Melonjs UI And Text Agent Skill

UI, HUD and text

Outlined text: the stroke eats into the glyph

Text draws fillText and then strokeText, so the outline lands **on top of

the fill** and is centred on the glyph edge — half of it inward. On a small or

chunky face a lineWidth of 3 leaves the letters solid black. Keep it to 1,

and raise the font size rather than the stroke.

The stroke used to cost you the top row of pixels as well, and a display face

whose glyphs overshoot the nominal ascent did the same on its own: the render

box was sized from the line height alone, with no allowance for either. **That

is fixed** — the bake is padded by the ink's real extent and the blit shifts

back by the same amount, so nothing is clipped and the reported bounds are

unchanged. Write the string you mean:

new Text(x, y, { font: "Display", lineWidth: 1, lineHeight: 1.45, text: "" });
hud.setText("SCORE 100");

If you have a label carrying a leading \n to buy headroom, that workaround is

now dead weight — drop it, and take the line height back off the position you

shifted it by.

Gradient text

fillStyle takes a Gradient as well as a colour — the same object

Renderer#setColor accepts, built the way the canvas API builds one:

const ramp = renderer.createLinearGradient(0, 0, 0, 24);  // top to bottom
ramp.addColorStop(0, "#fffdf0");
ramp.addColorStop(1, "#f0a020");

new Text(x, y, { font: "Display", size: 24, fillStyle: ramp });

Coordinates are the label's own bake: (0, 0) is the top-left of the render

box, so the ramp above runs down one line. A multi-line label **restarts it on

every line**, reading like one Text per line — you do not have to author the

gradient over the block height. Pass gradientPerLine: false for a single ramp

spanning the whole block, which is what a plain canvas does and what a

deliberate fade across a two-line title wants.

The ramp colours the fill only — Text strokes in a separate pass, so

an outline keeps its own colour without any luminance trickery — and it works on

Canvas2D, which a post effect does not.

The gradient does not live in fillStyle

fillStyle ACCEPTS a Gradient; it never HOLDS one. The constructor routes it

to fillGradient and leaves fillStyle as the pooled Color it always is,

which is what keeps fillStyle.alpha gating the fill and keeps the colour

owned by the pool.

So changing the ramp later means assigning the field it actually landed in:

label.fillGradient = ramp;   // ✓
label.fillStyle = ramp;      // ✗ TS2740: Gradient is not a Color

The ramp is authored for ONE line height, so a label that changes size needs a

new one built at the new size; there is nothing that rescales it for you.

BitmapText: size is a RATIO, not pixels

The trap when moving over from Text:

new Text(x, y,       { font: "Arial", size: 24, text: "SCORE" });  // 24 pixels
new BitmapText(x, y, { font: "arial", size: 24, text: "SCORE" });  // 24 TIMES

size scales the font's authored size, so 1 is native and 2 is double —

resize(scale) and set(textAlign, scale) take the same ratio. Whole numbers

keep pixel art crisp; fractional ones resample the page image.

It has no stroke — there is no strokeStyle or lineWidth here, which is

part of why it stays sharp. Colour comes from the tint instead:

const score = new BitmapText(8, 8, {
    font: "arial", text: "1000", fillStyle: "#ffd700",
});
score.fillStyle = "#ff4040";                  // recolour at any time
score.fillStyle = new Color(255, 255, 255);   // UNTINTED, not "white text"

fillStyle is Renderable#tint under another name: white is the *absence* of a

tint and every other colour tints away from it, so author the page in white to

keep every colour available to you.

Load the descriptor as binary and its page as image under the same name.

Both BMFont flavours — text (.fnt) and XML — are auto-detected, so an .xml

descriptor loads as-is:

loader.preload([
    { name: "arial", type: "binary", src: "data/font/arial.fnt" },
    { name: "arial", type: "image",  src: "data/font/arial.png" },
]);

Reach for it over Text when the text is mostly static, has to stay crisp at

integer scales, or wants recolouring without a re-bake.

Do not build a throwaway label to measure a string

A Text owns a canvas bake and a GPU texture, and only destroy() frees

them — which is @internal and absent from the published types, so a game

cannot free a label it owns outside a container. A label built purely to

measure something therefore strands both, and if it is built in a stage's

onResetEvent it strands another on every entry to that state.

measureText(text?) measures any string on a label that already exists,

including one it is not currently holding:

const widest = hudScore.measureText("00:00").width;

Advance widths scale linearly with the point size for one family, so measuring

on an existing label and scaling the result is exact. For labels you genuinely

spawn in bulk, getPool("text") recycles the bake.

fillStyle means two different things

The name is shared and the behaviour is not, which is the trap when moving a

label from one class to the other:

| | Text | BitmapText |

|---|---|---|

| what fillStyle holds | a real Color, the fill | Renderable#tint, under another name |

| white | white glyphs | the ABSENCE of a tint |

| gradient | yes, via fillGradient | no: a tint is one colour over a page image |

| recolour it later | label.fillStyle.parseCSS("#ff4040") — the field is a Color with no setter, and ASSIGNING a string replaces it, after which the glyphs stop filling entirely and nothing is logged | label.fillStyle = "#ff4040" — a real setter that parses the string |

| stroke | strokeStyle + lineWidth | none |

| size | pixels | a RATIO of the authored size |

Per-glyph wave, shake and colour: glyphEffect

Animating individual letters does not mean one renderable per character.

BitmapText#glyphEffect offsets and tints each glyph inside the batch the

label already draws, so an animated 100-character line stays one draw call and

costs no extra objects:

label.glyphEffect = (out, ctx) => {
    // phase from the glyph's POSITION, not its index — see below
    out.offsetY = Math.sin(ctx.time * 0.005 + ctx.x * 0.02) * 6;  // wave
    out.tint.setColor(255, 128, 128);                             // per glyph
};
label.glyphEffect = null;    // back to the plain path

out and ctx are reused: read ctx during the call and write out, but

never keep either. out.offsetX / offsetY start at zero and out.tint at

opaque white on every glyph, so an effect that only touches some glyphs leaves

the rest alone. ctx carries index (across lines, line breaks excluded),

char, code, x, y and time.

Four things that catch people:

• time advances in update, not in draw. A label drawn through two

cameras animates once, not twice. It is the milliseconds ACCUMULATED across

updates while an effect was set, and nothing resets it, so swapping the

callback continues the same clock rather than starting a new one.

• Effects move pixels, not layout. Metrics, wrapping, alignment,

visibleCharacters and the bounds all ignore the offsets, so the pen advance

and kerning are exactly as without one. The flip side is that **offsets do

not extend the culling bounds**: a big wave near the screen edge can clip, so

keep the measured text in view.

• Canvas pays per distinct colour. WebGL and WebGPU carry the tint in the

per-vertex colour, so batching survives. The Canvas renderer instead caches a

tinted copy of the whole font page per colour, and that cache is unbounded

for the renderer's life, so a colour driven by ctx.time allocates a

page-sized canvas every frame there. On Canvas, keep the palette finite or

animate only the offsets.

• tint is multiplicative, like fillStyle above: white leaves the glyph

alone, and out.tint multiplies whatever the label's own fillStyle is.

Text has no equivalent, since it rasterises the whole string into one texture.

Making it read as motion

Three things decide whether a wave looks alive or mechanical, and none of them

is the amplitude.

• Take the phase from ctx.x, not ctx.index. This is the big one.

Stepping per glyph advances the wave as far across a narrow I as across a

wide M, so the wavelength stretches and squashes with whatever the line

happens to spell, and the eye reads the unevenness as a mechanism. ctx.x

gives the wave one fixed length in pixels and lets it travel through the

text at a constant rate, whatever the letters are.

• Round the offsets if the frame is upscaled. They are added to the glyph

destination unchanged, so a fractional offset in a small design resolution

that is then magnified with nearest sampling makes the glyph's OWN pixels

come out uneven — and differently uneven each frame. That is a shimmer, not

movement. Math.round the offset and it steps cleanly, for the same reason

subPixel defaults to false. Whole pixels mean few steps, so the amplitude

has to be large enough to give you several.

• A shake is not a sine. A sine moves the whole line as one body, which is

a wave however fast you run it. A shake gives each glyph a direction

unrelated to its neighbours: pick a random angle per glyph, hold it for a

tick, interpolate to the next, and scale the whole thing by a decay if it is

an impact rather than a state.

Starting points that look right across sizes, as fractions of the font size:

an amplitude around an eighth of it, a wavelength around ten times it, and a

little under 5 radians per second. For a shake, re-target each glyph roughly

20 to 40 times a second.

ctx.time starts when the effect is assigned, so attach it at the moment

you want to be t=0 — on the frame the hit lands, not when the label was built.

A label reused across lives carries the old clock, so a decay attached once

will already be spent the second time.

Both, whole:

import { BitmapText } from "melonjs";

const title = new BitmapText(app.viewport.width / 2, 60, {
    font: "display",
    textAlign: "center",
    text: "JUNGLE RABBIT",
});
title.floating = true;
app.world.addChild(title);

// A wave: four pixels deep, a ~1.3s cycle, and a wavelength of roughly ten
// times the font size. Phase from ctx.x, offset rounded.
title.glyphEffect = (out, ctx) => {
    out.offsetY = Math.round(Math.sin(ctx.time * 0.005 + ctx.x * 0.02) * 4);
};

// A shake: each glyph picks its own direction, holds it for a tick, slides to
// the next, and the whole thing decays. Deterministic, so the same hit shakes
// the same way every time.
const angle = (i, tick) => {
    const h = Math.sin(i * 127.1 + tick * 311.7) * 43758.5453;
    return (h - Math.floor(h)) * Math.PI * 2;
};
const impact = (amplitude, durationMs, rate) => (out, ctx) => {
    const left = 1 - ctx.time / durationMs;
    if (left <= 0) {
        return;                              // settled; `out` is already zeroed
    }
    const amp = amplitude * left * left;
    const tick = (ctx.time / 1000) * rate;
    const held = Math.floor(tick);
    const k = tick - held;
    const a = angle(ctx.index, held);
    const b = angle(ctx.index, held + 1);
    out.offsetX = Math.round((Math.sin(a) + (Math.sin(b) - Math.sin(a)) * k) * amp);
    out.offsetY = Math.round((Math.cos(a) + (Math.cos(b) - Math.cos(a)) * k) * amp);
};

// assigned at the moment of the hit, which is what makes ctx.time start here
banner.setText("CAPSIZED!");
banner.glyphEffect = impact(3, 650, 24);

The HUD pattern

A HUD is a floating container at a high z, built once and re-added:

class HUD extends Container {
    constructor(app) {
        super(0, 0, app.viewport.width, app.viewport.height);
        this.floating = true;       // screen space, not world space
        this.isPersistent = true;   // survives a world reset
    }
}

app.world.addChild(new HUD(app), 100);   // explicit z — not `.z = Infinity`

There is no UIContainer class. Use Container for a bare grouping,

UIPanel for anything with a background or a border (see below), or

UIBaseElement when you want the pointer handling and nothing drawn — it is a

Container that already sets floating = true and isKinematic = false.

Three things matter here:

• floating = true opts the container out of camera transforms, so it stays

put while the world scrolls. Without it the HUD scrolls away — a silent

failure.

• Only the parent container needs floating. addChild forces

child.floating = false under a floating parent, and the container applies one

projection swap for the whole subtree.

• Pass an explicit z. this.z = Number.POSITIVE_INFINITY appears in several

shipped examples and does nothing — renderable.z is not a real property. The

real accessor is depth (an alias for pos.z); addChild(child, z) sets it

for you.

Bars and gauges: ProgressBar

A health bar, a shield gauge, a cooldown and a loading bar are one object. Set

value; it redraws.

const hp = new ProgressBar(20, 20, {
    width: 200, height: 24,
    min: 0, max: 100, value: 100,      // min/max default to 0 and 1
    trackColor: "#0000001a",
    fillColor: "#c0392b",
    borderColor: "black", borderWidth: 2,
    padding: 2,                        // inset of the fill inside the track
    radius: 6,                         // 0 is square
});
hud.addChild(hp);

hp.value -= 15;                        // clamps; `ratio` reads back 0..1

direction takes "left-to-right" (the default), "right-to-left",

"top-to-bottom" or "bottom-to-top", each growing from its own edge.

The colour follows the value by mutating the Color you passed. This is

the one non-obvious part, and it is deliberate: every game wants a different

rule, and most want the colour somewhere else too, next to a shader uniform or

a glow. The bar re-reads fillColor every frame, so it never has to know:

const full = new Color().parseCSS("#7fe0ff");
const empty = new Color().parseCSS("#ff2a1e");
const mixed = new Color();

const shield = new ProgressBar(x, y, {
    width: 190, height: 13,
    trackColor: null,                  // hollow: the scene shows through
    borderColor: "rgba(122, 190, 235, 0.8)",
    padding: 2,
    fillColor: mixed,                  // KEPT, not copied
});

// each frame
mixed.copy(empty).lerp(full, hp);      // the bar follows, and so can a shader
shield.value = hp;

Blinking is the same story: setOpacity() already does it, on whatever clock

the game is already keeping. Neither is built in.

bindEvent takes the value from an event instead, and the subscription

then lives exactly as long as the bar:

const bar = new ProgressBar(0, y, {
    width: renderer.width, height: 4,
    trackColor: "black", fillColor: "#55aa00",
    bindEvent: event.LOADER_PROGRESS,
});

Hold that subscription anywhere else and it outlives the bar, firing into a

renderable whose pos has already been released. The engine's own loading

screen is built on exactly this.

Two more worth knowing:

• The border is four filled rects, not a stroke (unless the bar is

rounded). A stroke is centred on the path, so half of it falls outside and a

bordered bar paints over its neighbour; and strokeRect adds corner joins

only above a line width of 1, so at exactly 1 a corner pixel goes missing.

• trackColor: null leaves the bar hollow, which is what a HUD over a busy

scene usually wants: the border is then the only chrome.

Panels and cards: UIPanel

A dialog, a menu card, a HUD frame: a rectangle with a fill, an optional

border and children on top. UIPanel is a UIBaseElement, so it is a

container that carries pointer events.

import { UIPanel, Text } from "melonjs";

const card = new UIPanel(cx - 200, cy - 120, {
    width: 400, height: 240,
    fillColor: "#0a1526",                       // null leaves it transparent
    borderColor: "rgba(130, 200, 240, 0.55)",   // null for none
    borderWidth: 2,
    radius: 14,
});
card.addChild(new Text(200, 24, { text: "PAUSED", textAlign: "center" }));
app.world.addChild(card, -150);   // small depth: see the 3D section below

fillColor, not backgroundColor. Container already has a

backgroundColor and it does something else entirely: it is handed to

clearColor and wipes the whole render target, so a panel setting it repaints

the screen. The names here match ProgressBarSettings.

The border is drawn INSIDE the box. width and height are the outer

size, border included, so a panel never paints past the rectangle you gave it

and its getBounds(), which is what the pointer dispatcher hit-tests, stays

that rectangle. One thing to know with a radius: keep borderWidth at 2 or

more. A rounded border has to be stroked, and a stroke at exactly 1 goes out as

plain lines, which drops a pixel where each arc meets its straight edge. A

square border is four fills and is fine at 1.

blocking (default true) makes the panel consume clicks and hovers that

miss its children, which is what a dialog wants — a click on the card, or on

the dimmed scene behind it, must not reach the game. Children are unaffected:

the dispatcher asks a child before the container holding it. Set it false for

a frame drawn over a playfield that still has to receive clicks.

A full-screen panel with a translucent fillColor and borderColor: null is

the veil behind a modal, and it blocks for free.

bobAmplitude / bobPeriod give the panel the slow float an idle

dialogue or pause panel usually has, so a screen waiting for input does not

look frozen:

card.bobAmplitude = 3;      // pixels either side; 2 to 4 is the usual range
card.bobPeriod = 2500;      // ms for a full up-and-down cycle

Three things to know. The panel is clickable where it is drawn: the offset

goes into getAbsolutePosition(), which is what the bounds pipeline reads, so

the hit area of the panel and of every widget inside it follows the float at

any depth. pos is left alone, so it still reads and writes the resting place.

It is fed from update(dt), so it stops with the rest of the game on

state.pause() unless the panel sets updateWhenPaused, which a pause menu

does anyway. And do not go below an amplitude of 2: the offset is rounded to whole

pixels, so an amplitude of 1 has three positions to move through and twitches

instead of drifting.

One consequence worth knowing: getBounds() on a bobbing panel answers

differently each frame, so do not cache it or compare two children's bounds

while a bob is running.

A pause overlay: updateWhenPaused cascades

state.pause() stops every world child stepping while rendering carries on, so

the overlay that reads the key to UNPAUSE has to opt out:

const overlay = new UIPanel(0, 0, { width: w, height: h, fillColor: "rgba(4,8,16,0.72)" });
overlay.updateWhenPaused = true;   // the whole subtree, not just this one

The flag covers the container's entire subtree, so nested labels, buttons

and the card inside the veil all keep stepping without being flagged one by

one. That matters for more than their update: inViewport is computed in the

same loop that honours the pause, so a skipped child keeps whatever it had —

false for one built while paused — and Container.draw then declines to draw

it, silently. A panel whose labels never appear while paused is this.

Buttons: extend the handlers, do not bind listeners

UISpriteElement is a Sprite that already registers itself for pointer

events, so a button is made by overriding methods rather than by wiring

registerPointerEvent:

| handler | fires | returns |

|---|---|---|

| onClick(event) | pressed | false to stop the event propagating |

| onRelease(event) | pressed and released | false to stop propagating |

| onOver(event) | pointer enters | false to stop propagating |

| onOut(event) | pointer leaves | — |

| onHold() | pressed and held | — |

class MuteButton extends UISpriteElement {
    constructor(x, y) {
        super(x, y, { image: atlas, region: "speaker.png" });
        this.setOpacity(0.5);
    }
    onOver() { this.setOpacity(1.0); }
    onOut()  { this.setOpacity(0.5); }
    onClick() {
        audio.muteAll();
        return false;          // consumed — do not fall through to the world
    }
}

The pointer still has to reach it: a renderable with isKinematic = true — the

default on a plain Renderable — is skipped by the broadphase and receives

nothing. UISpriteElement and UIBaseElement clear it for you; anything else

you make clickable has to clear it itself.

Keyboard shortcuts on a button: bindKey

A UITextButton can be activated from the keyboard as well as the pointer:

const resume = new UITextButton(x, y, {
    font: "PressStart2P", text: "RESUME",
    bindKey: input.KEY.ENTER,     // a keycode, or an action name you bound
});

It takes either an input.KEY keycode or the NAME of an action the game has

already bound. Neither form touches the global bindings, so a keycode works

even when nothing has called input.bindKey for it.

The handlers fire with no Pointer. This is the one that breaks existing

code: a button whose onClick(event) reads event.gameX worked for years

because only a pointer could ever press it, and the first key press throws.

Guard the argument, or keep pointer-only work out of the handler:

onClick(event) {
    if (event) { this.ripple(event.gameX, event.gameY); }  // pointer only
    this.resumeGame();                                     // either input
}

Three more rules, each with a reason you can feel in a game:

• The first input to press owns the button. A pointer cannot release a key

press and a key cannot release a pointer press, so a button cannot be left

half pressed by using both at once.

• Losing focus cancels a held key WITHOUT calling onRelease(). If you

use the pair as press and depress, firing while held, also stop on

event.BLUR or the gun is still firing when the player alt-tabs back.

• isClickable gates the press, not the release. Disabling a button while

its key is down still releases it, so it cannot stick.

Shortcuts live only while the button is activated, and never start dragging or

tap-and-hold: those stay pointer gestures.

A fullscreen option

Three methods on the Application, not on device (they moved there in

19.7.0):

app.isFullscreen();        // boolean
app.requestFullscreen();   // needs a USER GESTURE
app.exitFullscreen();      // needs none

requestFullscreen only works from a click or a key press. Called at

startup, or from a timer, the browser rejects it and nothing happens. That is

why a fullscreen option lives on a menu entry or a key, never in a scene's

setup:

import { event, input } from "melonjs";

input.bindKey(input.KEY.F, "fullscreen");
event.on(event.KEYDOWN, (action, keyCode, edge) => {
    // `edge` only: a held key auto-repeats, and a repeat toggles straight
    // back out of what it just turned on
    if (action !== "fullscreen" || !edge) {
        return;
    }
    if (app.isFullscreen()) {
        app.exitFullscreen();
    } else {
        app.requestFullscreen();
    }
});

One subscription on KEYDOWN covers every screen, where polling

isKeyPressed would need a copy in each menu that is up.

It fullscreens the canvas's PARENT, not the canvas, so a DOM HUD or

overlay inside that container goes with it. Pass an element to override.

A menu entry showing ON/OFF must read the state back, not cache it:

the player can leave fullscreen with Escape or the browser's own control

without your code hearing about it, and a remembered flag then lies.

entry.setLabel(`FULLSCREEN: ${app.isFullscreen() ? "ON" : "OFF"}`);

Nothing to branch on for support. requestFullscreen checks

device.hasFullscreenSupport itself and no-ops without it, and

exitFullscreen no-ops when not fullscreen. Query

device.hasFullscreenSupport only to decide whether to *offer* the option at

all.

Fullscreen does not resize a fixed-resolution game: with

new Application(w, h, { scale: "auto" }) the internal resolution is

unchanged and only the canvas scales, so a HUD built from

renderer.width/height needs no repositioning.

z is container-local

addChild(child, z) writes a depth that means something **only among that

container's own children**. autoDepth numbers them from 1, so a button at

z = 8 inside one panel and a whole second panel at z = 2 are not comparable

numbers, and the second panel is still the one on top. Drawing and hit-testing

both resolve a cross-container pair the same way: walk up to the common parent

and compare the two siblings there.

So a panel is raised by raising the panel, not its contents:

this.ancestor.moveToTop(panel);   // reorders, and sets z past its neighbour

Giving a child a huge z to "put it in front" lifts it only within its own

panel. This is also why a widget's z never has to be coordinated with

anything outside its own container.

Panels and boxes that stretch: NineSliceSprite

A panel, dialogue box or button background drawn from one small image wants

NineSliceSprite, not a Sprite scaled up. It cuts the image into a 3×3 grid,

keeps the four corners at their own size, and stretches only the edges and the

middle, so the border stays crisp at any size. Do not hand-roll it from nine

sprites.

From a texture atlas, the third argument of createSpriteFromName asks for the

9-slice version. This is how the UI example's draggable panel draws itself:

class Panel extends UIBaseElement {
    constructor(x, y, width, height) {
        super(x, y, width, height);
        // `true` returns a NineSliceSprite stretched to the panel's size
        this.addChild(
            texture.createSpriteFromName("grey_panel", { width, height }, true),
        );
    }
}

From a loaded image, construct it directly (the Text example's dialogue box):

const box = new NineSliceSprite(48, 640, {
    image: "panel",
    width: 900,
    height: 256,
    insetx: 36,          // the border thickness in the source image
    insety: 36,
    tint: "#3a3f58",     // tint works as on any sprite
});
box.anchorPoint.set(0, 0); // place it by its top-left corner

Three things trip people up:

• width and height are mandatory. They are the size to stretch *to*, not

the image's size, and the constructor throws without them.

• Set insetx / insety to the art's border. Left out, each defaults to a

quarter of the source image, which on most panel art cuts through the border:

the corners smear or the border thins out when the box grows.

• The anchor is the centre, as for any Sprite. Set it to (0, 0) to place

the box by its corner, as a layout usually wants.

It resizes live: set width / height and the next draw re-slices, so a box

that grows with its text, or a panel the player drags larger, needs no new

sprite.

An opaque panel has to say so

The hit test asks the topmost renderable first and then keeps walking down, so

consuming is what stops a widget underneath answering too. A panel that can

overlap another one therefore wants both:

class Panel extends UIBaseElement {
    onClick() {
        this.ancestor.moveToTop(this);  // and bring it to the front
        return false;                   // consumed
    }
    onOver() {
        return false;                   // the frame the pointer crosses in
    }
    onActivateEvent() {
        super.onActivateEvent();
        // and every frame after that, when the pointer is ALREADY inside and
        // `onOver` no longer fires
        input.registerPointerEvent("pointermove", this, () => false);
    }
    onDeactivateEvent() {
        input.releasePointerEvent("pointermove", this);
        super.onDeactivateEvent();
    }
}

Leave onOut alone: an element that suppresses its own leave stays lit after

the pointer has gone. It fires for the covered element too, not only when the

pointer leaves its bounds, so a button half under the panel goes dark as soon as

the pointer slides onto the panel, which never takes it out of the button's

bounds.

In a 3D scene, a HUD needs a SMALL depth

floating opts a renderable out of the camera transform. It does not opt it

out of the depth sort, and the two sorts read z differently:

| container sortOn | ordered by | on top |

| --- | --- | --- |

| "z" (default, 2D) | pos.z | highest z |

| "depth" (what Camera3d sets) | distance from the camera | nearest the camera |

Under "depth" a floating child is ordered by |pos.z| alone — its pos.x/y

are screen pixels, not a place in the world, and the camera does not move

relative to it. So the *magnitude* is the distance, and the sign is ignored:

world.addChild(hud, -150);        // small -> in front of the whole scene
world.addChild(backdrop, -10000); // large -> behind the whole scene
world.addChild(backdrop, 100000); // equally far: sign does not matter

Give a HUD the huge z that would put it on top in 2D and it lands at the far end

of the level instead, with the scenery drawing over it. Both shipped idioms are

the same rule: afterBurner's HUD sits at -150, and the glTF, Billboard, Night

City and Instanced Forest examples park a floating sky at -10000 or 100000.

Related skills

• melonjs-scenes-and-state — Tween and the easing families, for animating any of this
• melonjs-input — which object gets a pointer event, and how to consume it
• melonjs-loading-assets — web fonts have to be loaded before a Text bakes