FlashSR Lowpass Lab

Super-resolution models have a dirty secret: they reconstruct missing bandwidth from context, so if you give FlashSR a source whose top end is already a mess, it faithfully invents a mess in high fidelity. The fix is to control what FlashSR sees before it runs and what it's allowed to leave behind after. FlashSR Lowpass Lab is a configurable Butterworth low-pass that does both jobs with the same node - pick a PRE preset to clean up before enhancement, or a POST preset to sweep away the artificial air after.

The presets are genuinely useful, not decorative: PRE 12 kHz - recommended is the starting point for a 32 kHz MiniMax source, and the POST presets are gentler, designed to shave extreme reconstructed high-frequency energy without darkening everything. If you want to think of it as a "spectral leash" on the enhancer, you're not far off.

Where it sits

Two copies of this node normally appear in the chain: one before FlashSR (PRE), one after the hybrid crossover and HF repair (POST). The example workflow uses the sibling FlashSR Processing Settings node to drive both consistently, but you can drop standalone copies anywhere you need a clean low-pass with reproducible settings.

How it works

It's a Butterworth low-pass with three knobs plus a master bypass:

• preset - the seven choices: CUSTOM, four PRE presets (14 kHz light, 12 kHz recommended, 10 kHz strong, 8 kHz aggressive) and two POST presets (20 kHz gentle, 19 kHz slightly stronger). Selecting a preset updates the custom fields to show its effective values, which is a nice touch.
• custom_cutoff_hz, custom_order, custom_phase_mode - used only in CUSTOM. Cutoff is self-explanatory; order (1–12) steepens the slope; zero_phase runs the filter forward and backward offline to avoid phase rotation (effectively steepening the response further), while causal is a normal one-way filter with phase shift, useful for gentle post-processing.
• bypass - returns audio unchanged while still emitting settings/info, which is the A/B testing workflow's best friend.

Inputs and outputs

Beyond the widgets, every filter knob has a matching optional override input (preset_override, custom_cutoff_override, custom_order_override, custom_phase_override, bypass_override) meant to be driven from a centralized settings node. When connected, the override wins at execution time. Outputs are audio, info, preset (the effective preset as a string), and settings_json - the last two are exactly the reproducibility receipts this pack loves.

Installing it

Standard pack install - ComfyUI Manager, 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 (which does the filtering), soundfile, imageio-ffmpeg, mutagen, Pillow. Restart ComfyUI and hard-refresh the browser once.

Gotchas

The recurring mistake is picking the strongest filter "to be safe." Cutting too low before FlashSR discards real transient and cymbal information, and the README explicitly says not to do that automatically - for acoustic or cymbal-heavy material, A/B against the lighter presets or a bypass. And remember the difference between the two stages: PRE filters force FlashSR to reconstruct more, POST filters suppress what it invented. They pull in opposite directions, so don't treat them as interchangeable.