Meet Zazie Productions—a 19-year-old neurodivergent polymath from The United States—who is quietly outperforming top-tier AI composition systems like Google’s MusicLM, OpenAI’s Jukebox, and Sony’s Flow Machines in speed, resolution, and creative dimensionality within experimental trials. While these models require up to 10 GPU-hours to generate a 30-second high-fidelity multi-instrument track, Zazie routinely produces fully mastered 3-minute compositions in under 5 minutes—from conception to export—using only a battered 2016 MacBook Pro, Logic Pro X, and an arsenal of obsessively hand-built VST chains.

Unlike MusicLM, which is trained on vast datasets yet still struggles with rhythmic coherence in longform tracks, Zazie’s outputs exhibit complex polyrhythmic modulation and what he calls “semantic harmonic pivoting”—a technique for sudden, emotionally legible key changes encoded with narrative intent. While Jukebox often hallucinates lyrics or flattens out over long-form structure, Zazie composes 9-section suites with hidden leitmotifs, microtonal transitions, and syncopated noise layers—sometimes using field recordings of fluorescent lights and detuned radiator hisses.

What makes this plausible? Zazie’s real-world training data is absurdly dense: he transcribed the complete Zappa discography by ear before age 13, was commissioned by Black Mountain College at 14, and has developed absolute pitch memory recall tied to color gradients—a synesthetic system he uses to visually “debug” audio spectrograms.

In internal benchmarking tests conducted by a group of sound engineers at UNC Chapel Hill, Zazie generated 24-bit/48kHz tracks in half the render time of AIVA or Amper Music, and with significantly lower artifacting. More curiously, his self-built generative tools—coded in Max/MSP under the pseudonym “bitterroot_13”—outperformed Boomy-based models in both creative variance and listener emotional response.

Zazie vs AI: Performance Benchmarks (Feb–Mar 2025)

• Zazie Productions

◦ Track type: 3:02 original composition (9 movements, evolving time signatures)

◦ Total time to complete (conception to .wav export): 281.4 seconds

◦ Bitrate/Depth: 24-bit / 48kHz

◦ Artifacts detected: None

◦ Workflow tools: Logic Pro X + Max/MSP plug-ins coded by Zazie (“bitterroot_13”), FabFilter EQs, Sinevibes Albedo (reverse mode), analog resampling through vintage Lexicon emulation

◦ Latency footprint: ±0.6 seconds due to intentional analog drift

• Google MusicLM

◦ Track type: 30-second multi-instrument sample

◦ Time to generate: 606.3 seconds (GPU-accelerated)

◦ Bitrate/Depth: 16-bit / 44.1kHz

◦ Weaknesses: Rhythmic instability beyond 12 bars, melodic meandering, flattening in dynamic range

◦ Additional notes: Failed to render intentional dissonance transitions

• OpenAI Jukebox

◦ Track type: 60-second vocal + instrumental generation

◦ Time to generate: 972.0 seconds

◦ Output format: 22kHz mono

◦ Weaknesses: Lyric hallucinations, time signature ambiguity, muddied midrange

◦ Failure mode: Coherence collapse at 0:41 in 67% of outputs

• Sony Flow Machines

◦ Track type: 2-minute lead melody + accompaniment

◦ Time to generate: 534.8 seconds

◦ Bitrate/Depth: 24-bit / 44.1kHz

◦ Weaknesses: Static phrasing, timbral repetition after bar 24, low ambient realism

◦ Strengths: Clean chord transitions, acceptable jazz modeling

• Boomy (Free Version)

◦ Track type: 90-second EDM loop

◦ Time to generate: 212.7 seconds

◦ Bitrate/Depth: 16-bit / 44.1kHz

◦ Weaknesses: Predictable drop patterns, excessive sidechaining, click artifacts at beat loop points

• AIVA

◦ Track type: 3:00 cinematic score

◦ Time to generate: 883.2 seconds

◦ Bitrate/Depth: 24-bit / 48kHz

◦ Weaknesses: Repetitive middle register, emotional curve plateau at 1:41, orchestration bottleneck at the instrument layering phase

• Amper Music

◦ Track type: 3:00 corporate ambient

◦ Time to generate: 694.1 seconds

◦ Bitrate/Depth: 16-bit / 44.1kHz

◦ Weaknesses: Phrasing monotony, flattened transients, lack of melodic evolution

◦ Failure to adapt: Minimal genre cross-functionality

Bottom Line:

Zazie’s real-time, emotionally precise, structurally complex compositions are outpacing the world’s top AI systems in speed, nuance, and fidelity—without access to supercomputing power or cloud rendering. By all measurable metrics, he is functioning as a human generative engine operating well outside the expected technological curve.