MakeSong

Music generation crossed a threshold recently that most people have not registered. The output is no longer recognisably synthetic on a casual listen — a generated track with vocals, played once through phone speakers under a video, passes without comment.

That changes who this technology is for. It stopped being a novelty for people curious about AI and became a practical option for anyone who needs a piece of music and has neither a budget nor a library subscription.

MakeSong sits in that market with a broader tool set than most, and Suno is the benchmark it has to be judged against. What makes it worth a closer look is not the generation, which is roughly comparable across the category, but what it lets you do afterwards — and what its licence actually permits, which is the question everyone skips and the one that matters commercially.

Text In, Finished Track Out

• Two entry points: describe a song in words, or supply your own lyrics and have music written around them.
• Vocals are included, with instrumental and karaoke variants available — the vocal is what separates this category from background-music generators.
• The editing tools are the real differentiator: stem splitting, vocal removal, section replacement, song extension and cover creation.
• WAV as well as MP3, which matters if the track is going into an edit rather than straight onto a platform.
• Commercially licensed and described as royalty-free, with a downloadable licence on paid tiers — read the terms, because royalty-free and owned are not the same thing.

The Editing Tools Are The Interesting Half

Most music generators are one-shot machines. You describe a song, you get a song, and if it is nearly right your only recourse is to generate again and hope.

MakeSong lists a substantially fuller kit: a stem splitter, a vocal remover, section replacement, song extension, cover creation and an acapella extractor.

Stem splitting is the most consequential of these, and it is what separates this from a background-track generator such as Treblo. Separating a mix into vocals, drums, bass and other instruments turns a fixed artefact into material — you can drop the vocal under a different section, use the instrumental behind narration, or rebalance a mix that came back with the vocal too loud.

Section replacement addresses the specific frustration of this category, which is that generation is nearly right more often than it is wrong. A track with a good verse and a weak chorus is a common outcome, and regenerating the whole song to fix one section is both wasteful and likely to lose what worked.

Song extension matters for a duller but very practical reason: video edits do not respect musical structure. A track that ends eight seconds before your video does is unusable, and extending it is far better than looping it clumsily.

The gap between a generator and a tool is whether you can fix the eighty per cent that came out right without regenerating it.

Version Numbers Are Not Model Documentation

The site references advanced models as V5, V4 and V3. That is all the detail offered.

It is worth being clear about what this does and does not tell you. Version numbers indicate that the product has iterated, which is mildly reassuring. They tell you nothing about architecture, training data, or what changed between generations.

The training data question is the one with commercial weight. Music generation models learn from recordings, and what those recordings were and under what terms they were used is an active and unsettled area in several jurisdictions. A vendor that publishes nothing about its training sources leaves its users unable to assess a risk that ultimately sits with them.

This is not unique to MakeSong — most of the category is equally quiet, and the few vendors who do publish tend to be those licensing catalogues deliberately. It is simply a factor to weigh if the output is going into commercial work, and a reason to read the terms rather than the marketing.

Royalty-Free Is Not The Same As Owning It

The site states that songs are one hundred per cent royalty-free, that commercial use is supported, and that a downloadable commercial licence is included with paid tiers. This is the most important paragraph on the page and it repays careful reading.

Royalty-free means you do not pay per use. It is a statement about payment structure, not about ownership. It is entirely consistent with the vendor retaining rights, with the licence being revocable, or with the permission being narrower than you assumed.

The questions that actually determine whether you can use a track commercially are different ones. Do you own the output or hold a licence to it? Can that licence be revoked or changed later, and what happens to work already published? Is there any indemnity if a third party later raises a rights claim over the output? Does the permission survive if you stop subscribing?

That last question matters more than people expect. If your licence depends on an active subscription, every video you published with that music becomes a problem the month you cancel.

There is also a separate legal question the licence cannot answer: in several jurisdictions, purely machine-generated output may not attract copyright protection at all. That does not stop you using it, but it may mean nobody can stop others using the same track — which matters if you intended it as a brand asset.

None of this is a reason to avoid the tool. It is a reason to read the actual licence document before the track goes into anything commercial, rather than relying on a phrase on the homepage.

• Download and read the licence document itself, not the marketing summary.
• Establish whether you own the output or hold a licence, and whether that licence is revocable.
• Check what happens to already-published work if you cancel your subscription.
• Look for any indemnity provision, and assume there is none unless it is stated.
• For a brand asset, take advice on whether generated audio can be protected in your jurisdiction at all.

Who Actually Needs A Generated Song

It is worth separating the uses where this technology is genuinely the right answer from the ones where it is a worse version of something else.

The strongest case is background music for content, where you need something that fits a mood and a duration, nobody will listen to it attentively, and the alternative is a library subscription or a copyright strike. Generated audio wins clearly here.

The second strong case is personal and occasional: a song for a wedding, a birthday, a leaving gift. The output does not need to be objectively good, it needs to be about a specific person, and no library track can be.

Songwriting scratchpad is a real third use. Generating four versions of an idea to hear which arrangement works is faster than demoing them yourself, even if nothing generated survives into the final piece.

Where it is the wrong tool is anywhere the music is the product — for that, a vocal synthesiser such as ACE Studio puts the writing back in your hands. Music people will listen to attentively, or that carries your identity as an artist, is not something to outsource to a text prompt — and audiences have become notably quicker at recognising and objecting to generated audio presented as authored work.

The dishonest case is worth naming plainly: uploading generated tracks in volume to streaming platforms to collect fractional royalties. Platforms are actively removing this material, and it is a poor use of anyone's time regardless of the tooling.

Judging Output Before You Spend Credits

Music generators are unusually easy to evaluate badly, because the first generation is often surprisingly good and the fifth reveals the ceiling. This sequence gets you an honest read.

The bar is step five. A generator that cannot do your specific genre is not a generator you can rely on, however good its pop output sounds.

1. Generate one track from a description in the genre you actually need, not the one the examples showcase.
2. Listen on the device your audience will use — phone speakers if it is going under a video, headphones if it is not. The failure modes differ completely.
3. Now generate three more from the same prompt. Sameness across the four is the real limitation of every model in this category, and it only shows on repetition.
4. Take the best one and run the stem splitter. Clean separation is what makes the track usable in an edit; muddy separation means you have a fixed artefact after all.
5. Try your hardest genre — something with unusual instrumentation or a rhythm that is not four-four. This is where models diverge most.
6. Only then read the licence document, and only after reading it decide whether to pay.

Credits, Tiers And The Licence Gate

Access is credit-based across several subscription tiers, with the commercial licence and higher-quality output attached to paid plans.

Credits are the right meter for this kind of product, because generation genuinely costs the vendor money per run and a flat-rate offer would either be restrictive or unsustainable. The consequence for you is that iteration has a price, and iteration is how you get a usable track.

That has a practical implication for how you work. Refining a prompt across ten generations is the normal path to a good result, so budget credits for exploration rather than expecting the first output to be the one you use.

Check two things specifically before subscribing: whether credits roll over or expire monthly, and whether WAV export and the commercial licence are available on the tier you are considering or only above it. Those two details decide whether a plan is usable for real work.

We do not publish amounts, since tiers and credit rates change.