Neural Frames founder Nicolai Klemke on building a profitable, bootstrapped AI music video startup to $5M ARR with 12 people in Berlin

Neural Frames: $5M ARR, 12 People, Berlin

Neural Frames is a bootstrapped, profitable AI music video platform generating $5M ARR with a team of 12 people in Berlin. Founded by Nicolai Klemke in early 2023, the company grew 6x year-over-year and has never raised outside capital.

Klemke's background is unusual: a PhD in physics and a long history as a hobby musician. He started playing with Stable Diffusion-based animation at the end of 2022, built a prototype around an image-to-image loop he could steer with text, and launched in January 2023. It went viral on Hacker News and Reddit, and Neural Frames grew from there.

“We are a team of 12 people at the moment in Berlin. We are bootstrapped still. We are profitable... We're doing 5,000,000 in run rate. We grew six on year and so we are very excited for what's to come... One year ago, we were, to the best of my knowledge, the first music video tool out there that had character consistency.”

Why music videos specifically

The focus is deliberate. Competing with general-purpose AI video companies as a bootstrapped startup is a losing bet, so Neural Frames narrowed to a specific use case and built the harness around it. The core challenge is non-trivial: a three-to-six-minute song requires generated imagery that matches the beat, the mood, the characters, and tells a coherent story across what are typically five-to-ten-second AI video clips. Klemke argues the underlying video models matter less than people assume — the orchestration layer, image generation workflow, and timeline editing tools are at least equally important.

Character consistency was a meaningful differentiator. Klemke says Neural Frames was the first music video tool to offer it, roughly a year ago, when it was technically very difficult. Today users can upload a few photos, create a character, and tag it across scenes. Power users spend significant time in the platform refining individual scenes via a drag-and-drop timeline, changing characters, outfits, and pacing — a workflow Klemke says would be harder to replicate in a pure chat-based interface.

Model landscape

On the current AI video models Neural Frames integrates:

• CogVideoX 2.5 (CDense 2.5) — Klemke calls it the best model available right now. It supports clips up to 30 seconds, a multi-shot feature that generates multiple shots simultaneously, reference image uploads for object consistency across scenes, and experimental audio input. It is also the most expensive model on the market, and cost is a real constraint.
• Kling 3 — a strong all-rounder, significantly cheaper than CDense 2.5, and a practical workhorse.
• Google's Nano/Banana Light — a lightweight image model, fast and affordable, useful at specific steps in the workflow even if not highest quality.
• Gemini OmniFlash — a video model with editing capabilities, but in Neural Frames' testing it doesn't reach CDense 2.5's quality.

Who's using it

The platform is attracting use cases beyond individual creators. A cosmetics brand built an AI influencer, generated songs with an AI music tool, and produced music videos with Neural Frames. Musicians are using the platform to create Instagram and Facebook ads that link back to Spotify. The corporate novelty angle also lands — one host described using Neural Frames to produce a six-minute, Netflix-style true crime documentary by chaining ChatGPT, Suno, and Neural Frames together.

At $5M ARR on 12 people, the unit economics are the story. That's roughly $415K ARR per employee — a number most venture-backed AI companies running 10x the headcount would envy.