menivor

Inspiration

Every creator we know has the same bottleneck: they have something to say — a product to show, a screen to record — but turning that raw capture into a post-worthy vertical reel means hours of scripting, voiceover, avatar work, and caption timing. Tools exist for each step, but nobody stitches them into one tap-to-post flow on the device where creators actually work: their phone.

So we built Menivor — drop in a screen recording, and it comes back a viral, ready-to-post UGC reel.

What it does

You give Menivor a screen recording and a prompt. It:

1. Scripts it — a hook-driven script (TR/EN) tuned for short-form retention.
2. Voices it — natural TTS timed to the words.
3. Adds a lip-synced AI avatar presenter — a talking-head that delivers the hook.
4. Burns CapCut-style word-by-word karaoke captions.
5. Renders a 1080×1920 MP4 — ready for TikTok, Reels, and Shorts.

No timeline, no manual editing. Create → render → post, from your phone.

How we built it

• Mobile: Expo / React Native (24+ screens: capture → compose → render → jobs library → publish), with device-session auth and a persistent, on-device export receipt so every render shows its real file, duration, and share result.
• Monetization — RevenueCat: Pro exports are entitlement-gated. We mount RevenueCatProvider at the app root, Purchases.configure with the platform API keys, verify entitlements on getCustomerInfo, and present the paywall via RevenueCatUI — with restorePurchases and update listeners wired. The entitlement policy is unit-tested.
• Backend: FastAPI + Celery orchestrate the render chain (script → tts → {avatar ‖ captions} → compose → upload); Cloudflare R2 for storage; a microdollar wallet with success-only settlement so a failed render never charges the user.
• Model routing: ElevenLabs (voice), HeyGen / RunPod & diffusion providers (avatar/video), FFmpeg + libass for the final compositing (chroma PiP + karaoke captions, hardware-accelerated).
• Web: a live marketing + product site at menivor.com (Next.js on Cloudflare Workers).

Challenges we ran into

• Timing is everything. Voice, captions, and avatar all have to line up to the same millisecond. We made the TTS timestamps the single source of truth and drove captions + avatar off them, instead of guessing.
• Failures shouldn't cost money or progress. Renders can fail mid-chain, so we made every step resume on the same job and made billing success-only — the wallet reserves, then only settles when a real MP4 lands.
• RevenueCat across a real render economy. Gating exports (not just a flat subscription) meant entitlements had to be checked at the exact moment of value — the export — and degrade gracefully when unconfigured.

What we learned

• Consistency and timing, not raw model quality, are what make a generated reel feel "real."
• Success-only billing + entitlement-gated exports is a cleaner monetization story than a blunt paywall — you charge at the moment you deliver value.
• Shipping a mobile creative pipeline forces brutal simplicity: one tap, one job, one receipt.

What's next

Character consistency (the same avatar/voice across every reel), direct publish to TikTok/IG/Shorts, and a template marketplace.

Built With

• celery
• cloudflare-r2
• cloudflare-workers
• expo.io
• fastapi
• next.js
• python
• react-native
• redis
• revenuecat
• typescript