Mureka generates full songs from lyrics, prompts or a reference track, then lets you split them into as many as 12 stems and export WAV, MIDI and DAW sessions. Built by Skywork AI, it also ships a developer API priced per song.
Used this tool? Rate it
Used this tool? Rate it
Mureka is an AI music generation platform that builds complete songs from a set of lyrics, a written prompt, a hummed melody idea or an uploaded reference track. Where many text-to-music tools treat lyrics as an afterthought, Mureka is organised the other way round: you supply or generate the words first, and the model composes vocals and instrumentation around them. The service is operated by SKYWORK AI PTE. LTD., a Singapore-registered company that its own terms of service name as the contracting party, and it sits inside the wider Skywork AI product line built by the Chinese technology group Kunlun Tech.
The product spans three surfaces that share one model family. A consumer web app at mureka.ai handles generation, editing and publishing. Mobile apps on iOS and Android carry the same catalogue plus an AI radio mode. A separate developer platform at platform.mureka.ai exposes the same models through a per-song API, with nine capability groups covering music generation, custom model fine-tuning, lyrics, vocal cloning, song extension, region editing, music analysis and text-to-speech.
Positioning-wise, Mureka competes directly with Suno and Udio. Its differentiation is less about raw generation quality — which no independent benchmark has verified — and more about what happens after a track exists: splitting a song into as many as twelve stems, exporting MIDI and DAW session files, remixing an existing track into another genre, and driving a local digital audio workstation through natural language. That post-generation toolkit is the clearest reason a working producer might choose it over a pure generate-and-download service.
Mureka is a weaker fit for anyone who needs guaranteed originality, provable rights clearance for high-stakes commercial use, or a vendor with published training-data provenance — none of which this platform currently offers.
A free tier exists and is genuinely usable for evaluation: four song creations per day, restricted to the v7.5-all free model, two concurrent generations in a shared queue. Its hard limits are the ones that matter commercially — no audio download, no stem separation, no commercial rights. Two paid tiers sit above it, both quoted at annual billing with a stated twenty percent saving over monthly. Pro works out to $8.00 per month, billed $96 annually, and grants 5,000 Gold monthly — enough for up to 500 songs or 250 minutes of speech — with the advanced model line-up, unlimited MP3 downloads, reference-track creation, remixing, a ten-slot priority queue and commercial rights covering both new and past creations. Premier works out to $24.00 per month, billed $288 annually, with 20,000 Gold covering up to 2,000 songs or 1,000 minutes of speech.
Premier is where the production features live: twelve-stem splitting, Mureka Studio access, unlimited WAV, stem, MIDI and DAW session downloads, professional voice cloning, using any voice as the singer, and full Mureka Co access to your DAW and local files. The developer API is billed separately on a top-up model — $0.03 per song on V7.6, $0.045 on O2, V8 and V9, and a list price of $0.15 on V9.5 with a beta rate of $0.045 running until 28 August 2026, while text-to-speech is metered at $0.013 to $0.62 per output second. API concurrency is bought in bands tied to a single top-up: $10 for one concurrent request, $1,000 for five, $3,000 for fifteen, $5,000 for twenty-five and $30,000 for a hundred and fifty. Prices and promotional rates observed here are subject to change, and monthly-billing figures were not displayed in the default view.
The honest framing is that no AI music generator, Mureka included, currently offers the rights certainty a traditional production library does. Choosing between them is largely a choice about how much licensing risk a project can carry.
Ownership is tier-dependent, and less absolute than the marketing implies. The homepage advertises "full commercial rights", but the governing terms are more precise. Only users who have purchased credits or subscribed to a paid tier own the output; free-tier output is owned by Mureka and licensed back for non-commercial use with mandatory attribution. Section 3(a) of the same document licenses the service itself "solely for your personal, non-commercial use" — a tension with the homepage language that anyone planning commercial release should read carefully.
Even paid users grant the platform sweeping rights to their own tracks. Regardless of tier, users grant Mureka an unrestricted, assignable, sublicensable, irrevocable, royalty-free worldwide licence to reproduce, distribute, publicly perform, create derivative works from and otherwise exploit their output, explicitly including commercial use. Owning your song and having exclusive control of it are not the same thing here.
The company does not guarantee originality or copyright registrability. The terms state plainly that output may bear similarities to work generated by other users, that uniqueness is not guaranteed, and that because of the uncertainty of copyright law around AI-generated content, Mureka does not guarantee output can be registered for copyright — with those legal risks assigned to the user.
Training-data provenance is undisclosed. Music Business Worldwide's reporting noted that Kunlun said nothing about training data sources or licensing when launching the model, contrasting that silence with Suno's court-documented position. For anyone releasing commercially, this is unresolved downstream exposure rather than a settled question.
Performance claims are unverified. The company's assertion that Mureka O1 outperformed Suno V4 across mixing quality, vocal texture and instrumentation was published without specific metrics or independent verification, as that same trade coverage observed. No credible independent benchmark of this model family has been published.
Billing and cancellation are the dominant user complaint. On Trustpilot the platform holds 1.4 out of 5 across 173 reviews, with the aggregated themes centring on difficulty cancelling subscriptions, unexpected annual charges following a trial, unanswered refund requests, and inconsistent output quality. Two caveats are needed in both directions: the profile is unclaimed and the company has never solicited reviews, so the sample skews toward users motivated to complain; and the same page shows Suno at 1.6 across 753 reviews and Udio at 1.5 across 57, indicating this is a category-wide pattern on that channel rather than something unique to Mureka.
App store ratings tell a very different story, and both are real. The iOS app rates 4.79 from 10,872 ratings and the Android app 4.66 from 159,415 ratings against more than five million downloads. These channels measure different things — product satisfaction among installed users versus billing disputes among people motivated to file them — and neither cancels the other out. The reasonable reading is that most users like the output and a meaningful minority have serious problems with the subscription mechanics.
The published terms are out of step with themselves. The version rendered at mureka.ai/terms is dated 18 October 2024, while the PDF linked from the sign-in dialog is dated 13 May 2026 and contains materially different provisions, including the uniqueness and copyright-registration disclaimers. Users should treat the PDF as controlling. The legal text is English-only across every locale prefix, and the agreement states the English version prevails over any translation.
Corporate context is easily overstated. Kunlun Tech's announced group ARR above $800 million is dominated by its AI short-drama platform at over $700 million, with the AI tools segment that contains Mureka at over $100 million. That figure is company-reported and not independently audited, and it should not be read as the scale of the music product itself.
Disputes go to arbitration in Singapore. The agreement is governed by Singapore law, with disputes referred to the Singapore International Arbitration Centre before a three-arbitrator tribunal seated in Singapore, conducted in English — a meaningful cost consideration for individual users.
There is a free tier that allows four song creations per day using the v7.5-all model, with two concurrent generations in a shared queue. It is suitable for evaluating output quality, but it excludes audio downloads, stem separation and commercial rights, so it cannot support published work.
Only on a paid plan. Under the current terms, users who have purchased credits or subscribed own their output including its intellectual property rights, while free-tier output is owned by Mureka and licensed back for non-commercial use only, with attribution required. The subscription page states that commercial rights on paid plans cover previously created songs as well as new ones. Note that even paid users separately grant Mureka an irrevocable, royalty-free licence to exploit the same output commercially.
No. The terms state explicitly that because of how machine learning works, output may have similarities to music created by other users on the platform, and that uniqueness is not guaranteed. They further state that Mureka does not guarantee output can be used for copyright registration, and assign that legal risk to the user.
Free accounts cannot download audio. Pro provides unlimited MP3 downloads. Premier adds WAV, individual stems, MIDI and DAW session files, along with watermark-free lyric videos on both paid tiers.
Up to twelve, using the Splitter tool. Stem separation is exclusive to the Premier tier — neither the free nor the Pro plan includes it.
Free accounts are restricted to v7.5-all. Paid accounts can select among V9.5, V9, V8, O2 and V7.6. V9.5 is the newest and is marketed as more natural, more expressive and more faithful to the prompt. On the API these are priced differently — $0.03 per song for V7.6, $0.045 for O2, V8 and V9, and $0.15 list for V9.5.
Yes, on a separate platform with nine capability groups covering music generation, custom model fine-tuning, content services, lyrics, vocal cloning, song extension, region editing, music analysis and text-to-speech. Pricing is per song by model, with text-to-speech billed at $0.013 to $0.62 per output second. Concurrency is granted per individual top-up rather than cumulatively, so repeated smaller purchases do not stack.
The terms name SKYWORK AI PTE. LTD., a Singapore company at 2 Science Park Drive, as the operator; the same entity appears as the developer on both the iOS and Android app listings. It belongs to the Skywork AI line built by Kunlun Tech. The agreement is governed by Singapore law, with disputes resolved by SIAC arbitration seated in Singapore.
Professional voice cloning and using any voice as the singer are Premier-exclusive features. Because these capabilities involve a real person's vocal likeness, you need the rights or consent to use any voice you upload, and the terms place responsibility for that on the user.
The current terms require users to comply with applicable laws and platform policies to disclose, mark or label output as AI-generated where that applies. Several distribution platforms now require such labelling independently, so check the requirements of wherever you intend to publish.
