AI image compositor · commercial real estate marketing

Raw building photos in.
Marketing-ready images out.

Relight is a node-based AI image machine that already runs— watch it work below. It's built to be integrated into your business and owned, with full customization — not a per-seat subscription. And you can try it yourself now: invited test runs on a budget we fund.

Try it on an invite Watch the machine run

The real recorded run lands here — same graph, real pixels, real logged cost.

How it works

A node graph does the editing — you approve the results.

1

Drop the shoot

Raw photos are filed, captioned, and bucketed by what each image needs.

2

Describe the change

Remove scaffolding, stage an empty floor, relight a dull sky — plain words.

3

The graph runs

Each edit is a node pipeline on gpt-image-2, with cost estimated before and logged after.

4

Only the edit changes

Edits are composited back — every pixel outside the masked region stays identical to the shot.

Structure preservation is the whole point: a listing photo that comes back with a warped facade is worthless. The compositor's merge step keeps the untouched region of the original byte-identical — verified in 768 unit tests and 31 browser tests that run before every change ships.

Before / after

The three edits every listing needs, done by the machine.

Scaffolding removed — before — Scaffolding removed
Sidewalk shed and scaffold gone; facade, entrance and street kept as shot.

Scaffolding removed — after — Scaffolding removed
Sidewalk shed and scaffold gone; facade, entrance and street kept as shot.

Empty floor staged — before — Empty floor staged
Warm move-in-ready furniture; windows, columns and the view kept as shot.

Empty floor staged — after — Empty floor staged
Warm move-in-ready furniture; windows, columns and the view kept as shot.

Sky and light lifted — before — Sky and light lifted
Flat gray day turned warm; the building itself untouched.

Sky and light lifted — after — Sky and light lifted
Flat gray day turned warm; the building itself untouched.

Honest label: every photo on this page is synthetic — generated and edited by the machine itself (gpt-image-2, draft quality, ~$0.02 per pair). No client photography appears here.

The Claude seat

Ask for changes in plain words — inside the app.

A Claude agent (Claude Opus) lives in the web app. It builds and tests the edit pipelines for you: “clean this set up and stage the empty floors” becomes a working node graph you watch it author, run, and self-review.

Money stays gated: before any paid generation the agent shows an approval card with a hard dollar cap. You approve or deny in the chat; a kill switch aborts anything mid-run. The agent runs on our own model subscription — no per-call model charge to you.

Your browser never holds credentials, and the agent's file access is locked to the open project — enforced server-side, not by prompt.

from a real session

clean up stage 1 and run it live, cap it at 10 cents

Approve paid run?

live_run — capped at $0.10

✓ approved by operator

Live run succeeded — $0.0057 spent, 1 image, under the cap. Result is in the project; nothing auto-selected — your pick.

What it costs to run

Cents per image, shown before you spend them.

Edit tier	Per image	Source
Draft quality (low)	~$0.005–0.006	published token rates
Production quality (medium)	~$0.04–0.05	real logged edits
Hero / print (high)	~$0.17–0.21	published token rates

Every batch shows its estimated cost before it runs and logs the actual after — a complete project showcase ran end-to-end for $0.235 of real model spend. Off-peak batching halves the model rates. A shared spend ledger caps every run; nothing generates unattended without a cap and an explicit confirmation.

The lego wall

Built and running today. Extended for your workflow on request.

BUILT — you can see each one run

■Node canvas — wire edits visually, rerun any branch
■Six goal presets: staging, object removal, cleanup, sky & lighting, life-adding, rendering enhancement
■Masked edits with auto-alignment — untouched pixels stay identical
■In-app Claude chat that authors and runs pipelines (approval-gated)
■Ratings on every result + a learning pass that sharpens future prompts
■Bounded auto-loop: run → AI-judge → keep/defer, inside a hard budget envelope
■Full provenance: every image carries the recipe that made it; every run replays from its journal

BUILD-ON-REQUEST — honest label: not built yet

□Hookup to your DAM / CRM / listing feed
□Your brand style baked in as house presets
□AI upscaling and background-removal providers
□Multi-user cloud deployment (today it runs single-team, local-first)
□Off-peak batch tier at half model cost

We don't sell novelty — we sell a battle-tested machine plus the work of fitting it to your business. If a piece you need is missing, that's a build item, not a secret.

Three doors

See it, try it, or make it yours.

Watch the full run

The recorded end-to-end session: a real shoot in, approved exports out, costs on screen.

Watch above →

Try it on an invite

We fund a capped test budget; you bring a few photos and run real edits yourself. Invites are personal, by hand.

Request an invite →

Integrate it

The system, adjusted to your pipeline and owned by you — scoped and built by pravda.systems.

Talk integration →

Raw building photos in.Marketing-ready images out.