FaceKit, the face-rating app built by a 19-year-old — how ‘40 AI-generated influencers’ drove 100M views a month and $100k in 2 months with zero ad spend
A ‘scan your face and score it’ looksmaxing app built by 19-year-old Rafael Kramer: an iPhone TrueDepth 3D face scan quantifies symmetry and golden ratios. The growth engine isn’t human — it’s ‘40 AI-influencer accounts generated with Nano Banana.’ A few clippers mass-post them on TikTok, and 100M+ monthly views drove $100k in cumulative sales in two months with zero ad spend.
The pain point, and how they found it
‘Objectively, how many points is my face?’ — for the Gen Z youth sinking into ‘looksmaxing,’ this is an urgent, repeating obsession. Until now there were only mirrors, other people’s subjective takes, or sketchy rating sites — nothing that returned, in numbers, ‘what exactly to fix to improve.’ FaceKit hit that gap: an iPhone TrueDepth 3D face scan measures interocular distance, symmetry, and various ratios, then translates them into a ‘score’ plus improvement tips (skincare, hydration, gym). The very anxiety young people feel daily — ‘what’s my face’s score?’ — was itself the market entry point.
FaceKit (full name ‘FaceKit – 3D Face Analysis’) is a ‘looksmaxing’ app that 3D-scans and analyzes your face. Using the iPhone’s TrueDepth camera (the same sensor behind Face ID), it reads your face in three dimensions, measures interocular distance, symmetry, and various ratios, and returns an overall ‘face analysis’ with a score plus skincare and lifestyle improvement tips. It’s published by the founder’s own app studio, Mindmush, in iOS utilities/self-improvement.
It was built by Rafael (Raf) Kramer, 19. He runs Mindmush, a studio that ships apps solo and, per reporting, has produced 10M+ downloads across 50+ apps. FaceKit is one of them — but what drew attention was less the product than *how it was marketed*.
Kramer leaned on no human influencers and no paid ads. Instead, he generated realistic human avatars with the image tool ‘Nano Banana,’ made them talk and move with OpenAI’s video model ‘Sora,’ and stood up ‘influencers who don’t exist’ across ~40 TikTok accounts. A few ‘clippers’ ran those accounts, mass-producing before/after and viral short videos that together generated 100M+ views a month. The result, as reported: ~$100,000 in cumulative gross sales about two months after launch, with video production costing ~$2,000/month and net profit after Apple’s cut around $35,000/month. After the peak it cooled to roughly $17,000/month recently (TrustMRR).
In other words, FaceKit sold ‘an AI-made face (the product)’ using ‘AI-made faces (the influencers)’ — a very 2026 snapshot of indie development that runs generative AI on both the product and the distribution side.
The repeatable playbook
- 1Generate the distributor itself with AI: make realistic avatars (fictional people) with Nano Banana and animate them into video with Sora, etc.
- 2Stand up dozens of ‘influencers who don’t exist’ (FaceKit ran ~40) and mass-post them on TikTok
- 3Coordinate a few clippers in a group chat and instantly copy any winning format across all accounts
- 4Spread accounts against bans/stalls and fan winning angles out dozens of times faster than humans (100M+ views/month at zero ad spend)
- 5Pick a ‘flammable’ topic (looksmaxing = acute anxiety × controversy) and use the product’s core value (face scoring) directly as the video hook
- 6On the product, add ‘science-y’ credibility with real TrueDepth measurements; bridge free scan → paid detailed analysis into a subscription
- 7Assume synthetic virality decays — keep the ‘factory’ running by refreshing formats, injecting new avatars, and rotating topics
This case has a shadow side. First, durability — virality-sourced revenue is fickle and drops after the peak. After a stretch of ~$35,000/month in net profit, TrustMRR tracks it settling to roughly $17,000/month recently. Second, ethics and regulation — having ‘people who don’t exist’ recommend products as if real sits next to platform crackdowns on AI ‘slop’ and questions of trust. Behind the flashy launch numbers, a synthetic-influencer factory decays if left alone. Sustaining it takes the unglamorous work of ‘restarting’ — refreshing winning formats, injecting new avatars, rotating topics. It’s a lesson that launch force and staying power are different things.
Deep dive
【Deep dive】FaceKit matters less as a face-rating app than as an ad factory a 19-year-old built solo — one that replaced human influencers wholesale with generative AI. Here is Rafael Kramer’s method, taken apart step by step.
■ Core idea: synthesize the distribution channel itself. Classic UGC/influencer growth means ‘real human creators distribute for you.’ Kramer dropped that premise and generated the *distributors* themselves with AI. He made realistic avatars (fictional people) with the image tool ‘Nano Banana,’ then made them talk and move with OpenAI’s ‘Sora,’ standing up ‘influencers who don’t exist’ across ~40 permanent TikTok accounts. Where humans would need negotiation, fees, and scheduling, he collapsed it to generation cost alone — on the order of ~$2,000/month in production.
■ Decentralize × imitate: instantly copy one hit across 40 accounts. The 40 accounts are run by a few ‘clippers’ coordinating in a group chat. The crux is *instant lateral replication*: when a format (before/after, reaction, provocative hook, etc.) breaks out, everyone immediately clones that pattern on their own accounts. This (1) spreads the risk of any one account being banned or stalling, (2) fans a winning angle out across the surface dozens of times faster than a human could, and (3) learns which algorithmic hooks land, at many points simultaneously. The result: 100M+ views a month, at zero ad spend.
■ Topic selection: looksmaxing, a ‘flammable’ vein. The theme is part of the strategy. Looksmaxing carries young people’s acute anxiety, self-reference, and controversy — a ‘flammable’ subject where a heated comment section actually lifts reach. The hook ‘scan your face and get a score’ instantly triggers the viewer’s ‘what’s *my* score?’, shortening the path from video to install. The product’s core value (scoring) doubles as its strongest thumbnail/hook.
■ Product ‘plausibility’: put numbers on it with TrueDepth. Virality alone won’t keep people paying. FaceKit actually 3D-scans with the iPhone TrueDepth camera and presents symmetry and ratios as *measured values*. That ‘science-y’ feel (measured, not guessed) differentiates it from a mere novelty/horoscope app and bridges score-chasers into recurring subscriptions. Monetization is reported to use weekly/monthly/annual tiers — free at the door (scan + rough score), detailed analysis and improvement tips behind the paywall (we don’t assert this specific app’s exact prices here).
■ How the numbers were built — and their ‘seasonality.’ ~$100,000 cumulative gross in about two months, ~$35,000/month net profit. But virality-sourced revenue is fickle and drops once the peak passes — TrustMRR tracks ~$17,000/month recently. There’s an honest lesson here: an AI-influencer factory has spectacular *launch* force but decays if left alone. Sustaining it requires ‘restarting the factory’ — refreshing winning formats, injecting new avatars, rotating topics.
■ The studio behind it: Mindmush’s rate of fire. FaceKit isn’t a one-off stroke of luck; it came out of Mindmush, a ‘shoot-a-lot’ parent studio. Precisely because there’s a portfolio of 50+ apps and 10M+ downloads, a winning template (scan-style hooks × AI-influencer distribution) can be redeployed to the next app immediately. The moat accrues not in ‘the app called FaceKit’ but in ‘a distribution system that manufactures cheap mass exposure with synthetic influencers.’
■ The shadow side (worth stating): ethics and durability of synthetic growth. This method has a dark side: the propriety of ‘people who don’t exist’ recommending products as if real, platform crackdowns on AI ‘slop,’ and revenue instability from depending on virality. The takeaway for indie builders is clear: (1) generative AI can now collapse the cost not only of *building* but of *distributing* by an order of magnitude; (2) but assume synthetic virality decays — the real edge is the operational stamina to keep the factory running and to keep refreshing topics and formats.