Hunyuan1.5 use attention masks with variable-length sequences. For best performance, we recommend using an attention backend that handles padding efficiently.

We recommend installing kernels (pip install kernels) to access prebuilt attention kernels.

You can check our documentation to learn more about all the different attention backends we support.

import torch

dtype = torch.bfloat16
device = "cuda:0"
from diffusers import HunyuanVideo15Pipeline, attention_backend
from diffusers.utils import export_to_video

pipe = HunyuanVideo15Pipeline.from_pretrained("hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-480p_t2v", torch_dtype=dtype)
pipe.enable_model_cpu_offload()
pipe.vae.enable_tiling()

generator = torch.Generator(device=device).manual_seed(seed)
with attention_backend("_flash_3_hub"): # or '"flash_hub"' if you are not using H100/H800
    video = pipe(
        prompt=prompt,
        generator=generator,
        num_frames=121,
        num_inference_steps=50,
    ).frames[0]
    export_to_video(video, "output.mp4", fps=24)

Of course, you can still run inference with default attention backend.

import torch

dtype = torch.bfloat16
device = "cuda:0"
from diffusers import HunyuanVideo15Pipeline
from diffusers.utils import export_to_video

pipe = HunyuanVideo15Pipeline.from_pretrained("hunyuanvideo-community/HunyuanVideo-1.5-Diffusers-480p_t2v", torch_dtype=dtype)
pipe.enable_model_cpu_offload()
pipe.vae.enable_tiling()

generator = torch.Generator(device=device).manual_seed(seed)

video = pipe(
    prompt=prompt,
    generator=generator,
    num_frames=121,
    num_inference_steps=50,
).frames[0]
export_to_video(video, "output.mp4", fps=24)