Skip to content
MakerGain
← Back to project

Putting pictures on the beat: neural frames turns synchronization into a business

Edited by MakerGain

neural frames handles the synchronization involved in setting music to pictures: it separates elements such as drums and vocals so generated visuals land on the beat. Founder Nicolai Klemke began with general text-to-video, then followed advice to narrow the product to music videos because the product lacked a well-defined use case. The small team buys models and APIs externally, concentrating its work on separating songs and passing parameters to the models.

Setting a song to pictures involves an unavoidable move: zoom into the waveform, listen for the snare, then shift the visual transition onto the beat. Miss it and the music and pictures go their separate ways.

That is the work neural frames wants to take over. As of 2026-11-17, its website positions it as an audio-reactive AI music-video platform for musicians: upload a song, let an audio engine analyze rhythm, energy, and structure, choose style, characters, and beat-sync intensity, then download a video for YouTube, TikTok, or Reels (website[1]). It automatically separates 8 stems—drums, bass, vocals, melody, and others—and provides more than 10 modulation parameters so visuals can follow these elements separately. Lyrics, key, and tempo are also extracted for synchronization and lyric videos (website[1]).

Image: from the product website[1], obtained on 2026-11-18. An official music-video illustration.

It did not begin in this form.

The website's About page says neural frames started in late 2022, when founder Nicolai Klemke began building out of his interest in AI animation generation. He ran it alone for almost a year before it grew into a small team (Our Story[2]). His previous path was a physics PhD and academia. In an interview dated 2025-07-30, he said he had continued with physics because he was good at it rather than particularly passionate about it, but wanted to do something that affected the real world (Starter Story[3]).

The original landing page said “text to video for everyone.” He later admitted in the interview that this was “really not good.” It lacked a clear use case. Visitors had to make a mental turn—“Maybe I could make music videos here”—and some simply left instead. A friend suggested narrowing the scope, leading to music videos, a step he called a huge unlock (Starter Story[3]).

The product now tackles timing. To put music videos on the beat, the system first reads rhythm, energy, and structure from a song. Stem separation, lyrics, and tempo become computational inputs (website[1]). The site currently lists three creation modes: Autopilot turns a song directly into a full video; a frame-by-frame editor gives control over each frame; and a text-to-video editor uses a timeline to adjust sections. Models and APIs are bought externally, with different models charged per generated second. The website's table estimates users' spending by model and duration; it is not the platform's operating cost (pricing[4]). Uploaded images can train character and style models for consistency across projects (website[1]).

For a small team, this is a practical tradeoff. Buy models and APIs, then refine the part that separates the song and passes parameters to the models. Users can hear whether that work is good: do the visuals arrive when they should?

The cold start was somewhat chaotic. About a week after launch, he posted to Hacker News, when the product still looked rough. He recalled reaching the top six that Sunday, with about 350 people arriving within 30 minutes. He rushed home to keep the GPU running. That traffic brought what he described as his first internet income and backlinks (Starter Story[3]).

Acquisition afterward was manual: SEO, dedicated landing pages, free tools, and a deliberate emphasis on being an indie developer. The footer said there was no VC, just a small company; he put his photograph on the page and recorded YouTube tutorials. His view was that people buy from people, not companies (Starter Story[3]). It is a plain belief, but parallels niching down: narrow the product until one person can explain it and users can recognize it.

In the 2025-07-30 interview, he reported about 1,500 customers, $100,000 monthly revenue at that time, and 1.5 million videos completed cumulatively on the platform (Starter Story[3]).

On costs, he said roughly $45,000 a month went to GPUs and text-to-video APIs, $5,000 to servers, storage, and hosting, and $1,000 to Cursor, leaving little profit (Starter Story[3]). These are historical figures, not current operating performance. He did not disclose complete costs, and there are no current customer or subscription counts or cost structure available to check. Public information therefore does not establish the amount of profit.

Even that historical set of figures shows the business's shape: subscriptions are sold, while its major costs—GPUs and external video-model APIs—are consumed with generation usage. Product design has room to reduce wasted renders. This is why a rule such as “failed generations do not consume credits” matters: users at least do not spend their allowance again on a failed attempt.

As of 2026-11-17, the website's three monthly tiers cost $39, $99, and $299 for 2,400, 7,200, and 24,000 credits a month. Annual payment is collected upfront at $312, $792, and $2,388, displayed as monthly equivalents of $26, $66, and $199 (pricing[4]). All tiers include 1080p upscaling; only Neural Ninja and Neural Nirvana include 4K upscaling (pricing[4]).

Credits are not a fixed count of clicks. The help center defines rendering credits in terms of seconds used for rendering: frame-by-frame animation costs 1 credit per second, text-to-video is charged per second for each model, and Autopilot shows the cost on the Storyboard page before generation (help center[5]). The pricing page also estimates the cost of a 1-minute video across models, with more expensive models generally producing greater detail (pricing[4]).

Two conditions deserve attention. Subscriptions issue fresh credits monthly. Unused subscription credits roll over while an active subscription remains at the same or a higher tier, capped at 300% of the monthly allowance; downgrading or canceling removes rollover credits (pricing[4], help center[5]). Failed generations do not consume credits (pricing[4]). For a service charging by rendered seconds, that reduces the user's credit loss from a failed attempt to zero; successful renders still consume credits per second. Refunds are less generous. The website offers no general money-back guarantee. Someone who subscribed to try the service, created only one project, and is genuinely dissatisfied can contact support for case-by-case review. Cancellation takes effect at the end of the paid period, when unused subscription credits expire (pricing[4]).

Runway's pricing page is a natural reference, but only a partial comparison. As of 2026-11-17, its monthly plans are Standard, Pro, and Max; annual plans display lower monthly equivalents, with different monthly credit allowances. Standard and Pro monthly credits expire on the billing date, unused Max credits roll over for at most 1 month, and purchased credits do not expire (Runway pricing[6], Runway help center[7]). Both sell generation allowances by the month, but their inputs, outputs, and manual editing workflows differ. Runway's page also does not describe a complete audio-reactive music-video workflow. These figures therefore compare pricing structures, not prices for an identical task.

There are already many AI video tools. neural frames' narrow opening is to generate visuals with beat synchronization built in, while retaining frame-by-frame and text-to-video editors for section-by-section adjustment. It sells not the pictures themselves, but the beat on which each picture lands.

Continue reading

AI video
Free video ideas, finished videos by the minute — how Subscribr runs a YouTube workflow

Subscribr handles the long-form YouTube workflow from ideas and scripts to thumbnails and finished videos. Video ideas and scheduling are free, and finished video is priced by the minute. Founder Gil Hildebrand left a funded startup and bootstrapped Subscribr, which launched in 2024. He sold lifetime licenses before the product was built, with delivery promised within 60 days and refunds available through two weeks after delivery. The tool later grew from ideation and scripting to the full production workflow.

AI video
Zero monthly fees, with pricing by the second: VOSU unbundles AI video, then asks who will deliver it

VOSU aims to take a campaign from concept to an editable workflow: users describe the requirement, and agents plan concepts, build prompts, select models and generate variants in bulk. Its three full-time creator founders chose pay-per-use billing and non-expiring balances, while betting growth on a marketplace connecting brands with AI creative talent.

AI video
Break down the script before opening AI: how Layer8 sells films

Layer8 is an AI film studio co-founded by Ali, Kevin, and James. It breaks down scripts before opening AI tools, with directors, writers, artists, and AI experts working together while filmmakers retain visual judgment. After the team sent early samples to contacts, California Highway Patrol became its first paying client. An earlier attempt at templated ads failed to take off, and a trademark dispute prompted the Layer8 name. Agencies, production companies, and brands buy finished films.

AI video
Study what makes ads work, then generate your own: Starpop puts four steps in one conversation

Starpop handles the stretch of performance advertising between understanding an effective ad and turning it into material ready to produce. It researches ads, then develops strategy, scripts, and storyboards; approved storyboards can lead into image, video, voiceover, and music generation in the same conversation, with final decisions still made by people. In the interview, Alexander Van Le described his previous VC-backed FinTech as too focused on terms and strategy, and too little on marketing and understanding customers. He shifted to marketing-first and later bootstrapped AI Flow Chat and Starpop. The business puts research before generation and lets other agents call its generation tools through MCP: connection and chat are free, while media generation costs credits.