Own product · Real estate · iOS + AI · 2025–26

Luster

Magazine-grade listing photos from your phone

Role: Co-founder, all software
When: Aug 2025 – now
Built with: React Native, Python, OpenAI
Status: Live on App Store

~3 min: phone photo to finished
$0.18: to enhance one photo
97.2%: of jobs succeed

App Store ↗lusterapp.ai ↗

Front of a house, after — One phone photo in, Afterglow style out. Drag to compare.

Front of a house, before — One phone photo in, Afterglow style out. Drag to compare.

The before

Pros cost hundreds and take days. Phone photos look dark. AI editors invent furniture.

What I built

One tap, photos an agent can stand behind.

Kitchen sink window: original phone photo next to the Luster result — My kitchen, with the real view outside.

Features

01A camera that helps

A guided camera keeps shots level and sorts rooms right on the phone.

02Pick one look

Eight styles, no prompt box, and every room in the listing matches.

03Tap once, walk away

Up to 50 photos per batch, and anything that fails refunds itself.

04Proof built in

A small QR code on each photo links to the untouched original.

05Flyers, ready to post

Fourteen listing-flyer templates with the same disclosure QR, ready to launch.

Six listing flyer designs for a sample home

Hard part 01

The room kept changing.

Agents noticed lamps changing size. In real estate, that's a liability.

I built a lab to score how much of the room survives. Declutter prompts redrew the whole room. Style-only prompts held steady.

Variance lab: the same photo and prompt run repeatedly, scored for faithfulness — Same photo, same prompt, scored each run.

Hard part 02 · in testing

Windows that show the real view.

Phones blow windows out to white. A made-up view on a listing is a lie.

So the camera grabs a darker frame, and those real pixels go into the glass first.

Window Pull, step by step, on a kitchen window — Real pixels first, then the AI blends inside the window.

Hard part 03

Never charge someone twice.

Credits can't go negative, duplicates get caught, and failed jobs refund themselves.

Zero refunds or chargebacks in the first two months.

Why it matters

Before	After	So what
Photographer: hundreds, days	About $2 a photo	Every listing looks its best
AI that invents rooms	Style-only, tested edits	Photos agents trust
Posting an ugly raw copy	One QR to original	Honest and still beautiful

Declutter prompts redrew the room. Style alone kept it.

Declutter prompt, run 1183
Declutter prompt, run 2107
Declutter prompt, run 350
Style-only prompt, run 1755
Style-only prompt, run 2781
Style-only prompt, run 3752

Source: Matching points kept from the original room. Sep 2026

Six free credits at signup: activation went from 32% to 78% (7 of 9).

How it works

iPhone app
Upload + credit check
Job queue
AI enhancement
Verified original kept

The phone uploads, a server queues the job, a worker edits it.

Technical details +

Under the hood

Postgres is the job queue. No Redis to run.
Credits come off in one statement that can't go negative.
On-device classifier sorts interior from exterior.
One worker handles about 220 photos an hour.

Stack

React Native / Expo + Swift
FastAPI on Railway
Postgres on Supabase
OpenAI gpt-image-2

How it came together

Aug 2025
Formed the company with Sam. First commit.
Sep–Nov 2025
Built the iPhone app, payments and sign-in.
Apr 2026
Disclosure trail: QR export and verify page.
Jun 24, 2026
Live on the App Store.
Aug 2026
Version 1.1. Verified Export free for everyone.
Sep 2026
Variance lab: why the AI redrew rooms.

What I learned

—Never trust one run. AI output is random, so I run everything three times.
—The people sell it. Most users came through one Coldwell Banker office.

Video

See it on a real listing.

Watch on Instagram ↗

Know an agent drowning in listing photos? Let's grab coffee.

Co-founded with Sam Loya. Working with Chuck Whitehead.

Grab coffee App Store ↗lusterapp.ai ↗

Next projectGo Further Group →