HeartMuLa — генерация музыки и песен из текста

★ 7.1 · content

heartmula is a Claude Code skill that generates complete songs from lyrics and tags using the open-source HeartMuLa model family (Apache-2.0), a self-hosted alternative to Suno for AI music generation. The system includes four components: the HeartMuLa language model (3B/7B), the HeartCodec 12.5 Hz audio codec, the Whisper-based HeartTranscriptor, and the HeartCLAP audio-text alignment model. It supports multilingual input, structured lyrics with [Verse]/[Chorus]/[Bridge] tags, and exposes parameters like temperature, cfg_scale, and top-k sampling. Minimum hardware is 8 GB VRAM with --lazy_load enabled; 16 GB+ is recommended for single-GPU comfort, with multi-GPU splits also supported. Output is 48 kHz stereo MP3 at up to 4 minutes, making this skill suited for local, offline music creation with fine-grained quality control.