Lyria 3.5 music generation lands on the Gemini API: clip + full song
Text or an image in, 44.1kHz stereo out. The clip model loops 30-second beds; pro writes full vocals. Pricing's unposted, and SynthID rides every file.
Copy markdown
Two models, two jobs
`lyria-3.5-clip-preview` is tuned for 30-second clips, loops, and fast iteration; `lyria-3.5-pro-preview` writes full-length songs with tighter structure and natural vocals. Both hit public preview on the Gemini API on Sept 3.
Text or an image goes in
Both take text prompts and images as input and return high-fidelity 44.1kHz stereo audio, with lyrics as a text output. Hand it a mood board or a brand still and get a scored track back — no DAW in the loop.
What you could ship this weekend
The clip model's loop focus is built for background beds: adaptive game music, a photo-to-soundtrack app, on-brand jingles, or ambience that regenerates per session. Reach for pro when you need a finished, vocal-led track.
Pricing for 3.5 isn't posted yet
Google hasn't published 3.5 rates. Use the shipping Lyria 3 preview tier as your reference — reportedly $0.04 per 30-second clip and $0.08 per full song — and budget around that until the 3.5 numbers land.
SynthID rides every track
Every generation carries an inaudible SynthID watermark by default. It matters if you plan to resell output or need provenance you can verify downstream.