Enhancing real estate listings with AI—without changing the space

For an agent, editing is only one step among securing the listing, visiting the property, writing the description and distributing it. Yet the work repeats every time: select the files, correct orientation, brighten a dark room, balance colour, remove temporary clutter and export a consistent gallery.

Sparkle already runs that pass in a few minutes. The agent uploads a gallery, starts enhancement, compares every result with its source and downloads web-ready JPEGs. Originals remain available, so the decision to publish always belongs to the user.

Preservation is not only marketing copy. The production prompt limits enhancement to removable clutter, cleaning, light and colour. It forbids inventing, replacing, moving or redesigning elements, and requires the exact camera position, crop, architecture, geometry and spatial proportions to remain fixed.

Those rules give the current mode a clear boundary: improve what exists. They cannot guarantee that a generative model will never drift, so the interface keeps the original and presents every result as something to review—not as an automatically publishable truth.

Before

After

Trust therefore comes less from a spectacular image than from an understandable rule: improve the presentation without reinventing the property.

The source work now in progress introduces one contract for local imports, Google Drive, Dropbox and S3-compatible storage including AWS S3, MinIO and Cloudflare R2. Each adapter can traverse files, normalise errors and send photos through the same import and enhancement pipeline. Automatic synchronisation applies to connected sources.

A user connects an account in Settings and chooses one parent folder. Each direct child folder then represents one property; compatible photos inside it are discovered recursively. The first scan records existing folders and presents them as import candidates, with previews loaded from the source but no silent listing creation. After that initial inventory, every new child folder can automatically create a draft listing.

Two guardrails keep convenience from becoming destructive: deleting a remote file never deletes anything in Sparkle, and an automatic import never consumes a separately purchased credit.

The automation adopts a convention many agencies already use: one folder per property. Sparkle does not ask them to replace their storage tool or repeat a file selection for every listing.

1. 01

   Connect a source

   The user authorises Drive or Dropbox, or configures compatible S3 storage. Persistent tokens and credentials are encrypted.

2. 02

   Choose the parent folder

   Sparkle records existing child folders and makes them browsable. Importing one remains an explicit action.

3. 03

   Add the next property

   After the initial inventory, a new child folder is queued automatically and becomes a draft with its photos.

4. 04

   Enhance, then publish

   The shared pipeline normalises, enhances and versions the images. The agent compares, selects and exports.

Automatic import can be disabled independently for each synchronised root. The saving comes from removing repetition, not removing the agent’s control.

Enhancement is the product available today: one short action improves an existing photo within a strict boundary. Connected sources and automatic synchronisation are still in preview and are not presented as already shipped.

Home staging and renovation are also in development, with no public release date. They require more user direction because their intent is different: staging adds movable furniture; renovation proposes new finishes or permanent equipment. Both continue to hold the room envelope, openings, proportions and viewpoint as the reference.