Help & FAQ

The Refiner offers neural restoration, AI-artifact cleanup and mastering for music and voice recordings. Enhanced synthesizes plausible detail; the result can change texture and timbre, so compare it with your original. The typical track workflow is:

New verified accounts start with 5 welcome credits and 30 days of Pro feature access. After that, the account uses its normal plan.

1. If you already have an invitation, sign in at /sign-in. Otherwise, create an account for access.
2. Drop a track into the dashboard. MP3, WAV, or FLAC, up to 8 minutes long.
3. Pick Enhanced, the recommended default. Flow Matching remains visible but is temporarily unavailable while we review output quality. Optionally flip on AI artifact removal and Finalize to try the recommended complete workflow. These features are available during the welcome Pro trial and on Hobby, Pro and Studio afterward. Enhanced costs 2 credits; cleanup adds 1. Finalize adds 1 credit. Pro, Studio and active welcome trials can also enable Advanced mode to rebalance the cleaned stems.
4. Follow the stage shown on your track. With Finalize, choose optional Analogue Colour, then compare your preview with the original at final listening. Previewing does not use a download slot.
5. Choose Render & download for the versions you want to keep, then Confirm and close session. A simple refinement without a final-listening stage finishes directly in your results.

Need isolated stems instead of a refined mix? Head to the Stems page for vocals / drums / bass / other or speech / background splits. See the stem separation section below for details.

Albums is temporarily unavailable on all plans. Refine tracks individually while we update the album workflow. Bulk downloads of completed individual tracks remain available.

The Refiner ships two families of models. They take very different paths to the same destination, and their trade-offs are worth understanding before you spend credits.

Enhanced — neural audio reconstruction

Enhanced uses a trained audio model and GPU processing to synthesize high-frequency detail from your recording. It goes beyond an EQ lift or a noise filter: the model constructs a plausible restored version, aiming for more definition and air. It cannot know the exact detail that was lost, and its reconstruction can change texture and timbre.

Enhanced is our recommended starting point and costs 2 credits. Its processing path is shorter than Flow Matching, but it is still model inference, not a lightweight filter. End-to-end time varies with track length, queue demand, and worker startup. Compare the result with the original at matched listening levels; generated detail is not always preferable on every source.

Flow Matching — compute-intensive iterative reconstruction

Temporarily unavailable. Flow Matching is paused for new jobs while we review output quality. Its card remains visible but disabled in the app. Use Enhanced for now; existing results and downloads are unaffected.

Flow Matching uses a different generative restoration approach, guided by your source audio. It rebuilds detail through repeated refinement rather than a simple filtering pass. This is intensive GPU computation, and it can reshape fine texture, transients and ambience—not just reduce noise.

Normal and High use different processing depths, with High requiring more compute. That produces an alternative reconstruction, not a guaranteed quality upgrade or an exact recovery of the original master. Before the pause, Normal cost 3 credits and High cost 4. Enabling AI artifact removal added a cleaning stage first. No preset guarantees any particular AI-detector result; processing can change scores, sometimes much lower.

AI Artifacts — standalone cleaning

The AI-artifact cleaner running on its own, with no refinement model after it. It uses the same whole-mix cleanup and targeted stem cleanup stages as Enhanced with AI artifact removal, while skipping Enhanced restoration. Compare with the original: cleanup can change texture as well as reduce haze and noise. It costs 2 credits and adds a cleaning stage that can take roughly the track's length, plus queue and worker startup time. It is available on Hobby, Pro, Studio and active welcome trials.

Vocal De-mush & Vocal Restore — vocal and voice cleanup

Vocal De-mush (1 credit) cleans AI-generated vocal stems and reduces background noise in podcasts and spoken-word recordings. Upload a voice recording directly; for a full song, separate the vocal on the Stems page first. Vocal Restore (2 credits) adds an Enhanced restoration pass after vocal cleanup to rebuild detail in singing vocal stems. Both are available on Pro and Studio.

Vocal De-mush runs AI artifact cleanup first, then blends in vocal restoration. The Restoration strength fader starts at 25%; lower it to retain more soft word endings and breaths, or raise it for more restoration. At 0%, you hear the cleaned vocal alone. More of the original noise may remain at lower amounts. The result is a centered mono voice, even when the uploaded file has two channels.

Finalize — mastering with final listening

Finalize is a mastering stage, not a restoration model. Enable it with Enhanced on Hobby, Pro, Studio or an active welcome trial for 1 additional credit, or select standalone Finalize for an already-mixed track (1 credit). Standalone Finalize starts from your uploaded mix and takes you through optional Analogue Colour and final listening.

Choose Gentle, Medium or Strong mastering and a loudness target: −9 LUFS by default, or −7, −11, −14 or −16. You can also choose a reference, Glue Compressor, Resonance Mix, Stereo Air and Stereo Width. Colour provides the optional hardware character; there is no separate Harmonics Off/Standard/Hot selector in this workflow. Check the measured LUFS of each render: the requested target may not be reached.

The default is a listening preview. It does not automatically create a download or use one of your three version slots. See final listening and downloads below for how to preview, render and confirm your chosen result.

Output format

All presets output WAV. Every new setup starts at 44.1 kHz, including Pro and Studio. Choose 48 kHz explicitly on those plans when needed; a previous 48 kHz choice is not carried into the next setup. If you cancel the confirmation or need to sign in again before starting, the next setup returns to 44.1 kHz. Your choice stays with the current track or confirmed batch, and existing saved 48 kHz versions keep their recorded rate. Bit depth is 24-bit. A 32-bit float option normally belongs to Flow Matching on Pro and Studio. That option is temporarily unavailable with the paused preset.

Which to pick

Default to Enhanced for most material — it is fast, inexpensive, and our recommended starting point. Flow Matching is paused while we review output quality, so it cannot be selected for new work. Finalize can master an Enhanced result — that is the difference between a clean file and a release-loud one.

Advanced mode is available on Pro, Studio and active welcome trials for Enhanced with AI artifact removal. It adds 1 credit. Select Finalize before upload if you also want mastering (+1 credit), making the full Enhanced workflow 5 credits. Advanced mode starts off for each new submission: enable AI artifact removal, then Advanced mode before upload. When enabled, the job pauses at the stem mixer before Analogue Colour and final listening. With Advanced off, you skip the stem mixer and its vocal controls; Finalize still includes optional Colour and final listening.

1. Use Preview mix to listen to bass, drums, other and vocals together, and adjust their level faders independently. Output protection helps manage clipping. The preview uses encoded listening files; the final mix uses the lossless working stems.
2. Choose Original vocal stem (the default) to keep its stereo ambience, Process full vocal stem to reduce reverb on the whole vocal, or Separate lead and backing for a centered mono lead with independent backing-vocal level.
3. Vocal noise cleanup is experimental and off by default. It treats the selected vocal before new reverb; in Separate lead and backing mode, only the lead is affected. Compare both ways: it can change vocal texture.
4. Reverb works with all three vocal treatments and starts at zero. Adjust its amount and style in real time. Low cut and High cut shape the wet return; Listen to reverb only isolates that return.
5. Vocal harmonizer adds a wide stereo double of the vocal beside the reverb, using the classic studio harmonizer technique: two slightly detuned copies panned hard left and right. The dry vocal is unchanged. De-esser softens harsh “s” sounds across the whole vocal, including its reverb and harmonizer. It finds each “s” and turns down only the part of the treble it occupies, so there is no voice setting to choose. Mixes saved with the earlier de-esser keep their Low voice or High voice choice.
6. On the drums, Saturation adds Tape punch (odd harmonics) or Tube warmth (even harmonics) below 5 kHz, so cymbals stay clean. On the bass, Harmonics adds upper harmonics so the bass is heard on phones and small speakers.
7. Each effect has one amount control that starts at 0% (off). You hear changes live without playback restarting, and the saved mix uses the same processing. Reset returns every effect to off. A job prepared before the harmonizer was added may not offer it. The De-esser, Saturation and Harmonics don't depend on when the job was prepared.
8. Choose Continue to Analogue Colour to accept the mix and move on. Closing the mixer leaves the job paused; reopen it from the track on your dashboard.

Activity notices stay visible while stems are loading, buffering or the mix is being saved. Loading your stems… remains while the audio loads, even after the controls appear. If Synchronizing all stems… appears, the player is waiting until the stems can play together. Use Cancel loading to stop waiting.

Advanced mode does not provide individual stem downloads. Its private working files remain while you choose your result and are cleared after you confirm and close the session. For separate WAV stems, use the Stems page.

Read the Advanced mode walkthrough.

Finalize and Advanced workflows offer Analogue Colour before final listening. Choose a curated hardware capture, such as SSL Fusion/SSL Bus, to shape the accepted mix. You can also continue without colour. Colour processes the track from the previous stage, before mastering.

Level matching starts off and manual Volume starts at 0 dB. Raise Input Drive to hear the selected device's character, use Volume to compensate the level, then adjust Delta amount for the amount of colour you want. Recheck Volume as you change Drive or Delta.

Automatic Level matching is an alternative when you want the app to compensate the level. It bypasses the manual Volume control while on. During the session, manual and matched modes each remember their own Drive and Delta settings. The first matched audition starts with your current settings; switching matching off restores your manual Drive, Delta and Volume. Listening level is separate: it changes your playback volume and is not saved in the colour render.

On first use, Preparing level matching… appears while the app measures the audio. If you are already listening, playback continues, and Play and Pause remain available. If you pause during preparation, playback stays paused when matching is ready.

Activity notices also show when Colour controls are loading, colour audio is being saved, or final listening is being prepared. Colour error messages remain visible as you scroll, even after loading stops. Follow the message's instructions to retry. Retrying source preparation… shows when the app is preparing the Colour source again.

Volume compensates the device output before Delta is calculated. Delta amount adds that change to the dry signal: 0% keeps the source unchanged, 100% gives the adjusted device output, and 200% doubles its difference from the dry track. Delta audition lets you hear the change on its own. Compare with the original at a similar listening level before continuing.

Compare your preview with the original upload using the A/B controls. The shared signal meter follows the active audio before your listening volume. You can adjust the following controls before rendering:

• Reference: choose a factory preset, a saved reference or a reference file here when mastering is selected.
• Glue Compressor: inspired by classic British console bus compressors. Adjust the threshold in dBFS; 0 leaves signal below full scale uncompressed. Fixed 2:1 ratio, 10 ms attack and Auto release. Link makeup estimates a fixed gain when you adjust Threshold; unlink it to set Makeup independently. Glue Compressor off bypasses both compression and makeup. The meter shows gain reduction.
• Resonance Mix: blend from the untreated mix at 0% to resonance reduction at 100%. Instrument Humanizer is available earlier in Analogue Colour.
• Stereo Air: an optional high-frequency lift in the stereo sides, off by default.
• Stereo Width: narrow the whole track toward mono or widen it beyond the original stereo width. 100% keeps its normal width. This setting is saved in the render.
• Mono check: temporarily hear the entire preview in mono. It is a listening aid and does not change the exported file. Watch phase correlation and listen for sounds disappearing.
• Output: every new setup starts at 44.1 kHz. Pro and Studio can explicitly select 48 kHz here for a version. Reopening a saved version keeps its recorded rate.

Listening controls prepare when you open the preview. Once ready, Glue Compressor and Stereo Width adjust during playback without loading another preview. The original loads separately for A/B comparison; you can listen to the processed preview while it loads.

Original loading is shown beside the A/B controls. An activity notice identifies playback starting, buffering, preparation, saving and rendering. Buffering activity remains visible until playback recovers or you pause, including after moving the playhead. The notice stays visible as you scroll through the final-listening controls. Errors while adding a reference or preparing a result remain visible after loading stops. If a reference cannot be added, its error stays visible until you retry, choose a saved reference or turn Reference off. If Original cannot load, you can keep using the processed preview. Playback and loading errors clear as the affected operation recovers; unrelated errors still need attention.

Arriving here prepares a listening preview with the defaults. It does not spend a download slot. Use Render & download when you want to keep the chosen settings. You can download up to three distinct mastered versions at no additional credit cost. Downloading the same version again is free and uses no extra slot; the unmastered comparison also uses no slot.

Confirm and close session keeps your chosen main result and downloaded versions for 7 days and clears other previews and temporary working stems. Confirmation alone does not use a slot. Read the warning if you have unrendered changes: render them first if you want to keep that sound.

Closing the window does not confirm the export. Generated previews are saved automatically; use Open final listening on the track to return. Unfinished sessions stay on the dashboard until you open them; they do not pop up automatically. Unrendered control changes reset on close. A render that has already started continues.

With mastering selected, choose the reference at final listening. Without mastering, enable Reference before submission to open the earlier reference panel. Its free 10-second audition applies only the reference's tonal shape to your uploaded excerpt; it does not preview the full refinement chain. With Reference Off, it plays the original excerpt.

If you continue with a reference and no mastering, tonal matching is applied to the refined output without a mastering loudness target. Reference matching shapes tone; it does not copy another track's arrangement, performance or mastering settings.

New album work is paused on all plans while its workflow is updated for the revised track pipeline. Please refine tracks individually. See final listening and downloads for the current mastering workflow. Bulk downloads of completed individual tracks remain available.

The Refiner can also split a track into isolated stems — useful for remixing, dropping the instrumental into a DAW, building a karaoke version, isolating a vocal to use elsewhere, or pulling speech out of a podcast for transcription. Stem separation is a Pro feature; Free users get one lifetime trial split.

Open the Stems page from the dashboard, drop a track, and pick one of three engines. Output is one WAV per stem plus a single ZIP containing all of them — handy for one-click download.

Demucs — fast 4-stem classic

The original go-to source-separation model. Splits the track into vocals, drums, bass, and “other” (everything else). Fast, well-tested, and the right default for most music.

Melband — newer transformer

Mel-Band RoFormer is a more recent architecture that produces cleaner separations — less bleed between stems and tighter isolation on vocals and percussion in particular. The trade-off is runtime: it takes longer than Demucs. Reach for it when stem quality matters more than speed (final remix, sampling, lifting a vocal you actually plan to use prominently).

Voice Cleanup — speech vs. background

A 2-stem split optimized for spoken-word audio: speech on one stem, everything else (music, room noise, applause) on the other. Useful for podcast editing, lifting an interview voice out of a noisy recording, or preparing speech-only audio for transcription.

Cost and limits

Each split costs 3 credits regardless of which engine you pick. Track length limit is the same as refinement (8 minutes). After the split completes you can preview every stem in the in-app multitrack player before downloading, and the convenience ZIP stays available alongside the individual WAVs.

Refinements are paid for with credits. New verified accounts receive 5 welcome credits and 30 days of Pro feature access. After that, Free provides 5 credits per month for Enhanced. Paid subscriptions (Hobby, Pro, Studio) add monthly credits and plan features.

Monthly subscription credits reset at the end of each billing cycle and do not roll over. Credit-pack credits do not expire and stack on top of subscription credits. A Credit Pack adds balance only: it does not change the account's plan, extend the welcome trial, or unlock features outside that plan.

If a refinement comes out worse than the source, email support@therefiner.app with the job ID (visible in the dashboard). We'll review the result and refund the credits when appropriate.

Pricing details and plan comparisons are on the pricing page.

Manage your plan, payment method, and credit balance from Account → Billing. Cancel any time — your subscription remains active until the end of the current billing period.

Source files and refined outputs stay in your private workspace for 7 days after refinement so you can preview or re-download them. After 7 days the files are deleted, while the job record remains in your history. Temporary working stems and unselected previews are cleared after you confirm and close the session. Closing final listening does not confirm it. We never use your audio to train models.

To remove a finished result sooner, choose Delete beside it on the dashboard and confirm. This removes its job record, uploaded audio and reference files, processed versions, previews and any stem splits created from it. Separate reruns keep their own files. Active or paused processing must finish first. Deletion cannot be undone and does not refund credits; your credit history is retained. If file cleanup is interrupted, use Retry delete to finish.

Upload fails: check the file is in a supported format (MP3, WAV, or FLAC) and under 8 minutes long. Tracks longer than 8 minutes are rejected at upload — split them and refine in chunks.

Job appears stuck: refresh the dashboard once to load the latest server state. If the status still does not change, email support with the job ID. Do not start repeated copies of the same job; failed jobs automatically return their reserved credits.

Result sounds wrong: start with Enhanced and compare one setting at a time. Compare with Instrument Humanizer off in Analogue Colour, then audition the mastering controls one at a time. Try Reference Off before matching another track. Extreme inputs such as very short clips or unusual instruments can still vary. Email support with the job ID if the result remains off — we read every report.

Email support@therefiner.app with the job ID (if relevant) and a short description. We typically respond within one business day.