How to build a scene

1. Click a key on the rail — Inspect, Scene Graph, Composite, Archetype, Machine, Clips, Bus or Diag. The panel opens in the dock and its key turns violet.
2. Open a second one and the two sit side by side, each the full height of the dock.
3. Open a third and the workshop asks which of the two to replace: both light up, and you click the one to swap out. Esc, or a click anywhere else, calls it off.
4. Click a tool's own rail key again to close it. That key is the only way to close a docked panel.
5. Swap, at the foot of the rail, moves the whole dock to the other side of the window if you would rather work right-handed.

Panels are working chrome, not scene content: nothing about which ones you had open is saved into what you publish.

1. Double-click empty space in the scene — anywhere that is not an object. A matte grey wall fades up in front of you: the shape browser.

2. Scroll anywhere over the wall to turn the ring. One notch of the wheel is one shape, on every mouse and trackpad.
3. Click the shape facing you — or the big turning copy in the middle. It is placed in your scene, the wall fades, and the camera turns to what you just made.
4. To leave without placing anything, press Esc or double-click again.

Where it lands is beside whatever the camera was looking at, so frame something first (double-click it) to choose where the next thing arrives. One click places one object: for ten of something, or a particular size, set it afterwards in the Inspector.

The shape you place arrives in the colour it had on the ring — every primitive there has its own. Change it afterwards like anything else.

1. Double-click any shape or lamp in the scene. The Inspector opens on it and the camera turns to face it, keeping the distance you were at.
2. Or click Scene Graph on the rail and click a node in the tree. The Inspector opens on it and the camera stays where it is, so you can pick something to change without losing your view. Double-click the node instead and the camera turns to face it too.
3. In the Inspector, find the label row and type a name.

Two things deliberately do not answer a double-click: a group (click the child instead — a group would swallow every click), and the insides of a word or an imported model, which are picked whole.

1. Open Scene Graph.
2. Drag a node onto a group to move it inside. It lands at the group's centre rather than keeping where it was — otherwise it would very often land outside the group's box and vanish as you dropped it.
3. Drag a node onto the top-level bar above the tree to pull it out of whatever it is in. It becomes a tree of its own, at the centre of the world.

A move that cannot work says why instead of failing quietly: a group cannot go inside itself, only groups hold things, a group holds at most 32 things, and imported geometry is read-only.

Making a new, empty group to gather things into is an asking job — the tree moves things between groups that already exist.

1. Double-click the object, or pick it in the Scene Graph, so the Inspector is showing it.
2. Find the Material section and open the picker. The whole catalog is listed by name, each with its cost badge; click one and it is applied.
3. The colour row is a colour picker shown outright — nothing to expand, no edit button to find. Pick a colour to tint the material.
4. alpha sets opacity. On a coloured glass the same number also sets how deeply the colour reaches into what you see through it.
5. roughness runs from mirror-sharp (0) to matte (1); metallic and index of refraction are there too.

What the picker does not give you: there is no way to search or filter the catalog by hand, no tags, no swatch before you click, and no sight of the approximation note some materials carry. You scroll the list and read the names. Asking — what materials are there?, is that one exact? — is how you get the rest.

Rows the engine works out for itself — a letter's texture, a model's root flag — are shown but not editable, marked engine-set. Showing them without offering them is the honest version: three of them used to be draggable and did nothing at all.

1. Select the object so the Inspector shows it.
2. Type the numbers. Position, rotation and size are three boxes each — one per direction — coloured to match the handles in the scene. Rotations are shown in degrees, because nobody thinks in radians.
3. Or drag it. Tick position and three thin glowing lines appear through the object, one per axis. Drag it and it travels along one line — whichever your drag most follows — and stays on that line until you let go.

4. To turn it, tick rotation instead. Three circles appear around it; the one nearest your pointer lights up, and dragging turns the object about that circle.

The handles come with the object, ready for the next drag. Untick to put them away.

There is no Add light button, but there is a real route, because in this engine a light is a sphere that emits. That is literally how the lamp in a new project is stored.

1. Place a sphere from the shape browser where you want the light.
2. With it selected, in the Inspector tick emitsLight.
3. Give it glow above zero, and set glowIntensity for how strong it is. Set its colour for a warm or cool light, and its size for how soft the shadows are — a bigger lamp makes a broader, softer edge.
4. To retune the lamp a new project starts with, just double-click it in the scene — you can click a lamp by its glow — and edit the same rows.

To stop something casting or receiving shadows, the participation switches are in the Inspector on every object.

Worth knowing: a glowing object neither casts a shadow nor takes one, which is why the switch pair and the glow value are best set together.

1. Double-click the model or the picture — both are picked as a whole, never facet by facet — or click it in the Scene Graph. The Inspector opens on it.
2. Position, rotation and the move and rotate handles work exactly as they do on a sphere.
3. A picture has a single size row — its longest side. It goes up to four times the picture's own resolution; type a bigger number and you get the largest it can be, and hovering the row says why. On a large picture the slider moves in big steps, so type a number for a finer or smaller size.
4. The colour row tints it. Leave it white to see it exactly as imported.
5. The Material picker gives it that material's finish. The colour stays yours.
6. The participation switches — casts shadows, receives shadows, appears in reflections — are there as on anything else.

Three things are shown but not editable, marked engine-set: the pixel dimensions a picture was stored at, and the internals the engine works out for a baked model. The inside of an imported model is not in the Scene Graph either — it is not yours to hand-edit.

Deleting from your library is yours on both plans — the assistant can never do it, and there is no button, file chooser or address anywhere that changes that.

1. With a picture card selected, the Inspector carries a warning-coloured delete button. Click it once and it asks you to confirm, naming what will break.
2. Or open the Canvas panel → Manage pictures…, which lists every stored picture with its size and a ✕ beside it. It also shows how full the library is — it holds 200 pictures or 1 GB, whichever fills first.

What survives and what does not. A zip you have already downloaded and a scene you have already published are safe: each carries its own copy and never reads your library again. But a saved scene that still uses the picture loads that card blank, with a warning naming it, and will not export until you fix it — drop the file again, or delete the card. Every card printed from that picture goes with the file.

1. Double-click empty space to open the shape browser, and scroll to the end of the ring — the cream T.
2. Click it. A word reading text is placed upright, facing you.
3. With it selected, the Inspector gives a word its own rows: the text itself with an apply button, its size, and its font.
4. The font row is a drop-down of everything in your library. Click add… to bring in a new one by family name, by a fontsource.org link, or by a direct .ttf / .otf / .woff address. It takes about two seconds and is then permanently in your library, shared by every project.

The font list is not offered on a label inside a control or a component, where the change would not be saved, nor on a checkbox or radio mark, whose character lives in an icon font.

This one is yours on both plans — the assistant can place copies of what you have already stored, but it can never bring a file in and can never delete one.

1. Drag the file onto the canvas. Models: .glb, .gltf, .obj. Pictures: PNG, JPEG, WebP, GIF, BMP or AVIF, up to eight at once.
2. For an .obj, drop its .mtl and its texture images together with it — open the folder and drag the files out; a folder itself cannot be dropped. Same for a .gltf that keeps its geometry in a separate .bin.
3. The model is converted right there in your browser and arrives keeping its colours. A picture arrives as a card at the thing the camera is looking at, and several arrive in a row.
4. Everything you drop is stored in your library, which every project of yours shares.

SVG, HEIC and TIFF are refused, each with its fix — rasterize an SVG to PNG first.

1. Double-click empty space to open the shape browser.
2. Above the wall, click Imports for your stored models or Images for your stored pictures — the 32 newest of each, from every project. Each button appears only once you have something behind it.
3. Pause on one. After a moment its full version loads; if the download will take more than about a second you get a progress bar and a count ("5.1 of 16.8 MB"). Scrolling on does not cancel it — it finishes in the background, so it is ready next time.
4. Click it to place a copy. A click on one that is still loading does nothing; click again once it appears.
5. Shapes puts the ring back.

To see how full your library is, or to delete from it, open the Canvas panel and click Manage pictures…. It lists every stored picture with its size and a ✕ to delete it, and asks first.

The boolean operators are not in the Inspector — they are on the Composite panel, which is the thing that builds a combined solid in the first place.

1. Place the shapes you want to combine and arrange them so they overlap the way you want the cut to land.
2. Click Composite on the rail, then click each shape in the scene to collect it. The order matters: the first one you click is the base, and the rest fold into it in the order you picked them.
3. Leave combine members switched on — that is what makes the group a single solid rather than a plain container — and press Group.
4. Each member now has its own row with a union / subtract / intersect drop-down and a smoothing box from 0 to 1. Subtract carves that shape out of the base; smoothing blends the seam like clay instead of welding it.

Two limits worth knowing before you start: you need at least two objects and they must already share the same parent, and you cannot add a fifth shape to a fold of four afterwards — that one is an asking job. Glass and matte cannot be fold siblings, so a combined solid takes one material throughout.

1. Scroll to circle whatever the camera is aimed at. Space+scroll goes over and under; Z+scroll moves closer and further. The horizon stays level however you combine them.
2. Click the scroll mode — the small drawing and word under the scene, at its left edge — to change what the wheel does. It steps from circular (going around, as above) to vertical (sliding the view up and down) to horizontal (sliding it side to side; a sideways trackpad swipe works too), then back to circular. The slides follow the camera, so whatever you are looking at stays straight ahead. Space and Z only apply in circular.
3. Drag the slider under the word to set how far each wheel step goes, from a quarter of the usual to twice it. It applies in every mode.
4. Double-click a thing to aim at it without moving closer or further, landing level. It is the fastest way to straighten a crooked view.
5. Where you left the camera is saved with the scene, so park it where you want the scene to open.

Orbiting is a building tool and stays in the workshop: a visitor to a published scene does not free-scroll, because the author chose the viewpoint. The scroll mode and the slider are yours too — they stay on this computer and are never saved into a scene or published. The switch hides while a scene rides a rail, because the rail owns the wheel then, and in Fullscreen.

What you cannot do here by hand: type an exact camera position or target, or change the field of view. Orbiting and framing get you anywhere you can see; a number, or a wider lens, is an asking job — pull the camera back to fourteen, widen the lens to 40°.

1. Drag any of the four guide lines around the scene. The two lines of a pair move together, so the picture stays centred in its space; the strip underneath is a little taller than the other three sides, because it holds the readout and the scroll mode. The readout tells you what you have — 1280 × 720 · 16:9.
2. Dragging near a common shape snaps to it (16:9, 3:2, 4:3, 1:1 and the three portrait ones). Hold Alt to say no. With a guide focused, the arrow keys nudge, Page Up/Page Down step through the common shapes, Home goes back to 16:9, and double-clicking a guide grows the frame as large as it fits.
3. Click Canvas, beside the readout, to open the Canvas panel.
4. Set a max width or max height — 160 to 16384 px, or No limit. The panel tells you the result in words: "On a large screen this scene publishes at 1067 × 600". Unlike the frame's size on your screen, this reaches the export.
5. Leave Fallback image on Automatic and every export carries a still of the scene's first frame; set it to None for a scene that opens on something you would rather not show as a still.
6. Under Behind the scene, tick Use a page colour and pick the background colour of the page your scene will sit on. It shows through every empty part of the scene while you work — and it is what glass and metal reflect and refract, so it is saved and exported too.
7. Or set Fill to a linear or radial gradient and choose its start and end colours, where each stop sits, and the angle (linear) or the centre and radius (radial). A gradient is drawn by the scene itself, edge to edge — not by the page behind it — so the glass refracts the same glow the empty pixels show, and it exports exactly as you see it. One colour puts it back.

Set the page colour to your own site's background and the scene stops looking like a black rectangle cut into your page. Black is a real answer, not an empty one: if your page really is black, tick the box and leave the colour black.

1. Click Fullscreen at the foot of the Canvas panel — or Fullscreen scene in the Video window's Layout row.
2. The rail and the dock fade away and the scene grows across all three columns. Nothing about the scene changes, and your panels stay open behind the fade.
3. Leave by the same button, which now reads Exit fullscreen, or by pressing Esc. Both always work.

Everything you do in there is still kept and saved — a stray double-click still selects, and orbiting still re-aims and saves the scene's camera. Park the camera before you record, or put it back after. Fullscreen ignores your max size, which is why Preview, not Fullscreen, is the proof of what you publish.

1. There is nothing to press. Every change marks the project dirty and one save writes it up about two seconds after you pause; saved · rev N flashes on the status line.
2. Click Scene at the top of the rail for this project's scenes. Pick one to open it; start a new one from the same place.
3. Close the tab and come back tomorrow: the workshop reopens the scene, and your panel layout, as you left them.
4. If the same project was saved somewhere else in between, you are told there was a conflict rather than silently overwritten.

1. Preview, near the foot of the rail, uploads a throwaway copy of the exact package and opens it in a new tab, running as the published player rather than as the workshop. It answers "will the export actually work?" before you ship it. A preview link dies on its own after three days, so it can never quietly become hosting. Preview shows its loading screen while the scene compiles; the fallback image is only in the export.
2. Export downloads one zip that is the complete runnable thing: the player, your scene, its models, its lettering, its pictures, a fallback image of its first frame, and its two notice files. Every reference inside points inward. The exported page shows that picture at once and swaps in the live scene the moment it is ready.
3. Unzip it onto any static host — or open it locally over any small web server — and it plays there with zero requests back to us.

Every export carries LICENSE.txt and NOTICES.txt at its root, on either plan. The licence says what you may do with the package: everything, because it is yours — host it, copy it, change it, embed it, sell it, give it away. The one exception is the fonts under fonts/, which keep their own licences; keep those files with the folder wherever it goes.

1. Click Diag on the rail. The list keeps its history, so opening it after the confusing thing happened still shows it.
2. Click Bus for every message the engine sent itself, newest first — a press, a state change, a value arriving. It filters to whatever you have selected, so "what did my switch just do" is one page.