How to Build Viral AI Apps With Simple Interfaces

TL;DR

Build a viral AI app by finding an emerging model, wrapping it in an interface simple enough for almost anyone, and making the output naturally shareable. WOMBO followed this approach with four core screens, deliberately selected songs, and a content loop that generated more users without paid marketing, eventually contributing to more than 250 million downloads across Ben Benkhin's apps.

Transcript

This is not your standard SIP episode. I brought on this guy who went from zero to 250 million mobile app installs using some open-source AI models. He tells the whole story how he did it. You know, it definitely fired me up around this whole new wave of creating, you know, multi-million dollar uh mobile apps in under 7 months. Ben, you've created ... Read More

Key Insights

• Mimesis is the practice of copying a proven pattern and adding a personal variation. Ben first applied it in League of Legends by watching stronger players in the same role, copying their decisions, and adapting those decisions to his own play.
• WOMBO emerged from combining two visible signals: Reface had demonstrated demand for simple AI face applications, and memes made with the open-source first order motion model were spreading online. Together, those signals suggested that a more accessible mobile experience could attract a broad audience.
• Competition is evidence that people want a category, not proof that every opportunity has disappeared. Ben did not view WOMBO as a direct attempt to defeat Reface, because he believed an early creative market could support many different products, creators, and interpretations.
• WOMBO reduced a technical workflow to four core screens: selfie input, song selection, loading, and output. This simplicity replaced a Google Colab process that required substantially more technical familiarity, allowing even very young or inexperienced users to understand the product quickly.
• Virality is strongest when the product creates content that users naturally want to share. WOMBO let people animate themselves, relatives, politicians, or other subjects, so each result could become both entertainment and a demonstration that encouraged another person to try the app.
• Song selection was a core product decision rather than a minor content detail. The team launched with 15 recognizable, iconic, or meme-oriented songs and deliberately created matching driving videos, including choices that people could enjoy without understanding English.
• WOMBO spent no money on marketing and directed its spending toward AI inference. Organic distribution came from users creating and sharing outputs, demonstrating that product behavior itself can serve as acquisition when every completed creation invites viewing, imitation, and further experimentation.
• Monetization combined subscriptions and advertisements, while about 2 percent of users paid and the other 98 percent supported distribution. This structure treated nonpaying users as valuable participants because their creations and sharing behavior helped expand the audience and sustain the viral loop.

Questions & Answers

Q: How can you find an idea for a viral AI app?

Look for multiple signs that a product pattern is already working. Ben noticed that Reface had hundreds of millions of downloads and used an interface he could easily understand. He also saw an open-source motion model producing viral memes through a difficult Google Colab workflow. Combining proven demand, emerging technology, and poor accessibility led to the WOMBO concept.

Q: What does the copy what works strategy mean for app builders?

The strategy, described as mimesis, means observing a successful behavior or product, identifying its useful structure, and recreating that structure with a distinct variation. Ben used this approach first in gaming and later in app development. WOMBO did not reproduce Reface directly. It applied a similar accessibility pattern to a different open-source model and entertainment format.

Q: Why did WOMBO become easier to use than the original AI workflow?

The original process relied on a Google Colab notebook, a selfie, a driving video, and technical steps that ordinary users were unlikely to complete. WOMBO converted that workflow into four core screens: users provided a selfie, selected a song, waited during processing, and received an output. This reduction made the technology accessible without requiring Python or model knowledge.

Q: How did WOMBO engineer a viral content creation loop?

WOMBO made creation fast, simple, and amusing, then produced an output that people naturally wanted to share. Users could animate themselves, family members, politicians, or other subjects singing familiar songs. Shared outputs exposed new viewers to the app, and those viewers could easily create their own versions, producing a repeating cycle of creation, distribution, and acquisition.

Q: Why was song selection important to WOMBO's growth?

The team deliberately launched with 15 songs selected for recognizability, cultural reach, or existing meme appeal. It also created a specific driving video for each song, which controlled the animated choreography. Songs that remained enjoyable without understanding English could travel more easily across audiences, while familiar music made each generated result immediately understandable and more likely to be shared.

Q: Should AI app founders avoid markets with established competitors?

An existing successful product can demonstrate that users already value the category. Ben saw Reface's popularity but did not conclude that it had eliminated every opportunity. He viewed the market as early and spacious, comparing creative apps to movies and music, where audiences continue seeking new creators and formats. The opportunity was differentiation, not direct duplication.

Q: How did WOMBO grow without spending money on marketing?

WOMBO spent no money on marketing and instead spent its available resources on inference. Its users performed the distribution by sharing generated singing animations. Because the content was funny, recognizable, and easy to reproduce, each shared result could function as a product demonstration. The app's acquisition mechanism was therefore embedded in its core creation experience.

Q: How can a viral AI app monetize mostly free users?

The monetization strategy described for the apps used subscriptions and advertisements. About 2 percent of users paid, while the other 98 percent still created substantial value by making and sharing content that drove virality. This model does not treat free users as irrelevant. Their activity expands distribution, attracts more users, and increases the audience available for advertising and subscription conversion.

Summary & Key Takeaways

• Ben Benkhin describes mimesis as studying something that already works, understanding why it succeeds, and applying the pattern with a distinct spin. He used signals from Reface, viral AI memes, and an open-source motion model to identify an opportunity for a simple consumer app that animated selfies to sing recognizable songs.

• WOMBO transformed a process involving a Google Colab notebook into four basic stages: selfie input, song selection, loading, and output. Its accessibility allowed users to create amusing content quickly. The product encouraged sharing and repeated creation, while carefully selected songs and driving videos made its results recognizable across languages and cultures.

• The broader strategy begins with attention rather than monetization. A simple, funny experience acts like a small piece of candy that introduces users to the product. Subscriptions and advertising can generate revenue, while the large majority of nonpaying users create and distribute content that supports virality and attracts additional users organically.