Question about prompt / input structure.

[Goldlionren]

Hi HeartMuLa team, thank you for the great work on heartlib 🙏

I’m currently experimenting with the models in heartlib and had a question regarding prompt / input structure.

Specifically, is there any recommended reference, example, or template for:

• how the style field is expected to be structured, and
• how lyrics should be formatted (e.g. sections, line breaks, tags, constraints, etc.)

I couldn’t find a clear example or guideline in the documentation, so I’m wondering whether there is:

• an internal prompt format you recommend,
• example inputs used during training, or
• any best-practice conventions that work particularly well with the model.

As a bit of background, I’ve been integrating heartlib into ComfyUI and recently released a custom node for it:
https://github.com/Goldlionren/ComfyUI_JR_heartlib

The node is designed to work across different hardware backends and currently supports both CUDA (NVIDIA GPUs) and XPU (Intel GPUs).
Having a more explicit reference for style / lyrics formatting would help ensure the node aligns better with the model’s intended usage across these environments.

Thanks again for open-sourcing the project, and looking forward to your guidance!