RepoPicks Radar

Updated every day around 23:30 UTC

Guides

HeartMuLa Studio

fspecii/HeartMuLa-StudioReviewed by AI

Checked by RepoPicks Radar on

What is HeartMuLa Studio?

HeartMuLa Studio is a web page you run on your own computer to generate complete songs, with vocals and lyrics up to about four minutes, or instrumental tracks. You type a topic and style tags, or upload an audio file as a style reference. It is for people with a gaming or workstation PC: it needs an NVIDIA card with 10 GB or more of video memory, Docker or Python plus Node, and it downloads about 5 GB of AI models on first run. Lyrics can be written by a local Ollama model or by the paid OpenRouter service.

HeartMuLa Studio is free and can replace Suno and Udio.

How easy is it to install HeartMuLa Studio?

D · terminal, several tools or a specific graphics card · Windows

  • Windows D
  • Linux D

Linux is the smoothest path and still grade D: you clone the repository and run "docker compose up -d", which needs Docker plus the NVIDIA Container Toolkit and an NVIDIA card with 10 GB or more of video memory, then waits for a ~10 GB image and ~5 GB of models. Windows is the same grade through Docker, but the from-source route there asks for Python, Node and, for the faster torch.compile mode, a separate "triton-windows" package.

Coming soon: install in 1 click with the RepoPicks app.

HeartMuLa Studio on GitHub

Stars
647on Nov 10, 2026
Downloads
—all releases
Growth
+1stars in 6 days

Stars over time

647648649 Nov 7Nov 8Nov 9Nov 10 647648649 Nov 7Nov 9

Facts

Runs on
Windows, Linux
You need to install
Docker, NVIDIA graphics card
To build from source
Docker, NVIDIA graphics card
Disk space
not determined
Accounts or keys
OpenRouter
Cost
Free · The software and the music models cost nothing and run on your own card. Only the optional OpenRouter route for writing lyrics is paid; the local Ollama route is free.
License
No license (No license)
Can replace
Suno, Udio
Last change on GitHub
Feb 25, 2026

HeartMuLa Studio risk signals (9)

  • AttentionInstall command runs a script from the internet

    Install command pipes a remote script into a shell: curl -fsSL https://ollama.com/install.sh | sh

    automatic check

  • Attentionreadme says mit but no license file

    README: "This project is open source under the [MIT License](LICENSE)" and a badge reading "License-MIT", but the top-level file list is ".dockerignore, .github/, Dockerfile, README.md, backend/, docker-compose.yml, frontend/, preview.gif, start.sh" with no LICENSE file, and the checks detected no license. Without that file the MIT permission is only a claim in the text.

    AI review

  • Attentionnvidia card with 10gb required

    README, Docker prerequisites: "NVIDIA GPU with 10GB+ VRAM (12GB recommended for reliable generation)", and the VRAM table marks "<8GB" as "Not supported". AMD cards are listed only as "Flash Attention: Disabled (compatibility varies)". A computer without a suitable NVIDIA card cannot run it.

    AI review

  • InfoNo license detected

    No license detected by GitHub, the license file, the LICENSES folder or the winget manifest

    automatic check

  • Infoprebuilt image from personal docker hub account

    README offers "ambsd/heartmula-studio:latest" on Docker Hub next to "ghcr.io/fspecii/heartmula-studio:latest". The Docker Hub name "ambsd" does not match the GitHub owner "fspecii" (the README does link to the X account @AmbsdOP), so that image is not published under the repository's own name.

    AI review

  • Infomodels downloaded from the internet on first run

    README: "Models are automatically downloaded from HuggingFace Hub on first run (~5GB)". The AI music models are not in the repository, so what you actually run comes from HuggingFace at start-up.

    AI review

  • Infoweb page opens on all network interfaces

    The documented start command is "python -m uvicorn backend.app.main:app --host 0.0.0.0 --port 8000" and the Docker file publishes "8000:8000". With 0.0.0.0 the page answers anyone who can reach your computer on that port, and the README mentions no password or login.

    AI review

  • Infogenerated songs may copy a style you upload

    README, "Reference Audio (Style Transfer)" marked "Experimental": "Use any audio file as a style reference" with a "Style Influence" slider, and "Coming Soon: LoRA Voice Training ... voice consistency that surpasses Suno". Uploading someone else's recording or voice to copy its style can run into the rights of that recording or performer, and the repository gives no terms of use.

    AI review

  • Infono commits for about eight months

    The checks record the last commit on 2026-02-25, 254 days before this review, while the README still announces LoRA voice training as "Coming Soon" and the reference-audio feature as "Experimental".

    AI review

Evidence is shown in its original language.

What Radar could not check about HeartMuLa Studio

  • Whether generation really works on each card in the VRAM table, and the quality of the songs it produces
  • Real disk use: the README mentions a ~10 GB Docker image plus ~5 GB of models, and the checks could not measure it
  • What is inside the pre-built images on Docker Hub and GitHub Container Registry, since they are built elsewhere
  • Whether the MIT permission claimed in the README applies, as no license file exists in the repository
  • The cost of using OpenRouter for lyrics, which depends on the model you pick
  • Whether any text or audio you type leaves your computer when the Ollama or OpenRouter lyrics feature is on

Similar projects

The RepoPicks app

Coming soon: install in 1 click with the RepoPicks app.

Before you download

See the release page on GitHub