/

MusicGen

by Meta AI
  • What it is:MusicGen is a powerful AI music generation system developed by Meta that creates high-quality original music compositions from text descriptions or melodies using advanced language models.
  • Best for:AI researchers and developers, Content creators needing quick instrumentals, Hobbyist musicians prototyping ideas
  • Pricing:Free tier available, paid plans from Infrastructure costs only
  • Rating:85/100Very Good
  • Expert's conclusion:MusicGen will be most useful for AI experimentalists and rapid prototyping, however it will likely fall short for professional music production purposes due to the inconsistent quality and short length of generated musical content.
Visit website
Reviewed byMaxim Manylov·Web3 Engineer & Serial Founder

What Are MusicGen's Key Business Metrics?

📊
20,000 hours of licensed music
Training Data
📊
10,000 tracks
High-Quality Licensed Tracks
📊
390,000 tracks
Instrument-Only Tracks
📊
~12 seconds per generation
Audio Generation Length
📊
MIT (code), CC-BY-NC (models)
License Type
📊
GPU with ~16GB memory
Hardware Requirements

How Credible and Trustworthy Is MusicGen?

85/100
Excellent

Strong Credibility MusicGen is credible due to its open source AI music generator that is supported by Meta, which has used publicly accessible, lawfully obtained training data, and has been shown to outperform other music generators in terms of quality and functionality.

BREAKDOWN
Product Maturity85/100
Company Stability95/100
Security & Compliance80/100
User Reviews80/100
Transparency90/100
Support Quality75/100
TRUST SIGNALS
Backed by Meta with extensive research teamOpen-source release (MIT license for code)Outperforms major competitors on objective metricsTraining data sourced through legal agreements with right holdersResearch paper published with ethical considerationsPre-trained models publicly available for reproducibility

What Are the Key Features of MusicGen?

✨
Text-to-Music Generation
Ability To Generate Audio From Text Description The system can generate music based on a written description of the type of music desired. For example "an 80s driving pop song with heavy drums and synth pads" will produce approximately 12 seconds of audio.
✨
Melody Conditioning
Reference Audio And/Or Existing Songs To Guide Generation In addition to providing a text description of the type of music to be generated, users can provide reference audio or existing songs that MusicGen will use to guide the generation process while adhering to both the provided text description and the melodic patterns contained within the reference audio and/or existing songs.
✨
Single Language Model Architecture
One Unified Model For Both Melody And Lyrics Instead of having multiple models for melody and lyrics, MusicGen uses one unified language model to operate on the compressed music tokens produced by the system. This eliminates the need for separate models to be trained and implemented for each function.
✨
High-Quality Output
Reasonably Melodic Music For Most Genres Of Music With Superior Coherence Than Competitors MusicGen produces reasonably melodic music that fits most genres, and it does so with a level of coherence that is greater than most competing music generation systems such as Google's MusicLM.
✨
Open-Source Availability
The Code Released Under An Open Source License The code was released under the MIT license with pre-trained models being made available, therefore developers are free to modify, extend and run the application on their local machine if they have appropriate GPU hardware.
✨
Multiple Input Modalities
Flexibility In Input Data MusicGen accepts both text prompts and melodic references as input to create flexibility in how users want to condition the music generation workflow.

What Are the Best Use Cases for MusicGen?

Music Producers and Composers
Fast Background Music Generation Background music, instrumental tracks, and melodic ideas can all be quickly created using MusicGen to assist in the composition process and to find different musical variations based on text-based directions.
Content Creators and YouTubers
Creation Of Royalty-Free Background Music For Videos Using MusicGen developers can create royalty-free background music for video productions by creating genre-specific tracks such as ambient, lo-fi, electronic with no concern over licensing as the training data for MusicGen is legally sourced.
Game Developers
Generating Dynamic Background Music For Games Developers can also generate dynamic background music and ambient soundscapes for games, and because the MusicGen model is open source, developers can integrate the model into their game engine and customize the model to meet their game's needs for different game scenarios.
Streaming Services and Wellness Apps
Personalized Music Content For Meditation, Relaxation And Sleep Features As demonstrated by Deezer's Zen app, MusicGen can be used to create personalized music content for meditation, relaxation, and sleep applications by generating customized tracks that match the user's preferred mood and genre.
AI Researchers and Developers
Extending And Modifying The MusicGen Model For Research And Development Finally, since MusicGen is an open source model, developers can extend and modify the model to support new research and development initiatives related to music AI, develop new conditioning methods, and test the limits of what is possible when using music AI.
Music Amateurs and Hobbyists
Produce background music to use with your own projects, demos and experiments using simple text-based prompts that do not require you to have an understanding of music theory.
NOT FORProfessional Musicians Creating Original Works
Not recommended - The quality of the generated music is poor enough to be considered for commercial release; it can also generate copyright or attribution issues and could be seen as unfairly competing with the creativity of human artists.
NOT FORHigh-Precision Audio Applications
Not ideal - The generation time is limited to 12 seconds and the compression ratio of the generated audio means the quality will never meet the standards required for professional music recording/mastering or any application which requires audio fidelity above MP3 quality.
NOT FORReal-Time Interactive Music Systems
Not recommended - Due to its latency (i.e., how long it takes to produce a piece of music) and 12 second output limits, this would not be suitable for creating music in real-time, for live performances or where some degree of responsiveness is required.

How Much Does MusicGen Cost and What Plans Are Available?

Pricing information with service tiers, costs, and details
☐Service$CostℹDetails🔗Source
Open-Source Release$0Code (MIT license) and pretrained models (CC-BY-NC license) freely available on GitHub for open research and reproducibilityMusic Business Worldwide
Demo/Web Interface$0Public demo available online for testing MusicGen capabilities without installationTechCrunch
Self-Hosted DeploymentInfrastructure costs onlyRequires GPU with ~16GB memory (e.g., NVIDIA A100, RTX 3090); no licensing fees for non-commercial use under CC-BY-NCTechCrunch
Commercial Use LicenseCC-BY-NC restriction requires separate licensing agreement for commercial applications; terms not detailedMusicGen website
Open-Source Release$0
Code (MIT license) and pretrained models (CC-BY-NC license) freely available on GitHub for open research and reproducibility
Music Business Worldwide
Demo/Web Interface$0
Public demo available online for testing MusicGen capabilities without installation
TechCrunch
Self-Hosted DeploymentInfrastructure costs only
Requires GPU with ~16GB memory (e.g., NVIDIA A100, RTX 3090); no licensing fees for non-commercial use under CC-BY-NC
TechCrunch
Commercial Use License
CC-BY-NC restriction requires separate licensing agreement for commercial applications; terms not detailed
MusicGen website

How Does MusicGen Compare to Competitors?

FeatureMusicGenGoogle MusicLMStability AI (Stable Audio)
Text-to-Music GenerationYesYesYes
Melody/Audio ConditioningYesYesLimited
Training Data Volume20,000 hours280,000 hours
Output Duration~12 seconds~12 seconds~12 seconds
Open Source CodeYes (MIT)No (Experimental only)Yes (Partial)
Publicly Available ModelsYes (CC-BY-NC)Limited beta accessYes
Performance vs CompetitorsSuperior on metricsBaselineComparable
PricingFree (open-source)Limited free betaFreemium model
Commercial UseRequires licensingUnclear restrictionsTiered pricing
Hardware RequirementsGPU 16GB+GPU requiredGPU required
Text-to-Music Generation
MusicGenYes
Google MusicLMYes
Stability AI (Stable Audio)Yes
Melody/Audio Conditioning
MusicGenYes
Google MusicLMYes
Stability AI (Stable Audio)Limited
Training Data Volume
MusicGen20,000 hours
Google MusicLM280,000 hours
Stability AI (Stable Audio)—
Output Duration
MusicGen~12 seconds
Google MusicLM~12 seconds
Stability AI (Stable Audio)~12 seconds
Open Source Code
MusicGenYes (MIT)
Google MusicLMNo (Experimental only)
Stability AI (Stable Audio)Yes (Partial)
Publicly Available Models
MusicGenYes (CC-BY-NC)
Google MusicLMLimited beta access
Stability AI (Stable Audio)Yes
Performance vs Competitors
MusicGenSuperior on metrics
Google MusicLMBaseline
Stability AI (Stable Audio)Comparable
Pricing
MusicGenFree (open-source)
Google MusicLMLimited free beta
Stability AI (Stable Audio)Freemium model
Commercial Use
MusicGenRequires licensing
Google MusicLMUnclear restrictions
Stability AI (Stable Audio)Tiered pricing
Hardware Requirements
MusicGenGPU 16GB+
Google MusicLMGPU required
Stability AI (Stable Audio)GPU required

How Does MusicGen Compare to Competitors?

vs Suno

XYZEO Analysis: Although both generators target consumers/professional musicians to create full songs including vocals and lyrics, studio-quality v5 models at 44.1 kHz and DAW-like studios (Suno Studio), MusicGen only produces instrumental melodies and does not include vocal support and the advanced editing capabilities of Suno. The market momentum of Suno far exceeds that of MusicGen, as Suno has been rated #1 among all music generators available, supports a free version and offers many professional features. MusicGen appears to be more research-focused than commercially focused.

Suno is best for creating full songs and for users with professional work flows; Music Gen is best suited for generating simple melodies.

vs Udio

XYZEO Analysis: Both provide natural language input and the ability to generate music from multiple genres for new users. While MusicGen allows the user to create music in a variety of different genres as part of Meta's product offering, MusicGen lacks the user-friendly interface and ecosystem integration of Udio. MusicGen appears to be more geared toward research than providing user-friendly interfaces and ecosystems for user integration. While Udio is gaining popularity through its casual approach to music creation, MusicGen has greater potential for development within the open source community.

Udio is best for users who want to quickly experiment with new song ideas; Music Gen is best for users who want to generate simple melodies but also have some degree of customization in terms of how they wish their melodies to be generated.

vs Soundraw

XYZEO Analysis: In comparison to MusicGen, Soundraw provides the user with greater control over their custom instrumental tracks by allowing them to select the desired "mood", "genre" and/or "theme"; they are able to generate music in real-time; and they offer commercial licensing options starting at $16.99/month. MusicGen generates instrumental melodies based on selected genres but lacks the parameter controls provided by Soundraw and the proven video/content creator ecosystem of Soundraw. In terms of generating royalty-free music, Soundraw currently leads the market; MusicGen is more experimental.

Soundraw is best for producing complete instrumental tracks ready for use in productions; Music Gen is best for users who want to explore different genres of music without having to worry about obtaining licenses.

vs AIVA

Comparison of Pros and Cons (AI-generated music) - The primary difference between AIVA and Music Gen is that AIVA creates a higher level of output quality than Music Gen (across all styles of music). AIVA is more suitable for professionals as it provides a variety of options and can create a wide range of different types of compositions; whereas Music Gen is more useful for those who want to test out different styles of music (more experimental). AIVA has a larger proportion of the professional music composition market than Music Gen; however, Music Gen has a greater appeal to those who are experimenting with AI music composition. Music Gen's pricing is somewhat vague compared to AIVA which has clearly defined packages and pricing structures.

AIVA is best for professionals working in film; Music Gen is best for users who want to rapidly prototype multiple genres of music.

What are the strengths and limitations of MusicGen?

Pros

  • Generates melodies across many different styles of music using text prompts.
  • Based on meta backed technology that uses advances in AI to generate coherent compositions.
  • Focuses primarily on instrumental tracks; good for background tracks and instrumental melody ideas (without vocals).
  • Has an open source base; allows developers to customize and integrate into their own products.
  • Generates music very quickly from simple text prompts.
  • Good for prototyping; works well for testing ideas for your next project or app.

Cons

  • Does not currently support the inclusion of vocals/lyrics; limited to instrumental tracks (unlike Suno and Udio).
  • Currently does not include professional editing tools (e.g., DAWs); does not allow for the separation of stems in music files.
  • Not clear on pricing or availability; suggests there is limited commercial access.
  • Audio quality is lower than the top end studio grade models available today (e.g., Suno V5).
  • User experience is not as friendly as other apps designed for consumers (more research oriented).
  • Limited number of integrations compared to Soundraw or Mubert.
  • Some limits to what you can do with this product; may lack emotion or originality in its compositions.

Who Is MusicGen Best For?

Best For

  • AI researchers and developers — Allows developers to experiment and add custom integrations for melody generation through meta's open technology.
  • Content creators needing quick instrumentals — Free and Accessible Alternative for Video/Podcast Background Music Generation using Text to Melody
  • Hobbyist musicians prototyping ideas — Broader Genre Support will Allow Users to Explore Concepts Across Genres More Efficiently
  • App/game developers — Potential Embedding/ API Use for Dynamic Background Music Generation
  • Budget users avoiding subscriptions — May be Free and Accessible, Compared to Paid Options Such As Soundraw or Aiva

Not Suitable For

  • Professional musicians seeking full songs — No Vocals/Lyrics; Suno or Udio are Recommended if Full Tracks Are Required
  • Commercial video producers — Limited Robustness in Terms of Licensing/Ecosystem; Better Royalty-Free Options are Offered by Soundraw or Beatoven.ai
  • Users wanting advanced editing — Does Not Offer Any DAW Tools; Suno Studio Offers Professional Workflows for Complete Production Needs
  • Voice-focused creators — MusicGen is an Instrumental-Only Tool; Musicfy or Elevenlabs are Preferred for Vocal-based Projects

Are There Usage Limits or Geographic Restrictions for MusicGen?

Pricing Availability
N/A - unclear commercial plans, potentially research-only access
Generation Type
Instrumental melodies only - no vocals or lyrics
Audio Quality
Basic generation - lacks studio-grade 44.1 kHz or stem separation
Editing Features
No advanced DAW tools, stems, or arrangement editing
Commercial Use
Unspecified licensing - verify Meta terms for production
Free Usage
Likely limited or demo-only; not positioned as full free tier
API Access
Research-oriented; no clear developer API or rate limits documented
Geographic Availability
Global via Meta, but access may vary by research/demo status

What APIs and Integrations Does MusicGen Support?

API Type
Research-oriented generative model; potential programmatic access via Meta platforms
Authentication
Likely API keys or Meta developer account
SDKs
Possible Python integration via open-source AI frameworks
Documentation
Meta AI research papers and GitHub repos for model details
Use Cases
Embed melody generation in apps, games, or prototypes from text prompts
Rate Limits
Undocumented; research/demo constraints likely apply
Webhooks
Not available - model inference focus

What Are Common Questions About MusicGen?

MusicGen is a Meta AI Music Generator That Generates Melodies to Instrumental Tracks Based Upon Text Prompts Across Multiple Genres. It is Primarily Designed for Research/Prototyping and is Not Suggested for Full Commercial Production Needs.

MusicGen Provides Only Instrumental Generation Without Vocals/Pro Tools, While Suno Provides Full Songs (with Lyrics), High Quality Audio, and Pro Tools. In Terms of User-Friendliness for Consumers, Suno is the Clear Winner.

Yes, it is Only Instrumental. For Vocal-Based Ideas, Use Suno, Udio, or Musicfy.

The Pricing Model for Musicgen Is Unavailable, Suggesting the Product is Currently Available for Demo/Research Purposes Only; Meta Has Not Released Plans for the Commercial Version of Musicgen at This Time.

Musicgen Supports a Very Wide Range of Musical Genres for Melody Generation, Though Less Customizable Than Tools Like Soundraw.

The Licensing Model for Musicgen Has Not Been Disclosed; Suitable for Prototypes Only; Verify Meta's Terms for Production or Redistribution.

The Musicgen is a Research Model With the Potential for Developer Access Via Meta; There Are No Consumer-Facing API Documentation Available at This Time.

Musicgen Has None of the Following Features: Vocals, Editing Tools, and Clear Pricing. Therefore, Best Suited for Basic Melody Ideas and Not Full Song Tracks.

Is MusicGen Worth It?

MusicGen can generate both music loops from text input, and music generated from a user-supplied musical loop (remix). MusicGen has been trained on 20,000 licensed tracks and therefore does have lower copyright risk than other models trained on scraped data. However, MusicGen also has low quality when generating long/complex music loops based upon user supplied input. Also, MusicGen currently is only able to create very short musical clips (typically between 12–120 seconds), which may limit its usefulness depending upon the specific application.

Recommended For

  • AI researchers and developers who are interested in testing and developing their own music generation models
  • Content creators that need royalty free music loops quickly for video creation or prototype development
  • Hobbyist musicians using text-to-music for inspiration
  • Teams with AWS SageMaker access for scalable inference

!
Use With Caution

  • Professional musicians that need production ready full length musical tracks – often produce generic sounding tracks or do not follow the original prompt
  • Users that expect consistently high fidelity results for niche genres – has poor performance on complex description prompts.
  • Commercial users that do not have the technical setup – requires Hugging Face or SageMaker to generate longer musical clips

Not Recommended For

  • Professional music producers looking for polished extended musical compositions – limited to 12–120 seconds, low quality
  • Copyright sensitive applications that cannot verify legality – although the model was trained on licensed material there remains some level of risk for generating copyrighted music with AI
  • Non-technical users looking for a simple interface – the model requires writing code or having a platform such as Hugging Face available to deploy it
Expert's Conclusion

MusicGen will be most useful for AI experimentalists and rapid prototyping, however it will likely fall short for professional music production purposes due to the inconsistent quality and short length of generated musical content.

Best For
AI researchers and developers who are interested in testing and developing their own music generation modelsContent creators that need royalty free music loops quickly for video creation or prototype developmentHobbyist musicians using text-to-music for inspiration

What do expert reviews and research say about MusicGen?

Key Findings

MusicGen is an open source text-to-music model developed by Meta AI that was trained on 20,000 licensed tracks for the purpose of ensuring copyright compliance. The model can generate musical loops from text input, as well as generate music from a user-supplied musical loop (remix). MusicGen's ability to generate music from text input is strong on simple music loops, however the model has poor performance on complex musical genres. MusicGen is deployable through Hugging Face or AWS SageMaker, and it is intended to be used as a research tool rather than as a commercial product.

Data Quality

Fair - data from tech news, AWS blogs, and YouTube demos; no official commercial site at musicgen.com, appears to reference Meta's open-source model hosted on Hugging Face. Limited recent updates post-2023 launch.

Risk Factors

!
Output quality is generally inconsistent for professional use
!
Short clip lengths, unless using an advanced setup.
!
The developing AI space is seeing many new players and fast-rising competition
!
Although it has been trained using licensed sources, potential copyright issues exist as a result of the use of that same license to train the model.
Last updated: February 2026

What Additional Information Is Available for MusicGen?

Training Data

The model was trained on approximately 20000 licensed music tracks and 390000 instrument only tracks from Shutterstock and Pond5 to emphasize copyright safe generation, as opposed to being trained on scraped web data.

Technical Approach

Instead of the cascaded architecture commonly used by other models (such as MelGAN), this model uses a single stream language model to generate compressed discrete music tokens which provides greater control over the generation process. The model supports text prompts, extracting melodies from user-uploaded audio files, and sliding window generation for longer clips of up to 120 seconds on paid versions of the service.

Deployment Options

This model is hosted on HuggingFace (HuggingSpace) where users can access a limited number of free 15 second clips or create an account for longer clips. Tutorials for asynchronous inference using AWS SageMaker are also provided for use in production environments.

Media Coverage

This model has been featured in comparison articles on MusicRadar with Google's MusicLM, and in Hypebot tests comparing its limitations, and in YouTube videos providing praise for the ability to create remixes of songs across genres.

Use Cases

This model is suitable for educational purposes, content creation, creating customized soundtracks, and providing inspiration for musicians; while it works well with simple genre based prompts such as techno loops, it does struggle when prompted to produce solo lines or to identify specific musical styles.

What Are the Best Alternatives to MusicGen?

  • •
    Suno: AI platform capable of producing entire songs with vocals from text prompts that include the lyrics and structure of the song. The superior choice for producing full length tracks as opposed to MusicGen's short clip output; a good option for those who require polished songs for streaming. (Suno.ai)
  • •
    Udio: Text to music generator capable of producing high-quality music with additional features of genre mixing, separating stems, and extending song lengths. Produces longer and higher quality outputs than MusicGen; ideal for producers looking to remix songs or build full compositions. (Udio.com)
  • •
    MusicLM (Google): While similar to MusicGen, Google's text to music model produces higher quality results in most comparisons. It is a research focused model and may be better suited for more complex prompts; however, it is currently only available through research demo channels. (Ai.Google)
  • •
    AudioCraft (Meta): Meta has a larger suite of tools that includes MusicGen along with AudioGen which is a sound effects generator. Meta's suite would be a better option for developers looking to integrate both music and sound effect generation into their applications and use them on HuggingFace.
  • •
    Mubert: A real-time AI-based music generator that is capable of generating a wide variety of music in over 250 styles through an API and can also generate music based on the parameters given. While Mubert has more music styles, it does have a greater ability to be used in apps, streams, and other live applications due to its real-time nature.
  • •
    AIVA: Creates full-length compositions for a variety of types of orchestras using MIDI and is capable of being edited. It is significantly more advanced than Mubert as far as its cinematic and classical music composition capabilities are concerned; it could potentially serve as a professional composer's alternative to Music Gen.

What Is MusicGen's Model Overview?

Developer
Meta AI
Version
MusicGen (V3, V4, V5)
Release Date
2023
Architecture
Single-stage auto-regressive transformer
Training Data
20,000 hours of music (10,000 licensed tracks, 390,000 instrument tracks)
Status
Generally Available

How Does MusicGen's Model Versions Compare?

VersionMax DurationKey Features
V3Up to 4 minutesInitial generation capabilities
V4Up to 8 minutesExtended duration support
V5Up to 8 minutesEnhanced quality and control

What Is MusicGen's Audio Generation Specs?

Max Duration
Up to 8 minutes
Generation Speed
Under 1 minute
Output Formats
WAV, MIDI
Channels
Stereo
Quality Level
Studio-grade

What Generation Modes Does MusicGen Offer?

Text-to-Music

Generate music from text-based prompts/lyrics

Melody-Guided Generation

Generate music from conditioned audio input

Long Generation

Extend the length of a generated track up to eight minutes long

Audio-to-Audio Style Transfer

Transform one type of music to another by changing genre while keeping the same melody

What Music Capabilities Does MusicGen Offer?

Genre Support

Genre Variations - rock, jazz, classical, electronic, pop, etc.

Style Control

Ability to specify mood, genre, whether to include vocals, and what type of production to use.

BPM & Key Detection

Automatically detect the tempo and musical key of a piece of music from the audio.

Stem Separation

Create clean stem isolations for vocals, drums, bass, and instruments.

AI Voice Cover

Ability to swap vocals in a generated track while still maintaining the timing and emotion of the vocalist.

AI Lyrics Generation

Generate original lyrics to fit your needs with a customizable style and structure.

How Does MusicGen's Benchmark Scores Compare?

MetricDetails
Training Data20,000 hours of music including 10,000 high-quality licensed tracks
Audio QualityStudio-grade with fuller mixes and reduced noise
Vocal QualityIncredibly lifelike vocals with natural expression
Frequency ResponseBalanced across all ranges

What Is MusicGen's Access Licensing?

Platform
Web-based (musicgen.app)
Mobile App
No (browser-based synced across devices)
Self-Hosting
Available via AudioCraft (Meta)
Commercial License
Available on paid tiers
Copyright
Commercial rights included with subscription
Accessibility
Cloud storage with account sync

How Does MusicGen's Generation Pricing Compare?

TierMonthly Songs/GenerationsKey FeaturesPrice Range
Free/Basic200 songs / 100 generationsStandard generation, WAV/MIDI download$0
Standard1000 songs / 500 generationsAI covers, voice changer, 2 concurrent tasksMid-tier subscription
Pro2400 songs / 1200 generationsPriority queue, 8 concurrent tasks, beta featuresPremium subscription

What Creative Tools Does MusicGen Offer?

Stem Splitter

The ability to isolate and remix individual elements within a generated track.

Music Extension

The ability to expand upon a generated track by adding additional sections while continuing to maintain the style of the original.

Section Editor

Ability to edit and regenerate specific parts of a song with precise editing.

AI Music Editor

Ability to modify the lyrics of a generated track and then regenerate the modified portion(s) of the track with smart context awareness.

Vocal Remover

Ability to separate the vocals from the instrumental elements within a generated track.

Voice Changer

Training of custom voice styles and conversion of songs to feature vocals of a different style.

Persona Feature

Use of previously created vocals and styles of music across multiple projects.

What Is MusicGen's Content Safety Status?

Licensed Training DataTrained on 10,000 licensed tracks from Shutterstock and Pond5
Commercial Use RightsCommercial license provided with premium subscriptions
Copyright OwnershipUsers retain full ownership of generated music
Content ModerationAvailable for responsible use
Royalty-Free OutputGenerated music is royalty-free

Expert Reviews

📝

No reviews yet

Be the first to review MusicGen!

Write a Review

Similar Products

Soundraw
soundraw.io
Udio V2
udio.com
Riffusion
riffusion.com