What Changes
Inside Your Track

Suno and Udio both produce a narrow-band resonance centered around 3 to 5 kHz that reads as "shiny" or "metallic" on headphones. It is the giveaway. It does not exist on acoustic instruments in nature and trained ears flag it within seconds.

A dynamic EQ cut at the problem frequency, with a Q around 1.4 to keep it surgical. The cut is matched to the level of the resonance; on the worst cases we use a soft de-esser-like behavior to catch the peaks. No broadband cuts, that would dull the whole track.

The bass region on AI tracks is often a perfect sine subwoofer signal with almost no harmonic content above the fundamental. It feels "there and not there" at the same time: pressure without texture.

Saturation with even harmonic emphasis, then a gentle compressor on the bass band. The harmonics give the ear something to lock onto, so the bass sounds weighted instead of sterile. We do not add sub harmonic content; that lies about what your track is.

AI vocals have inconsistent sibilance. The "s" on one word sounds natural, the next "s" hits like a snare. Sustained vowels drift in pitch by a few cents where a human singer would not.

Multi-band de-essing with lookahead, set above 6 kHz only. Targeted micro-pitch correction on problematic sustains, less than 20 cents. We do not run autotune-style processing that makes everything "in tune" at the cost of expression.

Drums and rhythmic elements hit the grid with sub-2ms deviation. Real musicians drift within a 15 to 30ms window. The perfection is what makes it sound unnatural.

Manual swing adjustment on a per-section basis. We do not run random quantization shuffle plugins; the swing has to feel right in context of the song, not statistically.

Sampled kits, often a single velocity layer, no room sound. Snares are the same sample on every hit.

Live drum kit recorded in a treated room. Multiple takes, real hits, real bleed between mics, real human velocity variance. We choose the kit and the heads to match the song. Not a sample library swap, an actual recording.

Sterile sine sub or amp-sim with no room. Bass notes are present but the audience does not feel them.

Electric bass or upright bass, recorded direct plus a room mic, with a real amp and a real cabinet. We do not model that. We plug into the hardware.

Stereo shimmering pads with brittle high end. Acoustic piano emulation that sounds like sampled piano in a fishbowl.

Hardware synths (the analog ones, with actual pitch drift and overshoot), recorded at line level. Acoustic keys are real piano or real rhodes, mic'd with proper stereo technique. The artifacts we are trying to remove cannot be modeled away, so we do not try.

Amp-sim distortion plumes with no interaction. Clean guitars have phasey stereo that screams "plugin."

Real guitar amps, mics in front of them, room mics for the air. Distortion comes from pushed speakers and tubes, not algorithms. Choice of guitar, pickup position, and amp is decided per section.