AI Image Compositor for CRE Marketing

A NYC commercial real estate marketing agency was losing 2–3 days per project to image editing done by hand, one picture at a time.

Every office-building website needs photo work before launch: empty offices staged with furniture, scaffolding removed from exteriors, rough renderings polished. The client did all of it manually — 5–10 prompt iterations per image in the OpenAI playground, with no batch and no record of what worked. The AI output was already good enough; the system around it was the bottleneck.

A node-based image compositor batches the entire edit queue in one supervised sitting instead of grinding through each image by hand.

The operator wires a visual pipeline — Image → Generate → Mask → Merge → Export — then runs the whole project at once. Claude reads the client brief and photos and authors the pipeline itself, so the operator reviews and runs rather than building from scratch. After each session, Claude distills what worked into a living rulebook, so the next project starts sharper.

The AI changes only the masked region — every other pixel comes back byte-for-byte identical to the source.

The Merge step registers the AI edit back onto the original using computer-vision feature matching (ORB/AKAZE) and composites through the mask in one pass. A confidence valve degrades to manual placement rather than forcing a bad blend. This solves the core failure of masked AI edits: the untouched region distorting.

A hard spend cap plus an append-only journal means a crash mid-batch resumes with zero re-billing.

Before every paid API call, the engine writes the request to disk; on restart it skips completed work via idempotency keys. An opt-in full-auto tier can generate, judge, and retry on its own — bounded by a dollar cap, a time limit, and a kill switch — but human taste stays in control by default.

409 commits in one focused build, validated end-to-end on live OpenAI for $0.235.

The full edit path ran against real gpt-image-2, confirming the spend cap and retry loop in production for a fraction of a dollar. The codebase is a five-package TypeScript monorepo (32k TypeScript, 7.7k TSX UI) plus a Python computer-vision sidecar. Keyless replay paths stay green in CI without ever spending.