Ghost Town

No more ghost towns.

Type an address in Blender and get the city around it: buildings, terrain, roads, sidewalks, water, parks, trees and lot lines, built from open data as clean geometry you can take into Revit or any BIM tool. In Toronto it can also show what is proposed and going up around the site.

1140 Dundas St W at Ossington, Toronto, 150 m radius: find the address, build, and switch the ground between colours by kind and the aerial photo. The build is shortened here. Data: City of Toronto and Natural Resources Canada (see Data and credits).

Ghost Town is a free, open-source extension for Blender 5.2 and later. It is at an early stage (version 0.5): Toronto is covered in depth, and the rest of the world gets buildings only.

What you get

Where the site is	Buildings	Terrain	Ground, trees, lot lines	Address search
City of Toronto	City 3D Massing model, newest edition: stepped parts with measured heights	NRCan elevation, 2 m grid	Roads, sidewalks, parking, rail, water, parks, trees, parcels	Yes
Elsewhere in Canada	OpenStreetMap	NRCan elevation where available	Plain ground only	Not yet (type coordinates)
Rest of the world	OpenStreetMap	Flat	Plain ground only	Not yet (type coordinates)

Each build makes one collection, Context · <site>:

• Buildings: one object per building. Each tier is a closed prism standing on the terrain. Buildings that cross the circle come in whole. In Toronto, parts of the City's model that touch or overlap form one building object, and where parts overlap the taller one wins. On a main street a whole block face can become one object; Separate › By Loose Parts splits it in Blender. Elsewhere in Ontario, with LiDAR roofs ticked, each building also gets two roofs measured from the province's LiDAR: a fitted roof of a few planar faces, which a build shows, and the sampled LiDAR surface. You can switch between them and the flat top. Buildings from the City of Toronto's model keep its massing, the same flat-topped blocks the City publishes as SketchUp and AutoCAD files.
• Ground: one draped surface per kind (road, sidewalk, parking, rail, water, green, plain ground). The pieces share their edges, so there are no cracks or overlaps. Water lies flat at its shoreline.
• Trees: a trunk and a low-poly crown each, sized from the City's tree heights, merged into one object.
• Parcels: lot lines draped 15 cm above the ground.
• Aerial photo (Toronto): the City's newest aerial photo of the site, kept inside the .blend file, for the ground and low roofs. See Use.
• Development applications (Toronto, when ticked): a see-through box on every nearby site with an open development application, a building going up or one just finished, coloured by where it stands: under construction, recently built, appealed, under review, approved, or Committee of Adjustment. The boxes are in an Applications · <site> collection. Each box sits on its lot and is as tall as the building permit or the application's description says, else as the permit's floor area spread over the lot, else one storey. Applications and permits filed on the same lot become one box, as tall as the newest of them that states a height, a building permit's own height first. See Use.
• Street Look (a step after the build, with your own Mapillary token): facades coloured from street photos, with storefront, wall, glass and cap zones and window patterns, plus floor bands and mullions on a few buildings you pick. See Street Look.
• Materials: one per kind, named Context - Building, Context - Road, and so on. A building whose height had to be guessed is orange (Context - Building (height guessed)).
• Location: a Context origin empty at 0,0,0 holding the latitude, longitude, ground elevation above sea level, the origin's survey grid coordinates and the data credits.

Everything is in metres, with x east, y north, the address at the origin and z = 0 at its ground level.

235 Queens Quay W, Toronto, 300 m radius, built in about 8 seconds.

320 Bay St, Toronto, 300 m radius.

Install

Ghost Town needs Blender 5.2 or later on macOS (Apple silicon or Intel), Windows x64 or Linux x64. It bundles the libraries it needs (shapely, and Pillow for Street Look), so there is nothing to pip install.

1. Download the zip for your computer from the latest release:

   Your computer	File
   macOS, Apple silicon	ghosttown-<version>-macos_arm64.zip
   macOS, Intel	ghosttown-<version>-macos_x64.zip
   Windows	ghosttown-<version>-windows_x64.zip
   Linux	ghosttown-<version>-linux_x64.zip

2. In Blender, go to Edit › Preferences › Get Extensions › ⌄ › Install from Disk… and pick the zip.

3. Go to Edit › Preferences › System › Network and turn on Allow Online Access.

Building it yourself

git clone https://github.com/winterweken/GhostTown.git
cd GhostTown
tools/build.sh

tools/build.sh downloads the shapely and Pillow wheels from PyPI and writes one zip per platform into dist/. It expects Blender at /Applications/Blender.app; on other systems, set BLENDER to your Blender executable first. Install the zip as above.

Use

1. In the 3D View, open the sidebar (N) and the Ghost Town tab.
2. Location: type a Toronto address such as 320 Bay St and press the search button beside it. If several addresses match, pick one. Anywhere else, type latitude, longitude, for example 51.5074, -0.1278.
3. Site name (optional) names the collection. A found address fills it in.
4. Radius: 150, 300, 500 or 1000 m.
5. Fetch: tick Aerial photo (on by default) to fetch the City of Toronto's newest aerial photo of the site with the build. It is kept inside the .blend file. Tick LiDAR roofs (slower) in Ontario to fetch the province's LiDAR and give buildings outside the City of Toronto's 3D Massing model measured roofs. It is a large download, and the province's server can take a minute to answer the first request. Tick Development applications to put a box on every nearby site with an open development application, under construction or finished since the City's 3D Massing (City of Toronto applications and building permits; houses left out). The first build of the day downloads the City's tables and can take up to a minute more.
6. Press Build Context. A 300 m site takes about 10 seconds and a 1000 m site about 20. With LiDAR roofs ticked a build takes a few minutes more, because the province's server is slow to send the LiDAR (a 300 m site in Hamilton took about 4 minutes). Cancel or Esc stops it, and Ctrl+Z removes a finished build in one step.

Building the same site again replaces what Ghost Town made and keeps anything you added, including your own objects and collections inside the context collection.

The Site section under the panel shows one site at a time; pick it at the top. It lists the site's survey point, and, when the site has an aerial photo:

• Ground: Colours | Photo shows the photo on the ground, or the colours by kind. Shows in Material Preview, or Solid view with Color: Texture.
• Roofs: Plain | Photo puts the photo on the roofs of buildings up to a height you set (20 m to start). Taller buildings lean in the photo, so their roof texture would be offset.
• Save Site Photo… writes the photo and a world file, for an underlay in Revit or CAD.

When the site has LiDAR roofs:

• Roof shapes: Flat | Fitted | LiDAR switches every building between its flat-topped prism, its fitted roof and its LiDAR roof. Fitted shows after a build: a house gets the flat, shed, gable or hip roof that best explains the LiDAR, its planes running out to the walls, and a larger building gets stepped tiers at its measured roof levels, at least a storey apart. Every fitted roof is a closed solid. Sites built before fitted roofs existed show Flat | LiDAR.
• Roof detail simplifies LiDAR roofs, from 100 % down to 5 %, without moving walls or eaves.
• For Revit shows the site's triangle count. Past the budget in Preferences (500,000 to start) it says: Heavy for Revit: use Fitted or Flat roofs, or lower Roof detail, before exporting.

Wismer, Markham, 150 m radius with LiDAR roofs: the houses switch between flat tops, the LiDAR surface and fitted roofs, and the triangle count for Revit follows. The LiDAR download, a few minutes, is cut here. Data: OpenStreetMap, Natural Resources Canada and Geospatial Ontario.

Switching never downloads anything again, and each site in a file keeps its own choices.

When the site has development application boxes, Development applications lists them by status, each with its colour (change it there to restyle every box of that status). Select a box to see its applications and permits, their status, date and description, and Open in City AIC for the City's page about each application. Reshape a box as you like: grab, rotate, scale (the Scale Cage tool moves one side at a time) or edit its mesh. Building the site again refreshes each box's status and keeps the shape you gave it; a box whose applications all closed is removed, or turned grey (Closed) if you changed it. A box you delete stays deleted; Bring Back Deleted Boxes makes the next build put them back.

Street Look

Street Look gives a site's buildings colours, glass and storefronts read from Mapillary's street photos: a dark storefront, a brick-coloured or concrete-grey body, glass with mullions, a cap, at measured heights, with windows spaced by the floor height it finds. It is a separate step after Build Context and needs your own free Mapillary token: create one at mapillary.com/dashboard/developers and paste it in Preferences › Add-ons › Ghost Town › Mapillary token (or set GHOSTTOWN_MAPILLARY_TOKEN). The token is saved in Blender's preferences like any other setting; the field only hides it on screen.

1. Build Context, then open Street Look under the Site panel. It works on the site picked there.
2. Optionally select up to 20 buildings (about 5 is plenty) and tick Detail for selected to model floor bands, a storefront band and mullion fins on them, in a Detail · <site> collection.
3. Press Apply Street Look. A 300 m site in Toronto takes from about 10 seconds to 2 minutes (the densest downtown blocks are the slow end), and up to 2 minutes more the first time, while the photos download. Cancel or Esc stops it; Ctrl+Z undoes it.
4. Show street look, Show detail and Photo brightness change the result without fetching again. Add Sky, shown while the scene has no world or Blender's default one, adds a sky texture and a sun so glass has something to reflect.

A building no photo shows clearly gets a plain storefront-and-body look, and the panel counts it as guessed: "Look from photos: 23 buildings · guessed: 1 · 38 photos (2014–2025)". Photo budget (150 to start) caps the photos read, nearest buildings first; Not before leaves out older photos. With Keep street look on rebuild (on), Build Context puts the look back on the buildings that are still there. The look is a shader driven by values on each building, so material names stay as they are, roof shapes and the roof photo keep working, and no photo is stored in the .blend file.

Taking it into Revit

Ghost Town does not export. It keeps the geometry friendly to exporters and to Revit instead: metres, closed building solids, outlines cleaned of edges under 3 mm, and stable material names.

• OBJ: set Up Axis: Z (Blender's default is Y).
• FBX: tick Loose Edges if you want the parcel lines.
• In Revit, import origin to origin. The material names become the layers or materials you control under Object Styles › Imported Objects.
• Before exporting, set the site's Ground to Colours and Roofs to Plain, so every piece keeps its Context - … material. While the photo shows, exporters see the photo material on those faces instead.
• Fitted roofs export as closed solids of a few faces each: the lightest measured roofs, and the ones to take into Revit.
• Street Look changes how buildings render, not their materials' names. Its detail meshes sit in the Detail · <site> collection with the Context - Facade detail material and count toward the triangle figure; exclude that collection (untick it in the Outliner) to leave them out of an export.
• Development application boxes are closed boxes with a Context - Application (<status>) material each, in the Applications · <site> collection; exclude that collection to leave them out of an export.
• LiDAR roofs export as meshes, which Revit imports as DirectShapes: heavier than prisms. Exporters apply Roof detail (modifiers are applied by default), so keep the site under the panel's budget, or use Fitted or Flat roofs.
• To use the aerial photo in Revit, Save Site Photo…, then in a site plan use Insert › Image, set the image's width to the width Ghost Town reports (twice the radius, in metres), and centre it on the origin.
• Survey point. The model stays at the origin, which is the site's latitude and longitude at ground level. After a build, the panel shows where that origin sits on the survey grid, and its copy button puts the values on the clipboard. The grid is the City's own in Toronto (NAD83(CSRS) / MTM zone 10, EPSG:2952) and the site's UTM zone elsewhere. The same values are on the Context origin empty as survey_easting_m, survey_northing_m, survey_elevation_m and survey_grid_angle_deg.
• To set Revit's survey point, go to Manage › Coordinates › Specify Coordinates at Point, pick the model's origin, and enter the northing, easting and elevation. Ghost Town's +y is true north, while Revit's true north is the survey grid's north; enter the grid angle as the Angle from Project North to True North, East or West as the panel says.
• The elevation is the height above sea level of z = 0 (ground_at_centre_m). Latitude and longitude are used as given, which matches the City's data; positions from OpenStreetMap or a phone can be 1–2 m off. The survey point lines up context, not a legal survey.

Data and credits

Ghost Town downloads open data when you build. If you publish anything made with it, credit the sources you used. The panel lists them after every build.

Source	Used for	Credit
City of Toronto open data	Buildings (3D Massing, newest yearly edition), aerial photo, ground, trees, parcels, addresses, city boundary, development applications, building permits	Contains information licensed under the Open Government Licence – Toronto
Natural Resources Canada (HRDEM)	Terrain	Contains information licensed under the Open Government Licence – Canada
Geospatial Ontario (lidar-derived surface and terrain models)	LiDAR roofs in Ontario	Contains information licensed under the Open Government Licence – Ontario
OpenStreetMap	Buildings outside Toronto	© OpenStreetMap contributors (ODbL)
Mapillary	Street Look: facade colours read from street photos	Street photos © Mapillary contributors, CC BY-SA 4.0 Labels from Mapillary · https://www.mapillary.com

Street Look's photos and labels come from Mapillary:

More detail is in CREDITS.md.

Accuracy. This is context for early design, massing and presentation, not survey data. Building heights are derived from aerial data, some are guessed, and lot lines are approximate. Check anything you rely on against a survey.

Privacy. Apart from the Mapillary servers Street Look adds (below), Ghost Town contacts only gis.toronto.ca and the City's open data portal (ckan0.cf.opendata.inter.prod-toronto.ca), datacube.services.geo.ca and overpass-api.de, and, only with LiDAR roofs ticked, ws.geoservices.lrc.gov.on.ca, and only when you press the search button, Build Context or Apply Street Look. It sends what the query needs (the address you search for, or the location and radius you build) and nothing else. Answers are cached on your computer for 30 days; with Development applications ticked, the City's application points and its applications and building permits tables are kept for a day (about 17 MB, cleared by the next day's build) and its address points for a week. The City's 3D Massing model is downloaded once per yearly edition (81 MB, about 300 MB unpacked in the cache folder) and kept until a newer edition comes out. It lives in the cache folder (Preferences › Add-ons › Ghost Town › Cache folder, which shows where it is; by default ~/Library/Application Support/Blender/<version>/extensions/.user/user_default/ghosttown/cache on macOS, with <version> 5.2 today, a hidden folder, and the matching Blender extensions folder on other systems); deleting its toronto_massing folder is safe, and the next Toronto build downloads it again. With Aerial photo ticked, the photo comes from gis.toronto.ca with the build and is kept inside the .blend file, which adds up to about 5 MB. LiDAR answers (10–30 MB a site) are cached like the rest, and LiDAR roofs are kept inside the .blend file, about 20 to 30 MB for a dense 300 m site (less with Compress, in File › Save As or Preferences › Save & Load).

Apply Street Look contacts graph.mapillary.com (photo listings and labels) and Mapillary's image servers on fbcdn.net, and only then. Inside Canada, when the site was built with ground heights, it also asks datacube.services.geo.ca for ground heights over the photo search area, as Build Context does for the site. graph.mapillary.com receives the area around the site, the ids of the photos read and your Mapillary token, which Ghost Town sends only in a request header and never stores in a .blend file; the image servers receive only the link for each photo read. Each user registers their own free application with Mapillary, so the request limits are theirs. Photos and labels are cached on your computer for 30 days, in the cache folder set in Preferences › Add-ons › Ghost Town › Cache folder (delete its mapillary folder to clear them; the whole cache folder also holds the City's massing model); the photos are the blurred thumbnails Mapillary serves, kept as they are; only the colours and heights read from them reach the .blend file.

Ghost Town is an independent project. It is not affiliated with or endorsed by the City of Toronto, Natural Resources Canada, the Province of Ontario, OpenStreetMap or Mapillary.

Known limits

• Outside Toronto there are no roads, trees or parcels yet, and outside Canada the ground is flat.
• Buildings are flat-topped prisms unless they have measured roofs (Ontario, outside the City of Toronto's 3D Massing model). LiDAR roofs carry no roof planes, just measured points.
• Fitted roofs give a house one roof for its whole outline, so an L-shaped house whose wings have their own ridges gets the main roof's planes, and a house under thick canopy may keep a flat top at its measured height. Larger buildings become tiers at least 3 m apart, so their sloped or curved roofs become steps.
• LiDAR roofs on house-sized buildings (up to 400 m² and 20 m tall) are trimmed to their own typical height plus 3 m, which removes a tree crown over part of the roof; trees over most of a roof still raise it.
• A building newer than the province's LiDAR survey keeps its flat top in Fitted and LiDAR modes, and one surveyed while under construction can show a partly built roof.
• Bridges and elevated rail are draped onto the ground.
• A stream is flat at one level along its length instead of following its valley.
• Rebuilding a site resets colours you changed on the Context - … materials (the development application colours stay).
• Development application boxes are starting volumes, not the proposals' designs: the largest rectangle that fits the lot, one height for the whole box. On a lot as big as a block, separate projects become one box, so a finished building there can take the height of a tower proposed beside it.
• Building permits for houses are left out, so a house going up or just finished gets a box only when it also has an open application.
• Street Look reads facades from street photos, many from 2014–2019: glass in the lower floors that mirrors the street, or tall glass seen in only a few photos, can read as an opaque wall; sky at a roofline can tint a building's top; a building changed or built since the photos gets what they show.
• Street Look's windows and mullions are evenly spaced (1.5 m on glass, 3 m on walls), not measured on each building; only the floor height is measured, and it is 3.5 m where the photos don't settle on one.

How it works

The Blender side only draws. All downloading and geometry work happens in a separate process running Blender's own Python, so the interface stays responsive, Cancel is immediate, and a problem there cannot take Blender down.

sidebar panel ─▶ request.json ─▶ fetcher process ─▶ context.json ─▶ collections, meshes, materials
                                  (ghosttown_fetch)

• ghosttown/ghosttown_fetch/ is plain Python (numpy, shapely and Pillow, no Blender). Each data source is one module in sources/, which is where new cities and layers go.
• ghosttown/*.py is the extension: panel, operators, the process runner and the scene builder.
• Street Look is another fetcher command, look: it reads look_request.json and writes look.json, with the Mapillary token in the process's environment only.

Roadmap

• Worldwide address search.
• OpenStreetMap roads, green space, water and trees outside Toronto.
• Terrain outside Canada.
• A site outline: cut your site out of the ground and mark its buildings and parcel.

Development

uv sync                    # Python 3.13 with numpy, shapely and Pillow, matching Blender 5.2
uv run pytest              # fetcher tests; they run offline against recorded answers
/Applications/Blender.app/Contents/MacOS/Blender --background --factory-startup \
  --python-exit-code 1 --python tests/blender/run.py     # Blender-side tests
tools/build.sh             # per-platform packages in dist/
tools/smoke_installed.sh "320 Bay St" 300   # macOS: install into a throwaway profile, build a live site
                                            # (and apply Street Look when GHOSTTOWN_MAPILLARY_TOKEN is set)
uv run python tools/look_accuracy.py        # Street Look on five Toronto sites (needs the token)

Issues and pull requests are welcome. New data sources are the most useful contribution: a source is one module that turns a public dataset into polygons, points or lines.

Licence

GPL-3.0-or-later. See LICENSE. The data Ghost Town downloads keeps its own licence, listed under Data and credits.

Recorded test data in tests/fetch/fixtures/ keeps its sources' licences (see tests/fetch/fixtures/README.md): OpenStreetMap data under the ODbL, City of Toronto, NRCan and Ontario data under their Open Government Licences, and six Mapillary street photos under CC BY-SA 4.0 with, beside them, Mapillary's own data about those photos (labels and listing records), for which Mapillary states no licence. The GPL does not cover them, and the extension package does not include them.

The Ghost Town name and logo are © winterweken and are not covered by the GPL. That includes the files in media/brand and the icon in ghosttown/icons/. You may use them to refer to this project; please ask before using them for a modified version or another product.