Lyria 3.5 is Google DeepMind's newest music model. From a single text description it creates a complete song — verses, chorus, vocals and a full mix. Tracks run up to 3 minutes, and generation takes just 30–60 seconds.
You can guide the model in different ways. Paste in your own lyrics and Lyria 3.5 will sing them — clearly, in many languages. Leave the lyrics out, and it will write its own based on your description. Ask for a vocal-free version, and you'll get a clean instrumental for background use.
One standout feature is music generated from a photo. Upload an image and the model matches the mood, tempo and sound to the picture: a calm landscape becomes ambient, a night cityscape turns into electronic music. Every track costs 13 credits, regardless of length.