Refine Suno tracks. Studio audio quality.

How it works

Built for AI-generated music, end to end

AI tracks have a specific shape of damage: high-frequency rolloff, compression haze, and the spectral artifacts the neural decoder bakes in. Generic audio restorers were not built for this. We were.

Pro feature · 3 credits per split, any engine · Free users get one lifetime trial split to try it.

Pricing

Start with 5 credits and 30 days of Pro features. $9/mo when you need more.

Enhanced costs 2 credits. AI artifact removal, mastering and Advanced mode each add 1: 5 credits for the full workflow.

What is The Refiner?

+

The Refiner is a web app for refining audio, from Suno and Udio tracks to demos, podcasts and older recordings. Enhanced uses neural restoration to synthesize plausible high-frequency detail; texture and timbre can change. You can add AI-artifact cleanup and, with Finalize, audition Analogue Colour and mastering before choosing what to download. Compare with your original to decide what works for your track.

What's the difference between Enhanced and Flow Matching?

+

Both use trained audio models, not just filters. Enhanced uses GPU-powered neural restoration to synthesize high-frequency detail and is our recommended default. Flow Matching uses compute-intensive, iterative reconstruction, with Normal and High processing depths. Both generate plausible detail rather than recover an original master, so texture and timbre can change; more compute does not guarantee a better result. Flow Matching is temporarily unavailable while we review output quality, so use Enhanced for new refinements.

Can The Refiner master my track?

+

Finalize is the mastering stage, available on Hobby, Pro, Studio and active welcome trials. Enable it with Enhanced for 1 additional credit, or choose standalone Finalize for an already-mixed track (1 credit). Choose optional Analogue Colour, then use final listening to compare with your original and set your reference, strength, loudness, Glue Compressor, Resonance Mix, Stereo Air and Stereo Width. The default listening preview does not use a download slot. Choose Render & download to keep a version: up to three distinct mastered versions, with free repeat downloads and no additional credits. Confirm and close session keeps your main result and downloaded versions; confirmation alone uses no slot. The loudness target starts at −9 LUFS, with −7, −11, −14 and −16 options. Check the measured result: a requested target is not guaranteed.

What audio formats can I upload?

+

MP3, WAV, or FLAC. Tracks can be up to 8 minutes long. Current outputs are 44.1 or 48 kHz WAV at 24-bit. The Flow Matching-only 32-bit float option is temporarily unavailable with that preset.

How long does a refinement take?

+

Timing depends on track length, selected stages, queue demand and worker startup. Enhanced is the shortest path; cleanup adds another processing stage. A complete refinement can take several minutes. Advanced mode, Analogue Colour and final listening also pause for your choices; a track waiting for you is not still rendering. The dashboard shows the current stage. Once an upload has finished and processing has started, work continues if you leave the page.

Do I need a subscription?

+

No. New verified accounts get 5 welcome credits and 30 days of Pro feature access, then the account uses its chosen plan. Free provides 5 credits each month for Enhanced. Hobby, Pro, and Studio add monthly credits and plan features. A one-time Credit Pack adds credits but does not change your plan or unlock features outside it.

Can I split a track into stems?

+

Yes — stem separation is a Pro feature. Drop a track on the Stems page and pick an engine: Demucs (fast 4-stem classic), Melband (newer transformer, cleaner separation), or Voice Cleanup (speech vs. background, great for podcasts and interviews). Output is one WAV per stem plus a ZIP of all of them. 3 credits per split. Free users get one lifetime trial split to try it.

What is Rebalance stems?

+

Rebalance is the stem mixer in Advanced mode, available on Pro and Studio for Enhanced with AI artifact removal. It adds 1 credit. Adjust bass, drums, other and vocals. Choose Original vocal stem to keep its ambience, Process full vocal stem to reduce reverb on the whole vocal, or Separate lead and backing for independent backing-vocal level. Reverb works with all three and starts at zero. Four effects, each starting at 0% and heard live, shape the mix: Vocal harmonizer adds a wide stereo double beside the reverb; De-esser softens harsh "s" sounds across the vocal, finding each "s" and turning down only the part of the treble it occupies; Saturation adds Tape or Tube character to the drums; and Harmonics helps the bass come through on phones and small speakers. Experimental vocal noise cleanup is optional and off by default. Continue to Analogue Colour, then final listening. Advanced mode starts off for each new submission: enable AI artifact removal and Advanced mode before upload to get the mixer. With Advanced off, Finalize still includes optional Colour and final listening. Active welcome trials also include Advanced mode. Its working stems are private, not separate stem downloads, and remain available while you choose your mix; they are cleared after you confirm and close the session.

Can The Refiner detect AI-generated music?

+

Yes — open Detect AI from the main navigation to check a track separately from refinement, free and without spending a credit. It detects both Suno and Udio and tells you which one, and reliably separates their unprocessed output from human recordings; scores on refined or mastered tracks are less reliable. Support for more generators is on the way. It's a guidance tool, not a verdict — a low score means no AI signals were detected, not proof of human authorship, and no detector is 100% accurate, so please don't use the scores to make accusations.

Will refining change AI-detector scores?

+

It can. Cleanup, resynthesis and mastering all change the audio that AI-content detectors analyze, so a refined track can score very differently from its source, sometimes much lower. No detector result is guaranteed in either direction. Treat scores on processed audio as unreliable evidence of how the track was made: a lower score doesn't change how it was made, and AI disclosure is still your job.