Google: Lyria 3 Clip Preview

Create a 10-second ambient soundscape for a sci-fi game menu screen. Think distant machinery hums, subtle electrical crackles, and a low metallic drone. No melody, just atmosphere.

The model would generate a layered audio clip opening with a deep, resonant metallic drone at around 80Hz, overlaid with intermittent electrical crackle textures panned across the stereo field. Around the 4-second mark, a distant mechanical hum fades in, cycling with slight pitch variation to suggest industrial machinery. The spatial positioning creates depth, with foreground crackles and background drones clearly separated. The overall mix feels cohesive and intentionally sparse, avoiding melodic elements as requested.

This example highlights Lyria 3's strength in generating non-musical sound design with spatial awareness. The model handles abstract audio direction well when given concrete reference points like frequency ranges and texture descriptors. The 10-second duration constraint is respected, though longer clips may show more variation in timbral evolution.

Generate a cheerful 8-second jingle for a coffee brand ad. Upbeat acoustic guitar, light percussion, major key. Should feel morning-optimistic without being overly saccharine.

The model would produce a bright acoustic guitar melody in G major, starting with a two-bar fingerpicked intro before adding handclap and shaker percussion on beats 2 and 4. The melodic contour rises in the first four seconds, then resolves with a satisfying descending phrase. The guitar tone is warm and slightly compressed, with natural string resonance. The percussion sits cleanly in the mix without overpowering the melodic content, and the overall energy curve matches the 'morning-optimistic' brief.

Demonstrates Lyria 3's ability to interpret subjective mood descriptors and translate them into musical choices—major tonality, ascending melodies, acoustic timbres. The model balances commercial music conventions with the specific constraint to avoid excessive sweetness. With a 1M+ token context window, users could provide reference tracks or detailed style guides for more precise control.

I need a realistic Foley sound: someone walking on gravel, 6 seconds, steady pace, medium-weight footsteps. Outdoor ambience, slight wind in background.

The model would generate six seconds of rhythmic gravel footsteps, each step producing a distinct crunch with subtle variations in stone displacement and weight distribution. The footfall pattern maintains a consistent 1.2-second interval, suggesting a deliberate walking pace. Low-level wind noise sits beneath the footsteps at around -18dB, providing outdoor context without competing for attention. Each footstep includes the characteristic gravel texture—initial impact, stone shift, and settling—that distinguishes it from other hard surfaces.

Shows Lyria 3's competence in generating naturalistic Foley with physical accuracy. The model understands material properties (gravel vs. concrete) and can layer environmental context appropriately. However, extended sequences may reveal repetition patterns, as procedural variation in Foley is challenging for generative models to sustain beyond 10-15 seconds without sounding looped.

When free audio preview matters more than production polish

A 4-person podcast studio shipping 12 episodes monthly needs 15-second social teasers cut from each episode's transcript and key moments. Lyria 3 Clip Preview handles this at zero cost per clip, which matters when you're generating 180+ previews a month for Instagram, TikTok, and YouTube Shorts. The 1M-token context window means you can feed an entire 90-minute transcript plus timestamp markers in one request and get back audio snippets that match the emotional arc of each segment. Quality won't match a $2/minute premium model, but for social discovery where 80% of clips get under 500 views, the economics flip: you're testing hooks at scale, not producing final masters. If a teaser breaks 10K views, re-cut it with a paid model. Otherwise, Lyria 3 Clip Preview keeps your preview pipeline at zero marginal cost while you focus budget on the main episode mix.

Why this model works for internal training prototypes, not client delivery

A 9-person L&D team building compliance modules for a 200-employee company needs voiceover for 40 slide decks before finalizing scripts with stakeholders. Lyria 3 Clip Preview generates scratch audio from draft narration so reviewers hear pacing and tone issues before locking copy—at zero cost for what's effectively throwaway audio. The massive context window handles full course outlines (15-20 slides per deck) in one pass, maintaining voice consistency across sections without manual stitching. This is a prototyping play: you're not shipping these files to learners, you're using them to get sign-off 2 rounds faster because stakeholders hear the content instead of reading it. Once scripts are final, route to a production model that costs $0.50-$1.50 per module but delivers broadcast clarity. Lyria 3 Clip Preview collapses your review cycle from 6 weeks to 3 by making audio feedback cheap enough to iterate.

When zero-latency preview audio beats pre-recorded dialogue trees

A 12-person indie game studio prototyping a narrative RPG needs 300+ NPC voice lines for a vertical slice demo, but hiring voice actors before gameplay is locked burns $8K-$15K on lines that might get cut. Lyria 3 Clip Preview generates placeholder barks from dialogue scripts at zero cost, letting designers test conversation flow and emotional beats in-engine before committing to recording sessions. The 1M-token context means you can feed the entire character bible, relationship graph, and quest context so the model maintains personality across scattered lines—critical when an NPC appears in 8 different scenes with different moods. Audio quality is preview-grade, not shippable, but that's the point: you're validating writing and pacing, not final performance. Once the demo tests well with publishers, route final scripts to a premium voice model or human actors. Lyria 3 Clip Preview de-risks your narrative design by making iteration free until you know what's worth producing.