QRForge

Minimal chat that builds AI-powered, scannable QR codes — with ElevenLabs voice.

A fullscreen chat (no chrome, no sidebar). You type or talk; the assistant asks for the redirect URL and returns a stylized, scannable QR. Drop in a logo or reference image — it's read locally in the browser and steered into the generation.

Stack

• Next.js 16 (App Router, Turbopack) + TypeScript + Tailwind v4
• lucide-react
• fal.ai
  • openai/gpt-image-2 — text-to-image backdrop
  • openai/gpt-image-2/edit — image-to-image when a reference is attached
  • fal-ai/elevenlabs/tts/turbo-v2.5 — voice for assistant replies
• qrcode — server-side QR matrix (error-correction H, guaranteed scannable)
• Web Speech API — STT in the browser (free, no key)

How it works

1. Chat fullscreen. Empty state → "Genera il tuo QR" + composer. No sidebar, no header, no plan/credits.
2. Voice ON / Voice OFF toggle in the composer. When ON, every assistant message is auto-spoken via fal → ElevenLabs Turbo v2.5. Each message also has a 🔊 button to replay.
3. 🎙️ Mic dictates into the composer (browser SpeechRecognition, it-IT).
4. 📎 Paperclip opens a local file picker. The image is read with FileReader as a base64 data URI — never leaves the browser until the user clicks send. On generate it's uploaded once to fal storage and used as a reference image for gpt-image-2/edit.
5. The composed QR is returned as a data:image/svg+xml;base64,… URI — no storage required.

Env vars (set on Vercel)

Name	Required	Notes
FAL_KEY	yes (for AI + voice)	If absent: QR still generated locally, no voice

No DB, no auth, no Blob — demo-ready.

Run

bun install
bun run dev   # http://localhost:3000

Deploy

vercel --prod

Or push to main — Vercel auto-deploys (GitHub linked).

Cost notes

• TTS: ElevenLabs Turbo v2.5 via fal — ~$0.05 / 1,000 chars.
• Image: fal openai/gpt-image-2 — pay-per-image.

License

MIT — built for an investor pitch by Loer Studio.