Google Dropped Lyria 3 AI Music but Sunauto v3 STOLE the show!

Google’s Lyria 3 (spelled “LIIA/Lyria 3” in the transcript) lands inside Gemini with a clear tradeoff: it’s a polished, safety-locked music generator that can produce only up to 30 seconds, but it brings a notable “watermarked” output that’s detectable by computers as AI-made. In Gemini, Lyria 3 delivers catchy, genre-flexible tracks—90s rap, Latin pop, and eight-bit metal—while refusing requests for specific artists or artist-specific sounds. The model’s 30-second limit also shapes expectations: prompts can yield something that sounds professional and clinical, yet it’s inherently built for short, gimmicky results rather than full-length songwriting.

The broader Gemini push matters because the transcript ties Lyria 3’s release to Gemini 3.1 Pro, described as a significant intelligence upgrade with measurable benchmark gains. The standout capability highlighted for Gemini 3.1 Pro is its ability to generate SVGs—illustrated with an Xbox controller rendered entirely via code—plus a claim that it produces far more detailed 3D-style outputs than the prior Gemini 3 Pro. That matters to music users because it signals tighter multimodal tooling around the same ecosystem: music generation in Gemini is paired with stronger general creative generation elsewhere in the platform.

Still, the transcript’s main comparison is less about Google’s model and more about what competitors can do with longer, more permissive generation. Suno v3 is positioned as the better overall music engine, with the creator arguing it can do “way more” than Lyria 3. The biggest disruption comes from Tsunado AI’s upcoming V3 preview, which the transcript frames as unusually unrestricted. In that preview, the model can be steered with tags that include direct artist inputs—explicitly demonstrated with Juice Wrld—while producing voice-like results that the creator claims are highly accurate. The transcript also notes a legal gray area: U.S. copyright law would likely treat profit-making from voice likeness as problematic, while Tsunado’s terms of service reportedly push responsibility onto users (“don’t generate anything that infringes… and then go trying to make money”).

Beyond artist mimicry, Tsunado V3 is shown generating long-form-feeling tracks from simple prompts: a “eat the periodic table” rock/rap concept, style swaps that mimic Queen, Eminem, Taylor Swift, and Michael Jackson, and even genre mashups like country ballads about herding jellyfish or arena-rock lyrics that drift into trap-like rhythms. The transcript repeatedly returns to a practical point: each model has strengths and weaknesses tied to how it handles tags, tokens, and style nuance. Lyria 3 is praised for polish and safety, but limited by length and restrictions; Tsunado V3 is praised for breadth and creative freedom, but sometimes produces muddier audio or quality drop-offs when the prompt demands complex rendering.

Overall, the transcript paints AI music as a fast-moving marketplace where “best” depends on the job: short, clean, safety-locked tracks in Gemini versus more permissive, tag-driven generation elsewhere—plus a reminder that the legal and quality tradeoffs are still evolving.