Save Production JSON

Generating a song you like is one thing. Being able to recreate it six weeks later - same prompt, same seeds, same settings - is the thing that actually separates a demo from a production habit. MiniMaxSaveProductionJSON is the pack's answer: it writes one canonical, atomic JSON per song that records the entire generation, enough that the pack's metadata loader can rebuild the track (with modified settings) from that file alone.

It's the last node in the chain for a reason, and it's an output node - everything funnels into it. After the original FLAC, the release FLAC, the release MP3 and the album cover have all been saved, this node gathers the evidence and writes the record. Note the deliberate ordering: it refuses to run until those upstream saves have happened, because a configuration file that claims artifacts exist when they don't is worse than no record at all.

What goes in the JSON

The whole story, in one place:

• the LLM stage - system prompt, user prompt, raw output, status, even the thinking;
• the structured-prompt summary and the parsed Caption / Lyrics / Title / Image_Prompt with provenance;
• seeds and generation settings - MiniMax text seed/CFG/top-k, sampler seed/steps/CFG/denoise, duration;
• every audio-enhancement report: declip, PRE/POST low-pass, FlashSR settings, hybrid crossover, HF repair, release prep;
• standard audio tags, and the actual saved file paths for the original, both release formats, and the cover.

That's the reproducibility story: together with the outputs section it's a complete generation record, not a sidecar of loose metadata.

The inputs you'll actually touch

The required list looks intimidating but most of it is pre-wired in the example workflow:

• original_audio_save_json, release_flac_save_json, release_mp3_save_json - connect these to the save_info_json outputs of the three audio savers. These connections are what make the node wait for the files to exist.
• artwork_path - the saved JPG path, same idea for the cover.
• configuration_prefix - the destination base, fed from the MiniMax Output Paths node; its directory is controlled by that node's configuration_subdir (default json).
• audio_tags_json and title - used for the consistent Album - Title filename and copied into the record.
• collision_mode - auto_increment (recommended for batches), overwrite, or error_if_exists.
• filename_mode - album - title is recommended so the JSON matches your release audio naming; it doesn't touch embedded title metadata.
• create_directories - just leave it on.

Everything else - the llm_*, caption/lyrics/image_prompt, the seed fields, and the long list of *_json enhancement reports - is optional and fed by the corresponding nodes in the chain. Wire what you have; the payload only includes what you connect.

Outputs

• saved_path - where the JSON landed.
• configuration_json - the rendered JSON itself, if you want to pipe it somewhere.

Where people get confused

First: no JSON beside your FLAC files is not a bug. Since v1.0.4 the pack deliberately writes one centralized file per song into the json/ subdirectory instead of duplicate sidecars - look in MiniMax Output Paths → configuration_subdir. Second, if the JSON never appears, the odds are an upstream save failed; check the ComfyUI log for the first failed saver, because this node won't paper over it. And the file is written atomically (temp file, then rename), so an interrupted write leaves you with the old state rather than a half-written record.

Installing

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

Restart ComfyUI and hard-refresh the browser. If you're ever tempted to skip this node for a quick test, that's fine - it's pure bookkeeping, and everything before it still works. But for anything you might want to revisit, the five minutes of wiring is worth it. Future-you, six weeks out, will thank present-you for the record.