors.tencent.com/pypi/simple/ --upgrade tencentcloud-sdk-python

   Copy
3. 3
   Install an attention library such as Flash Attention, Flex-Block-Attention, SageAttention or SGL-Kernel
4. 4
   Download the checkpoints by following checkpoints-download.md
5. 5
   Run generate.py with an image path for image-to-video, or IMAGE_PATH=none for text-to-video
6. 6
   Add --enable_step_distill to use the faster 8 or 12 step model