zurk / hunyuan-video-8bit

Hunyuan Video 8bit model API for video generation

  • Public
  • 232 runs
  • GitHub
Run with an API
Playground API Examples README Versions

Examples

View more

Run time and cost

This model costs approximately $0.27 to run on Replicate, or 3 runs per $1, but this varies depending on your inputs. It is also open source and you can run it on your own computer with Docker.

This model runs on Nvidia L40S GPU hardware. Predictions typically complete within 5 minutes. The predict time for this model varies significantly based on the inputs.

Readme

Here’s the improved version of your README:

HunyuanVideo API (8-bit Version)

HunyuanVideo is a cutting-edge text-to-video generation model capable of creating high-quality videos from text descriptions. It surpasses many closed-source alternatives in text alignment, motion quality, and overall visual fidelity.

This API provides access to the 8-bit version of the model, which is optimized for performance on less expensive GPUs and offers faster inference compared to the full HunyuanVideo model.

Examples

import replicate

output = replicate.run(
    "zurk/hunyuan-video-8bit:main",
    input={
        "prompt": "A cat walks on the grass, realistic style.",
        "negative_prompt": "Ugly",
        "width": 960,
        "height": 544,
        "video_length": 65,
        "embedded_guidance_scale": 6.0,
        "num_inference_steps": 40,
        "seed": 43,
    }
)

Parameters

  • prompt (string, required): Text description of the video you want to generate.
  • negative_prompt (string, optional): Text describing elements you want to exclude from the video.
  • width (integer, default: 960): Video width in pixels.
  • height (integer, default: 544): Video height in pixels.
  • video_length (integer, default: 65): Number of frames (maximum 129).
  • seed (integer, optional): Random seed for reproducibility. If not specified, you can find its value in the logs.
  • embedded_guidance_scale (float, default: 6.0): Scale for embedded guidance during generation.
  • num_inference_steps (integer, default: 40): Number of denoising steps.
  • flow_shift (float, default: 7.0): Parameter for motion control (flow shift).

Limitations

  • The maximum video length is 129 frames (approximately 5.3 seconds).
  • The video_length parameter must follow the formula 4*n+1 (e.g., 17, 21, 25, etc.).

Feedback

If you encounter any issues while using this API, please report them by creating an issue at GitHub Issues. I will address them as soon as possible.

For further details, visit the HunyuanVideo GitHub repository or explore the ComfyUI wrapper nodes for HunyuanVideo.