How I test AI music generators
Ten tests, the same for every generator, each graded from 1 to 5 on a fixed scale. The overall grade is their average plus my verdict, which can move it by up to one point. Every test is run by hand on an account I use, from one fixed set of prompts written before any tool is opened, and the prompts, the generated tracks, the counts, screenshots, the saved audio and the timings behind each grade go in the full test results published with the review.
Tested and written by Miles Foster, Owner and tester, RankDude.

Each generator gets the same brief, built before any tool is opened: 12 fixed prompts that between them name a genre, a mood, a tempo and the instruments wanted (an upbeat pop song with vocals about a road trip; a slow lo-fi hip-hop beat at about 70 BPM; a tense orchestral cue for a chase scene; a 30-second corporate background loop; an acoustic folk ballad with a male vocal; an EDM drop at 128 BPM; a jazz piano trio; an 8-bit chiptune level theme; an ambient drone with no percussion; a cinematic trailer build with choir; a reggaeton track with a Spanish vocal; and a children's nursery tune). Each prompt is pasted once, the first full result is taken, and the generated audio is saved so every grade can be listened back to rather than taken on trust. A prompt that asks for vocals is only used to grade vocals on a tool that offers them; an instrumental-only tool is told so and is not marked down for a feature it does not claim.
Each generator is run on the best plan the account offers, on a free or trial account where one exists, at the default settings, with each prompt submitted once and the first finished track taken. Where a tool offers a choice of model or quality the one on the tested plan is used and named. Nothing is graded from a vendor's demo reel, a sample on the home page or a marketing claim. Each review states how the account was obtained: used on a free tier, bought at the public price, or provided by the vendor. A vendor-provided account never changes a grade, and a vendor never sees a review before it is published.
How I rate
Every tool gets one score out of 5.
First, I run the same 10 tests on every tool and give each one a grade from 1 to 5. The average of the graded tests is the starting point; a test that cannot be graded is left out of it, and the review says how many tests the score is based on.
Then I add my verdict: what the tests cannot measure, such as how the tool holds up after the test weeks, how billing and cancelling work in practice, and how it compares with the others in the category. My verdict can move the score up or down by one point at most.
Every review shows how much my verdict changed the score, and why, in its scores table.
What a grade means
| 5 | Excellent |
| 4 | Good |
| 3 | Average |
| 2 | Below average |
| 1 | Poor |
The final score is the average of the graded tests (ten when every test can be graded), to one decimal place, plus my verdict, which can move it by up to one point. If a test couldn't be finished properly it gets no grade, and the review says how many tests the score is based on. A generator that makes no vocals, offers no stems or has no editing is not penalised on a capability it does not claim; that test is scored on what it can attempt, with the fact stated, or left out and noted. Some tests on a free plan use fewer prompts than usual because of daily or credit limits; I run them again on a paid plan when I can.
The 10 tests
1.Audio Quality & Fidelity
All 12 generated tracks are saved at the highest quality the plan offers and played back on the same headphones and the same speakers. Each is marked for clarity, stereo width, dynamic range and freedom from the tell-tale faults of generated audio: hiss and noise, a muddy or boxy mix, clipping or distortion on loud passages, metallic or watery artefacts, and abrupt drops. The export format and bit rate or sample rate on the tested plan are recorded, because a 128 kbps MP3 and a WAV are not the same thing, and any watermark tone or voice-over on the free tier is noted. The grade is the quality across the whole set, not the best single track, and the two weakest tracks are named.
- 5: studio-clean across the set, no audible artefacts
- 4: clean with occasional minor artefacts
- 3: usable but with noticeable noise, muddiness or a low export quality
- 2: frequent artefacts, clipping or a heavy watermark that spoils the track
- 1: tracks sound broken, distorted or unusable
2.Prompt Adherence
Each of the 12 prompts states a genre, a mood, a tempo and at least one instrument. The finished track is scored out of four: is it the genre asked for; does it carry the mood; is the tempo within about 10 BPM of the number given (checked with a tap-tempo or BPM reader); and are the named instruments actually present. A prompt counts as fully met only on four of four. The total across the 12 prompts is the grade, and the points the tool most often misses (tempo is a common one) are reported. A tool that ignores the brief and returns a generic track of its own, however pleasant, is marked down here.
- 5: all four points on 10 or more of 12 prompts
- 4: all four on 7 to 9 prompts
- 3: all four on 4 to 6 prompts, or three of four across the set
- 2: all four on 2 or 3 prompts
- 1: rarely matches the brief
3.Vocals & Lyrics
The four vocal prompts (pop, folk ballad, Spanish-language reggaeton and the nursery tune) are used. For each the vocal is marked on: is it in tune and free of the warble and smearing generated voices are prone to; are the words intelligible; are they the lyrics asked for or supplied, rather than invented filler or 'gibberish' syllables; is the language and pronunciation right, which the Spanish prompt checks directly; and does the voice suit the genre. Whether lyrics can be written by the user or are only auto-generated is recorded, as is any one-word or made-up-syllable fallback. A tool that states it makes instrumental music only is marked as not offering vocals and this test is left out of its score rather than counted against it.
- 5: natural, in-tune vocals, the right words clearly sung, language correct
- 4: mostly clear with occasional artefacts or a word lost
- 3: understandable but plainly synthetic, or lyric control limited
- 2: frequently garbled, off the brief or wrong language
- 1: gibberish or no usable vocal (instrumental-only tools are not graded here)
4.Musical Coherence & Structure
Every track is listened to from start to finish and marked for musical sense: a recognisable structure (an intro, a build or verse-and-chorus, and an ending rather than a fade or a hard cut); staying in one key and tempo unless a change is called for; transitions between sections that do not jar; and freedom from an obvious four-bar loop repeated to fill time, a stutter, or a track that simply stops mid-phrase when it reaches the length cap. The share of the 12 tracks that hold together is the grade, and the most common failure (an abrupt ending is typical) is named.
- 5: coherent structure and clean endings across the set
- 4: mostly coherent, the odd weak transition or ending
- 3: listenable but often looping, aimless or cut off
- 2: frequent glitches, key or tempo drift, or hard stops
- 1: tracks do not hold together as music
5.Length & Customisation Control
Each point is checked by using it on the test date: set or extend the track length past the default; set or change the tempo (BPM); set or change the key or scale; choose or exclude named instruments; pick a mood or energy level; choose a song structure or section layout; upload or hum a reference melody, or continue from an audio clip; set an intro and an outro; exclude vocals or make an instrumental version; and steer the output with advanced or negative prompting. A point counts only when it works and audibly changes the result on the account used, not when it is merely listed, and each is screenshotted. A tool that is one prompt box and a go button scores low here however good the single track is.
- 5: all ten points
- 4: 8 or 9
- 3: 6 or 7
- 2: 4 or 5
- 1: 3 or fewer, or a single prompt box with no controls
6.Editing, Stems & Regeneration
On a track already generated, each point is pass or fail on the test date: regenerate a fresh take from the same prompt; make several variations at once; extend or continue the track; shorten or trim it; re-roll or replace a single section while keeping the rest; edit or re-sing the lyrics in place; swap an instrument or the vocal; download separated stems (drums, bass, melody, vocal); adjust mix levels or a built-in master; and keep a project or history so earlier takes can be returned to. Each working point is screenshotted and, where it produces audio, the file is saved. Stem download in particular is recorded as free, paid or absent, because it is what separates a finished track from a workable one.
- 5: all ten points
- 4: 8 or 9
- 3: 6 or 7
- 2: 4 or 5
- 1: 3 or fewer, or generate-and-download only
7.Format, Download & Export
Each point is checked by doing it on the test date: download an MP3; download a lossless WAV or FLAC; download individual stems; export a MIDI file; choose the bit rate or sample rate; download without a watermark or voice-over on the plan tested; a batch or bulk download of several tracks; a direct share link or embed; a push to an editor or a cloud drive; and a documented API. A point counts only when it works end to end on the account used, and the free-tier limit (download count, watermark, format) is recorded against the paid tier. Each download is opened and listened to, so a file that is silent, truncated or the wrong format is caught.
- 5: all ten points
- 4: 8 or 9
- 3: 6 or 7
- 2: 4 or 5
- 1: 3 or fewer, or watermarked MP3 only
8.Commercial-Use Rights & Licensing
The licence is the thing that decides whether a track can be used at all, so it is read from the tool's own terms and licence page on a stated date, not from a marketing line, and confirmed against the plan the account is on. Recorded: whether the user owns the output or holds a royalty-free licence; whether commercial use (adverts, client work) is allowed, and whether monetised YouTube, TikTok and streaming on Spotify and Apple Music are allowed; whether that needs a paid plan while the free tier is personal-use only; whether credit or attribution is required; whether the tool keeps the right to re-use or resell the same track to others (non-exclusive output); and any training-data or copyright risk the tool itself flags. The exact wording is quoted and dated, because these terms change and a stale reading is worse than none.
- 5: full commercial and streaming rights with clear ownership on a usable free or cheap plan
- 4: commercial rights clear but gated behind a reasonable paid plan
- 3: personal use free, commercial use costly or with limits, or attribution required
- 2: rights unclear, non-exclusive in a way that hurts, or streaming disallowed
- 1: no commercial rights, or terms that leave the user exposed
9.Generation Speed
All 12 generations are timed on the same connection, from pressing create until a full track is finished and playable; queue messages, progress bars and 'your track is rendering' screens all count as waiting, not as done. The median is the grade; the average, fastest, slowest, any failed or timed-out generations, and the free-tier queue or slow-down when it differs from the paid tier are reported. Where a tool returns several variations at once the time to the first playable one is taken. Longer tracks and extensions are timed separately and noted, since a three-minute track naturally takes longer than a thirty-second loop.
- 5: median under 30 seconds
- 4: 30 to 59 seconds
- 3: 1 to 2 minutes
- 2: 2 to 5 minutes
- 1: over 5 minutes, or generations time out
10.Free-Tier Limits & Pricing
On the free tier I generate until the tool stops me and record the real daily or monthly cap (tracks, credits or minutes), whether an account is needed, whether the free tier runs the same model as the paid one, and whether download, watermark removal, stems, commercial rights and the higher quality are free or paid. Every plan is then read from the checkout, not the marketing page, with the date: the monthly cost billed yearly, the credits or tracks each plan gives, what the paid plan lifts or adds, and the cost of one finished, downloadable, commercially usable track worked out in dollars on the cheapest plan that removes the free limits. Any pay-as-you-go or credit pricing is worked out as the cost of a typical month's use, and any auto-renew or hard-to-cancel trap is flagged.
- 5: a usable free tier with downloads, or a paid plan under $12 that adds commercial rights, stems and clean export
- 4: a usable free tier and a paid plan $12 to $19.99
- 3: a tight free tier, or a paid plan $20 to $29.99
- 2: $30 or more, or no usable free tier
- 1: poor value or unclear credit pricing
Documented but not graded
A prompt or lyric assistant, genre and mood presets, a sample and loop library, voice cloning or custom voices, text-to-speech or narration, sound-effects generation, a built-in digital audio workstation or timeline editor, a mobile app, a Discord or community gallery, royalty and distribution tools that put a track on streaming services, team and brand workspaces, and the range of languages the vocals support are checked and described in every review. When they work well they count in the written assessment, but they never add points to the overall grade.
Rules that apply to every test
- Real use of the product on a real account, never vendor demo reels, home-page samples or promotional claims.
- Identical prompts, settings and sample sizes for every generator in the category, with the first finished track counted every time and the best model the plan offers.
- Every generated track is saved so each grade can be listened back to, and the two weakest tracks in a set are named.
- The licence is read from the tool's own terms on a stated date and quoted, because a track that cannot be used legally is worth nothing however good it sounds.
- Every grade has evidence: the prompt, the saved audio, a count, a measurement, a screenshot or a table, in the full test results.
- A grade changes only when a retest produces new evidence; the review records the date, the model version and the reason.
Questions about a specific grade go to the contact page. Makers can submit an AI music generator for testing. The tested ranking is on the AI music generators page.