If you liked AceJAM (Originally created by Victor and ported to run locally by me), you will love MiniJam. This is basically the exact same UI as AceJAM, but instead of using ACE-Step as the song generation model, uses MiniMax Music (just released).

The original Huggingface version of MiniMax Music JAM is here: https://huggingface.co/spaces/victor/MiniMax-Music3-Jam

But the original version requires a lot of VRAM to run, and could only generate up to 2 minutes (while the Minimax Music model officially supports up to 5 minutes), so I had to make a major swap to the backend while keeping the frontend the same.

System Support

• Cross Platform: I made sure it works on ALL OS (Mac, Linux, Windows)
• Best Performance: On top of that, I looked for the most optimized backend possible for each OS.
  • Windows/Linux: Uses ComfyUI as backend, and
  • Mac (Apple Silcon): Uses @ambsd's optimized fork of Audio.cpp at https://github.com/fspecii/audio.cpp-lowend-gpu, with a slight modification for Macs.

How it works

1. It comes with a built-in LLM (Qwen 3.5) via llama.cpp (also built-in)
2. It dynamically uses the correct backend (Audio.cpp or ComfyUI) based on the device.

Notes

1. Song duration: You can change the song duration from 30 seconds up to 5 minutes, but note that this is "Maximum duration", not precise duration. It will try its best to fit the song into that frame, but sometimes it may generate a 3 minute song even if you set 5 minutes as the maximum duration (which is still technically correct)
2. Lyrics auto-generated: The whole point of this app is you simply enter what you want WITHOUT having to write any structured lyrics yourself, and it will surprise you. Which is great but sometimes you may want to edit the lyrics. This app is not for that purpose.

Example