honenuki骨抜き

AI-GENERATED MUSIC ARTIFACT REMOVAL / STEM BY STEM

The music sounds fine. So why does it catch in your ear?

Start cleaning your stems
Press to hear the difference
SCROLL

What you hear: sibilant ring. What we see: a repeating grid.

GRID MATCH92.4%
ARTIFACT LINES17
MUSIC CONTENTPRESERVED

EXAMPLE ANALYSIS — sample figures

We don't shave every high note. The grain of the music stays. Only the repeating bones come out.

01Read all stems together.
02Confirm the repeating grid.
03Attenuate only the lines that need it.

When the verdict isn't conclusive, honenuki doesn't touch a thing.

Hear what we removed. And everything we kept.

CLEAN — bones removed

A demo of the states. Nothing auto-plays.

Even after the pull, the music should still be itself.

No broad EQ wash.
No full-band denoise.
STEM-level, surgical processing.

STUDIO

Analysis comes first. Processing only where a stem needs it.

midnight_drive_v3 / 4 STEMS48kHz · WAV · EXAMPLE SESSION
VOCALS — A/B COMPARE
REMOVED SIGNAL
ARTIFACT SEVERITY / STEM
Vocals17 LINES → attenuate
Drums6 LINES → attenuate
BassInconclusive — left as is
Other11 LINES → attenuate

Stems without a conclusive verdict are not processed.

PROCESSING — VOCALS
REMOVAL STRENGTH62
BRIGHTNESS RESTORE30
ARTIFACT LINES17
Start with your own files

CM SONG

The honenuki CM song.

0:00 / 1:37
Track
Compare

REMOVED is the removed component on its own, amplified +12 dB so it can be heard at all — at its natural level it sits about 46 dB below the mix.

Our official CM song was written with AI, then cleaned with honenuki. Press play and switch between the original, the processed result, and the residual — the part honenuki took out. Switching is instant and keeps its place, and the three master files share one gain, so the level difference you hear is the real one.

Keep the music. Pull only the bones.

Start honenuki
WAV · MP3 · FLACAnalyze first. Process only when needed.

FAQ

Frequently asked questions

How does honenuki detect AI-generated music?

The neural codecs inside AI music generators leave a measurable fingerprint: metallic spectral peaks locked to a 200 Hz grid. honenuki analyzes every stem, measures how much of its energy snaps to that grid, and returns a per-upload AI verdict before touching a single sample.

Which AI music generators does it work with?

honenuki is tuned on Suno stems first. The artifacts come from the codec family rather than any one product, so stems from other generators that use similar neural codecs clean up the same way — if the grid fingerprint is there, honenuki can pull the bones.

Is human-recorded music safe to run through it?

Yes. The AI verdict acts as a gate: if analysis finds no codec-grid fingerprint, honenuki does not process the audio. Your live takes and human mixes pass through untouched.

Will removing artifacts dull my mix?

No. Removal is surgical — narrow cuts placed exactly on the grid peaks, scaled by a strength you control — and adaptive restoration rebuilds high-frequency detail afterwards. You lose the metallic sound, not the brightness.

What file formats are supported?

Upload stems as WAV, MP3, or FLAC. Cleaned stems are rendered as WAV and delivered together in a single ZIP.

How much does it cost?

honenuki is free to use while in beta — no signup required. Paid plans for heavier workloads are planned, and pricing will be announced before anything changes.

honenuki is an AI music artifact removal tool for producers: upload your stems, remove the metallic sound that AI music generators bake into the audio, and download clean, release-ready files.