MiniMax Song Metadata

This is the node the whole pack orbits. MiniMax Song Metadata collects every setting that went into a song - prompts, seeds, MiniMax generation parameters, FlashSR filters, repair and release-prep reports, even the LLM system prompt - and assembles them into one complete JSON sidecar that gets saved next to your audio. It's the "what exactly did I do here" record, written while you can still answer the question instead of six months later when you can't.

The sidecar philosophy is worth understanding because it's unusual and it's the pack's superpower. Audio file tags (title, artist) are fragile and lossy - players rewrite them, and there's no standard place for a text_cfg_scale. So this pack splits the job: MiniMax Standard Audio Tags handles the boring interoperable tags players actually read, and this node handles the everything-else JSON that makes a render reproducible. Connect the *_json outputs from the repair/release/prep stages and you get a complete audit trail.

The inputs that matter

Most inputs are straightforward pass-throughs you'll already have wired: title, caption, lyrics, image_prompt, source_name, source_path, prompt_origin, prompt_provenance_json, then the whole MiniMax settings block (generation_seed, max_duration, text_seed, text_cfg_scale, text_top_k, ksampler_seed, ksampler_steps, ksampler_cfg, sampler_name, scheduler, denoise), then the filter block (pre_preset, pre_settings_json, post_preset, post_settings_json, flashsr_lowpass_input), and run_index/variant_count/workflow_name.

The interesting optional inputs are the four report JSONs - declip_json, hybrid_crossover_json, hf_repair_json, release_prep_json - plus llm_system_prompt. Connecting them is what turns a good sidecar into a complete one. The llm_system_prompt one is notable: storing the full system prompt makes later prompt regeneration and auditing possible, at the cost of a big string in the JSON.

Outputs

metadata_json (the full sidecar as a STRING, for the saver) and summary (a human-readable one-liner). Wire metadata_json into Save Audio Smart Prefix, which writes it next to the file.

Installing it

Pack install via ComfyUI Manager (search "MiniMax Music Production Toolkit") or:

cd ComfyUI/custom_nodes
git clone https://github.com/jplenio/ComfyUI-MiniMax-Music-Production-Toolkit.git
cd ComfyUI-MiniMax-Music-Production-Toolkit
python -m pip install -r requirements.txt

Dependencies: scipy, soundfile, imageio-ffmpeg, mutagen, Pillow. Restart ComfyUI and hard-refresh once.

Gotchas

Don't confuse it with the loader's job - this node writes the recipe, MiniMax Metadata Loader reads it back. And remember the sidecar is only as complete as its inputs: skip wiring the *_json reports and you've shipped a sidecar that claims the release prep happened without recording what it did. The pack's README calls the whole thing "reproducible generation metadata" - but the metadata is only reproducible if you connect the receipts.