Adds doors and windows
Will routinely add a second window or French door because "rooms like this usually have one." On MLS that is a material misrepresentation. The seller, the buyer's agent, and the listing brokerage all share the liability.
/assets/_shared/backgrounds/landing-page.v1.jpg)
Most virtual staging tools were built for pretty pixels. Edensign was built for the MLS — walls, windows, ceilings, and floors stay pixel-locked between source and output, disclosure footers attach automatically, and every render carries an audit log your compliance officer can pull. No hallucinated rooms. No added windows. No misrepresentation risk.
Used by compliance-sensitive brokerages and MLS-affiliated photographers across the US
/assets/gallery/bedroom/case-1-before.v1.jpg)
/assets/gallery/bedroom/case-1-after.v1.jpg)
The MLS-grade dataset
Generic image models train on aspirational imagery. We trained on the corpus an MLS compliance officer would recognize: real listing photography, with structural elements labeled, disclosure rules annotated, and edits explicitly bounded to staged content only.
Real listing photography sourced from MLS partners, brokerage archives, and professional listing photographers — every image licensed for AI training.
Walls, windows, doors, ceilings, floors, built-ins — each labeled per image so the model is explicitly trained to leave them untouched.
Disclosure language, watermark requirements, and prohibited modifications captured per state and per major MLS board.
Every training pair was reviewed for whether the staging-vs-source pair would meet MLS disclosure standards in California, New York, Florida, and Texas.
How the model is trained for compliance
Stage 1
The model is pre-trained with explicit structural element labels — walls, windows, doors, ceilings, floors, baseboards, built-ins. It learns to identify and protect these elements before it ever learns to add furniture.
Stage 2
Trained on millions of before/after listing pairs with a loss function that explicitly penalizes any structural drift. The model is rewarded for staging-only edits and punished for any change to load-bearing visual elements.
Stage 3
Disclosure language, prohibited modification lists, and state-specific staging rules are encoded into the training and inference pipeline. The output ships with the right footer for the listing's jurisdiction.
Stage 4
Final alignment pass reviewed by working MLS compliance officers, listing photographers, and broker-of-record professionals. They reject outputs that would create disclosure liability — added rooms, removed structural elements, fabricated views.
MLS compliance checklist
Staging does not introduce rooms, alcoves, or alcove openings that do not exist in the source.
Load-bearing walls, columns, beams, and structural posts stay byte-for-byte the same.
Window count, door count, and opening positions match source exactly.
Ceiling planes and heights preserved; no implied higher-than-reality vaulting.
View through windows stays the actual view; no fabricated ocean / mountain / skyline.
"Virtually staged" footer applied per MLS jurisdiction, in the position the local board requires.
Output resolution matches source; no downscaling that would hide source-vs-output diff at zoom.
Every render carries a timestamped log of source hash, model version, and operations applied.
Original photo retained alongside the staged output for as long as the listing is active.
Optional persistent watermark for MLSs that require visible virtual-staging identification.
Benchmarks · MLS compliance
Measured on a held-out set of 1,000 real listing photos against MLS-relevant geometry and compliance metrics.
| Metric | Edensign | Generic image AI |
|---|---|---|
| Wall-line preservation | 99.4% | 71% |
| Window geometry fidelity | 98.1% | 54% |
| No hallucinated rooms/openings | 99.8% | 78% |
| Built-in / fixture preservation | 99.2% | 63% |
| MLS compliance rate (reviewed) | 99.5% | — |
vs. Generic image AI
Generic image models were not built with the MLS rule book in mind. The failures are predictable.
Will routinely add a second window or French door because "rooms like this usually have one." On MLS that is a material misrepresentation. The seller, the buyer's agent, and the listing brokerage all share the liability.
Subtly resizes a window, tilts a ceiling, or shortens a wall so the generated furniture fits. The room is recognizable but no longer the listing — and a buyer comparing online to in-person will notice.
Output ships with no virtual-staging disclosure, no audit log, no source retention. When a buyer disputes the listing 6 months later, there's no documentation chain. The brokerage's compliance officer has nothing to pull.
Examples
Each pair below was edited by Edensign in a single pass. Walls, windows, ceiling, and floor stay pixel-locked.
/assets/gallery/bedroom/case-1-before.v1.jpg)
/assets/gallery/bedroom/case-1-after.v1.jpg)
/assets/gallery/living-room/case-1-before.v1.jpg)
/assets/gallery/living-room/case-1-after.v1.jpg)
/assets/gallery/kitchen/case-1-before.v1.jpg)
/assets/gallery/kitchen/case-1-after.v1.jpg)
State-by-state MLS rules
State pages roll out monthly. Need rules for a state not listed? Book a demo and our compliance team can deliver a per-state brief.
Yes — virtual staging is allowed on every major US MLS, provided the listing discloses that the photos are virtually staged and provided the staging does not misrepresent the property. Disclosure language and placement vary by board; Edensign applies the right footer per jurisdiction automatically.
It means the model is explicitly trained and constrained to preserve walls, windows, doors, ceilings, floors, baseboards, and built-ins between source and output. We benchmark this at 99%+ pixel-level fidelity on a held-out set of 1,000 real listing photos. The diff is downloadable on every render.
State-by-state rule pages are rolling out — California, New York, Florida, Texas, and Wisconsin (ahead of Act 69 in 2027) are first. Until those ship, our compliance team can provide the per-state summary on a brokerage demo call.
Yes. Every render carries a timestamped log with source hash, model version, operations applied, and the staging team member who triggered it. Compliance officers can pull the full chain on demand. Source photos are retained for the active life of the listing.
Wisconsin Act 69 (effective 2027) tightens disclosure language for virtually staged listing photos. Edensign tracks the act ahead of effective date and will auto-apply the required disclosure footer for WI-MLS listings on the day the act takes effect. Brokerage compliance officers can preview the disclosure flow on a demo call.
Yes. Edensign supports a brokerage-side approval gate: staging renders enter a queue that the compliance officer can review and approve / reject before the photo can be published to the MLS. This is part of our brokerage rollout — book a demo to see the workflow.
Individual agents can use the standard product (free trial, no compliance gate). Brokerages with compliance officers, audit requirements, or SSO needs should book a demo — the brokerage tier includes the approval gate, audit log access, and SSO/SAML.
Book a 20-minute walkthrough with our compliance team. We'll walk through the approval gate, audit log, and state-specific disclosure flow on a sample listing from your brokerage.
The technical foundation: a model trained end-to-end on architectural photography.
Editorial-grade output for $2M+ listings, on the same compliance backbone.
Custom brokerage rollout with white-label, SSO, and compliance gates.
How we measure and guarantee pixel-locked edits.