Google DeepMind Launches Lyria 3: Generate 30-Second Music Tracks from Photos and Text

Overview

Google DeepMind dropped Lyria 3 on February 18, 2026 — its most advanced generative music model, now available in the Gemini app. Users can input text prompts or upload images to generate complete 30-second tracks including instrumentals, vocals, and auto-generated lyrics.

Key Features

Multimodal input: Generate music from text descriptions or photo uploads
Auto lyrics and vocals: Lyria 3 automatically writes and sings lyrics based on your prompt
Creative controls: Fine-tune genre, vocal style, and tempo
Multilingual vocals: Supports English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese
SynthID watermarking: All generated tracks are invisibly watermarked for AI detection

Availability

Lyria 3 is available in beta in the Gemini app, with priority access for Google AI Pro and Ultra subscribers. Developers can access the model via the Gemini API and Google AI Studio.

Safety and Ethics

Every track generated by Lyria 3 is watermarked with SynthID technology, allowing users and platforms to verify whether audio was created by AI. Gemini can analyze uploaded audio files to detect SynthID watermarks.

Source: Google DeepMind (@GoogleDeepMind) on X

AI 3d ago 2 min read

Vision Banana turns image generators into all-purpose vision models

This paper argues that image generators may be turning into the vision equivalent of large language models. DeepMind says Vision Banana, built on Nano Banana Pro, beats or rivals specialist systems such as Segment Anything and Depth Anything on 2D and 3D tasks after lightweight instruction tuning.

#google-deepmind #computer-vision #vision-banana

AI sources.twitter 2d ago 1 min read

DeepMind trains a 12B model across four regions 20x faster

Google DeepMind’s new training stack matters because datacenter boundaries are turning into frontier bottlenecks. Decoupled DiLoCo trained a 12B Gemma model across four U.S. regions on 2-5 Gbps links, more than 20x faster than conventional synchronization while holding 64.1% average accuracy versus a 64.4% baseline.

#google-deepmind #gemma #distributed-training

AI Feb 21, 2026 2 min read

Google Brings Lyria 3 to the Gemini App in Beta for 30-Second AI Music Generation

Google announced on 2026-02-18 that Lyria 3 is rolling out in beta in the Gemini app. The feature generates 30-second tracks from text or images, attaches generated cover art, and embeds SynthID watermarking for AI-content identification.

#lyria-3 #gemini-app #music-generation