TypeScript icon, indicating that this package has built-in type declarations

5.12.0 • Public • Published



At the core of it, a 3D model viewer for MDX and M3 models used by the games Warcraft 3 and Starcraft 2 respectively.

The viewer part handles the following formats:

  • MDX (Warcraft 3 model): extensive support, almost everything should work.
  • M3 (Starcraft 2 model): partial support.
  • W3M/W3X (Warcraft 3 map): partial support.
  • BLP1 (Warcraft 3 texture): extensive support, almost everything should work.
  • TGA (image): extensive support, almost everything should work.
  • DDS (compressed texture): partial support - DXT1/DXT3/DXT5/RGTC.
  • PNG/JPG/GIF/WebP: supported by the browser.

There are file parsers that the viewer depends on.
These don't rely on the viewer or indeed on even running in a browser.
They include:

  • MDX/MDL: read/write.
  • M3: read.
  • BLP1: read.
  • INI: read/write.
  • SLK: read/write.
  • MPQ1: read/write.
  • W3M/W3X/W3N: read/write, including all of the internal files.
  • DDS: read (DXT1/DXT3/DXT5/RGTC).
  • TGA: read.

There are all sorts of utilities that were made over the years.
These include things like...

  • The library's unit tester, which compares rendered results against stored images that were generated in the same way.
  • The MDX sanity test, which looks for errors and weird things in MDX models.
  • A Jass context that can...well, run Jass code. That being said, it really runs Lua code converted from Jass, on a JS Lua VM. What a tongue twiser. While it supports some Warcraft 3 natives, don't expect it run whole maps. Maybe in the future 😉
  • A utility that makes it possible to open Warcraft 3 maps in the vanilla World Editor, in cases were said maps used a non-vanilla editor with extended GUI, in which case they crash upon opening in the official editor.
  • etc.

Finally, the library also comes with a bunch of clients.
A "client" in this context means external code that uses the library.
Most of these clients are simple and messy, since they were made as side projects while working on the library.
Most of these clients are also just wrappers around the viewer and the utilities, merely giving them an interface on a web page.
These include things like...

  • A simple example client.
  • The unit tester's page, which allows to run the unit tests, and to download the results.
  • The MDX sanity test's page, which visually shows the results of sanity tests, and other nifty things.
  • etc.


npm install mdx-m3-viewer
npm run build

This will generate the CommonJS, UMD, and the packed clients under dist.


Run the Webpack dev server with:

npm run serve

Once it compiled the code, open http://localhost:8080/clients/example/ in the browser, and play with the client's source to see how things change.

You can also check out the other available clients.


You can import the viewer in different ways:

// UMD export in the browser.
new ModelViewer.viewer.ModelViewer(canvas);

// require/import the library.
const ModelViewer = require('mdx-m3-viewer'); // CommonJS.
import ModelViewer from 'mdx-m3-viewer'; // ES6.
new ModelViewer.viewer.ModelViewer(canvas);

// require/import something directly.
const ModelViewer = require('mdx-m3-viewer/src/viewer/viewer'); // CommonJS.
import ModelViewer from 'mdx-m3-viewer/src/viewer/viewer'; // ES6.
new ModelViewer(canvas);

When developing with TypeScript, nothing needs to be done to get correct typings. This is true also when importing specific files, which means a client can import only what it needs from the library.

All code snippets will use the names as if you imported them directly to avoid some mess. See the examples for actual namespacing.


First, let's create the viewer:

let canvas = ...; // A <canvas> aka HTMLCanvasElement object.

let viewer = new ModelViewer(canvas);

If the client doesn't have the WebGL requierments to run the viewer, an exception will be thrown when trying to create it.

Now that we have a viewer, a scene can be created. Each scene has its own camera and viewport, and holds a list of things to update and render.

let scene = viewer.addScene();

// Move the camera backwards a bit, so we can actually see the origin.[0, 0, 500]);

Finally, we need to actually let the viewer update and render:

(function step() {


Models and textures are loaded with the load function.

To load models and textures with load, the viewer must have handlers that tell it how to load the different file formats.
If you want to load an MDX model, the MDX handler must be added to the viewer, and so on.
This is done with the addHandler function, and the different handlers are exported as a part of the library.

Let's add the MDX and BLP handlers:


Now MDX (and MDL) and BLP files will be accepted by the viewer.

Suppose we have the following directory structure, where model.mdx uses texture.blp:

├── index.html
└── Resources
    ├── model.mdx
    └── texture.blp

Loading the model is simple:

let modelPromise = viewer.load("Resources/model.mdx");

You get back a promise, which will resolve to either the MDX model, or to undefined if any error occured.

When the MDX model loads, it also loads internal resources, like its textures, so the viewer will attempt to fetch texture.blp.
If the server knows this is a relative path to Resources/ then all is fine.
It is typically a lot easier and more dynamic to control the paths on the client though.
This is done with "path solvers" - functions that, given a source to load from, such as a path, can modify it and return the actual source to load from.
It will probably make more sense with code - let's load the model again, with the texture fetch asking for the correct path: Resources/texture.blp.

function pathSolver(path) {
  return "Resources/" + path;

let modelPromise = viewer.load("model.mdx", pathSolver);

Here's the short version of what happens:

  1. pathSolver is called with "model.mdx" and returns "Resources/model.mdx".
  2. The viewer starts the fetch, and emits the loadstart event.
  3. A promise is returned.
  4. ...time passes until the file finishes loading...
  5. The viewer detects the format as MDX based on the file data (the url is irrelevant to this process).
  6. The model is constructed successfuly, or not, with a load or error event sent respectively, followed by the loadend event.
  7. In the case of an MDX model, the previous step will also cause it to load its textures.
  8. pathSolver is called with "texture.blp", which returns "Resources/texture.blp", and we loop back to step 2, but with a texture this time.

Path solvers can return promises which will be waited upon, and they can return models and textures directly for injections.

Generally speaking, you'll need a simple path solver that expects urls and prepends them by some base directory or API url.
There are however times when this is not the case, such as loading models with custom textures, and handling both in-memory and fetches in the same solver as done in the map viewer.

Once the promise is resolved, we have a model, however a model in this context is simply a source of data.
The next step is to create an instance of this model.
Instances can be rendered, moved, rotated, scaled, parented to other instances or nodes, play animations, and so on.

let instance = model.addInstance();

And finally add the instance to the scene, so it's updated and rendered:

// Equivalent to:

Other resources, such as SLK tables and INI configurations, are loaded with loadGeneric.

let resourcePromise = viewer.loadGeneric(path, dataType[, callback]);


  • path is an url string.
  • dataType is a string with one of these values: text, arrayBuffer, blob, bytes, or image.
  • callback is an optional function that will be called with the data once the fetch is complete, and should return the resource's data.

If a callback is given, will be whatever the callback returns.
If a promise is returned, the loader waits for it to resolve, and uses whatever it resolved to.
If no callback is given, the data will be the fetch data itself, according to the given data type (bytes refers to a Uint8Array).

loadGeneric is a simple layer above the standard fetch function.
The purpose of loading other files through the viewer is to cache the results and avoid multiple loads, while also allowing the viewer itself to handle events correctly.

Events and Promises

As mentioned above, there are emitted events, and they can be used with the NodeJS EventEmitter API:

viewer.on(eventName, listener), listener)
viewer.once(eventName, listener)
viewer.emit(eventName[, ...args])

The built-in names are:

  • loadstart - a resource started loading.
  • load - a resource successfully loaded.
  • error - something bad happened.
  • loadend - a resource finished loading, follows both load and error when loading a resource.
  • idle - all loads finished for now.

For example:

viewer.on('error', (e) => console.log(e));

In addition there is viewer.whenAllLoaded([callback]), which can be used to run code when nothing is loading. If a callback is given, it will be called, otherwise a promise is returned. If there are no resources currently being loaded, this will happen instantly. Otherwise, it will happen once the idle event is emitted.

viewer.whenAllLoaded((viewer) => {
  // Nothing is loading!

  .then((viewer) => {
    // Nothing is loading!

And now some more specific information and tips.

Team colors, event objects, and Reforged

When loading an MDX model that uses team color/glow textures, it will tell the handler to load all of the team textures.

Similarly, if an MDX model has event objects, it will tell the handler to load the needed SLK files.

The handler uses load much like the client does, and thus the same implications apply - if the server is set for the relative paths, all is fine, otherwise a path solver should be used.

A path solver can be passed when adding the handler:

viewer.addHandler(handlers.mdx, wc3PathSolver);

The handler also selects between TFT (16) and Reforged (28) team colors.
These will be used regardless of whether any specific model being rendered is a TFT or Reforged model.
The default mode is TFT, and it can be changed by passing true as the third parameter when adding the handler:

viewer.addHandler(handlers.mdx, wc3PathSolver, true); // Reforged team colors and event objects.

Interacting with model instances

Model instances are nodes, and can be transformed as such:

instance.setLocation([50, 0, 0]); // Move to the given point.
instance.move([50, 0, 0]); // Move by the given offset.

instance.setRotation([0, 0, 0, 1]); // Set the rotation to the given quaternion.
instance.rotate([0, 0, 0, 1]); // Rotate by the given quaternion.

instance.setScale([2, 2, 2]); // Set the scale to the given vector.
instance.scale([2, 2, 2]); // Scale by the given vector.

instance.setUniformScale(2); // Set the scale to the given number.
instance.uniformScale(2); // Scale by the given number.

instance.face([50, 0, 0], [0, 0, 1]); // Face the given point, with the given "up" vector.

instance.setParent(nodeOrInstance); // Set a parent, making all other transformations relative to it.
instance.setParent(); // Remove the parent.

Both MDX and M3 instances can run animations, have team colors, etc.:

instance.setSequence(-1); // No animation.
instance.setSequence(0); // First animation.

instance.setSequenceLoopMode(0); // Never loop animations.
instance.setSequenceLoopMode(1); // Loop animations based on the model.
instance.setSequenceLoopMode(2); // Always loop animations.

instance.setTeamColor(0); // First team color.

instance.setVertexColor([1, 0, 0]); // Red vertex color.

let node = instance.getAttachment(0); // Get the first attachment point.

MDX instances have setTexture, setParticle2Texture, and setEventTexture, to override textures, particle emitter textures, and event object textures:

instance.setTexture(0, myTexture); // Override texture 0.
instance.setParticle2Texture(0, myTexture); // Override the texture of particle emitter 0.
instance.setEventTexture(0, myTexture); // Override the texture of event emitter 0.

instance.setTexture(0); // Remove the override, same with the other functions.

M3 instances have setTexture:

instance.setTexture(1, 0, myTexture); // Override texture 0 of standard material 1.

Solver Params: Reforged and the map viewer

It is in fact possible to send more data to path solvers with load. The full signature is as follows:

let resourcePromise = viewer.load(src[, pathSolver[, solverParams]]);

Where solverParams can be anything.

When solverParams exists, it will be sent to the path solver as the second argument:

function pathSolver(src, solverParams) {
  // ...

The MDX handler and the map viewer use solverParams to select between SD/HD Reforged resources, and to select specific tileset resources.

The MDX handler defines the parameters as such: {reforged?: boolean, hd?: boolean}.
The map viewer defines them as such: {reforged?: boolean, hd?: boolean, tileset: string}.

If reforged is falsy or doesn't exist, they want a TFT resource.
If reforged is true, they want a Reforged SD resource and...
If hd and reforged are true, they want a Reforged HD resource.

For example, when a new MDX model is being loaded, and it is detected as a Reforged model (version > 800), any internal resources like its textures will be loaded with solverParams = {reforged: true}, and if the model is detected to be HD: solverParams = {reforged: true, hd: true}.

You can also manually supply your own parameters.
For example, let's suppose we want to load the Warcraft 3 Footman model, but with a twist - we want all three versions of it - RoC/TFT, Reforged SD, and Reforged HD.
The loading code can be something along these lines:

let TFT = viewer.load('Units/Human/Footman/Footman.mdx', mySolver);
let SD = viewer.load('Units/Human/Footman/Footman.mdx', mySolver, {reforged: true});
let HD = viewer.load('Units/Human/Footman/Footman.mdx', mySolver, {reforged: true, hd: true});

So what does the path solver do with solverParams?
As always, that depends on the client.
For example, the solver may append the parameters as url parameters, and the server selects the game based on them.
The solver can also completely ignore these parameters and return whatever resources it wants.

Starcraft 2 models are tiny

SC2 models are tiny compared to WC3 models.
If a client needs models of both games to co-exist, it's suggested to scale SC2 models by 100.
This can be done with something along the lines of:

let instance = model.addInstance();

if (model instanceof handlers.m3.resource) {

Loading resources from memory

Resources don't have to be fetched - if you have the data, you can load it directly.
Nothing special is needed, simply use it as you would an url:

let resourcePromise = viewer.load(buffer);

Say a web page wants to load MDX models from local files that are dragged into it.
After some event handling, you end up with data such as a string or an ArrayBuffer.

Practically speaking, most MDX models will attempt to load Warcraft 3 textures.
This means that if we load the model directly, it will fail to load the textures, unless as always, the server is set for the relative paths.

A path solver can again simplify the load:

function pathSolver(src) {
  if (src === buffer) {
    return src;

  return wc3PathSolver(src);

viewer.load(buffer, pathSolver);

When the thing being loaded is the buffer, it will be used, otherwise, e.g. for the textures, the Warcraft 3 path solver will be used instead.

Primitive shapes

It is possible to construct primitive shapes with createPrimitive, which is available under utils.mdlx.createPrimitive.

The function expects an object describing a primitive geometry, which can be obtained via the different functions in utils.mdlx.primitives.

An optional material can be given, which can control the render mode between polygons and lines, the color, texture, and such.

For example:

let modelPromise = createPrimitive(viewer, primitives.createUnitCube(), { color: [1, 0, 0] });

Note that this loads a standard MDX model which can be used like any other MDX model.


MDX models have sound emitters, and the viewer supports them.

If sound is desired, viewer.audioEnabled should be set to true BEFORE loading models.
This signals to the MDX handler that sound is desired, and it will load the neccessary sound files when loading models.
If audioEnabled isn't true, the sound files aren't downloaded in the first place to reduce file fetching.

To get the sounds to actually run, you must call scene.enableAudio(), which returns a promise that resolves to whether audio was actually enabled.
There are two reasons for it to fail - either because the browser simply does not support audio, or because the browser did not want to enable audio.
The latter will happen if you attempt to enable audio before the user made any interaction with the page (like clicking something). This is a browser policy and there is no control over it.

If audio was enabled, you will hear familiar sounds when running animations, like attack sounds, death sounds, and so on.
Not enough work was put into them to have the same feel as Warcraft 3, but it's sometimes a fun surprise to suddenly hear a model making sounds in the browser.

Scene composition

You can have any number of scenes you want.

Each scene offers the following to control how it's composed on the canvas:

  • viewport - the position and size of the scene in pixels (defaults to the entire canvas).
  • alpha - determines whether the scene has a background, or can be seen through (defaults to false - has a background).
  • color - the background color, which is used when alpha is false (defaults to black).
scene.viewport[0] = 100; // X offset from the left side of the canvas.
scene.viewport[1] = 100; // Y offset from the bottom side of the canvas.
scene.viewport[2] = 200; // Width.
scene.viewport[3] = 200; // Height.

scene.alpha = false; // Opaque, i.e. has a background (and also the default).

scene.color[0] = 1; // Red background.
scene.color[1] = 0;
scene.color[2] = 0;

The order in which the scenes are drawn is based on the order of creation, but you can manually move scenes around in viewer.scenes.

For example, let's say we want to mimic how Warcraft 3 looks. This could be done with 3 scenes:

  1. The game world.
  2. The UI with alpha = true so that where the UI isn't drawn, the game world will be seen through.
  3. The selected portrait, moved and sized to be on the correct portion of the UI.

Everything is blurry

WebGL uses a canvas as its back buffer, meaning it has the same amount of pixels as the canvas does. Surprising, right?
What may actually surprise you, however, is that the canvas back buffer isn't neccassarily the size it is drawn at, due to CSS styling.
For example, you can have a canvas that is scaled via CSS to the entire page, but if you never set its actual size, it will probably be a 100x100 pixel canvas (or whatever default size the browser uses), stretched to the page size.
If you want to set the size of the back buffer, i.e. the real resolution of the canvas, use the width and height properties of the canvas, rather than CSS properties such as clientWidth and clientHeight.

// Could be a static size.
canvas.width = 512;
canvas.height = 512;

// Or perhaps scaled with CSS, if you put this in a resize event listener.
canvas.width = canvas.clientWidth;
canvas.height = canvas.clientHeight;

// This however only changes the CSS size, not the canvas resolution!
canvas.clientWidth = 512;
canvas.clientHeight = 512;

// Nor does this. = '512px'; = '512px';

Variable frames per second

ModelViewer.update() and ModelViewer.updateAndRender() have an optional dt argument.

dt controls how much time in miliseconds to advance the animations.

By default, dt is set for 60FPS, or 1000 / 60.

If a client runs on a >60Hz monitor, and uses requestAnimationFrame for its main loop as it should, animations will run faster than they should.

In other cases, a client might have too many things rendering and it slows down, causing animations to go slow motion.

To support a variable FPS while keeping the same animation speed, dt can be controlled dynamically, for example:

 let lastTime =;

(function step() {

  let now =;

  // The faster the FPS, the lower dt will be.
  // Twice the FPS? half the dt.
  // There are more frames per second, so every frame advances the animation less.
  // And the other way is also true.
  // Half the FPS? twice the dt.
  // There are less frames per second, so every frame advances the animation more.
  let dt = now - lastTime;

  lastTime = now;





Package Sidebar


npm i mdx-m3-viewer

Weekly Downloads






Unpacked Size

8.26 MB

Total Files


Last publish


  • spectralcanine