WORK IN PROGRESS, STAY TUNNED

HunyuanVideo API

HunyuanVideo is a state-of-the-art text-to-video generation model that creates high-quality videos from text descriptions. This model outperforms many closed-source alternatives in text alignment, motion quality, and visual quality.

Examples

import replicate

output = replicate.run(
    "zurk/hunyuanvideo:main",
    input={
        "prompt": "A cat walks on the grass, realistic style.",
        "width": 1280,
        "height": 720,
        "video_length": 129,
        "guidance_scale": 6.0,
        "num_inference_steps": 50,
        "seed": 43,
    }
)

Parameters

• prompt (string, required) - Text description of the video you want to generate
• negative_prompt (string, optional) - Text describing what you don’t want in the video
• width (integer, default: 1280) - Video width in pixels
• height (integer, default: 720) - Video height in pixels
• video_length (integer, default: 129) - Number of frames (max 129)
• seed (integer, optional) - Random seed for reproducibility
• guidance_scale (float, default: 6.0) - Controls how closely the model follows the prompt
• embedded_guidance_scale (float, default: 6.0) - Additional guidance scale for embedded features
• num_inference_steps (integer, default: 50) - Number of denoising steps
• flow_shift (float, default: 7.0) - Flow shift parameter for motion control
• flow_reverse (boolean, default: true) - Whether to reverse flow direction

Limitations

• Maximum video length is 129 frames (5.3 seconds)

Feedback

If you have any issues running this API, create an issue on https://github.com/zurk/replicate-hunyuan-video/issues I try to fix them as soon as possible. However, I do not have direct access to GPUs with 80GB VRAM, so it may take some time.

For more details, visit the HunyuanVideo GitHub repository.