After 18 Months of Failure He Borrowed $100K and Bet It All — How the “Snap a Photo, Type a Prompt” AI Editor Halo AI Hit $300K/mo in 45 Days

Halo AI lets you upload a photo, type a prompt describing the change you want, and get an AI-edited image back. Swapping backgrounds, removing unwanted objects, adjusting how things look — all without pro software or retouching skill — is the core experience.

It was built by Dillion Verma, a developer previously known for Magic UI, a library of animated UI components he designed, built and sold. His personal site (dillion.io) lists several products, from an email-capture SaaS for GPT Store builders to an AI customer-support bot. In other words, Halo AI isn’t a lucky one-off — it’s the current waypoint of someone who keeps shipping.

The road wasn’t smooth, though. By his own account, roughly 18 months of failed products preceded Halo AI. His bank balance was nearly gone, and the one solid signal he had was a single data point: one TikTok video that returned 6x its ad spend. So he took out a $100K personal loan and bet everything on that winning format.

The payoff: using one UGC (user-generated content) video format and about 85 creators, he stacked 1.2 billion views in 120 days. With no VC and no traditional performance ads, he reached ~$300K monthly recurring revenue 45 days after launch. Technically, the image editing is reported to lean on Google’s Gemini image model (a.k.a. Nano Banana), with the paywall optimized via Superwall — a stack an indie builder can realistically reproduce.

From the founder (primary source)

The repeatable playbook

1. 1Even at your quit line, if you hold one solid proof (e.g. a 6x-ROI hit), bet on the data, not a feeling
2. 2Lock onto one winning UGC format and go deep (mass-produce it) before going wide
3. 3Pay creators on view-milestone bonuses, not flat CPM, to align incentives
4. 4Transcribe and dissect hit videos (e.g. with Gemini) into a distributable, repeatable brief
5. 5Drop forced gating; A/B intent-based paywalls with Superwall to lift conversion
6. 6Monetize thin-and-wide with weekly subscriptions ($8.99/wk etc.), adoption over ARPU
7. 7Earn proof of a winning angle BEFORE you invest — don’t skip the order and start from debt

Deep dive

【Deep dive】Halo AI’s real reason for hitting $300K/mo in 45 days isn’t flashy tech — it’s that he believed in one winning format enough to borrow against it, then turned it into a repeatable machine. Step by step:

■ The setup: 18 months of failure and a $100K bet. Behind the headline numbers sat ~18 months of failed products. At the point most people quit, he could act not on a feeling but on one data point: a TikTok that returned 6x its ad spend. Many would dismiss a single hit as luck and move to the next idea; he read it the other way — ‘if this format is real, volume will scale it’ — and bet a $100K personal loan on it. The key is that he held *evidence of a winning angle* before betting. It wasn’t a blind upfront splurge.

■ Lock onto ONE UGC format and mass-produce it without drift. Growth came not from a buffet of tactics but from fixing one winning format and producing it relentlessly. Reinventing the structure, hook and framing every time scatters your learning; freezing the format narrows the only variable to ‘which input lands.’ With that single format he built the 1.2B-views-in-120-days denominator. The indie lesson is clear: when something hits, go *deep on the same format* before going wide.

■ Move 85 creators with view-milestone payouts. Distribution ran through a network of ~85 creators, and the lever was the comp design. Instead of a flat CPM (pay-per-1,000-views), he offered bonuses tied to view milestones (20k/100k/500k/1M) — and 29 of 30 creators reportedly chose the milestone model. Creators push harder because ‘the more it spreads, the more I earn,’ while the operator caps downside because ‘misses don’t cost much.’ Performance-linked pay is a device for aligning incentives between creator and founder.

■ Make hit videos reproducible with Gemini + a Discord agent. Rather than relying on individual taste, he reproduced winners systemically. Reporting indicates he used the Gemini API to transcribe and break down viral hits, then fed that into a Discord agent (a ‘marketing coach’ for creators) so all 85 could share the same winning formula. Converting personal skill into a *distributable brief* is exactly why it’s described as ‘turning 85 creators into one synchronized viral machine.’

■ Tighten paywall conversion with Superwall; monetize thin and wide. He converted the top-of-funnel into revenue via Superwall paywall optimization. Dropping forced gated onboarding in favor of intent-based paywalls — shown when the user actually wants to edit *now* — reportedly drove a 16.5% conversion rate. Pricing centers on $8.99/week across several tiers ($5.99–$6.99/week, $29.99/year, credit packs), prioritizing adoption over ARPU while still holding ~50% margins at $300K scale.

■ What ‘no ads’ really means — and a warning for copycats. He frames it as ‘no VC, no performance ads, no (traditional) influencers,’ but that doesn’t mean he spent nothing. He routed the $100K loan not into ad managers but into *performance-linked creator investment* — betting on a bonus-on-hit creator network instead of burning it in Meta Ads. And don’t forget the capital was *debt*. It paid off because the format worked; had it missed, only the $100K liability would remain. If you want to copy this, first earn the one piece of evidence he held before betting — a 6x-ROI signal of your own. Don’t skip the order and start from the loan.