Freedom251 › AI Music Generators
HeartMuLa
What HeartMuLa is
HeartMuLa is a free, open-source family of music foundation models for understanding and generating music. It can generate songs from lyrics and tags with multilingual support, and supports vocal generation with MP3 export. The family includes HeartMuLa, HeartCodec, HeartTranscriptor, and HeartCLAP. HeartCodec is a 12.5 Hz music codec described as having high reconstruction fidelity. HeartTranscriptor is a Whisper-based lyric transcription model, while HeartCLAP aligns audio and text for music descriptions and cross-modal retrieval. Generation can combine style descriptions, lyrics, and reference audio, with section-level style control and short music creation for videos described. Models support web demos and self-hosted deployment, with checkpoints available through Hugging Face and ModelScope. Local setup recommends Python 3.10 and CUDA, including multi-GPU or lazy-loading options. The Apache 2.0 license allows commercial use. A documented default maximum generated length is 240,000 milliseconds, and the 7B version was not yet released in the cited instructions.
Who it suits
It suits developers, musicians, and researchers seeking open-source music generation, lyric transcription, audio-text retrieval, or self-hosted deployment. Commercial use is allowed under the stated license.
What it does well
- Free and open source
- Multilingual lyric-and-tag music generation
- Supports web and self-hosted use
- MP3 audio export
- Commercial use allowed
What to know first
- Default generated audio limit is 240,000 milliseconds
- 7B version was not yet released
- Local setup recommends Python 3.10 and CUDA
Verdict
HeartMuLa offers a broad open-source model family spanning generation, codecs, transcription, and audio-text alignment. Users should account for its documented generation-length limit and local hardware requirements.
Facts from HeartMuLa's own pages
- Audio codec
- HeartCodec is described as a 12.5 Hz music codec with high reconstruction fidelity. · github.com, 12 Oct 2026
- Audio export formats
- mp3 · heartmula.github.io, 21 Sept 2026
- Audio-text alignment
- HeartCLAP aligns audio and text in a shared embedding space for music descriptions and cross-modal retrieval. · heartmula.github.io, 12 Oct 2026
- Audio-text retrieval
- HeartCLAP aligns audio and text in a shared embedding space for music descriptions and cross-modal retrieval. · heartmula.github.io, 12 Oct 2026
- Commercial use
- Yes · heartmula.github.io, 21 Sept 2026
- Downloads
- The repository provides pretrained checkpoints through Hugging Face and ModelScope for local deployment. · github.com, 12 Oct 2026
- Generation controls
- The models support multi-conditional creation using style descriptions, lyrics, and reference audio; the site also describes section-level style control and short music generation for videos. · heartmula.github.io, 12 Oct 2026
- Hardware
- The setup instructions use CUDA devices and describe splitting model parameters across multiple GPUs or enabling lazy loading to save GPU memory. · github.com, 12 Oct 2026
- Integration
- The repository links to a community-created ComfyUI custom node for HeartMuLa. · github.com, 12 Oct 2026
- Integrations
- The repository links to a community-created ComfyUI custom node for HeartMuLa. · github.com, 12 Oct 2026
- Known limitation
- The README says the 7B version was not yet released when its usage instructions were written. · github.com, 12 Oct 2026
- License
- The repository and related model weights are stated to use the Apache 2.0 license. · github.com, 12 Oct 2026
- Limit
- The repository documents a default maximum generated audio length of 240,000 milliseconds. · github.com, 12 Oct 2026
- Local deployment
- The repository provides instructions to install HeartMuLa locally with Python and run music generation from downloaded model checkpoints. · github.com, 12 Oct 2026
- Model components
- The family includes HeartMuLa, HeartCodec, HeartTranscriptor, and HeartCLAP. · github.com, 12 Oct 2026
- Model family
- The family includes HeartMuLa, HeartCodec, HeartTranscriptor, and HeartCLAP. · github.com, 12 Oct 2026
- Music codec
- HeartCodec is described as a 12.5 Hz music codec with high reconstruction fidelity. · github.com, 12 Oct 2026
- Online demos
- The maintainers say they launched online demo spaces on Hugging Face and ModelScope. · github.com, 12 Oct 2026
- Supported platforms
- web · heartmula.github.io, 21 Sept 2026
- Product
- HeartMuLa is an open-source family of music foundation models for music understanding and generation. · heartmula.github.io, 12 Oct 2026
- Requirements
- The local deployment instructions recommend Python 3.10 and document CUDA device settings for inference. · github.com, 12 Oct 2026
- Song generation
- HeartMuLa generates music conditioned on lyrics and tags, with multilingual support. · github.com, 12 Oct 2026
- Support
- The maintainers list [email protected] and invite users to join their Discord or WeChat group. · github.com, 12 Oct 2026
- Transcription
- HeartTranscriptor is a Whisper-based model tuned for lyric transcription. · github.com, 12 Oct 2026
Where HeartMuLa stands on Freedom251
- AI Music Generators#103 of 123