MiniMax Music Production Toolkit: Full Music Studio in ComfyUI

The full production chain: LLM song planning, Music 3 generation, audio enhancement and FLUX.2 cover art in one workflow.

Song prompts without learning the prompt format

Instead of hand-writing MiniMax Music 3 captions and lyrics, you pick what you want from dropdowns: genre, tempo, key, language, voice, lyrics mode and target length. A local LLM (any GGUF through a bundled llama.cpp chat node, for example Qwen or Gemma) then writes the full song plan: caption, lyrics, title and even a cover-art prompt.

The structured parser extracts [Caption], [Lyrics], [Title] and [Image_Prompt] from the LLM response, with manual fallback fields if you switch the LLM section off. A bundled library of 62 genre presets (EDM, house, metal, folk, classical and more) ships with metadata that prefills every field.

The audio enhancement chain

The toolkit's pitch is audibly better output than raw Music 3, through a fixed processing chain:

• Declip / overload repair: conservative reconstruction of hard-clipped waveform tops before further processing.
• PRE/POST low-pass filtering: band-limiting before and after enhancement to keep artifacts out.
• FlashSR super-resolution: a bundled FlashSR node upscales audio to 48 kHz, replacing the external Egregora node used in earlier versions; only the weights download on first use.
• Hybrid crossover and cymbal/shimmer repair: blends original and super-resolved signal and reduces watery high-frequency sustain while preserving attacks.
• Release prep: static LUFS/true-peak gain and resampling, with no compressor or time-varying loudness riding.

Output lands as FLAC/MP3/WAV with proper [Album] - [Title] naming, embedded tags and a production JSON that recreates or modifies every song later. The same JSON drives a 25-track demo catalog on GitHub Pages; the author's SoundCloud set has full-length examples.

Self-contained in v2.0

Version 2.0.1 removed the last external node dependencies: the integrated MiniMaxFlashSRAudio, MiniMaxLLMChat and MiniMaxLLMUnload nodes mean the example workflow uses only toolkit and core nodes. Other additions include a FLUX.2 cover branch (square artwork, size also controls embedded cover resolution), declarative model auto-download via models_config.json, multi-GPU support with the LLM on its own card, and automatic migration of pre-2.0 saved workflows.

Availability

Install from ComfyUI Manager by searching "MiniMax Music Production Toolkit". You need the MiniMax Music 3 model files as with any Music 3 setup, plus a local LLM GGUF for the prompt stage; everything else downloads itself.