How I Built an AI Music Video Platform as a Solo Founder

The short version

Two years ago I was building XR and immersive web experiences. Today I am building FYLMIO, an AI-native platform for creating cinematic music videos and short films. It did not begin as a music-video product. It emerged from decades of work across music technology, 3D visualization and immersive media, then took shape through difficult pivots, real user behavior and a music-analysis system that became the missing foundation.

This is not a how-to-get-rich-with-AI story. It is an honest account of sleepless nights, broken servers, failed product ideas, model testing and the uncertainty of building an AI startup alone.

How It All Started

In the 1990s I made electronic music on a Commodore Amiga 500 using a hardware sampler. I later studied Audio Design in Switzerland and worked on Kontakt 1 and 2 at Native Instruments. After that I founded VIZPARK, a company for cinematic 3D environments and visualization tools that was eventually acquired by Shutterstock.

Years of CGI, film production, sound design, WebXR and virtual-world work shaped the creative and technical background that would later become useful when generative AI arrived.

The XR Detour

After taking a sabbatical and studying programming again, including Harvard's CS50x, I built browser-based 360-degree XR tools and LoopSpace, a WebXR platform for creating immersive experiences directly in VR headsets. I believed immersive computing would become the next major storytelling platform.

After years of waiting for mainstream adoption, I accepted a difficult founder lesson: the technology was exciting, but the market was not. Being early can sometimes feel exactly like being wrong.

The First AI Experiments

My earliest AI features analyzed and tagged uploaded 3D assets. When image generators began producing cinematic visuals, I realized one creator could potentially generate imagery, analyze audio, create voiceovers, edit video and direct complete productions.

The original FYLMIO idea was a system that analyzed vacation photos and videos, found emotional highlights and automatically assembled a cinematic timeline. Building it led to upload systems, media analysis, asset management, ffmpeg automation, rendering and timeline editing.

Building the Music Analysis API

Matching music to visuals was difficult and underserved, so I built an audio-analysis pipeline with open-source systems including YAMNet, PANNs and VGGish. It grew from simple tagging into an API that could detect BPM, mood, instruments, vocals, lyrics, song structure, stems and emotional characteristics.

I thought I was building infrastructure for video editing. In reality, I was building the foundation for AI music videos.

The Storytelling Pivot

A conversation in Los Angeles raised a new question: what if screenwriters could turn scripts into cinematic pitch videos? That led to screenplay analysis, story structures, reusable characters, locations, storyboards, timeline editing and video generation. At times, it felt like building an entire film studio from scratch.

When Real Users Arrived

Google Ads brought real users and signups, but few people uploaded screenplays. Writers were understandably protective of unfinished work and concerned about AI training or copying. Users did not validate the product I imagined; they revealed the product they were prepared to use.

The Music Video Pivot

The next pivot brought FYLMIO back to music. The story engine, character-consistency system, timeline, render pipeline and audio-analysis infrastructure already existed. I connected them and expanded the audio system with stem separation, lyric extraction, beat synchronization, vocal detection, structure analysis and emotional tagging.

Music creates emotion immediately, and every song needs a visual identity. That made AI music videos a clearer and more natural direction for the platform.

The Hidden Complexity Behind AI Products

FYLMIO grew beyond 250,000 lines of Python, React and TypeScript. It combines AI orchestration, audio analysis, rendering, asset management, editing, cloud infrastructure, authentication, analytics, payments and email automation with services from OpenAI, Google, ElevenLabs, fal.ai, Stripe and ffmpeg.

Building an AI product is not only about prompting models. It is about building reliable systems around unreliable technology.

What Building an AI Startup Alone Taught Me

Technology alone is not enough. Infrastructure does not matter if people do not emotionally connect with the result. Music videos combine emotion, story, rhythm, visual identity and culture, which is why the format feels so powerful for generative AI.

For the first time, one creator can make work that previously required a studio. After the pivots, failed ideas, broken servers and long debugging sessions, FYLMIO finally feels like the right direction.