MiniMax Music — локальная генерация музыки

MiniMax-Music3-Studio is an open-source desktop studio for generating full-length AI music tracks locally on your own GPU — no Python, no Node.js, no launcher, just a single Windows executable. Reach for it when you want to produce original music from a caption and lyrics without cloud subscriptions, when keeping audio data off external servers matters, when you need deterministic reproduction of any track you have ever made, or when fine-grained control over generation parameters — DiT steps, LM CFG, top-k, separate seeds, multiple variations per prompt — is essential. Under the hood it runs minimaxmusic.cpp with CUDA, a Tauri 2 + Rust UI, bundled ffmpeg for karaoke video rendering, word-level lyric alignment via Parakeet or Whisper, six-stem separation (drums, bass, guitar, piano, vocals, other) through HT-Demucs, a model manager with manual quantisation selection, a local library with search and playlists, and a caption/lyrics writing assistant backed by a local GGUF model or OpenRouter. Output formats are MP3 and 16/24/32-bit WAV with full ID3v2.4 tags, embedded cover art and lyrics. The tool generates music — it does not transcribe existing recordings or process pre-recorded audio (use other catalogue tools for that).