Don't miss out on the action at this year's Chrome Dev Summit, streaming live on YouTube. Watch now.

Rendering Text in WebVR

By David Lyons, Lead WebVR Developer, Within

View the site

Within ( is a platform for storytelling in virtual reality. So when the team heard about WebVR in 2015 we were immediately interested in its potential. Today, that interest manifests into a unique subdomain of our Web platform, Anyone with a VR-enabled browser can go to the site, click a button and throw on a headset to be immersed in our portfolio of VR films.

Today that includes but is not limited to Chrome on Daydream View. For information on your device and head-mounted display check out

Like other virtual reality specific rendering environments, the web predominantly relies on a three-dimensional representation of a scene. This scene has a camera, your perspective, and any number of objects. To help manage this scene, camera, and objects we use a library called Three.js which leverages the <canvas> element to throw rendering onto your computer's GPU. There are many useful Three.js add-ons to make your scene viewable in WebVR. The main two are THREE.VREffect for creating a viewport for each eye and THREE.VRControls for translating the perspective (e.g the rotation and position of the head-mounted display) convincingly into your scene. There are many examples of how to implement this. Check out the Three.js WebVR examples for ways to get started.

As we got further into our exploration of WebVR we ran into an issue. If we look at the contents of the web, text is an integral part of it. While the majority of our content is video based, if you go to the Within site text surrounds the content; the user interface and additional information about a film or related films are all constructed with text. Furthermore all of this text is created in the DOM. Our WebVR explorations and are all in <canvas>.

Text used in WebVR for

What are my Options?

Luckily, there is work being done to make this possible. In fact in our research we found a number of effective ways to render text in a three-dimensional environment on a <canvas> element. Below is a matrix of a few we found marked with pros and cons for each:

Resolution Independent Typographic Features Performance Ease of Implementation
2D canvas text
Triangulated vector text
Extruded 3D text
Signed distance field bitmap text

Our Decision: SDF Bitmap Font

2D canvas with ctx.fillText() can do text wrapping, letter spacing and line height, but overflow gets cut off, and text will be blurry if you zoom in really far. You could increase the size of the canvas texture, but might hit an upper limit in texture size or performance could suffer if the texture is too big.

Extruded 3D text is essentially the same as triangulated vector text, but with depth and possibly a bevel so it has at least twice as much geometry. Either of these could work in small doses for titles or logos, but wouldn't perform as well for large quantities of text and neither has typographic features.

Font to SDF bitmap workflow

Bitmap fonts use one quad (two triangles) per character, so they use less geometry and perform better than triangulated vectors. They're still raster based since they use a texture map sprite, but with an SDF shader they're basically resolution-independent so they look nicer than a 2D canvas texture. Matt DesLauriers' three-bmfont-text also includes reliable typographic features for text wrapping, letter spacing, line height and alignment. Overflow doesn't get cut off. Font size is controlled through scale. We chose this route because it gave us the best options for design while staying performant. Unfortunately, it wasn't as easy to implement so we'll go through the steps in the hopes of helping fellow developers working in WebVR.

1. Generate a bitmap font (.png + .fnt)

Hiero interface
Hiero output (Bitmap PNG and .fnt file)

Hiero is a bitmap font packing tool that runs with Java. The Hiero documentation doesn't really explain how to run it without going through a complicated build process. First, install Java if you haven't already. Then, if double-click on the runnable-hiero.jar doesn't open Hiero, try running it with this command in the console:

java -jar runnable-hiero.jar

Once Hiero is running, open a .ttf or .otf desktop font, enter any extra characters you want included, change rendering to Java to enable effects, increase the size so that your characters fill up the entire glyph cache square, add a distance field effect, adjust the distance field's scale and spread. The scale value is like a resolution. The higher it is, the less blurry it will be, but the longer it will take for Hiero to render the preview. Then save your bitmap font. It generates a bitmap font consisting of a .png image and an AngelCode .fnt font description file.

2. Convert AngelCode to JSON

Now that the bitmap font has been generated, we have to load it into our javascript app with Matt DesLauriers' load-bmfont npm package.

We could browserify load-bmfont and use that on the front end, but instead we're going to run load-bmfont.js with Node to convert and save Hiero's AngelCode .fnt to a .json file:

npm install
node load-bmfont.js
Example of output JSON

Now we can bypass load-bmfont and just do an XHR (XMLHttpRequest) request on the .json font file.

var r = new XMLHttpRequest();'GET', 'fonts/roboto/bitmap/roboto-bold.json');

r.onreadystatechange = function() {
  if (r.readyState === 4 && r.status === 200) {


function setup(font) {
  // pass font into TextBitmap object

3. Browserify three-bmfont-text

Once we have the font loaded, Matt's three-bmfont-text will take care of the rest. Since we're not using Node for our own app, we're going to browserify three-bmfont-text.js into a usable three-bmfont-text-bundle.js

npm install -g browserify
browserify three-bmfont-text.js -o three-bmfont-text-bundle.js

4. SDF shader

Adjust the afwidth and threshold sliders on to see the affect of the signed distance field shader.

5. Usage

For convenience, I created a TextBitmap wrapper class for the browserified three-bmfont-text.

Text-sdf-bitmap in action
<script src="three-bmfont-text-bundle.js"></script>
<script src="sdf-shader.js"></script>
<script src="text-bitmap.js"></script>

Create an XHR request for the .json font file and create a text object in the callback:

var bmtext = new TextBitmap({ options });

To change text:

bmtext.text = 'The quick brown fox jumps over the lazy dog.';

scene.add( );
hitBoxes.push( bmtext.hitBox );

The bitmap font's .png is loaded with THREE.TextureLoader in text-bitmap.js

TextBitmap also includes an invisible hitbox for three.js raycast interaction through a mouse, camera, or hand tracked motion controllers like Oculus Touch or the Vive controllers. The hitbox's size auto-updates when you change the text options. is added to the three.js scene. If you need to access the children / Object3D's, the scene graph for the text looks like:

6. Unminify json and modify xoffsets

If your kerning look off, you may need to edit the xoffsets in the json. Paste the json into to get an unminified version of the file.

The xoffset is essentially global kerning for one character. Kerning is for two specific characters that appear next to each other. The default values in the kerning array don't actually make a difference, and it would be too tedious to edit, so you can empty that array to decrease the file size of the json. Then edit the xoffsets for kerning.

First you'll have to figure out which characters go with which char ID in the json. In three-bmfont-text-bundle.js, insert console.log after line 240:

var id = text.charCodeAt(i)
// console.log(id);

Then type into dat.gui text field on and check the console to find the corresponding ID of a character.

For example, in our bitmap font, "j" is consistently too far to the right. Its char ID is 106. So find "id": 106 in the json and change its xoffset from -1 to -10.

7. Layout

If you have multiple blocks of text and want it to flow from top to bottom like HTML, everything has to be manually positioned, similar to absolute positioning every dom element yourself with CSS. Can you imagine doing this in CSS?

* { position: absolute; }

That's what text layout in 3D is like. In the detail view: title, author, description, and duration are each a new TextBitmap object with their own styles, color, scale, etc.: = - title.height - padding; = - author.height - padding; = - description.height - padding;

This assumes that the local origin of each TextBitmap group is vertically aligned with the top of the TextBitmap mesh (see centering in text-bitmap.js update). If you change the text for any of those objects later, and the height of that object changes, you will also need to recalculate those positions. Here, only the y-position of the text is modified, but one opportunity of working in 3D is that we can push and pull the text in the z-direction, as well as rotate around the x, y and z axes.


Text and layout in WebVR have a long way to go before they're as easy and as widely-used as HTML and CSS. But working solutions exist and you can do way more in WebVR than you can with a traditional HTML web page. WebVR exists today. There will probably be better tools tomorrow. Until then, try it out and experiment. Developing without a ubiquitous framework leads to more unique projects, and that's exciting.