MD: Audio Auto EQ (Adaptive)

AI-generated audio is often muddy in the low-mids, muffled on top, or harsh on vocals. Fixing that by hand means an EQ with a dozen bands and an hour of knob-twiddling. MD_AudioAutoEQ_Adaptive is the one-shot version: analyze the spectrum, compare it to one of 27 target profiles, and apply corrective EQ automatically. It's the surgical first step of the pack's mastering story - Auto Master Pro is the whole chain, this is the EQ-focused piece.

How it works

The engine runs an FFT analysis of the incoming audio, measures how the spectrum deviates from the chosen target profile, and applies EQ correction toward it. Three modes matter:

• adaptive_mode: full - real-time FFT analysis drives a dynamic EQ curve shaped by the actual content. This is the true auto-EQ.
• hybrid - blends the analysis with a fixed curve.
• preset - applies a static EQ curve and ignores the audio content. Fast, but it's just a fixed filter.

strength (0–1, default 0.7) sets how aggressively it chases the target. The tooltip's guidance is 0.6–0.8 for a natural sound; 1.0 is maximum correction and can sound processed.

The target profiles

27 of them, organized into groups: voice (Podcast/Speech, Vocal Clarity, Radio Voice, De-esser, Warmth, Presence) and music (Music Master, EDM/Electronic, Rock/Metal, Hip-Hop/Trap, plus more). Default is Music Master - a standard balanced sound that the tooltip recommends for most generative music. Pick the profile that matches your genre and the engine aims at it.

The other required inputs are highpass_freq (default 20Hz - the tooltip suggests 30Hz to clean sub-mud without losing bass) and lowpass_freq (default 20000Hz, keep it there for transparency).

Outputs

• audio - the EQ'd audio, ready to feed a master chain.
• report_string - analysis text describing what was measured and changed.
• image_1 / image_2 / image_3 - three plot images showing before/after spectrum comparisons. These make it trivial to verify the EQ actually did what you wanted - run the spectrum visualizer's advice, except it's built in.

The honest take

For its purpose - one node that makes generative audio stop sounding like generative audio - it's effective, and the before/after plots are a genuinely good UX touch. The caveats: "auto" EQ is only as good as the target curve, so a profile mismatch (say, applying a vocal profile to a full mix) will fight the source. And it's EQ only - no loudness, no dynamics. Pair it with Auto Master Pro for the full polish, or use it standalone when the mix is basically fine but dull.

Installing

It's part of MD Nodes:

cd path/to/ComfyUI/custom_nodes
git clone https://github.com/MDMAchine/ComfyUI_MD_Nodes.git
cd ComfyUI_MD_Nodes && pip install -r requirements.txt

Or via ComfyUI Manager (search MD Nodes), then restart. This is one of the nodes that actually justifies the heavy requirements - librosa does the FFT analysis, soundfile handles the I/O, matplotlib draws the plots.