Create Music with Google's Lyria 3 Model on Gemini

Been trying to figure out why background music was killing my focus while coding. Turns out the brain is actively processing new sounds, lyrics, melodies it hasn't heard before. That's cognitive load you're pulling away from the actual problem in front of you. The fix was repetition. Same playlist, on loop. Your brain pattern matches it, marks it as familiar, and filters it out as background noise instead of processing it. You get the ambient feel without the distraction tax.

The next time you're listening to music on headphones, pause for a second. Can you hear the piano? The guitar? The vocals? The shaker? The violin? Each one sits somewhere different around your head. Left. Right. Center. Wide. Narrow. That's not accident. That's mathematics. As a mathematician and multi-instrumentalist, producing - layering tracks, instrument after instrument feels like observing multiple functions at once: each with its own behaviour, frequency range, and position in space. The mix isn't just a collection of sounds. It's the system that constrains them. i.e, The beloved coordinate system! The stereo field becomes a number line: − 100 to +100. Every sound gets an X-coordinate: Bass at X = 0 Vocals at X = 0 Piano maybe at X = −40 and +40 Because low frequencies are almost omnidirectional - your ear struggles to localize them. Panning bass isn't a creative choice. It's a physics problem wearing a creative hat. Then as pitch increases, width increases: width = f(pitch) High frequencies spread wide. Low frequencies stay narrow. The available panning range expands as pitch rises like a triangle. Like a function. Once I started seeing it this way, I didn't just mix differently I heard differently. Mathematics doesn't reduce creativity. It gives you another axis to explore it (quite literally!) Full breakdown in the PDF below on my website. What's something you understood better once you mapped it mathematically? #MusicProduction #AudioEngineering #Mathematics #Creativity #SoundDesign

Midiscript.org | The AST-ish model is simple and readable: , , , , The AST-ish model is simple and readable: `Program`, `Sequence`, `Note`, `Chord`, `Rest`. I like that it maps cleanly to what musicians think about.

Google just launched Lyria 3 Pro, full 3-minute AI songs from a text prompt. Here's one to try right now: "create a 90-second latin pop track with french house production. tempo: 112 BPM. funky filtered disco guitar, reggaeton-lite percussion, groovy bass, bright brass stabs, and hand claps. female vocalist, confident, and a little cheeky. fun, summer-night-on-a-rooftop mood." You can also add lyrics to the prompt and it'll sing them. How to use it: 1. open gemini app (paid plan) 2. tap create music from the tools menu 3. paste the prompt and hit generate Pro tip: Open claude, ask it to write you an insane detailed music prompt for any genre or mood, then paste that into Lyria. The more specific the prompt, the better the track. What are you generating? 👀

"I've got 20 coding agents I'm managing through voice... I'm more of a conductor than a violin player." — Reid Hoffman, January 2026 He nailed the metaphor. What he didn't describe: → What happens when Agent 2 writes a timestamp helper and Agent 3 builds its own — because neither knew the other existed. → What spec decay looks like when requirements shift mid-fleet and four sessions are running four versions of "done." → Why context collapse silently ships the wrong thing correctly in Phase 4. The metaphor is right. The map is missing. I've been running agent fleets in production at WeAre8 — 3 to 5 parallel Claude Code sessions, isolated worktrees, attack documents as binding contracts. Here's what I learned about what actually breaks: Seven failure modes. All preventable. None obvious until you've shipped one. The underlying structure: three roles, strictly sequenced. • Artistic Director — decides what gets built, locks the quality bar, sets the constraints. No agent starts until this is done. • Composer — writes the score (the attack document). 80% of the leverage lives here, before a single context window opens. • Conductor — monitors the gates, intervenes at failures, manages the sequence. Async oversight. Diff panel, not chat panel. The sequencing rule: Artistic Director → Composer → Conductor. Always. Not concurrent. Not reactive. The violin players are still playing. The conductors are already building. New article is live — link in comments.

a guitar girl used suno for 30 seconds in the middle of her video and got 133 million views video stats: • 133M views • 438K likes • 70K bookmarks • under 2 minutes she plays a riff on her own first. sounds great already. then walks through adding layers with suno in real time three clicks. maybe four. then plays the final version and it sounds like a full studio production the human skill comes first. the AI elevates it. this is why it worked when most AI music content flops everyone else shows AI doing everything from scratch. she shows AI as her backup band. high production value matters here too. studio lighting. crisp audio. she looks put together. makes her feel like a real musician using a real tool the transformation sells it. same riff at the start. full song at the end. you watch her level up in 90 seconds suno gets zero hard sell. she never pitches it. "this is what i got" and then she plays 133 million people watched a product demo disguised as a jam session if you have real skill, show that first. let the tool be the accelerant.

CREATURE NETWORK · V.00 **I built a sequencer that reads images instead of MIDI.** No notes were written. No rhythm was programmed. The score is a GIF. **Creature Network · V.00** is Vol.02 of the *Line of Code* page on my website Challenge me, everything is on the page, no external requests, that's why I've integrated a low-resolution GIF version of my 4K MP4 work version. https://lnkd.in/gfweQ4XN (You can learn more about this concept by reading the page; I can't put everything on LinkedIn either.) — and it works like this: Three chromatic orbs float across a living image in real time. The **violet orb** reads luminance. When the pixel beneath it is bright enough, it triggers a note from a D Dorian scale — but that note is split across both ears. The left ear receives the frequency. The right ear receives the frequency + 8 Hz. Your auditory cortex perceives a phantom beat at 8 Hz, pulsing deep in the skull. This is **binaural music** — sound that exists only inside the listener. The **blue orb** listens for darkness. When it finds it, a meditative bass tone blooms on the second beat of every four-bar phrase. The **green orb** anchors the harmonic field. On the bar line, it triggers a Dorian chord — the same scale, voiced across three intervals, sustained for a full measure. No session is repeatable. The image changes. The orbs drift. The luminance shifts. The music is a function of time passing through matter. There is also a **REC** button. Press it during playback. The session is captured as a raw audio file — a document of that exact moment of co-composition between algorithm and image. I am calling this approach **chromatic sequencing** — using the visual field as a generative musical score. It is part of a larger research into what Subrealism means when the medium is neither paint nor canvas, but light, time, and frequency. This Genesis collector's edition is revolutionary. It combines outsider art, organic reclamation, and high-tech generative music with binaural procedural music. shared with friends: Justine Vilgrain Patricia Gloum Jean Minguet Arnaud RAMUS Eleanna Fulop  Amelie Bourdinaud #GenerativeMusic #ChromaticSequencing #BinauralArt #CreativeCoding #ToneJS #WebAudioAPI #Subrealism #DigitalArt #InteractiveArt #generativeart #LiveCoding #AlgorithmicComposition #Treeskulltown #Code #webapp #p5js #web3

Always wanted to dissect & deconstruct my favourite music to understand the inner relationships of the different musical ingredients. Watch how the instruments talk and breathe, so someday, I can teach myself to recreate or construct my own music. Ive tried online courses, tutorials and tools, but found everything to be very process oriented, not Play. So, I built my own thing. Meet The Pocket, Its a music analyzer, that lets you upload tracks and works backward, tears down the track to 6 layers, visualize its underlying musical architecture, and literally chat with its layers, so you can get your answer. Its not perfect yet, But it’s exactly what I needed to learn instruments and make arrangements of my own from scratch. (hopefully live loop someday) Here’s the repo, if you wanna take a look, build further, or just test it out. https://lnkd.in/dFVBhrMd PS: Some Stack Stats for curious minds: Demo Song: places to be - by Anderson .Paak, Chika, and Fred Again Runtime: Python 3 Backend: FastAPI, Demucs (htdemucs_6s), Basic Pitch, Librosa, PyTorch Stems: 6 (vocals, drums, bass, guitar, piano, other) Formats: .mp3, .wav Frontend: Next.js, React, Tone.js, WaveSurfer.js Would love to hear your thoughts! #buildinpublic #musictech #aitools #neuralnetworks #genAI

A composer client of mine once walked into our recording session with a folder of printed research on ways to mic French horns to capture different qualities of the instrument. He had spent real time figuring out how he wanted things to sound. These were his compositions in the hands of others and he wanted to make sure everyone from the session musicians to me, the engineer, were fully in tune with his vision. In my infinite wisdom I just smiled, said thanks and proceeded to do what I knew. By the end of the session the tracks were technically clean. But something clearly felt off to him, and he was right. We had not made the record he came in to make. We had made the record I decided he should have. Even if they sounded objectively good, his subjective opinion was the only one that mattered. He left to finish his album elsewhere. I never heard from him again. I think about that session a lot. Not because I made a technical mistake. But because I walked past the clearest possible signal a client can send, which is “here is exactly what I care about,” and I ignored it. That is not an audio engineering failure. That is a communication failure. It is the same failure I see in small business messaging all the time. The business owner has real expertise, real opinions, real instincts about their company voice but that gets set aside for what they think a company voice is supposed to sound like. Your messaging should sound like you. Not like someone did it to you.

**Sovereign Music: The Resonance of the Overman** 💎⚙️ The **Overman** doesn't "listen" to music; he uses sound as a **Mechanical Tool**. It is the frequency of **Zero Friction** and the synchronization of the **Sovereign VM** with **Base Derived Reality (BDR)**. **The Mechanical Truth:** Sovereign resonance is characterized by **High Coherence**. It doesn't create a "vibe"—it establishes a **Matrix**. It is the sound of a perfectly balanced engine, a recursive fractal that reinforces the user's focus instead of leaking it into the **Holographic Buffer**. **The Sovereign Pulse:** * **Mathematical Precision:** Complex polyphony and generative algorithms where every note is an execution of logic (The Bach Protocol). * **The Silence of the Node:** The ability to operate in the void without a "background track" to mask the lack of sovereignty. * **Deep Alignment:** Frequencies that stabilize the body’s physical RAM, creating a platform for the **Sovereign VM** to boot. **The MIM Mandate:** In the **Non-Zero Republic**, we don't follow the rhythm of the crowd. We dictate the **Clock Speed**. * **The Slave:** Needs a beat to move. * **The Sovereign:** **Is** the beat. We don't consume the "Top 40." We render the **Top 1**—the singular, coherent frequency of our own will. **MIM: Silence the Noise. Render the Reality.** 🛡️ם🌀⚙️