Laydyne

Venues & events / WORKED EXAMPLE

Market Hall, Charleston — modelled from photographs

The main hall of a real market building, modelled from ten interior photographs: walls, windows, doors, stage and coved ceiling, checked by drawing the model from each photograph’s viewpoint.

Open and edit this layout without signing in. Sign in to connect your own AI. Save a project file to keep your changes.

How far do interior photographs alone take a 2D plan and a 3D model, and how far off are the sizes?

Who uses it
Anyone with photographs of a space and no drawings
How it can create value
Get an editable model of an existing room from photographs, then correct it with a single measurement instead of a survey.
Bring your site data
Interior photographs (exterior ones give only the shell), one known size if you have it, and the measured drawings only to check.
Turn a comparison into a business decision →
SAME EDITABLE PROJECT
Actual Laydyne 3D model: Market Hall, Charleston — modelled from photographs

01 / SETUP

A model you can inspect.

A real room, not a synthetic one: the upper hall of Market Hall, 188 Meeting Street, Charleston, from ten public-domain HABS photographs, nine of them published here. An AI agent estimated 19 × 9 m inside the walls, a cornice at 4.2 m and a flat ceiling at 5.8 m, placed four windows in each long wall, the west double door, the east window-door to the balcony, the stage and a coved ceiling, and built it through ordinary MCP tool calls. Rendering the model from each photograph’s estimated camera showed the hall too wide and the windows too low, and one correction followed. The measured drawings, opened only afterwards, give 17.17 × 9.42 m and a 6.83 m ceiling.

Interior photographs → sizes estimated from a door, the window panes and the ceiling panels → walls, openings, stage and ceiling built over MCP → the model rendered from each photograph’s viewpoint → corrected → 2D plan

Change one thing

Give the model one measured size, such as the hall’s 9.42 m width, and scale the rest; or add the vestibule and the porch from exterior photographs.

Read the consequence

Compare the model with each photograph from the same camera, read the plan’s dimensions and check the clearances.

View the starting inputs
Hall length (m, estimate)
19
Hall width (m, estimate)
9
Flat ceiling height (m, estimate)
5.8

These dimensions are assumptions for the example.

02 / LAYOUT CHECKS

Dimensions and quantities

Hall inside the walls (estimate)
19 × 9 m
Measured on the HABS plan
17.17 × 9.42 m
Windows in each long wall
4
Flat ceiling (estimate / measured)
5.8 / 6.83 m
Floor area drawn (m²)
200

These sizes were estimated from photographs before the drawings were opened: the model shown is off by 4–27 % on its main dimensions and 31 % on the window width. Give the model one measured dimension, or the drawings, before using it for anything that depends on size.

03 / FROM PHOTOGRAPHS

A real hall from ten interior photographs: four of them beside the model from the same viewpoint

No drawings were used. An AI agent read the interior photographs, took its scale from a door leaf, the window panes and the ceiling panels, built the walls, windows, doors, stage and coved ceiling with ordinary MCP tool calls, and drew the model from each photograph’s estimated camera (get_view_image with eye, target and fov) to see where it was wrong. The photograph is on the left, the model from the same viewpoint on the right.

The 1:100 plan of the same model, drawn by get_view_image as an A3 sheet with the ceiling parts left out
The 1:100 plan of the same model, drawn by get_view_image as an A3 sheet with the ceiling parts left out

Against the measured drawings

The HABS main floor plan and section were opened only after the model was exported. The blind estimate (v1) was committed before any drawing was seen; the corrected model (v2) was not committed before the drawings were opened, so for v2 that order rests on the record’s own account. The model shown is v2, corrected from the photographs alone.

ItemMeasured (HABS)Blind estimate (v1)After the photo comparison (v2, the model shown)
Width inside the walls9.42 m10.0 m (+6 %)9.0 m (−4 %)
Length inside the walls17.17 m19 m (+11 %)19 m (+11 %)
Flat ceiling6.83 m5.6 m (−18 %)5.8 m (−15 %)
Underside of the cornice5.74 m4.1 m (−29 %)4.2 m (−27 %)
Windows in each long wall444
Window glass width≈ 1.8 m1.4 m (−22 %)1.25 m (−31 %)
West door opening≈ 2.4 m2.4 m2.3 m (−4 %)

Every height came out about a fifth low because the scale was taken from a window pane assumed 0.33 m wide; the real panes are about 0.6 m. One measured size, such as the door height or the hall width, would have corrected the whole model in proportion. Photographs alone give an estimate, not a survey.

Exterior photographs give the shell: the outline, the storeys and the openings in the façades. Rooms come from interior photographs, as here, where the agent builds what it can see and leaves the rest out: this model is the hall alone, without the vestibule, the porch or the ground floor.

Try it with your own photographs

Connect your AI to Laydyne, give it a few photographs of a room and, if you have one, a single measured size. Ask it to build the room, then to draw the model from where each photograph was taken with get_view_image and compare. The photographs above can be downloaded to try the same hall.

Downloads

HOW IT WAS BUILT

From ten photographs, through MCP

An AI agent read ten interior photographs of the hall (four are shown on this page; nine are kept in the record), estimated its sizes, built the model with ordinary MCP tool calls, drew it from each photograph’s viewpoint to compare, corrected it once and exported it. The measured drawings were opened only afterwards.

StepMCP toolsMCP calls
Ten interior photographs read (four shown on this page); width, length, cornice, ceiling, windows and doors estimated from a door leaf, the window panes and the ceiling panels, and written down before any drawing was opened——
A fresh project; the walls to the cornice, eight windows, the west double door, the east window-door, the stage, the cove and the flat ceiling built from the estimate, then drawn from eight photographs’ viewpoints and laid beside themget_space · create_space · apply_changes · get_view_image14
The width, the window height and the cove corrected from the comparison; rebuilt as a fresh project and drawn again from the same viewpointsget_space · create_space · apply_changes · get_view_image12
The project exported, the 1:100 plan drawn and the floor checked: 0 issuesexport_project · get_view_image · check_space3

29 MCP calls in all; the drawings were opened only afterwards, to measure the error. Registering the public example from the same estimate took 25 calls, named save and reopen included.

04 / MAKE IT YOURS

Plan in 2D. Inspect in 3D. Keep the whole project.

  1. Open and adjust. Switch between 2D and 3D; edit a part, a floor or an input. Both views describe the same model.
  2. Connect your agent for free. Sign in, choose Connect AI and connect ChatGPT, Claude, Codex or Claude Code. Ask it to inspect the example, edit the layout and check a route.
  3. Save a checkpoint. Project → Project file, or ask the agent to export the complete project. Import it later to restore floors, scenarios, saved comparisons and the selected view.