HQ Mastering & Cover Art

Generated music has a tell, and it's up top. Above 8 kHz, AI takes leave a thin, glassy shimmer - a stereo "sizzle" at the edge of the image that isn't on any real recording. Every Plenio template ships a single switch that attacks exactly that, and it's off by default, because turning it on roughly doubles how long a song takes to render. Let me be blunt: it's the single biggest quality lever in the pack, and the reason people leave it off is also the reason they should try it once.

What it actually is

This is not a mastering suite. It's a switch - a small card at the top of group 1 in every Plenio template, next to the writer model, and its job is to flip one stage inside the Plenio · Master block further down the graph. That stage is HQ 2 · Regenerate Highs, which runs UniverSR: it wipes the song's top octave and has a super-resolution model grow a new one. On, you get clearly less AI shimmer and fewer artifacts. Off, the song passes through untouched and - importantly - UniverSR is never loaded, so a missing model file isn't an error.

Because the switch is wired only into the master block, toggling it re-runs the mastering and nothing else. ComfyUI reuses the render and everything before it. That's the whole reason it's a separate node at the top instead of a field buried in the master: you A/B the finish without regenerating the song.

Inputs and outputs

There is one input, hq_mastering - a Plenio finish switch, default off in the song templates and on in 4 · Enhance & Master since 1.2. And one output, hq_mastering, a plain boolean that wires into Plenio · Master's HQ mastering input, or straight into HQ 2 if you're building the graph by hand.

That's it. The output is just "on" - the node adds no tuning knobs, because the tuning lives in HQ 2's preset (and in the master block, where you can bypass individual stages with Ctrl+B).

One thing that confuses people: in the templates the card visibly shows a second Cover Art switch beside it. That's drawn by Plenio's frontend extension, not by this node - the node's schema has exactly one input. If you're hunting for where Cover Art lives in the API, it doesn't, not here.

How to install it

Same as the rest of the pack, once, for all of it. Search Plenio Music Production System in ComfyUI Manager → Custom Nodes Manager and install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/jplenio/Plenio-Music-Production-System

Restart, then open template 0 · System Check from the template browser and run it - it lists which model files your templates still need. Plenio installs no Python packages at all; requirements.txt is empty by design. The one file that matters here is UniverSR (pytorch_model.bin, 229 MB, CC BY 4.0 weights) into models/audio_sr, which ComfyUI offers to download when a template opens.

The time cost, honestly

Measured by the author on an RTX 5060 Ti with a 97-second song: about 280 seconds with the switch on versus roughly 130 without. That's the whole argument for the default. UniverSR wants about 3 GB of VRAM and runs in bfloat16, so it fits alongside plenty - this isn't a case where you need a 24 GB card.

Practical advice: leave it off while you're iterating on lyrics, arrangement and takes, then do one final render with it on. The seed inside HQ 2 is fixed after each run, so if nothing before it changed, ComfyUI will hand you back the same regenerated highs rather than recomputing them.

Where people get burned

• Expecting it to sound louder. It doesn't touch level. Loudness is handled later, at -14 LUFS / -1 dBTP, after the EQ.
• Waiting on a CPU. HQ 2 runs at about 1.5 s per second of music on a mid-range GPU and much slower without one. A four-minute song is roughly six extra minutes on the 5060 Ti.
• Treating it as a magic shimmer remover. It narrows the top end, because UniverSR makes both channels more alike. Paired with HQ 1 · Stereo Cleanup the effect is stronger; alone it's a modest, real improvement rather than a transformation.
• Wondering why nobody on Reddit warned you. There are essentially zero threads about this pack in the usual communities - it's a recent, single-author project with unusually good docs instead of a crowd-sourced knowledge base. Read docs/user/concepts/audio-hq.md in the repo; it beats most pack docs you'll meet.