Lyria 3 - Google DeepMind's next-generation AI music generation model

What is Lyria 3?

Lyria 3 is Google DeepMind's latest AI music generation model, now integrated into the Gemini app. Compared to its predecessor, Lyria 3 represents a significant breakthrough: users no longer need to write lyrics; they can generate a 30-second high-quality music clip with automatic lyrics, music composition, and vocals with a single click, simply by providing a text description, uploading a photo, or video. The system supports fine-grained style control, encompassing rhythm, mood, vocals, and other elements, and automatically generates a matching cover art for each track. Lyria 3 incorporates SynthID watermarking technology to track and verify AI-generated content, and also features copyright protection mechanisms to prevent direct imitation of existing artists' work. The model currently supports eight languages: English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese. It is free for Gemini users aged 18 and older and also provides AI-generated music for YouTube Dream Tracks, making it suitable for short video creation, personal entertainment, and creative expression.

Main features of Lyria 3

• Multimodal music generationIt supports generating 30-second high-quality music clips through three methods: text description, uploaded photos, or videos. AI automatically matches the mood and style.
• Automatic lyrics creationNo lyrics are required from the user; the system automatically generates complete lyrics and vocals based on the prompts.
• Fine-tuning of styleIt allows adjustment of music style, vocal performance, tempo, and other elements to achieve more realistic and complex music arrangements.
• Intelligent Cover GenerationEach song's cover art is automatically generated by Nano Banana AI.
• Multilingual supportCurrently, it supports eight languages: English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese.
• Copyright security protectionBuilt-in filters prevent the generation of content that is too similar to existing works, and when mentioning specific artists, it is only used as a style inspiration rather than a direct copy.
• SynthID watermark trackingAll generated music is embedded with an imperceptible digital watermark, and uploaded audio can be verified to determine whether it was generated by Google AI.
• Multi-platform integrationIt is integrated with the Gemini app and YouTube Dream Track, and supports downloading MP3 audio or MP4 video formats.

The technical principles of Lyria 3

• Multimodal understanding architectureIt can process text, image and video input simultaneously, and analyze the content's emotion and scene through a visual-language model, transforming it into music generation instructions.
• End-to-end music generationIt adopts a unified neural network architecture to integrate lyric generation, melody creation, arrangement, and vocal synthesis into a single process, rather than processing them in stages.
• SynthID audio watermarkThe generation process embeds a digital fingerprint that is imperceptible to the human ear. The identification information is hidden in the audio waveform through frequency domain transformation technology, which supports subsequent traceability and verification.
• Copyright protection mechanismBased on a large-scale audio fingerprint database and similarity detection algorithm, the generated content is compared with existing copyrighted works in real time, triggering filtering or adjustment mechanisms.
• Style control and constraint generationBy using conditional generation technology, user-specified parameters such as style, rhythm, and emotion are injected as constraints into the generation process to ensure that the output meets expectations.
• Nano Banana Visual GenerationIt integrates an image generation model to automatically generate matching cover artwork based on music style and lyric theme.

How to use Lyria 3

• Access pointOpen the Gemini app (web or mobile), find and click the "Music" option in the bottom toolbar.
• Select generation methodIt supports three input methods: directly entering a text description, uploading a local photo, or uploading a video file.
• Write promptsDescribe the desired music style, mood, theme, and scene in words, such as "a nostalgic Afrobeat song about childhood memories".
• Waiting to generateAfter submission, the system will automatically complete the lyrics, composition, arrangement, and vocal synthesis within approximately 10-60 seconds, generating a 30-second music clip.
• Preview and AdjustmentsPlay the generated music. If you are not satisfied, you can modify the prompts or adjust parameters such as style and rhythm to regenerate.
• Download and saveIt supports downloading works as MP3 audio format or MP4 video format with cover art, making it easy to share to social media platforms.
• Watermark VerificationTo verify whether an audio segment is AI-generated, you can upload it to Gemini for SynthID watermark detection.
• Usage restrictionsMust be 18 years of age or older. Free users have a limited number of generation attempts. Google AI Plus/Pro/Ultra subscribers can get a higher limit.

Lyria 3 project address

• Project official websitehttps://deepmind.google/models/lyria/

Application scenarios of Lyria 3

• Short video background musicQuickly generate personalized background music for platforms such as TikTok, YouTube Shorts, and Instagram Reels to enhance content appeal.
• Social media contentAutomatically generate personalized background music for photos or videos such as travel vlogs, pet daily life, and food reviews to enhance emotional expression.
• Personal entertainment creationOrdinary users can easily create personalized musical works such as birthday greetings and anniversary theme songs without any musical background.
• Podcasts and Audio ContentGenerate intro and outro music or transition sound effects for podcasts, audiobooks, audio commercials, etc.
• Games and interactive contentGenerate customized background music and ambient sound effects for indie games, interactive stories, and virtual scenes.
• Marketing and Brand ContentBusinesses can quickly generate original music that matches their style for brand events, product launches, and advertising videos, reducing copyright costs.