LookyMaster vs. SonicMaster — how they compare

Two different tools

One restores damaged recordings. One masters clean mixes with explainable, physiologically-anchored correction.

LookyMaster

Rule-based, explainable mastering

Deterministic signal processing — every correction traces to a measured artifact and a cited threshold
Seven physiologically-anchored frequency zones, harmonicity per zone
Patent-pending detectors: phase-coincidence spikes, detuned-harmonic families, static machine-residue lines
Dose-capped correction inside published listener-aversion and noise-damage bands
100% client-side — your audio never uploads
Assumes a reasonably-recorded mix; masters it

SonicMaster (AMAAI Lab)

Generative, text-conditioned restoration

Flow-matching generative model, trained on 25k paired clean/degraded tracks
Targets 19 degradation functions across 5 groups: EQ, dynamics, reverb, amplitude, stereo
Natural-language control — describe the fix you want, or run automatic mode
Runs on Hugging Face's own cloud compute (their infrastructure, not LookyMaster's)
Built to fix badly-recorded, degraded amateur audio — declipping, excess reverb, tonal imbalance
A published research result, not a commercial product

Where they actually overlap

Real overlap in EQ, dynamics, and stereo width — real gaps everywhere else.

We checked SonicMaster's own taxonomy from the paper against what LookyMaster's engine actually does. Some categories line up closely. Two don't line up at all — and we're naming that honestly rather than implying LookyMaster does something it doesn't.

Category	SonicMaster	LookyMaster
Equalization / tonal balance	✓ generative EQ correction	✓ 7-zone anchored EQ, presence-dip lift
Dynamics	✓ generative dynamics correction	✓ crest/PLR management, limiter, saturation
Stereo width	✓ restores narrowed stereo image	✓ mid/side widening, elliptical bass-mono
Amplitude / loudness	✓ generative amplitude correction	✓ loudness normalization, −1.0 dBTP ceiling
Declipping / distortion repair	✓ core restoration target	✗ not built — assumes clean source audio
Excessive reverb reduction	✓ core restoration target	✗ not built — adds reverb as an enhancement, doesn't remove it
Phase-coincidence / neural-generator tick detection	✗ not in their taxonomy	✓ patent-pending, LookyMaster-specific
Detuned harmonic-family detection	✗ not in their taxonomy	✓ LookyMaster-specific
Static machine-residue line detection	✗ not in their taxonomy	✓ LookyMaster-specific

Read plainly: if your recording is genuinely damaged — clipped, drowning in reverb, recorded on bad gear — SonicMaster is solving a problem LookyMaster doesn't attempt. If your mix is already reasonably clean and you want physiologically-precise, explainable, dose-capped mastering with patent-pending artifact detection — that's what LookyMaster is built for. A realistic workflow could use both: restore first, master second.

Try it yourself

SonicMaster's official live demo, embedded directly from Hugging Face.

This is AMAAI Lab's own demo, running on their own infrastructure — not rehosted, not modified. Upload a track and describe what you want fixed, or let it run automatically.

◉ NOTE

This demo runs on Hugging Face's servers, not LookyMaster's — audio you upload here goes to their infrastructure, not ours. This is a genuinely different privacy model from LookyMaster's own tool, which never uploads anything. We're linking to it transparently because it's real published research worth trying, not because it's part of LookyMaster.