HeartMuLa
Open-source AI song generator — a full song with vocals, or run its 3B model on your own GPU
About
HeartMuLa is an AI song generator: describe music or paste lyrics and it makes a full song with vocals in about 1–2 minutes, in 10+ languages, with two versions per generation. Separately, its open-source 3B music model can be run on your own GPU. The project also publishes related model work (HeartCLAP, HeartCodec, HeartTranscriptor) and has demos on its site. Do not repeat its self-published rating claim in a listing without independent verification.
Web app: sign up in the browser and generate. Local model (verified from the README): git clone https://github.com/HeartMuLa/heartlib.git, then cd heartlib and pip install -e ., download the model checkpoint from Hugging Face (HeartMuLa-oss-3B), then run the example generation script in the README. The site states about 24 GB VRAM for local use (RTX 3090 / 4090 / A100 recommended).
More in Voice & Audio
View allElevenLabs
The most realistic AI voices — clone yours in minutes.
Whisper
OpenAI's open-source speech-to-text — transcribe anything locally.
Anthropic Agent Skills
Anthropic's official skill collection — documents, spreadsheets, design and more.