AIVA: AI Music Composition, MIDI Editing, Licensing, and Lyra

AIVA's most important public product direction is Lyra, an instrumental foundation model designed for personalized music generation. The announcement describes a newer technical stack built around Lyra, the OmniCodec audio codec, the ORAQL quality model, and a reference-guided variant called Influence. This is more than a cosmetic update to the existing style picker, but users need to understand its status: AIVA's own page still invites people to register for a private beta. WhatAI therefore treats Lyra as an announced and testable roadmap for selected users, not as a generally available replacement for every feature in the current subscription.

AIVA says Lyra accepts natural-language prompts and can be conditioned to create complete instrumental compositions from 30 seconds to 10 minutes. That maximum is materially longer than the public individual plans, which currently advertise tracks up to three minutes on Free, five minutes on Standard, and five minutes and thirty seconds on Pro. The distinction matters. A ten-minute Lyra capability described in a research and beta announcement does not automatically raise the duration limit of the existing product or the number of tracks a subscriber may download.

The supporting technology is also notable. AIVA says OmniCodec compresses musical audio efficiently while preserving reconstruction quality and producing tokens that are easier for a language model to learn. Lyra then uses a customized transformer with multi-stage modelling and parallel token processing. AIVA reports that Lyra was trained with 44,000 GPU hours at an estimated compute cost of about $110,000. Those figures help explain the engineering approach, but they remain company disclosures rather than independently audited performance evidence.

ORAQL addresses a different problem: how to rank music when quality is subjective. AIVA reports a correlation of 0.78 between ORAQL predictions and human preference scores on its internal benchmarks. The company says it uses ORAQL as a reward model and to choose stronger samples from a batch, and claims that doubling inference-time compute doubles the likelihood that a listener will like the selected result. WhatAI treats those as vendor-reported internal findings. They are a reason to test Lyra carefully, not proof that one model will produce the best music for every listener, genre, brief, or production standard.

Influence may be the most practical part of the roadmap. It is designed to accept a reference audio file and generate a different instrumental composition sharing core musical characteristics. AIVA claims that its system outperforms open and closed alternatives on personalization, originality, quality, and faithfulness to the source intent. Again, that is AIVA's own assessment. More importantly for buyers, the current End User License Agreement already gives AIVA a perpetual, worldwide, commercial, transferable training licence over audio or MIDI uploaded through its influence feature. Only upload material you own or are clearly authorized to provide.

AIVA is also explicit about what it is not prioritizing. The company says generated vocals can make an impressive demonstration, but its focus is on tools that assist artists. Proposed derivative models include an arrangement assistant that could build a song stem by stem from a melody, another stem, text, or audio references. AIVA also discusses temp-track collaboration between composers and clients, plus music education. These are sensible extensions of its composition-first identity, but the arrangement assistant is described as future work rather than a finished public feature.

WhatAI's take: Lyra makes AIVA more strategically interesting, especially for instrumental creators who value long-form generation, reference guidance, composition learning, and eventual arrangement assistance. It does not remove the need to evaluate today's public product on today's terms. Before paying, test the existing editor with a real brief, export the formats your workflow requires, compare MIDI usefulness with the final audio, and choose the licence based on the destination of the music. Recheck whether Lyra has left private beta, whether its duration and download rules match the public plans, and whether its output receives the same copyright treatment. Until AIVA publishes that bridge clearly, the current pricing page and End User License Agreement remain the safest basis for the buying decision.

WhatAI Decision Box

AIVA generates instrumental music in more than 250 styles, supports custom style models and audio or MIDI influences, lets users edit generated tracks, and exports MP3, MIDI, WAV, and other formats according to plan. It is designed around composition and soundtrack workflows rather than lyric-led vocal song generation.

Free is for attributed, non-commercial use. Standard permits monetization only on YouTube, Twitch, TikTok, and Instagram while AIVA retains copyright. Pro assigns full copyright in downloaded compositions to the user and supports unrestricted monetization under the agreement. Choose the licence before judging the generation limit or export format.

AIVA results vary with the selected style, custom model, influence file, duration, instrumentation, editing choices, export format, and licence.

Join the discussion below and share the brief, plan, styles tested, reference rights, number of generations, edits, export path, final use, and licence you reviewed. Community evidence can show when AIVA's editable composition workflow is more useful than a vocal song generator or a simpler background-music service.

Use Cases

Pricing varies by plan and region — see current pricing.

Plan features change — last updated: 2026-07-29.

AIVA Pros & Cons

How to Get Results with AIVA: Step-by-Step Workflow

1. Define the final use

   State whether the music is for private learning, a monetized social channel, a client, a film, a game, a podcast, streaming distribution, or another commercial destination. Map that use to Free, Standard, Pro, or Enterprise before creating.

2. Write a musical brief

   Define the scene, listener, emotion, duration, tempo range, genre ingredients, instrumentation, structure, energy curve, loop requirement, and elements to avoid.

3. Choose the guidance route

   Start with a preset style, build a custom style, or upload an audio or MIDI influence that you own and are willing to license under AIVA's training terms.

4. Generate controlled options

   Create several versions while changing one variable at a time. Record the style, influence, duration, settings, generation count, and first-listen notes.

5. Edit the strongest composition

   Choose the version with the best musical structure and emotional fit, then use AIVA's available editor to correct arrangement, pacing, instrumentation, or other supported elements.

6. Export for the real workflow

   Test MP3 and MIDI on the available plan, or WAV and other formats on Pro. Confirm that the exported MIDI, audio quality, duration, and download allowance are sufficient before scaling production.

7. Review rights and restrictions

   Confirm copyright ownership, permitted monetization destinations, attribution, reference rights, prohibited bulk uses, renewal, refund, and termination terms for the exact account.

8. Compare and document

   Run the same instrumental brief in at least one alternative, then compare composition, personalization, editability, MIDI value, audio quality, time, price, ownership, and final project fit. Share the evidence with WhatAI.

AIVA Gotchas and Limits to Know Before You Start

• Free compositions are non-commercial and require AIVA attribution.
• AIVA owns copyright in compositions downloaded under Free and Standard.
• Standard monetization is limited to YouTube, Twitch, TikTok, and Instagram.
• Pro is the individual plan that assigns full copyright to the user.
• Plan rights are granted when the MIDI or audio composition is downloaded.
• Free allows only three downloads per month.
• Standard allows only fifteen downloads per month.
• Pro allows only three hundred downloads per month.
• Track duration is capped at three minutes on Free, five minutes on Standard, and five minutes and thirty seconds on Pro.
• High-quality WAV and all-format exports require Pro.
• Annual advertised prices exclude VAT.
• The public pricing page captured by WhatAI displays annual equivalents, while the agreement also identifies monthly plans. Verify live monthly checkout pricing.
• Individual accounts cannot be shared with multiple people.
• AIVA defines Individual and Enterprise eligibility through employee and revenue thresholds in the agreement.
• Users must own or control every audio or MIDI influence they upload.
• Uploading an influence grants AIVA a worldwide, non-exclusive, commercial, transferable training licence in perpetuity.
• Unauthorized private API calls are prohibited.
• Bots and software that automate fake human interaction with the interface are prohibited.
• High-volume composition and automated pipelines require a separate agreement.
• Large-scale upload or licensing of unmodified or uncurated compositions to third parties is prohibited without a separate arrangement.
• Compositions and samples cannot be used to create a competing service.
• Compositions and samples cannot be used as a machine-learning training dataset without a separate agreement.
• A user can generate no more than five compositions concurrently.
• A user can generate no more than one hundred compositions per hour.
• A user can generate no more than four hundred compositions in twenty-four hours.
• An account can store no more than four thousand compositions at one time.
• Refunds are available only before any premium feature is used.
• Forgetting to cancel or disable auto-renewal does not qualify for a refund under the agreement.
• The agreement says rights terminate for noncompliance and instructs the user to stop using and delete compositions after termination.
• Users remain responsible for checking whether a composition infringes another party's rights.
• A polished instrumental can still require human arrangement, performance, mixing, mastering, and originality review.
• AIVA's current public product does not promise a complete sung-vocal and lyric-generation workflow.
• Lyra, Influence, the proposed arrangement assistant, and other roadmap features should not be treated as generally available until AIVA confirms launch.
• AIVA's Lyra and ORAQL performance statements are vendor claims based on company evaluations.
• Do not confuse unrelated mobile apps using AIVA-like names with the official AIVA Technologies product.
• No official public affiliate program was verified.
• Pricing, VAT, plan limits, editor capabilities, licence terms, Lyra availability, and enterprise conditions can change. Verify all live sources before subscribing or publishing.

Which AIVA Feature Fits Your Use Case

Starter Prompts for AIVA

What is AIVA?

AIVA is an AI music composition platform. It generates instrumental music in more than 250 styles, supports custom style models and audio or MIDI influences, provides track editing, and offers plan-specific MP3, MIDI, WAV, and other exports.

Is AIVA free?

Yes. Free accounts can download three compositions per month, each up to three minutes, in MP3 and MIDI formats. AIVA owns the copyright, the user must credit AIVA, and the compositions cannot be monetized.

How much does AIVA cost?

The current public annual pricing is €11 per month plus VAT for Standard and €33 per month plus VAT for Pro, both billed annually. The agreement also identifies monthly Standard and Pro plans, but the public page captured by WhatAI displays annual pricing. Verify the live monthly checkout total, currency, VAT, and renewal terms.

Can I monetize AIVA music?

It depends on the plan. Free does not permit monetization. Standard permits monetization only on YouTube, Twitch, TikTok, and Instagram. Pro permits unrestricted monetization under the plan licence and assigns full copyright in downloaded compositions to the user.

Who owns music made with AIVA?

AIVA owns compositions downloaded under Free and Standard. Pro assigns full copyright in downloaded MIDI or audio compositions to the user. Users should still review the complete agreement, prohibited uses, account eligibility, and termination terms before relying on a composition for a valuable commercial project.

Does AIVA generate vocals and lyrics?

AIVA's current public positioning focuses on music composition, soundtracks, styles, influences, editing, and instrumental output. Its Lyra announcement explicitly describes that newer model as instrumental-only. Choose Suno or another vocal-focused service when sung lyrics are the main deliverable.

Can AIVA export MIDI and WAV?

Free and Standard include MP3 and MIDI downloads. Pro adds high-quality WAV and all available download formats. The number and duration of downloadable compositions also change by plan.

What is AIVA Upload Influence?

Upload Influence lets a user provide an audio or MIDI composition so AIVA can generate a new composition sharing some emotional qualities. The user must hold the necessary rights and grants AIVA a worldwide, non-exclusive, commercial, transferable licence to train its AI systems on the uploaded influence in perpetuity.

What is AIVA Lyra?

Lyra is AIVA's announced instrumental foundation model. AIVA says it supports natural-language prompting and can be conditioned for complete compositions from 30 seconds to 10 minutes. The accompanying Influence variant uses reference audio, but the official announcement still presents the system as a private beta.

Does AIVA have an API?

AIVA does not publish a self-serve public API. Its agreement prohibits unauthorized private API calls and says approved API access, high-volume composition, and automated pipelines require a separately negotiated licensing agreement.

What devices support AIVA?

AIVA provides a browser-based creator experience and desktop downloads for macOS, Windows, and Linux. WhatAI did not verify an official AIVA mobile app and warns against confusing unrelated mobile apps using similar names with AIVA Technologies.

Can a team share one AIVA account?

No. The End User License Agreement says individual accounts are intended for their owners exclusively and cannot be shared by multiple individuals. Eligible organizations should discuss an enterprise arrangement with AIVA.

Can I get a refund from AIVA?

AIVA's agreement says refunds are available only if the customer has not used any premium feature. Using a premium feature waives the withdrawal right, and AIVA says it does not refund users who forget to cancel or disable auto-renewal.

Does AIVA have an affiliate program?

WhatAI did not verify a public official AIVA affiliate program. Use the direct product link unless AIVA publishes a formal program with confirmed terms, tracking, eligibility, and disclosure requirements.

Related Audio & Voice Tools

Explore the Network