ExPlora4D is a web viewer for cultural heritage in 3D. Inspect the surface, annotate what you see, say how certain each part of a reconstruction is, and publish it all as an open IIIF manifest.
The model as it was published: photographic textures, scene lights and environment.shadingMode: "original"
Where the name comes from
Explore + Experience = ExPlora
The name joins two words that begin the same way. Explore is the researcher's side: searching a model, measuring it, questioning it. Experience is the visitor's side: walking round a building that no longer stands, or seeing an object up close. ExPlora4D is made for both.
Ex
Shared by both words, and drawn as the crossed X in the logo.
Plora
From explore, Latin explorare: to search out, to examine.
4D
The three dimensions of the model, and time, in which heritage changes.
Exploreto search out, to study closely
Experienceto see and live through for yourself
ExPlora4Dresearch and experience in one viewer
One manifest to share them all
Don’t share screenshots, share the ExPerience.
What is an AIM3Dex manifest?
A screenshot freezes one angle, one light, one moment. An AIM3Dex manifest captures the whole experience: the model, the exact view you set up, lighting and materials, annotations, renders and metadata, all in one lightweight, shareable file. Whoever opens it lands exactly where you were, and can keep exploring: rotate, zoom, inspect details and move through time. Built to work alongside IIIF, it turns a 3D object into a citable, reusable resource for research, teaching and heritage, not just a pretty picture.
Not a picture of the object, the object itself.
Screenshots show. Manifests ExPlain.
What a manifest holds, and under which key. Everything is plain JSON: readable, diffable, and kept in the same file as the IIIF scene.
The model and its scene
A IIIF Scene that paints the model by URL and format, with its units and, for several models, each one's position, rotation and scale.
items[].Scene
Cameras
Position, target and up vector, field of view or orthographic zoom: the exact view you set up. Several named scene cameras to switch between, the first the start view. Each says which screen it was set on, landscape or portrait, so a view made on a desktop still shows the whole model on a phone held upright.
AIM3DViewer.camera · IIIF cameras
Lights
Ambient, hemisphere, directional, point and spot lights, each with its colour, intensity, position, target, cone and falloff.
AIM3DViewer.lights
Rendering
Tone mapping and exposure, the environment map and its strength, the background, anti-aliasing and ambient occlusion.
viewer.rendering
Shading
Original materials, clay, matcap, toon, flat faces, normals, Phong or Lambert, or your own GLSL vertex and fragment shader; wireframe.
modelTransform.shadingMode
Raking light
A grazing light tied to the view: the direction it comes from, its height above the surface, and whether it sweeps.
AIM3DViewer.rakingLight
Place and sun
Where the object stands (latitude, longitude, north) and the day and hour of the sun that lights it, or the year of the timeline.
AIM3DViewer.location · sun
Sections
Clipping planes on three axes, where each one cuts, which side stays, and the section outline.
viewer.clipping
MeasurementsNew
Distances, angles and areas with their labels. The points are kept in the model's own coordinates, so the measurements stay on the model wherever it is placed and come back, in metres or feet, when the manifest opens.
AIM3DViewer.measurements
Annotations and tours
W3C Web Annotations on faces or points, titles and texts in several languages, a saved view for each; in order they make a guided tour.
annotations · viewer.tour
Level of certainty
The scale of a reconstruction, each level with its code, symbol, colour and label, and the assessments with their justification.
AIM3DViewer.certainty
4D timeline
Stages on a time axis: the year, whether it is documented or modelled, its label and source, the model parts it shows, and how playback runs.
AIM3DViewer.timeline
Variants and comparison
Alternative hypotheses with their parts, source and certainty, the one shown, and which two are compared, swiped or side by side; or two shading modes and lightings of the same model.
AIM3DViewer.variants · compare
Historical photos
Each photo or IIIF image with its year and source, and the camera it was taken from: position, target, field of view and roll.
AIM3DViewer.photos
Renders
The images for the gallery, in order, each with a label in any language.
AIM3DViewer.thumbnails
Viewer options
Theme, language, auto-rotation, hidden UI or metadata, notifications: how the page around the model behaves.
AIM3DViewer.viewer
The manifest is the view
A IIIF manifest with the viewer's state on board
Standard IIIF carries the scene. The AIM3DViewer block carries what a researcher set up on top of it. Viewers that don't know the block still show the model; ExPlora4D restores the whole reading.
Export from the toolbar after setting up the view.
Validate against the published schema; the viewer checks it on import too.
Share the file or its URL. Opening it restores cameras, lights, shading, annotations and measurements.
Reuse the look on another model: the viewer keeps the lights, rendering and shading and swaps only the model.
Each tool stores its state in the manifest, under the key shown beneath it. What you set up on screen is what the next reader opens.
Inspect
See the surface, not the texture.
Scan viewsOriginal materials, clay, matcap, normals, toon or your own GLSL, with ambient occlusion.modelTransform.shadingMode · postprocessing.ao
Raking lightA grazing light that follows the view, for inscriptions and tool marks.AIM3DViewer.rakingLight
Sections and measurementClipping planes and profiles; distance, angle, area and volume in real units. Section cuts and finished measurements are saved with the model and exported in the manifest, so they reopen where they were left.AIM3DViewer.clipping · measurements
Sun and comparisonThe sun of the model's place on any day, and two looks side by side. See Light.AIM3DViewer.sun · compare
Interpret
Say something about a part of the object.
AnnotationsClick or drag over faces; on a point cloud, a click annotates the point.W3C Web Annotation
Level of CertaintyEach element graded from surveyed fact to hypothesis, and the model coloured by it.AIM3DViewer.certainty
Compare hypothesesAlternative reconstructions or two years of the timeline, swiped or side by side.AIM3DViewer.variants · compare
Historical photosOld photographs laid over the model from where they were taken, to check it against its sources.AIM3DViewer.photos
Guided toursAnnotations as steps with saved views, played in order.viewer.tour
Publish
Hand the view on, intact.
IIIF manifestsExport and import IIIF Presentation 4 with the AIM3D block: cameras, lights, rendering, annotations, measurements.IIIF Scene + AIM3DViewer
Embed anywhereAn embed configurator writes the snippet; presentation mode hides the editor.viewer.presentationMode
Repository integrationA Drupal module for WissKI-based repositories, with metadata shown beside the model.
Screenshots and thumbnailsCapture the view, or render gallery thumbnails from nine angles.
Ingest
Bring in what the project already has.
ConversionTo glTF with Blender, IfcConvert (BIM) and OpenCASCADE (CAD), compressed with gltfpack, Draco and KTX2; ZIP, RAR and tar unpacked.
Point clouds and splatsLAS, LAZ, E57 and Gaussian splats, streamed by level of detail. See Point clouds.
VolumesMicroscope, CT and MRI stacks - Leica LIF, TIFF / OME-TIFF / Zeiss LSM, PNG / JPEG / BMP slices, NRRD, NIfTI, MetaImage, MRC, Volume Graphics VGI, Nikon XTEKCT, DICOM - drawn as a volume: maximum intensity, translucent or surface, with channel colours and, for CT, Hounsfield windows.
Sound and imagesAudio files - WAV / Broadcast Wave, MP3, FLAC, Ogg, M4A - and 2D images - JPEG, PNG, WebP, AVIF, GIF, BMP, alone or a ZIP of them - with their tags, markers, EXIF and resolution. See Sound and images.
BlenderAn add-on imports and exports AIM3D manifests: a scene set up in Blender opens in the viewer as it was.
Point clouds, from a quick look to a whole site
Laser scans and photogrammetric point clouds open in the browser as they are, or stream from an octree so that only the points the view needs are downloaded. Colour, size and lighting are worked out on the graphics card, and the tools work on the points themselves.
Eye-Dome Lighting
Each point is shaded by the depth of its neighbours, so edges and creases read without normals. A LAZ scan in UTM coordinates, opened directly.
viewer.pointCloud.edlStrength
By height, intensity or class
Colours are switched in the shader, never rewritten point by point. The adaptive point size follows the density of the scan and keeps surfaces closed at any distance.
viewer.pointCloud.colorMode · sizeMode
Streamed level of detail
The Wołpa synagogue as a streamed point cloud: the coarse levels arrive first, finer nodes load as you come closer.
Potree 2 · 3D Tiles
Profiles
Two clicks draw a line; a thin vertical slab of points along it stays, seen square from the side. Walls, floors and the roof in section, ready to measure.
point cloud panel · Profile
Gaussian splats, streamed like an octree
SPZ, SPLAT, KSPLAT, SOG or a PLY of splats, drawn with Spark. The server builds a level-of-detail RAD file once; the viewer fetches it in chunks, the coarse levels first, and only what is in view takes graphics memory.
SPZ · SPLAT · KSPLAT · SOG → RAD (build-lod)
Gaussian splats, streamed like an octree
3D Gaussian Splatting captures (SPZ, SPLAT, KSPLAT, SOG or a PLY of splats) sit next to meshes and point clouds, drawn with Spark. On upload the server builds their level-of-detail tree once, as a RAD file. The viewer reads its header and then fetches the tree in chunks of 64,000 splats by range requests: the coarse levels first, finer ones where you look. The capture shows after a few hundred kilobytes and sharpens as you look around, and only what is in view takes graphics memory.
SPZ · SPLAT · KSPLAT · SOG → RAD (build-lod)
Splats in the scene's light
Captures already hold the light they were taken in: the viewer does not replace their colours but scales them by the scene's lights. Normals, shadows and occlusion come from a volume of the surface built on upload (SPLAT, SPZ, 3DGS PLY).
Studio light
Three directional lights; a splat's normal is its shortest axis, turned to the camera.
AIM3DViewer.lights
Raking light
Soft shadows, marched through the voxel volume of the surface towards the light.
<model>.light
Golden hour
A low warm sun and a blue sky, whose light the volume darkens in hollows.
viewer.splatLighting.occlusion
Night
A floodlight from below casts shadows upwards; the exposure brightens splats as it does meshes.
AIM3DViewer.viewer.splatLighting
From the scanner to the screen
Open directly
LAS and LAZ decoded in a Web Worker while they download, on screen from the first 100,000 points.
Quick preview
Uploads show a shuffled sample of a million points within seconds, before the octree is built.
Octree
PotreeConverter makes a Potree 2 octree of LAS, LAZ, E57 and PLY: three files, every attribute kept.
Delivery
Nodes fetched by HTTP range requests, cached by nginx in 1 MB slices and served by a CDN.
Coordinates
UTM and Gauss–Krüger re-centred in double precision: millimetres kept, the scan upright.
Seeing
Adaptive point sizeIn world units, from the estimated spacing of the scan, or fixed in pixels.
Square, round or sphereSpheres write their depth, so overlapping points meet as a surface.
Filtering and measuring
Class filterA check box per class in the scan: hide the vegetation to measure the building behind it.
Picking on the GPUThe ruler and the camera's focus take the point under the cursor, read from a buffer of point ids.
Annotating
Point annotationsA click on the cloud annotates that point, stored as a IIIF point annotation; it works on streamed clouds too.
A large scan to try
A full laser scan streamed as a Potree 2 octree: open it, zoom in and watch the detail arrive, then draw a profile through it.
It explains, translates questions into filters and checks the documentation, always next to the data it came from. It never decides for you: every answer is marked as AI-generated, and the counts and sums are computed from the model itself.
The model check, in plain language
The validation report turns glTF-Validator codes into a verdict and a short list of what matters to viewers and what to fix first: a model in millimetres, textures too big for phones.
File › Validation report
Ask the building model
"Load-bearing walls on level 2", "total slab area": a question about an IFC model becomes a filter you can see, run on every element, then highlighted, isolated or hidden.
Model › Ask the building model
Variants and paradata
Reconstruction variants compared with the viewer's own measurements ("the roof of B is 3.2 m higher"), and each one's sources and Level of Certainty reviewed in the spirit of the London Charter.
Variants › Compare and review paradata
Point clouds, classified on upload
Raw scans get ground, vegetation, buildings and noise before they are streamed, so the class filter works from the first view. On the server, with your own segmentation model if you have one.
Point cloud settings › Classes
Business+The AI assistant comes with the Business and Enterprise plans, in the web viewer and in the app.
Only what a feature needs goes to the assistant (the validation report, the names of the IFC properties, the variants' texts and measurements), never the model files. A project can switch it off, or run it with its own key and model on its own server.
Time as the fourth dimension
A reconstruction is not one moment. The 4D manifest of the Wołpa synagogue follows its exterior from 1730 to 1936: fifteen stages on a time axis, each with its year, its kind and a description. Any year in between can be chosen; the stages around it fade into each other.
c. 1730178119031936
Documented event: a date the sources give
Modelled stage: weathering placed between the events
c. 1730documented
Construction
Fresh, light timber and shingles.
Source: the sources give only “first half of the 18th century” (or the mid-17th century), so the year is conventional.
1780modelled
Old, mossy shingles
50-year-old shingles, dark, with moss on the northern slopes.
1781documented
New shingles
The roof is light again; the walls stay 50 years old.
1902modelled
The roof before the repair
The 121-year-old roof just before the repair, heavily overgrown with moss.
1903documented
Roof repair
Shown as a complete replacement of the roofing.
Source: mentions only a repair of the roof.
1936modelled
Final state
Almost as in the original model (the surveyed state).
What a stage says
Each stage has a year (marked approximate when the sources only give a period), a type, a label and a description, in every language of the manifest. A documented stage can name its source, so the reader sees what the date rests on. nodes name the parts of the model that show the building at that point.
Moving along the axis
With "transition": "interpolate" the two neighbouring stages are faded by the year: 1841 is mostly the 1840 state, 1869 almost the 1870 one. "step" keeps one stage until the next year is reached and fades over transitionDuration seconds. A jump to another year plays as a short time-lapse.
Playback
Play runs through the years at yearsPerSecond (times the chosen speed) and holds on every documented event for pauseAtEvents seconds. The panel's arrows step from stage to stage; the slider picks any year.
The Wołpa synagogue from the same camera. The sun stands where it stood at the synagogue's place on midsummer day; the other images are raking light, two stages of the timeline set against each other, and the night lights of a manifest.
Morning, 8:30
Low sun from the east, long shadows across the ground.
AIM3DViewer.sun
Solar noon, 13:20
The sun high in the south, about 60° up: short shadows close to the walls.
AIM3DViewer.sun
Evening, 19:40
Low sun from the west, the other side of the building in shade.
AIM3DViewer.sun
Raking light
A grazing light tied to the view brings out the shingles and the timber joints.
AIM3DViewer.rakingLight
17801781
Comparison
1780 on the left, 1781 on the right of the swipe divider: the old, mossy shingles against the new ones. Drag the divider.
AIM3DViewer.compare
Night lights
The lights a manifest sets: spotlights on the facades, the sky dark.
AIM3DViewer.lights
Weather around the modelNew
Fog, rain, snow or a thunderstorm around the model, computed on the graphics card and kept in the manifest with the view.
Morning mist at sunrise
Banks of mist on the ground, gilded by the low sun of the sun study.
AIM3DViewer.weather
Mist
Its height, density and spread are settings: clumps of low cloud or an even blanket.
AIM3DViewer.weather
Downpour
No rain under the eaves; drops splash and bounce on the roofs, the shingles darken and shine.
AIM3DViewer.weather
Winter
Snow settles where the surfaces face up and melts again when it stops.
AIM3DViewer.weather
Thunderstorm
Wind-driven rain, lightning with its flash and, if you want it, the thunder.
AIM3DViewer.weather
A building is never seen in a studio. The viewer puts it in fog, rain, snow or a thunderstorm, under a sky with clouds and the sun where the sun study puts it.
On the model itself. Rain stops on the model's roofs and splashes there, the surfaces get wet and snow settles on them; mist lies in banks on the ground.
The seasons. The seasons of the intro come along too: petals in spring, pollen in summer, falling leaves in autumn and snow in winter, each with its own colouring of the model.
On the graphics card. Everything is computed on the graphics card, scales with the model, and works in WebGL and WebGPU, in the browser and in the app.
In the manifest. A manifest stores the weather with the view, so a curator can open the synagogue on a misty autumn morning.
Generated surroundingsNew
A typical setting of the period round the model - a village, a town, a stronghold, a shtetl, a monastery, a port, across Europe - drawn from its metadata and marked as generated.
A village, late Middle Ages
Thatched and shingled cottages along the roads, a brick church, strip fields, a river with willows; the shadows follow the sun.
AIM3DViewer.surroundings
A stronghold about 1150, in winter
Timber houses behind a palisade, a moat round the castle, snow on the roofs and ice on the water.
AIM3DViewer.surroundings
A town about 1700
Taller plastered houses, a church tower under a baroque dome, earthen ramparts with bastions instead of walls.
AIM3DViewer.surroundings
A town about 1890
Tenements, a boulevard with an avenue where the walls stood, a railway with a factory and its chimney, bridges over the river.
AIM3DViewer.surroundings
A shtetl about 1780
Narrow timber houses gable to the market square, a road out of every side, a church off the square, no walls.
AIM3DViewer.surroundings
A monastery's precinct
A wall with a gate tower, the cloister's ranges beside the church, a granary, an orchard and fishponds; a village farther on.
AIM3DViewer.surroundings
The Mediterranean
Stone and render in lime and ochre, low terracotta roofs or terraces, a bell tower beside the church, olive groves and cypresses.
AIM3DViewer.surroundings
A town in 1946
Burnt-out, roofless shells among the standing houses; no new blocks yet - the estates come in the late 1950s.
AIM3DViewer.surroundings
A town about 1975
Tenements among new blocks inside the boulevard, a housing estate of prefabricated slabs beyond it, the strips ploughed into wide fields.
AIM3DViewer.surroundings
A town about 2015
The estates renovated in pastels, detached houses along the roads round the old edge, a shop with its car park out of town.
AIM3DViewer.surroundings
The North
Timber painted Falu red and ochre under turf roofs, a whitewashed church with a tall spire, spruce and pine all round.
AIM3DViewer.surroundings
The British Isles
Red-brick terraces under slate, a stone church with a square tower, cottages under thatch and fields in hedgerows outside.
AIM3DViewer.surroundings
Eastern Europe
Log houses and whitewashed ones gable to the road under thatch, a church with an onion dome, birch and pine.
AIM3DViewer.surroundings
The plain look
The default: pale shapes without textures, so that nobody takes the setting for part of the model.
AIM3DViewer.surroundings
A building rarely stood alone. From what the model's metadata say - what it is, its century, its place - the viewer draws a typical setting round it. It is a hypothesis about the kind of place, not a reconstruction of this one, and the viewer says so.
Marked as generated. A badge over the view says what it is; the manifest keeps only its parameters, with generated: true; nothing of it is picked, measured or exported with the model.
The real place. Where the place is known: its terrain from open elevation data, its rivers and lakes from OpenStreetMap - or from OpenHistoricalMap for the year - and the region's brick or stone.
Old maps. An old map can be laid over the ground, and its roads, fields and buildings taken from a digitised file - or read from the map by AI, corrected over the map in the viewer and saved as GeoJSON.
Through time. From 900 to 2025: a year slider whose milestones - 1250, 1550, 1800, 1914, 1945, 1990 - say what changes, and a play button that lets the years go by; or the 4D timeline, the settlement growing house by house.
A scene that stays connectedNew
Change a number in a table, save a file, or turn the model on your phone: the scene follows at once, and nothing is loaded again.
A table drives the model
Each row is a part of the model. Its numbers give the part a colour, a label and an alert; a changed cell blends in and flashes.
AIM3DViewer.data
One scene on every screen
Send a link: whoever moves the camera, the year or the section moves them for everyone. A phone is the remote control.
?session=…
A manifest that is followed
Edit the manifest where it lives, or type its settings beside the model. Only what changed is applied; the camera flies there.
AIM3DViewer
A viewer usually shows a file as it was when the page opened. This one keeps listening: to a table of measurements, to the manifest, to the model's own file and to the other people looking at the same scene.
Live data. A CSV, a JSON feed or a published spreadsheet is read again every few seconds. Moisture on a wall, the condition of a column, a sensor over its limit: seen on the part it belongs to.
Shared session. A lecture, a meeting or an exhibition: the audience opens the link and sees what the speaker shows, live. Anyone may take the lead.
Followed manifest. Lights, weather, the year, comments: changed in the file, they change on screen, and the model is not loaded again.
Followed model file. Keep the viewer open beside the modelling program. Each save brings the new model in under a cross-fade, with the camera, comments and section kept.
Sound and 2D imagesNew
Not only models: sound files and photographs open in the viewer too, with tools of their own, and sound can be placed in a model's scene.
Sound as a landscape
An audio file opens like a model: its spectrogram as a terrain, the waveform in front, a playhead moving with the time. Markers, a part in a loop, filter, reverberation.
WAV · MP3 · FLAC · Ogg · M4A
Sound in the scene
Wind everywhere, a bell from a point, a voice in a cone: heard from where the camera is, placed with a click, tied to years of the 4D timeline.
AIM3DViewer.sounds · IIIF PointAudio
Narrated tours
An annotation can tell its story: the tour plays the narration when the camera arrives, shows its captions and goes on when it has been told.
annotation audio · WebVTT
How the room sounded
The reverberation of the reconstructed hall, traced through the model's own walls: wood or stone, empty benches or a full house, open windows or a lost roof. The scene's sounds are heard in it.
sounds.acoustics · Web Worker
Relief for worn surfaces
A 2D image is embossed in a light you turn: inscriptions, coins and tool marks read again, against the original on the other side of the line.
AIM3DViewer.image
Faded pigments
A decorrelation stretch, levels, equalisation and false colour on the graphics card; the result saved as PNG. A 300 dpi scan is measured in centimetres.
dstretch · autoLevels · equalize
Sets of images
A ZIP of photos or several images dropped together: a filmstrip, the arrow keys and a grid of all of them, the same settings for each.
ZIP · JPEG · PNG · WebP
A reconstruction has more sources than a model: recordings, photographs, scans and drawings open in the viewer too, and the manifest keeps them.
Field recordings. A recording opens as a landscape of its spectrogram, with its markers, tags and levels, and plays through a filter, a compressor or the reverberation of a church.
A scene that sounds. The wind, a bell on the roof, a voice from a doorway, heard as the camera moves and changing with the years of the 4D timeline; a guided tour can be narrated, with captions.
The room's acoustics. The sounds are heard as the reconstructed room would have sounded: its reverberation is traced through the model's walls, from what they were made of and what stood on the floor.
Photographs, scans and drawings. Relief, decorrelation stretch, levels, equalisation and false colour bring out what the eye no longer sees, compared with the original, saved as PNG, measured at their real size; a series with a filmstrip and a grid.
In the manifest. The manifest keeps all of it, in IIIF Presentation 4 where it has a word for it.
A scene suggested for the modelNew
The viewer suggests the light, the weather and the rendering from what it knows of the model, for documentation or for a presentation.
Wołpa: as opened
The lighting the model was published with.
The suggestions
For a presentation: the place's sun, a clear sky, AgX and AO, each with how sure it is and why.
Presentation
Wołpa's late-afternoon summer sun, with long shadows.
Documentation
An even, neutral light that hides nothing, with AO.
Tempietto: as opened
Its manifest's summer evening in Rome, in the timeline's year 1520.
The suggestions
The sun is already right; the window ticks only the sky and AgX.
Presentation
A clear Roman sky with haze towards the horizon.
Documentation
The sun off, a neutral light for the colours and the stone.
The viewer suggests the sun, the weather, the raking light, the tone mapping and the ambient occlusion from what it knows of the model: its texts, its place and year, its size and its shape. First it settles what the view is for: documentation, in an even light that hides nothing, or a presentation, in the sun of the place and the time. Each suggestion says how sure it is and why. Nothing changes until you apply it, and Undo puts the scene back.
Rules first
Words in the file name, the metadata and the manifest, in English, Polish and German: castle, interior, inscription, winter, fog. The year or century, the size in metres, and what the model is: a scan or splats keep the light they were captured in.
Look › Suggested scene
The place from its name
"Zamek w Malborku" puts the sun over Malbork. The server finds the towns the texts name in GeoNames, offline, with Polish and German endings ("w Krakowie", "pod Łęczycą"). An unsure match is shown as a guess to check.
POST /api/geo/places
A language model in the browser
Where no keyword decides ("nawa główna", "Kreuzgang"), a small multilingual model reads the texts in the browser and says whether they describe an interior, a building outside or an object, and relief to read. It is downloaded once, on your click.
multilingual-e5-small · transformers.js
The shape itself
Rays from inside the model tell walls seen from inside, a room or a nave, from the back of a closed building. A ground without a ceiling means the open air, and a thin plate gets the raking light. It takes a fraction of a second, even for half a million triangles.
orientations · enclosure · proportions
No key is needed, and the model never leaves your device: the rules, the language model and the shape run in the browser. Only the texts go to the server for the place search, to its own GeoNames copy.
How sure are we about this roof?
A reconstruction mixes what survives with what is inferred. ExPlora4D lets every annotation carry a certainty grade, colours the model by it, and keeps the scale in the manifest so others can read it the same way.
Level
Code
Basis
Score
✓
A
Preserved / surveyed
10
■
B
Direct documentation
8–9
▲
C
Indirect sources
6–7
≈
D
Analogy
4–5
?
E
Hypothesis
0–3
Each level has a colour, a letter and a symbol, so the scale reads without colour vision and in black-and-white print.
A grade applies to one object or to its whole group. The legend counts assessed objects per level; click a level to show only those parts.
Projects can define their own scale, labels and colours in the manifest; the default above is used otherwise.
Badges replace annotation numbers with the level's code and symbol while the certainty view is on.
What the model stands onNew
Six grounds under the model, from a barely visible shadow to a turntable with a compass and a scale bar. Each says something different about the model: that it stands, how big it is, where north is, or nothing but the object itself.
Contact shadow
The default: a soft shadow where the model touches the floor, and no floor. The model stands, without a word about the place.
?ground=contact
Mirror
A glossy floor with the model's reflection, fading out round it. For a presentation or an exhibition screen.
?ground=mirror
Grid
A grid without edges: lines every power of ten below the model's size, finer ones at a tenth, the axes in colour. For size, position and level.
?ground=grid
Turntable
A disc with a compass and a scale bar in the model's unit. For documentation: north and size in one picture.
?ground=turntable
Spotlight
A dark room with a pool of warm light; the background comes back when you choose another. For a single object, as in a museum case.
?ground=spotlight
Dot grid
Small dots on paper, fading out: a quiet scale for drawings and prints.
?ground=dots
It follows the model. Every ground is built again for each model, from its size.
Look › Ground: a window of tiles that stays open after a pick, to compare. It moves by its header, minimizes to an icon and closes.
Round numbers. The grid's step, the dots' spacing, the disc and the length of the scale bar, rounded to 1, 2 or 5 metres, centimetres or kilometres.
A shadow that keeps up. The contact shadow is drawn again when the model moves, a part is hidden or an animation plays.
Who chooses. A link with ?ground= first, then the manifest (viewer.ground, which the viewer's export also writes), then your pick in the window, then the site's settings. A pick wins over the manifest until the next model, so a curator can open the synagogue on the turntable and you can still try the others.
By mode. The sun study casts its shadows on every ground. Point clouds, splats and streamed 3D Tiles have no surfaces to cast a contact shadow, so they get a soft blob; with the WebGPU renderer the mirror falls back to that blob too and the grid is a plain one. In augmented reality the ground disappears and the model stands on your real floor.
Every model with a name, a type and its rightsNew
A model added to the repository is described in a short form: its name and type, then the author, copyright, licence, date, place and unit. Its owner can change them later.
Described when added
The form opens before the upload starts. The name starts as the file's, the type is chosen from a list; the rest is optional.
Found by name and type
The model list shows the given name instead of the file name, and the type as a label next to the date.
Read with the model
Model details lists them for every model; its owner edits them there or from the list.
The metadata is also what the viewer reasons from. The type says whether a model is a building, an interior or an object on its own; the place and the date put the sun where and when it stood; the unit gives the scale in metres. The suggested scene reads them with the rest of the model's texts, and they are the base for picking the type, scale and lighting on their own.
Required: name and type
Seven types: building, interior, town or settlement, site or landscape, object, sculpture or architectural detail, and other. Enough to tell an outdoor scene from a room or a showcase.
Upload › Describe the model
Rights stated up front
Author, copyright holder and a licence from the Creative Commons list, public domain or all rights reserved, so whoever opens the model knows how it may be used.
CC BY · CC BY-SA · CC0 · …
Only the owner edits
The person who uploaded the model, and the admins, can change its metadata; everyone else reads it. A private model's metadata stays as private as the model.
POST /api/jobs/{id}/metadata
Built on open standards
Nothing about a model or its interpretation is locked inside the viewer. The data stays in formats other tools already read.
Standard
Maintained by
Used for
IIIF Presentation API 4
IIIF Consortium
Scenes, models, cameras, lights, transforms and comments; import and export of whole manifests.
Web Annotation
W3C
Comments and assessments on faces and regions, with commenting and assessing motivations, point and WKT selectors.
glTF 2.0
Khronos Group
The delivery format, with Draco and Meshopt geometry compression, KTX2/Basis textures and unlit materials.
3D Tiles
OGC
Streaming large meshes and point clouds at the level of detail the view needs.
Potree 2
Potree project, TU Wien
The octree format for large point clouds: three files, each node read with a range request, intensity and classification kept.
RAD
Spark
The level-of-detail format for Gaussian splats: a header and chunks of 64,000 splats, read by range requests as the view needs them.
IFC
buildingSMART
Building models, read in the browser or converted with their property sets kept as metadata.
LAS / LAZ, E57
ASPRS · ASTM
Laser-scan point clouds: LAS 1.0–1.4 in every point format, decoded in the browser, with the ASPRS classes coloured and filtered; converted to Potree 2 or 3D Tiles.
HTTP range requests
IETF (RFC 9110)
Reading single nodes of an octree, or chunks of a splat tree, out of one large file; resumable downloads.
JSON Schema 2020-12
JSON Schema
The AIM3D manifest block, published and validated on import.
WissKI
WissKI community
Drupal-based, CIDOC CRM–aware repositories that host the models and their metadata.
Technology
A WebGL viewer in the browser, a conversion worker on the server, and the same code packaged for repositories and phones.
In the browser
three.js r187
WebGL 2
GTAO
3d-tiles-renderer
Web Workers
loaders.gl
laz-perf
Eye-Dome Lighting
GPU picking
Volume ray marching
Web Audio API
Ray-traced acoustics
WebVTT
geotiff.js
web-ifc
Draco
Meshopt
KTX2 / Basis
Logarithmic depth for city-scale and coin-scale models alike; a post-processing chain with SMAA, FXAA and seven tone mappings. Point clouds are decoded off the main thread and coloured, sized, filtered and picked in shaders.
On the server
Python worker
Blender + Cycles
IfcOpenShell
OpenCASCADE
trimesh
PotreeConverter 2
py3dtiles
laspy + lazrs
pye57
tifffile
pydicom
gltfpack
PHP
Docker
One docker compose up runs the viewer and the full conversion pipeline, with accounts and upload limits. Converted files go out with range requests, gzip and ETags.
Delivery and quality
Rollup
Drupal module
Capacitor
Android
iOS
nginx slice cache
Cloudflare
Playwright
GlitchTip
i18n EN · PL · DE
End-to-end tests in a real WebGL browser cover loading, editing, export and the manifest schema.
Follow your file
Every format the viewer reads, grouped. Pick yours: the steps it goes through on the server, what each one makes, and what the viewer shows in the end. Greyed-out steps are skipped.
1
Upload
The file, or a ZIP, RAR or tar archive with its textures and materials, unpacked on the server.
model.glb
2
Conversion
Already glTF: nothing to convert.
3
Checks and compression
glTF-Validator checks the model (specification, units, size); gltfpack compresses it: Meshopt geometry, KTX2 textures.
model.glb
4
Lighter copies and LoD
From 2 MB up, a preview GLB (about 10% of the triangles, textures up to 512 px) loads first and the full model replaces it.
model.preview.glb
5
Renders
Blender Cycles renders thumbnails from nine angles.
views/*.png
6
In the viewer
A mesh with its materials, in any shading mode, with sections, measurements and annotations.
Directly in the browser:Also opens in the browser as it is: glTF is the viewer's own format.
1
Upload
The file, or a ZIP, RAR or tar archive with its textures and materials, unpacked on the server.
2
Conversion
The server converts it with Blender: Blender imports the file and exports GLB (USD needs a Blender build with USD).
gltf/model.glb
3
Checks and compression
glTF-Validator checks the model (specification, units, size); gltfpack compresses it: Meshopt geometry, KTX2 textures.
model.glb
4
Lighter copies and LoD
From 2 MB up, a preview GLB (about 10% of the triangles, textures up to 512 px) loads first and the full model replaces it.
model.preview.glb
5
Renders
Blender Cycles renders thumbnails from nine angles.
views/*.png
6
In the viewer
A mesh with its materials, in any shading mode, with sections, measurements and annotations.
Directly in the browser:OBJ, PLY, STL, FBX, DAE, WRL and USD also open in the browser without the server (three.js); ABC, BLEND and X3D only through it, converted by Blender.
1
Upload
The file, or a ZIP, RAR or tar archive with its textures and materials, unpacked on the server.
2
Conversion
IfcConvert → GLB, each element named by its GlobalId; property sets, quantities and the spatial tree go to a metadata JSON.
model.glb · _ifc.json
3
Checks and compression
glTF-Validator checks the model (specification, units, size); gltfpack compresses it: Meshopt geometry, KTX2 textures.
model.glb
4
Lighter copies and LoD
From 2 MB up, a preview GLB (about 10% of the triangles, textures up to 512 px) loads first and the full model replaces it.
model.preview.glb
5
Renders
Blender Cycles renders thumbnails from nine angles.
views/*.png
6
In the viewer
The building as a mesh; a click on an element opens its properties.
Directly in the browser:Also opens in the browser without the server, with web-ifc.
1
Upload
The file, or a ZIP, RAR or tar archive with its textures and materials, unpacked on the server.
2
Conversion
CityGML2OBJ, then Blender → GLB.
model.glb
3
Checks and compression
glTF-Validator checks the model (specification, units, size); gltfpack compresses it: Meshopt geometry, KTX2 textures.
model.glb
4
Lighter copies and LoD
From 2 MB up, a preview GLB (about 10% of the triangles, textures up to 512 px) loads first and the full model replaces it.
model.preview.glb
5
Renders
Blender Cycles renders thumbnails from nine angles.
views/*.png
6
In the viewer
A mesh with its materials, in any shading mode, with sections, measurements and annotations.
Directly in the browser:Only through the server.
1
Upload
The file, or a ZIP, RAR or tar archive with its textures and materials, unpacked on the server.
2
Conversion
Cut into triangles → GLB: OpenCASCADE for STEP and IGES (its tolerance trades detail for size), trimesh for 3MF.
model.glb
3
Checks and compression
glTF-Validator checks the model (specification, units, size); gltfpack compresses it: Meshopt geometry, KTX2 textures.
model.glb
4
Lighter copies and LoD
From 2 MB up, a preview GLB (about 10% of the triangles, textures up to 512 px) loads first and the full model replaces it.
model.preview.glb
5
Renders
Blender Cycles renders thumbnails from nine angles.
views/*.png
6
In the viewer
A mesh with its materials, in any shading mode, with sections, measurements and annotations.
Directly in the browser:3MF also opens in the browser (three.js); STEP and IGES only through the server.
1
Upload
The file, or a ZIP, RAR or tar archive with its textures and materials, unpacked on the server.
2
Conversion
Kept as uploaded: the viewer reads it itself.
3
Checks and compression
Not needed.
4
Lighter copies and LoD
Not needed.
5
Renders
No thumbnails.
6
In the viewer
A mesh with its materials, in any shading mode, with sections, measurements and annotations.
Directly in the browser:The same in the browser, with or without the server (three.js).
1
Upload
The file, or a ZIP, RAR or tar archive with its textures and materials, unpacked on the server.
2
Conversion
LAS and LAZ go straight on; E57 scans are merged with their poses (pye57) and PLY points read with trimesh, into one LAS.
model.las
3
Checks and compression
If the server is set to, the points are classified first: ground, vegetation, buildings, noise.
4
Lighter copies and LoD
A quick preview of a million shuffled points at once, then PotreeConverter → a Potree 2 octree (or py3dtiles → 3D Tiles).
.preview.las · metadata.json
5
Renders
No thumbnails.
6
In the viewer
Streamed points with Eye-Dome Lighting, coloured by height, intensity or class; profiles and measurements.
Directly in the browser:LAS and LAZ also open without the server: decoded in a Web Worker while downloading, shown from the first points. E57 only through the server.
1
Upload
The file, or a ZIP, RAR or tar archive with its textures and materials, unpacked on the server.
2
Conversion
No conversion: the splats stay as they are.
3
Checks and compression
SPLAT, SPZ and 3DGS PLY get a voxel volume of their surface: normals, occlusion and shadows when the scene's lights shine on them.
.light
4
Lighter copies and LoD
Spark's build-lod → a RAD level-of-detail tree (a .rad already is one).
-lod.rad
5
Renders
No thumbnails.
6
In the viewer
Gaussian splats (Spark), streamed in chunks by range requests.
Directly in the browser:Also without the server: the level-of-detail tree is then built while loading.
1
Upload
The file, or a ZIP, RAR or tar archive with its textures and materials, unpacked on the server.
2
Conversion
Every Z stack becomes an 8-bit volume.
.dfgvol
3
Checks and compression
Not needed.
4
Lighter copies and LoD
Only the stack on screen is read, by range requests.
5
Renders
A maximum intensity projection as its thumbnail.
.png
6
In the viewer
A ray-marched volume.
Directly in the browser:Also without the server: read in a Web Worker, only the stack shown.
1
Upload
The file, or a ZIP, RAR or tar archive with its textures and materials, unpacked on the server.
2
Conversion
pydicom reads the series, also JPEG, JPEG 2000 and multi-frame, into an 8-bit volume; no patient data in the result.
.dfgvol
3
Checks and compression
Not needed.
4
Lighter copies and LoD
Only the stack on screen is read, by range requests.
5
Renders
A maximum intensity projection as its thumbnail.
.png
6
In the viewer
A ray-marched volume; CT in Hounsfield units, with the usual windows.
Directly in the browser:An uncompressed series in a ZIP also opens without the server; compressed ones only through it.
1
Upload
Not uploaded: opened from its URL, already cut up for streaming.
2
Conversion
Not needed.
3
Checks and compression
Not needed.
4
Lighter copies and LoD
Already a level-of-detail tree: tiles or octree nodes.
tileset.json · metadata.json
5
Renders
No thumbnails.
6
In the viewer
Streamed level of detail: only what the camera needs, sharper as it comes closer.
Directly in the browser:Always in the browser: 3d-tiles-renderer, Potree nodes by range requests.
How each format is handled
Format
In the browser
Converted on upload
Animation
glTF / GLB
Native format
–
✓
OBJ · PLY · STL
three.js
Blender → GLB
–
FBX · DAE
three.js
Blender → GLB
✓
IFC
web-ifc
IfcConvert → GLB, property sets kept as metadata
–
WRL (VRML)
three.js
Blender → GLB
–
USD · USDA · USDC · USDZ
three.js
Blender → GLB (USD-enabled build)
–
3MF
three.js
trimesh → GLB
–
KMZ
three.js
–
✓
three.js JSON
three.js
–
✓
3DS · AMF · VOX · XYZ · PCD
three.js
–
–
3D Tiles · Potree 2
Streamed level of detail (3d-tiles-renderer); Potree nodes read by range requests
–
–
LAS / LAZ
Web Worker: LAS decoded while downloading (point formats 0–10), LAZ with laz-perf; shown from the first points, thinned above a point limit
Quick preview, then PotreeConverter → Potree 2 (or py3dtiles → 3D Tiles)
–
E57 · PLY (points)
–
pye57 / trimesh → LAS → Potree 2, with a quick preview
–
SPZ · SPLAT · KSPLAT · SOG · PLY (splats)
Gaussian splats (Spark), drawn from a level-of-detail tree built while loading
build-lod → RAD level-of-detail tree
–
RAD
Gaussian splats (Spark), streamed: header, then chunks of the tree by range requests
–
–
LIF · TIFF · OME-TIFF · LSM · PNG/JPEG/BMP slices
Volume, ray-marched; only the stack shown is read, in a Web Worker (from the file, or by range requests)
8-bit volume and a maximum intensity thumbnail
–
NRRD · NIfTI · MetaImage · MRC · VGI · XTEKCT
Volume, ray-marched; only the stack shown is read, in a Web Worker (from the file, or by range requests)
8-bit volume and a maximum intensity thumbnail
–
DICOM
Uncompressed series from a ZIP: CT in Hounsfield units, with the files' and the usual windows
pydicom, also JPEG, JPEG 2000 and multi-frame → volume; never patient data in the result
–
ABC · BLEND · X3D · GML
–
Blender → GLB
–
STEP / STP · IGES / IGS
–
OpenCASCADE → GLB
–
Converted models are compressed with gltfpack and get a light preview that loads first; point clouds get a shuffled preview and a streamed octree, Gaussian splats a streamed level-of-detail tree, image stacks an 8-bit volume. Uploads can be ZIP, RAR or tar archives with textures and material files.
Where it runs
The same viewer, configured by one settings file or by the manifest itself.
In a repository
A Drupal module for WissKI: models on entity pages, conversion on upload, annotations saved back.
On your server
Docker images for the viewer and worker, with profiles for test, development and drag-and-drop sandbox.
In any page
An embed snippet with presentation mode, autorotation and a chosen camera. Settings follow the manifest. For the model alone, ?ultralight=1.
On a phone
Android and iOS builds of the same viewer, for fieldwork and teaching.
Three weights of the same viewerNew
What a visitor downloads before the model itself, compressed: the viewer's code, styles, fonts and the light the model is shown in.
Full
≈ 2.0 MB
Every tool, with saving to the server, the publishing checks and the file download.
Lightweight
≈ 1.2 MB
The same tools for viewing, without saving or downloads. Its light now comes from a 100 KB environment map instead of 1.4 MB, and the model looks the same.
viewer.lightweight
Ultralight
≈ 0.27 MB
The model and nothing else: no menu, intro or gallery, only turning and zooming. glTF and GLB, compressed or not, lit as in the other two.
?ultralight=1 · ultra.html
Why ultralight
The model sooner. Seven times less to download than the full viewer and four times less than the lightweight one: on mobile data the model is on screen seconds earlier.
Many objects on one page. A collection or a catalogue can show a model in every entry without each one bringing the whole viewer.
Nothing between the visitor and the object. No menu or intro: for museum labels and QR codes, kiosks, learning platforms and articles.
Less traffic. Smaller transfers for the visitor's data plan, the server and the CDN, and less work for the phone.
One parameter.?ultralight=1 on an existing viewer link opens it with the same model and options; ultra.html can be linked directly too.
Try it
Two public instances of the viewer, beside the main one: one for your own model, one for the example manifests.
Sandbox
Drag and drop a 3D model into the viewer and try every tool on it: scan views, sections, measurement, annotations.
The Android and iOS app is built from the same code as the web viewer, so it has the same tools: scan views, raking light, sections, measurement, annotations, the certainty view and IIIF manifests. Models and point clouds come from ExHUB, the shared repository, or from the device itself.
ExHUB
The shared repository of the project's models and point clouds, converted on upload: meshes as compressed glTF, large scans as streamed 3D Tiles.
On the device
Models picked from the phone, and models saved from ExHUB, stay on the device and open without a network.
Examples
The example manifests, from scan views to the certainty view, to try every tool.
One Models button switches between the three. The app reaches ExHUB at an address you can change, so a project can point it at its own repository.
What the phone adds
AR in your room
The model as it is on screen, with its materials and your changes, stands at its real size on the floor or a table, through ARCore, right in the app. On Android for now.
Motion look
Turn the phone and the view turns with it, as if the screen were a window onto the model. Dragging and pinching still work, and the motion carries on from the new view.
360° video
The camera goes once round the model and the app records it as a short video, ready to share in a messenger or by mail. AR, motion look and the video sit together at the end of the navigation bar, one tap away.
Pro+AR and motion look come with the Pro, Business and Enterprise plans; on the Free plan the video carries a small ExPlora4D mark. See Plans.
Android
iPhone and iPad
Scan a code with your phone's camera, or tap a badge.
On the phone
Welcome
The viewer
Section plane
Weather
Guided tour
Look
Room for the model on a tablet
On an iPad or an Android tablet held sideways, the app is laid out like the web viewer: the toolbar along the bottom, and the tools in windows beside the model rather than in sheets over it. Sections, weather and guided tours work as on the phone, with more of the building in view.
On its side
Section plane
Snow
Guided tour
Plans
Viewing is free. The paid plans take away the ads, add the phone's own tools, the repository and the AI assistant, and, for institutions, a setup made to measure.
Free
0 €
All viewing tools: scan views, sections, measurement, the certainty view
Models from the device and the examples
Ads in the app; three tries of AR
Pro
20 € one-time purchase
No ads
AR in your room and motion look
360° videos without the mark
Business
20 € per month
Everything in Pro
ExHUB models and your own repository address
Higher upload and conversion limits
Annotations, IIIF export and the materials editor
AI assistant: validation explained, questions to IFC models, variants and paradata reviewed
Enterprise
Individual price agreed with you
Everything in Business, the AI assistant included
Your own server or repository, set up with you
Limits, storage and AI usage agreed for your project
Pro and Business are bought in the app (Google Play, App Store), where the price shows in your currency. Enterprise is set up on your account after we agree the scope.
Author
Daniel Dworak
3D web developer · DevOps Team Lead · 16+ years
3D visualization
Clean architecture
CI/CD
Containers
Observability
Security
Hi, I'm Daniel. I build high-performance web applications with a focus on 3D visualization, interaction and clean architecture, and I'm a DevOps Team Lead with 16+ years of experience designing, automating and operating scalable infrastructure.
My background spans CI/CD, containerized platforms, observability and security. I lead teams, mentor developers and work across multinational, cross-functional groups, representing technical teams, facilitating decisions and aligning engineering work with business goals.
I care about reliable, secure, automated platforms and a culture of ownership, knowledge sharing and continuous improvement.