Work/ApeFX
Case Study
AI PlatformOwned + Built2024 – present

ApeFX

An AI cinematic content platform built from scratch, solo. Product, infrastructure, automation pipeline, and every AI model integration — all one person. Reached $200K+ MRR.

$200K+
MRR reached
Solo
One person built it
8+
AI model integrations
100%
Automated pipeline
ApeFX — character gallery view
ApeFX — generation queue / pipeline
ApeFX — model selector UI
What was built

A full-stack AI production platform. Nothing outsourced.

ApeFX started as a problem: the best AI cinematic content was locked behind manual workflows. Each model had different APIs, different quality profiles, and no way to run at volume without someone sitting in front of a screen.

The answer was to build the infrastructure myself. Platform, pipeline, routing layer, automation — everything. The system that resulted generates every image and video on the platform without a human touching individual assets.

Full-stack platform

Next.js 14 App Router, Supabase (auth, schema, RLS, real-time), Vercel (deploy, edge, env). Consumer-scale from day one — auth, payments, API routes, CDN-delivered assets.

Mass automation pipeline

n8n workflows handle character batch requests, generation queue management, QC gates, CDN upload, and state tracking. No human per asset. The pipeline runs unattended.

Multi-model routing

FAL, Replicate, ByteDance (Seedance), Kling, Higgsfield, Google (Imagen + Veo). Each model routed by task class and quality target. Fallback logic and retry queues for rate limits.

Character development + LoRA training

End-to-end character pipeline: concept brief → 20-image reference set → FAL LoRA training → dual-reference injection → generation → QC → delivery. Character identity consistent at scale.

The stack

Every layer built from scratch, not assembled from templates.

Platform

  • —Next.js 14 — App Router, TypeScript
  • —Supabase — schema, RLS, real-time
  • —Vercel — deploys, edge, env vars
  • —Cloudflare R2 — CDN asset delivery

AI Models — Image

  • —Flux Kontext (BFL) — character + editorial
  • —Nano Banana Pro — multi-char lifestyle
  • —Seedream 4.x — typography, texture
  • —Recraft V4 — vector + illustration
  • —Imagen 4 — nature, headshots

AI Models — Video

  • —Kling 3.0 Omni — lip-sync + motion
  • —Seedance 2.0 via kie.ai — narrative
  • —Veo 3.1 — environment scenes
  • —Higgsfield Soul ID — character anchor

Automation

  • —n8n self-hosted — 55 live workflows
  • —Generation queues — batch + retry logic
  • —QC gates — CSFD threshold checks
  • —CDN delivery — R2 + URL registry

Character Pipeline

  • —FAL LoRA trainers — 3 param schemes
  • —20-image reference bundles
  • —Dual-reference injection
  • —SAME anchor mechanic — lighting consistency

Infrastructure

  • —Docker — n8n, Qdrant, Grafana
  • —Sentry — error tracking + alerts
  • —Uptime Kuma — 15+ service monitoring
  • —Supabase RLS — multi-tenant isolation
ApeFX — n8n generation pipeline workflow
ApeFX — character batch output / CDN assets
The context

The rig is the product.

Most AI platforms are thin wrappers around an API call. The UI is nice, the generation happens once, and there's nothing holding identity consistent across runs. For character-driven AI content, that's a dead end.

ApeFX was built from the other direction: what would the pipeline need to produce consistent, high-quality character output at scale, with no manual steps after intake? The platform is the front door. The automation is the product.

The result is a system that runs unattended. Brief in. CDN assets out. $200K+ MRR on that loop.

First character batch deliveredWeek 1
External dependencies for core generation$0
AI APIs integrated and routed8+
Manual steps in the production loopZero

Want something similar built?

Tell me the brief. I'll tell you whether I can help and what it would take.

Visit apefx.ai ↗
← Back to all work