How Neural Frames hit $5M/yr on AI music videos

Nicolai Klemke, 34, spent his previous life as a physicist — a doctorate, lasers, the works — before falling in love with programming and AI. A long-time musician on the side, he kept running into the problem every indie artist knows: compelling visuals for your music cost a fortune or months in post-production. At the end of 2022, he started building the fix as a side project.

The first version launched in January 2023, built on Stable Diffusion mostly by himself; v2 followed in April. The first six months were rough — by his own account, he had barely any frontend or cloud-hosting experience. Then the breakthrough: after six months, the first really awesome music videos were created. That was the moment he knew it worked. By October 2023 the project was at $17K MRR with 130,000 visitors since April.

The product thesis was simple and has never changed: upload a song, get audio-reactive visuals that move with the music. Autopilot turns a track into a full video in two clicks; a frame-by-frame editor gives control freaks granular command over every scene. As the Amsterdam Dance Event put it: a synthesizer for the visual world.

The business model is credits, priced for working musicians rather than enterprises: roughly 100 to 500 credits a month for $20 to $100, with videos costing 200 to 4,000 credits depending on length — later formalized into $26, $66, and $199 monthly tiers. The workflow is deliberately simple: upload a song, the platform analyzes lyrics and mood, suggests a storyboard, and the rest is customization — change the visual style, upload your own characters, or put yourself in the video. Finished work embeds straight to YouTube, Spotify, and TikTok.

Klemke didn't launch with fanfare — he posted his first AI music videos in subreddits. The Stable Diffusion subreddit lit up: he posted his first content on Reddit and suddenly had 200 users. For a technical product made by a technical founder, the internet's weirdest corners were the perfect early-adopter pool.

The first money, he later said, came from a Hacker News post shortly after putting the product out — early signal before the company looked polished. It's a useful corrective to launch mythology: traction arrived as rough demos in front of the right nerds, not as a orchestrated Product Hunt campaign.

Community became the growth engine's first stage. Musicians are natural sharers — every video created is content, and every share is distribution. By mid-2026 the platform had hosted more than 2 million created videos from over 40,000 artists.

The most consequential strategic decision was what not to be. Neural Frames started as a generic text-to-video tool — the most crowded positioning imaginable in 2023. Klemke niched down to indie musicians specifically, and the product suddenly made sense: recurring visual needs around every release, a real workflow rather than a novelty, a customer who could name the problem.

Then he invested like an SEO maniac. In the first year the team wrote what Klemke calls a gigantic guide about AI video generation — 30,000 words, an incredible amount of love in it. It became an authority source in the AI video space and still drives organic traffic years later, supplemented by dedicated landing pages and free tools, with heavy use of Ahrefs to find and grow demand.

The lesson generalizes: creator tools can grow through search, not just social hype. Social gave Neural Frames its first 200 users; search gave it the next 40,000.

In August 2026 Klemke went on the TBPN podcast and gave the numbers directly: 12 employees in Berlin, still bootstrapped, still profitable, running at about $5 million in annualized revenue — roughly sixfold year-over-year growth. It is the biggest number ever celebrated on MilestoneWins, and no venture capital was involved at any step.

He is unusually honest about the economics, which is what makes the story believable. Generative video is expensive to serve: the business spends roughly $45K a month on GPUs and text-to-video APIs plus about $5K on hosting and infrastructure. Revenue in generative SaaS is not the same thing as gross margin — Klemke just refuses to pretend otherwise.

The market has noticed. Hazel Savage — founder of music-AI company Musiio, turned investor — joined as strategic advisor in 2026 after watching the numbers: she met Klemke at the Amsterdam Dance Event the previous September, when he already had what she called absolutely mad revenue, and when they caught up two months later he had doubled it again. Sometimes, she said, you feel like you are watching something become a rocket-ship right in front of your eyes — and obviously she wanted a ticket to board.

There is a technical insight underneath the business story, and Klemke articulated it on TBPN: the application harness and the image-generation pipeline matter at least as much as the underlying video model. Base models produce short clips; a three-to-six-minute song needs something else — coordinated images and clips that preserve characters, follow the beat and the mood, and tell a story across minutes, with power users refining scenes through an opinionated timeline rather than a chat box. That is why Neural Frames stays model-agnostic, letting musicians switch between Kling, Seedance, Runway, and Stable Diffusion inside its editor: the moat is the workflow built around the models, not any single model.