heartmula

To generate music locally, you provide a lyrics file and a comma-separated tags file to the music generation script. The system runs entirely on your machine using configurable GPU or CPU device parameters.

Yes, you can generate songs with multilingual lyrics on-device. The system processes bracketed section-structured lyrics like Intro, Verse, and Chorus to build full audio tracks without cloud services.

A GPU is not strictly required for local audio generation workflows, but it is recommended. You can configure the script to run on either a GPU or CPU, with options for splitting GPU memory between the music model and codec.

The best way to structure lyrics for AI music generation is using bracketed section markers like Intro, Verse, and Chorus. This structured format helps the model build complete, Suno-style song drafts accurately.

To install HeartMuLa for offline song generation, you must install the heartlib package and download both the HeartMuLa and HeartCodec checkpoints before running the provided music generation entry point script.

Generating Suno-style music tracks offline requires downloading large model checkpoints and demands significant hardware resources. You must manually configure device parameters to split GPU inference between the music model and the audio codec.