Research cut-off · 22 Sep 2026

AIVA

Artificial Intelligence Virtual Artist is a Luxembourg AI-music company and platform focused on AI-assisted composition, personalised instrumental generation and creator-controlled workflows.

[VERIFIED][COMPANY-REPORTED]CURRENT / HISTORICAL

1. Executive summary

AIVA started in 2016 as an AI-composition project and evolved into a commercial AI music generation assistant. Today it advertises 250+ styles, custom style models, audio/MIDI influence, editing and plan-specific licensing. Public research names OmniCodec, Lyra, ORAQL and Influence. Exact production model/version, current training corpus and current headcount are not publicly disclosed in sufficient detail. AIVA product.

2. Company identity & leadership

3. Pierre Barreau

AIVA describes Pierre Barreau as CEO/co-founder, computer scientist, award-nominated director and composer. TED describes him as an entrepreneur, computer scientist, composer and director and records his explanation that the idea for personalised AI-composed music was influenced by the film Her. A public professional profile records UCL computer-science study (2013–2016); that detail is secondary evidence. His documented company role is to lead AIVA toward its product and creative vision.

4. Founders & origin

Denis Shtefan is listed by AIVA as CTO/co-founder, published researcher and composer, leading R&D. Vincent Barreau is listed as COO/co-founder, engineer and musician, heading operations. The company's early concept was to use AI to learn musical structure and create new compositions for media. AIVA's historical material and EU project record connect the early system to classical-music analysis and film/game scoring.

5. Timeline

6. Mission & product positioning

AIVA's mission is to empower individuals by creating personalised soundtracks with AI. Its research page says it prefers models that support artists' creative journeys rather than simply replacing them. The current product therefore combines generation with user selection, reference material, editing and downloadable outputs.

7. Current product & human+AI workflow

Current public claims include 250+ styles, custom style models, audio/MIDI influence, editing and desktop apps. AIVA also showcases 150+ generated tracks arranged by humans.

8. AI models & research

[COMPANY-REPORTED RESEARCH RESULT] AIVA reports OmniCodec VISQOL 4.01 at 2.06 kbps versus DAC 3.95 at 2.6 kbps and Encodec 3.75 at 2.2 kbps, with reported win rates of 97.83% and 100% in its stated test. These are research results, not independent product benchmarks.

9. Training data & technical disclosure

The EU H2020 project description says the historical AIVA system had trained on 30,000+ classical pieces. That is HISTORICAL / PROJECT-REPORTED evidence, not a specification of current Lyra training data. Current dataset provenance, filtering, parameter counts and complete model architecture are Not publicly disclosed. AIVA's public research does disclose transformer use, audio tokenisation and neural-codec architecture at a high level.

10. Influence & legal terms

AIVA says Influence uses an uploaded reference to create an original piece sharing core musical characteristics. The EULA states that uploaded MIDI/audio gives AIVA a worldwide, non-exclusive, commercial and transferable licence to train its AI systems in perpetuity; the user warrants that the upload does not infringe third-party rights and accepts responsibility for use of the feature. The EULA also restricts reverse engineering, unauthorised private APIs, large-scale licensing and use of AIVA compositions as training datasets.

11. Pricing & licensing — current public individual plans

Verify live pricing before publication. AIVA can change prices and plan terms.

12. Copyright: three separate questions

13. Customers, projects & partnerships

AIVA's current site displays NVIDIA, Vodafone, Globant, TED, Animo TV, Nordic Business Forum, Lumitoro, Pioneers, World50 and Luxembourg Ministry of Culture. Logos are not proof of paying-customer status. Documented creations include NVIDIA GTC material, Luxembourg National Day music, Ode to Dubai, Vodafone's Tomorrow Street/Pioneers material and Globant ConVerge 2017. Each should be labelled customer, partner, event, demo or project only where the underlying evidence supports that classification.

14. Funding & financial disclosure

These located announcements establish €2.15m of disclosed 2017/2020 financing, excluding separate EU support. This is not a definitive lifetime-funding total. Revenue, ARR, valuation and current user count: Not publicly disclosed.

15. Business model & target users

Documented revenue mechanisms are individual subscriptions plus custom enterprise licensing. The product is relevant to beginners, creators, composers, film/game teams, advertisers, educators and businesses; the plan choice mainly changes monetisation, copyright and download rights. Enterprise/API/high-volume use requires separate commercial arrangements under the EULA.

16. Competitive landscape — no ranking

17. AIVA vs Suno — factual snapshot

18. Legal / responsible-AI context

AIVA's contractual terms are explicit about user responsibility for uploaded material and plan-based rights. Broader AI-copyright law remains jurisdiction-specific. The U.S. Copyright Office's Part 2 report (2025) emphasises human authorship; EU materials similarly focus on human creativity. EU AI Act transparency obligations for covered AI-generated/manipulated content apply from 2 Aug 2026, subject to transition rules. None of these sources establishes an AIVA-specific violation.

19. SWOT — evidence-led

20. Future roadmap

AIVA's Lyra research article describes planned derivative models for melody-to-arrangement, stem-by-stem generation, temp-track exploration for film/game composers and music education. These are research directions, not confirmed release dates. No authoritative public schedule for each item was located.

21. Current status & key facts

On 22 Sep 2026, AIVA's public website and product pages are live; the official status page reports the website and Music Engine as healthy. This establishes service availability only, not profitability or financial health.

22. Key lessons

• AIVA shows the shift from early AI-composition research to a commercial generative-music assistant.
• Its public differentiation is control: style models, MIDI/audio influence, editing and licensing—not simply one-click generation.
• Research benchmarks must remain separate from consumer-product performance claims.
• Contractual copyright ownership is distinct from statutory copyright protection.
• Customer logos must not automatically be described as paying customers.
• For publication, re-check live pricing, EULA, leadership and model availability immediately before release.

Sources

1. AIVA About
2. AIVA product/pricing
3. AIVA EULA
4. AIVA press
5. AIVA research blog
6. AIVA creations
7. AIVA downloads
8. AIVA status
9. EU CORDIS project
10. EU CORDIS results
11. NVIDIA on AIVA
12. TED Pierre Barreau
13. NetEase investment — MBW
14. NetEase investment — Tech.eu
15. U.S. Copyright Office AI report
16. EU IP Helpdesk
17. Suno
18. Udio
19. ElevenLabs Music
20. SOUNDRAW license
21. Beatoven terms
22. Luxembourg registry-derived record