If you've generated music with Suno, Udio, Stable Audio, or MusicGen, you've probably noticed it: a subtle metallic sheen, a watery swirl in the high mids, or a kind of hollow compression that makes the track feel synthetic even at first listen. These are AI music artifacts — and they come from the neural audio codec at the heart of every modern music generator.
This guide explains what they are, why they happen, and the most effective ways to remove them.
Same passage, level matched: Original → Gentle Reduction. Listen for texture and the musical detail you want to keep.
Recorded with de-artifact AU 0.4.4. Offline processing replay; this clip does not demonstrate live latency.
Watch this demo on YouTube · Watch the full walkthrough on YouTube
Explore demos and updates on the Intrect YouTube channel
In your browser. No account or installation.
Listening preview, not a full-track export.
Every preset and parameter. No card required.
macOS or Windows; a compatible DAW is required.
Pro Tools: AAX on Apple Silicon Macs, stereo inserts only. Product details and purchase options.
Nicerizer combines artifact reduction with positive Brightness: the 0.4 bandwidth-extension stage generates an estimated 8–22 kHz band from the cleaned signal. It is an optional way to add air, not a recovery of the lost original. Results depend on your source; compare with Bypass and check Residual for musical detail being removed.
9:50 — What Nicerizer adds · 10:15 — Hear the high-band comparison
Walkthrough recorded with 0.4.2; these presets ship from 0.4.3. The free web preview and the DAW trial are separate options; use the plug-in trial to adjust Brightness and try Nicerizer.
Same song, same moment. Original is the generator’s output, Cleaned is the same audio after de-artifact, and Diff plays what was removed, which may include musical detail as well as artifacts. Headphones help.
What to listen for: the metallic shimmer on cymbals and sibilants, and the watery swirl on sustained notes, in Original — then how much of it lives in Diff. If Diff sounds like music rather than hiss, the preset is too strong for that track.
Hear the A/B first, then follow the real upload, processing and comparison steps. The no-account preview is up to 30 seconds; the 14-day DAW plug-in trial is a separate option.
Hear a free 30-second preview on your own track · Watch the full A/B comparison
AI music artifacts are the residue a neural audio codec leaves behind when it decodes a generated track: a metallic sheen and watery swirl clustered above roughly 3 kHz, plus ringing around transients and sustained tones. EQ and compression can mask them but cannot remove them, because the damage is spread across the spectrum rather than sitting in one band. Removing it takes a model that estimates, per frequency bin, how much of the energy there is artifact rather than music, and subtracts that estimate. That is what de-artifact does — in real time inside a DAW, or in a browser at try.intrect.io with nothing to install.
| It targets | Codec and generation residue — RVQ ghosting from Suno, Udio, Stable Audio, MusicGen and Riffusion; the same artifact class in low-bitrate MP3, AAC and Opus; and the residue left on stems by a separator such as htdemucs or Spleeter. |
| It does not target | Bleed between stems. A vocal audible in the drum stem is content that does not belong to that track at all, which is a different problem — de-leak is the tool for it. The two are frequently used on the same source. |
| It is not | A guarantee of transparent results on every source. Reduction can remove musical detail, especially at strong settings. Start gently and compare with Bypass; if Residual contains melody or drum attack, back off. Optional positive Brightness generates an estimated high band rather than restoring the lost original. |
| It needs | A VST3, CLAP or Audio Unit host on macOS 11+, or VST3/CLAP on Windows 10 2004 or later with an AVX2 + FMA CPU (Intel Haswell 2013+ / AMD Ryzen 2017+) and the Microsoft Visual C++ 2015–2022 x64 runtime. Pro Tools loads it only through the AAX build, which is Apple Silicon only (stereo inserts); there is no Intel or Windows AAX, and no Linux build. |
Modern AI music generators — Suno, Udio, Stable Audio, MusicGen, Riffusion — don't generate audio directly. They generate a sequence of discrete tokens using a language model, then decode those tokens into audio using a neural audio codec such as EnCodec or DAC.
These codecs use a technique called Residual Vector Quantization (RVQ) to compress audio into tokens. RVQ introduces quantization residuals — errors that survive into the decoded audio as repeatable, structured artifacts. They're not random noise: they have a specific spectral signature that trained ears (and trained models) can identify.
| Artifact type | What it sounds like | Root cause |
|---|---|---|
| RVQ ghosting | Metallic shimmer, artificial brightness, "digital" sheen | High-frequency codec residuals from RVQ quantization error |
| Codec residue | Watery, swirly modulation on sustained notes and pads | Phase inconsistencies in the codec's overlap-add reconstruction |
| HF aliasing | Harsh, grainy texture above 8–12 kHz | Bandwidth limitation artifacts from the codec's learned spectral model |
| Hollow compression | Mid-range sounds thin, lacks body and presence | Token-domain dynamics that don't translate naturally to waveform dynamics |
Key point: AI music artifacts are not mastering problems. EQ and compression can mask them temporarily, but they're baked into the audio at the codec level. Effective removal requires understanding the artifact's spectral structure.
de-artifact is a VST3 / AU / CLAP plug-in powered by ArtifactNet, a forensic neural network trained specifically to isolate and subtract RVQ codec residuals from AI-generated music. It's designed to sit on the master bus (2-bus) and process the full mix in real time.
Working specifically with Suno? The Suno artifact guide goes deeper — including why the free plan's MP3 export adds a second layer of damage, and which preset to use depending on your export format.
If one passage is broken rather than the whole track — a garbled word, a smeared fill — try Suno's own editor first: Replace Section regenerates just that part, Get Stems lets you treat each part separately, and Remaster renders a variation aimed at sound quality. Each is a new generation from the same model, so it fixes a moment, not the metallic texture across the track. The guide compares the three in Start inside Suno.
Drag the MP3 or WAV file onto a new audio track. Route it to your master bus. Any VST3-, CLAP- or AU-capable DAW on macOS or Windows works — Logic Pro, Ableton, Reaper, Cubase, Studio One, Bitwig, FL Studio. Pro Tools needs AAX, which de-artifact ships for Apple Silicon Macs only (stereo inserts).
Place de-artifact after any existing master bus processing. If your chain is: compressor → limiter → de-artifact, that's correct. de-artifact should see the final, loudness-normalized signal.
Start with Default — it is the tuned starting point for AI-generated material. If a metallic sheen survives, step up to Aggressive. If the problem is watery sustains or smeared vocals rather than brightness, use More Harmonics instead of raising strength.
Strength controls how much the model subtracts, from 0–200% with a 60% default. 100% is the model's nominal estimate; above that it deliberately over-subtracts for stubborn material. Push it up until the residual starts containing music, then back off.
Engage Residual to solo the subtracted artifact signal. You should hear spectral noise and shimmer — not musical content. If you hear vocal body, drum attack or bass fundamentals, lower Strength.
In your browser. No account or installation.
Listening preview, not a full-track export.
Every preset and parameter. No card required.
macOS or Windows; a compatible DAW is required.
Pro Tools: AAX on Apple Silicon Macs, stereo inserts only. Product details and purchase options.
If you don't want a dedicated plug-in, you can reduce (not eliminate) AI music artifacts using standard DAW tools. These approaches mask the artifact rather than removing its source.
RVQ ghosting concentrates between 6–14 kHz. A dynamic EQ set to reduce gain by 2–4 dB when energy exceeds a threshold in that range can smooth the metallic sheen without dulling the top end. This works best on tracks where the artifact is mild. If the problem is instead one resonance that stays at the same pitch (a boxy vocal note, a whistling cymbal), that is not codec residue: a resonance suppressor such as ENZYME follows it more transparently than a fixed band.
Codec residuals tend to be wider in the stereo field than the underlying music. Applying mild saturation (0.5–1 dB drive) to the side channel only can add harmonic density that masks the artificiality. Use a mid-side EQ to roll off above 12 kHz on the side channel afterward.
Manual spectral repair changes the tonal character of the track. You're trading one kind of artificiality (RVQ shimmer) for another (over-processed top end). For production-ready results — distribution, sync licensing, client delivery — a model-based approach like de-artifact gives cleaner output because it subtracts the artifact specifically, not the frequency band it inhabits.
Pick by the problem you hear, not by the name. The full comparison is in Remove AI artifacts: plugins and tools compared.
| Generator | Start with | Notes |
|---|---|---|
| Suno full mixes | Default, then Aggressive | Start gently. For current Suno V6 material, also compare the Suno V6 factory preset; it is stronger, so check Residual. For cleanup with added air, audition Nicerizer. See the Suno guide for the MP3-vs-WAV distinction |
| Udio | Default, then More Harmonics | Udio artifacts carry more harmonic content; More Harmonics targets watery sustains without raising overall strength |
| Stable Audio | More Harmonics | Stable Audio's codec produces a different residual profile; More Harmonics preserves more top end |
| MusicGen / Riffusion | Conservative | These generators have a heavier codec imprint; the Conservative preset avoids over-subtraction |
| Separated vocal stem (any source) | Vocal | Tuned for narrowband signals; avoids removing sibilance alongside the artifact |
If you're working with stems separated by Demucs, htdemucs, or Spleeter — rather than full AI-generated mixes — you may be dealing with a different type of artifact: stem bleed. This is when audio from one stem (e.g., drums) leaks into another (e.g., vocals) because the separator's ML model didn't fully isolate the sources.
Stem bleed is not the same as RVQ codec residuals, and de-artifact is not designed to fix it. de-leak-rt is a separate plug-in built specifically for this: a real-time VST3 / CLAP leak gate with an ML classifier (F1 = 0.993) that detects inter-stem bleeding per-frame and applies multiband gain reduction only where leakage is detected. It works on vocal stems, instrument stems, and any Demucs / htdemucs / Spleeter output.
Yes — and it works by analyzing the same codec artifacts described above. ArtifactNet, the forensic model powering de-artifact, reported F1 = 0.9829 on the 2,263-track unseen test of ArtifactBench (v9.4, detector paper) — a historical result on that one partition. Other ArtifactBench partitions and protocols report different figures; the ArtifactNet page gives the protocol behind each figure it reports. It identifies AI-generated music by its residual fingerprint, not by generator-specific patterns — which means it generalizes to generators it hasn't seen during training.
The free online demo lets you upload any track and get a forensic verdict — AI-generated or human — along with a breakdown of the detected artifact signature. No account required.
de-artifact reports its latency to the host for PDC (Plugin Delay Compensation) — about 300 ms at 44.1 kHz (a little more at higher sample rates or with a larger host buffer). During a DAW freeze or offline bounce, some hosts signal "offline mode" mid-stream, causing the plug-in to re-prime its processing buffer. The output is correct from the first audible frame onward. This is a one-time priming artifact at the very start of the bounce, not a gap during playback.
Raise Brightness. Below 0, it reduces automatic high-frequency recovery; above 0, it blends in an estimated 8–22 kHz band generated from the cleaned signal. Nicerizer combines this bandwidth extension with firm reduction. The generated band is not the lost original restored. If cleanup is taking too much musical detail, lower Strength from its 60% default or try Gentle Reduction, then compare with Bypass.
Yes — the Conservative preset is specifically tuned for lossy-encoded sources where MP3 compression artifacts are layered on top of the codec residual. It applies lighter subtraction to avoid reinforcing the MP3 ringing.
de-artifact is trained to target the specific spectral signature of RVQ codec residuals. On real music without AI codec artifacts, Strength below 30% has minimal effect. Above 50% on a clean real-music source you may hear subtle tonal thinning — keep it below 20% if you're processing mixed sources.
de-artifact is built for music: AI-generated tracks, separated stems and lossy-codec audio. That is where it is tuned and measured. Speech, podcasts and AI voice clones were not part of its evaluation, so check a sample with the free online preview before relying on it there.
This page covers the general case. Each guide below takes one source and goes further into what that source actually does to the audio.
A ringing resonance or a boxy note is a different problem with a different tool — see removing room resonance from vocals and killing snare ring.
Compare the original and a free 30-second de-artifact preview. No installation or account needed. Keep the version that sounds right for your mix.