What's New in Magic Master: July 2026

July 2026 at Magic Master brought seven features in the app itself plus work around it: accessibility, mobile layout, the Telegram bot. Everything is collected here in one place: what's new, how to use it, and what changed in the sound. The sound changed once — on July 19, when the LR4 crossovers went live; everything else is about convenience and control.

Multipreview: your track in 4 styles at once

The most common beginner question is "which style should I pick for my track?". It used to cost several full mastering runs to answer; now it's one button: "⚡ Compare 4 styles".

1. Upload a track and hit the button — a 24-second snippet from the middle goes into parallel mastering in four styles: EDM (dense club), Pop (commercial balance), R&B (warm mids) and Streaming Standard (−14 LUFS).
2. In ~20 seconds four style chips appear with their LUFS — previews switch instantly, from the same playback position.
3. Like a style — clicking its chip selects it in the genre grid; just hit "Start mastering".

Honesty built in. Styles are mastered to their genre loudness (EDM −9, Standard −14), but playback is matched to the quietest version — otherwise the loud one would always seem better. Same principle as our A/B player. Free, up to 6 multipreviews a day; for developers — POST /api/v2/multipreview with your own set of 2–6 styles (docs).

The intensity slider: density in one move

Every genre has a signature compression amount: EDM squeezes harder, classical barely touches the dynamics. The slider in the settings scales it from 0.5× to 1.5×:

• Left (0.5–0.85×) — softer: more air, living dynamics, natural transients. For acoustic, jazz, indie — or when the master feels over-squeezed.
• Middle (1.0×) — the genre's signature setting, as before.
• Right (1.2–1.5×) — denser and more aggressive: the wall of sound for EDM, hip-hop, rock.

Loudness doesn't change: the target LUFS is guaranteed by a separate final stage; the slider shapes character only. It works in single mastering and in batch processing, including album mode.

Genre suggestion after analysis

Hit "Measure loudness" and a chip like "💡 Sounds like EDM (strong bass, dense dynamics) [Apply]" appears above the genre grid. The app measures objective features of your track:

• five spectral zones — from sub-bass to "air";
• spectral centroid — the brightness of the sound;
• dynamics — crest factor and frame-level dynamic range over 100 ms windows;
• rhythmic density — spectral-flux peaks per second;
• a speech detector based on 3–8 Hz envelope modulation — podcasts are recognized reliably;
• stereo width.

Genre profiles vote on these features; the runner-up is shown too. There's deliberately no neural net here: the analysis takes milliseconds, the same track always gets the same suggestion, and every suggestion comes with its reasons. The genre is never applied automatically — the Apply button is yours. Via the API, POST /api/v2/analyze returns a genre_suggestion field; analysis is free.

Express AI-trace cleanup — without mastering

Two users asked for the same thing through the "What do you need?" form: AI-trace cleanup without full mastering. Now it exists.

Tracks from Suno and Udio carry an inaudible digital trace: high-frequency steganography above 15.5 kHz, artifacts at generation block boundaries and machine-regular envelope periodicity. Until now the cleanup only ran as a module inside mastering.

1. In the app, enable the "AI-trace cleanup" card and the "Clean only, no mastering" checkbox.
2. Run it — we remove the digital trace and protect the peaks (True Peak ≤ −1 dBTP), and touch nothing else: no loudness, no EQ, no dynamics.
3. The job status carries a report: HF-marker energy before and after cleanup, in dB.

In tests the markers are suppressed by 10+ dB, and loudness changes by less than 1 LU. It costs 1 token; via the API — POST /api/v2/clean. More on what gets removed and how is on the AI-trace cleanup page; try it at magicmaster.pro/app?aiclean=1.

Linkwitz-Riley crossovers: tighter bass

Multiband dynamics splits a track into bands, processes each and sums them back — for a transparent sum, the bands have to meet perfectly. An engine audit showed the old implementation (double Butterworth filtering) didn't produce a correct LR4 sum: energy was lost where bands meet, most of all in the bass. The processing now uses a classic 4th-order Linkwitz-Riley, the industry standard.

We didn't switch it on by ear: test tracks in three genres were rendered by the old (A) and new (B) implementation and compared against a reference spectrum curve of commercial releases (based on AES research):

New masters get a more accurate, tighter low end — especially EDM, hip-hop and bass music. Target LUFS and True Peak are guaranteed as before, and already-downloaded files don't change. If a track seems to sound worse — hit 👎 in the app: we investigate every such case on the actual file.

Stem mastering: vocals, bass and drums separately

A finished stereo mix limits mastering: EQ and compression hit everything at once. Stems remove that limit: vocal buried — +1.5 dB on the vocal only; bass booming — −3 dB on the bass only.

In the app: upload 2–8 stems in the batch section (magicmaster.pro/app, PRO token mastering) and tick "These are stems of one track" — we sum them and master the result as one track, with a genre preset, target LUFS and a track passport.

Via the API — with a gain per stem:

curl -X POST https://magicmaster.pro/api/v2/master/stems \
  -H "X-API-Key: mm_live_YOUR_KEY" \
  -F "files=@vocals.wav" -F "files=@bass.wav" -F "files=@drums.wav" \
  -F 'gains_db=[1.5, -3, 0]' \
  -F "style=hiphop" -F "target_lufs=-10"

Different sample rates and lengths are aligned automatically, and the sum is protected from overload. It costs 1 token — one master of one track. Export stems from the same start point (bar 1): we align length but never shift phase. No stems? You can get them by splitting the track.

Reliability: live progress and automatic token refunds

• We interrupted your job — you get the token back. Service updates require restarts, and a mastering running at that moment could previously vanish silently. Now on every startup the server finds interrupted jobs itself and refunds the owner. The first run refunded tokens for every job ever interrupted.
• Tokens only for results: any processing error or failed delivery to Telegram triggers an automatic refund.
• Telegram bot: live progress ("🎛 45% · Studio EQ"), /cancel to exit any dialog, an honest warning about Telegram's limit (bots download files up to 20 MB — for bigger files, WAV up to 800 MB, use the website) and limits identical to the website: Free — 1 basic master per UTC day.
• The assistant in the app and the bot knows the new features and current token prices, and if the LLM provider is unavailable it says so and gives the support address.

Accessibility: fixed interface contrast

In light mode, some badges and pills (module indicators, statuses, labels) rendered as light text on a light background — hard to read. We brought the colors in line with WCAG AA guidelines (contrast ratio ≥ 4.5:1). Dark mode was unaffected.

• labels for PRO modules (de-esser, transient designer, parallel compression, etc.) are readable in light mode;
• "NEW" badges, optimal/elevated loudness indicators and feedback buttons use dark text instead of washed-out gray.

Mobile layout: the site scales correctly on phones

• the profile page and tool landing pages (Suno, Udio, the LUFS analyzer, podcasts) got proper breakpoints down to 380 px screens;
• controls got tap targets of at least 44 px — comfortable for thumbs.

The mastering app itself (open it here) was already responsive — this round polished the secondary pages around it.

Easier to find — in search and via AI assistants

We improved how the site gets found, both in regular search and in AI assistant answers (ChatGPT, Perplexity and others) — they now get an llms.txt file with a short description of the service and its key pages. If you're choosing a tool, see Best AI Mastering Services 2026.

Telegram bot verified and working

We tested the bot end to end: commands, file mastering, track analysis (LUFS, True Peak, LRA) and the AI assistant all respond in chat. If the bot went quiet on you before, try again — command /start.

Ready to master something? Upload a track — your first master is free. For more on the metrics, see What Is LUFS and What Is True Peak.