News · ai music — By WATMM Desk —

AI mastering, honestly: what LANDR and Ozone do well, and where an engineer still wins

LANDR and Ozone's Master Assistant both learn from finished records and push your mix toward a reference and a loudness target. That makes them fast and cheap at a good mix's last 10%, but a 472-listener public blind test still put two human engineers on top. Here is how the assistants work, how to test them fairly, and what they cost.

Key facts

  • What this is: LANDR (online, app and plugin) and Ozone 12's Master Assistant both analyse your mix and build a processing chain aimed at a genre profile or a reference track you upload, with a loudness setting on top.
  • How: LANDR charges \$10 to master a single track, with unlimited free previews; Ozone 12 is listed at \$55 (Elements), \$219 (Standard) and \$499 (Advanced), or comes in iZotope's \$15 and \$25 a month plans.
  • Why it matters: in Benn Jordan's 472-listener double-blind test of one track, two human engineers took first and second, Ozone 11 on its own came seventh of seven, and LANDR was cut before the final. A level-matched blind test of your own is the only result that counts for your music.

AI mastering is now cheap enough that most independent releases will pass through it at some point. The two names you'll meet first are LANDR, which says it is trusted by over 5 million musicians, and iZotope's Ozone, whose Master Assistant iZotope says launched in 2017. Both are good at one job. Neither is a mastering engineer, and the best evidence we have says the gap is real. Here's how to use them honestly.

What do the assistants actually do?

Both start by listening to your mix and comparing it with finished records.

LANDR says its engine is trained on the adjustments human mastering engineers make and builds "a unique mastering chain for every track". You pick from styles that vary in intensity and loudness, or upload a reference: LANDR's reference page says you can add up to three reference tracks to steer the result. Album mastering applies the same settings and loudness standards across a batch, which helps an EP sound like one record.

Ozone 12's Master Assistant runs inside your DAW. iZotope's own guide says it analyses a section you choose (it recommends the loudest part, usually the chorus), guesses the genre, then matches your track's tone, width and dynamics to targets built from recent chart hits. The guide lists 35 targets, from an all-purpose default to vocal jazz. Upload your own reference instead and it matches that file's tonal balance, width, dynamics and loudness. The new Custom flow lets you set a LUFS target, switch modules off and set processing strength.

In plain terms, both are reference matchers with a loudness knob. They move your spectrum, stereo width and density toward a target, then limit to a level. That is useful, because it's most of what a quick master does. It is also why a poor reference or a poor mix gives you a confident, polished version of the wrong thing.

Where are they genuinely good?

  • Speed and iteration. LANDR offers unlimited free previews, so you can hear several styles before paying.
  • Not making things worse. Sound On Sound's Ozone 12 review found Master Assistant "very effective" on material "in decent health", and said it makes an inexperienced user less likely to take a mix backwards.
  • A starting point you can open up. Sound On Sound describes Master Assistant as suggesting a chain "to get you started" while leaving you in control, and in Standard and Advanced you can dip into each module to see what it did. MusicRadar's review called the streamlined workflow "a real boon".

Where does an engineer still win?

Start with the vendors' own words. iZotope's Master Assistant guide says outright that "nothing replaces a dedicated mastering engineer", pointing to spec compliance, metadata and sequencing. Sound On Sound adds that the assistant probably won't rescue a poor mix the way an experienced engineer might.

Benn Jordan's 472-listener blind test agrees. In 2024 producer Benn Jordan gave one mix of his 2017 track Starlight to AI tools and human engineers and had 472 people rate the randomised results double-blind, MusicTech reported. LANDR was disqualified before the final seven. Human engineers Max Hosinger and Ed the Soundman took first and second. Ozone 11 on its own came seventh, while Ozone used together with Neutron came fourth.

Read that carefully. It's one track, one listener panel, and older versions (Ozone 11, not 12). It doesn't prove engineers always win. What it does suggest is that, on that track, listeners valued the human engineers' judgement over every automated result in the final.

An engineer also hears problems a reference matcher can't fix: a vocal that's too quiet in the bridge, a harsh cymbal, a muddy low end that the mix should have dealt with. Often the most valuable thing they do is send the mix back with a note.

How do you blind-test a master yourself?

  1. Bounce a clean mix. Take the limiter off your master bus and leave a few dB of headroom. The same applies wherever you mix, whether that's Logic or an AI DAW with a built-in co-producer such as Veena, which exports WAV as well as stems.
  2. Make the candidates. Run the same mix through each tool you're considering. Keep the free previews, and pay only for the one you'll release.
  3. Match the loudness. This matters more than anything else. iZotope's guide warns that when something is louder, "our ears are fooled into thinking it sounds better". Use Ozone's Gain Match or LANDR's volume matching, or measure each file's integrated LUFS and turn the louder ones down to match before you compare.
  4. Hide the labels. Ask someone to rename the files A, B and C, and don't find out which is which until you've chosen.
  5. Listen to the same section on three systems: headphones, monitors and a phone speaker. You'll see differences in the low end and the harshness of the top that one system hides.
  6. Check against the platform. Spotify normalises playback to -14 dB LUFS and turns louder masters down, so extra loudness buys you nothing there. It advises keeping true peaks below -1 dBTP, or below -2 dBTP for masters louder than -14 LUFS. Our loudness normalisation explainer covers Apple and YouTube too.
  7. Pick, then reveal. If you can't reliably tell two masters apart, buy the cheaper one.

Checklist before you release an AI master:

  • Compared level-matched, not at face value
  • Chosen blind, on at least two playback systems
  • True peak at or below -1 dBTP
  • Exported as 24-bit WAV for distribution, not MP3
  • Listened to the full track, not only the chorus the assistant analysed

What does it cost?

These were the prices on the vendors' own pages when we checked on 23 October 2026.

Watch the format trap. For Essentials, LANDR's FAQ lists MP3 downloads and puts WAV on the higher tiers, so "unlimited" may not mean a file you'd want to distribute. Don't upload an MP3 to a distributor that accepts WAV.

What we think

AI mastering is a good finishing tool for a good mix, and an honest second opinion on a bad one. It isn't a replacement for an engineer, and the vendors' own materials say as much.

For singles and demos, Ozone Elements is the better deal. It's a one-off $55, it runs on your own machine, and you can re-master as often as you like. That suits people who release regularly and want to learn. LANDR's per-track price makes sense for someone releasing one song who wants no software at all.

The real risk is loudness, not tone. Both tools offer a loudness control, and it's tempting to push it. Because Spotify turns loud masters down, a master pushed for loudness gives up dynamics and gets nothing back. Set the assistant to a moderate target and judge it level-matched.

Spend on an engineer where judgement matters. That means an album that needs sequencing, a release with vinyl or label specs, or a mix you aren't sure about. In our view, the best use of an assistant is the step before that: it tells you cheaply whether your mix is ready to pay a human for.

Sources

  1. AI Mastering | Online Audio Mastering, LANDR (primary source). Accessed 23 Oct 2026.
  2. Online Reference Mastering Software, LANDR (primary source). Accessed 23 Oct 2026.
  3. Ozone 12 Advanced, iZotope (primary source). Accessed 23 Oct 2026.
  4. Ozone 12 Elements, iZotope (primary source). Accessed 23 Oct 2026.
  5. Ozone 12 Standard, iZotope (primary source). Accessed 23 Oct 2026.
  6. iZotope subscriptions, iZotope (primary source). Accessed 23 Oct 2026.
  7. How to use Master Assistant in Ozone 12, iZotope (primary source). Accessed 23 Oct 2026.
  8. Loudness normalization, Spotify for Artists (primary source). Accessed 23 Oct 2026.
  9. Veena home page, Veena Studio (primary source). Accessed 23 Oct 2026.
  10. iZotope Ozone 12 Advanced (review), Sound On Sound. Accessed 23 Oct 2026.
  11. iZotope Ozone 12 Advanced review, MusicRadar. Accessed 23 Oct 2026.
  12. AI mastering vs human engineers: Benn Jordan conducts massive study to find which reigns supreme, MusicTech. Accessed 23 Oct 2026.

Researched with AI assistance, written and checked by WATMM Desk.

Published by the company behind Veena. Disclosures

Subscribe to We are the music makers

One email a week: the music-tech and AI-in-music news that mattered, and what we think about it.

Sign-ups open soon.