Favicon compression preview

Since a few months, RealFaviconGenerator can compress your favicon pictures for reduced bandwidth consumption. However, this option lacked three features:

  • You don’t know in advance the compression rate you will likely get.
  • You don’t know in advance if the quality loss will be acceptable.
  • It is a yes/no setting, you cannot choose the compression level.

This dark age is over with the new compression preview. The favicon editor now presents your master picture compressed with different quality settings.

Compression preview

If the differences between the original picture and the compress one are hard to spot, you can easily compare them with the zoom.

Compression preview zoom

For now, the “expected compression rate” is not that accurate. I expect to improve it in a few month when enough favicons will have been compressed to establish reliable stats.

SVG support

RealFaviconGenerator now accepts SVG pictures, in addition to PNG, JPG and GIF. Not a big deal, just the ability to jump right from Adobe Illustrator or Inkscape to RealFaviconGenerator.

When using a SVG picture, the scaling algorithm option is no longer accessible, since “scaling” is somehow replaced by “rendering”. And rendering is (supposed to) be way better than scaling.

That was issue #34.

More browserconfig.xml, less HTML

browserconfig.xml was introduced in Internet Explorer 11 and lets you define advanced web application tiles. Most of its features are too advanced for web applications that do not target Windows 8 platform capabilities. However, browserconfig.xml is very useful when it comes to favicon: it defines the various tile pictures and background color.

In fact, browserconfig.xml is not necessary. Tiles can be described directly from the HTML code. But the extra XML file has a sensible advantage: unlike HTML pages, it is loaded only by Windows 8 devices. And since most visitors do not use the Windows 8 + IE11 combo, we are glad not to clutter them with useless stuff.

By default, IE11 looks for browserconfig.xml at the root of the web site. RealFaviconGenerator takes advantage of this characteristic. Since a few months, when you choose to place the favicon’s files in the root directory, the XML file is generated. Yet, as Giuseppe Caruso pointed out, the location of this file can be overridden with the help of a <meta> tag. RealFaviconGenerator now uses this markup to rely on browserconfig.xml whatever its location. That was issue 32.

The gain? A few hundred of bytes. Nothing fancy, just the kind of little improvement that makes the service better, day after day.

Apple touch icon: The Good, the Bad and the Ugly

Warning: With the release of iOS 8, the recommended size is now 180×180. It was 152×152 when this article was published (updated 2014/12/26).

Whatever you use an iOS or Android device, the Apple touch icon appears as soon as you bookmark a web site or add it to your home screen. It is the only element still visible once you leave the site. It should be in the checklist of every web project. But this is often not the case.

A link to Apple web site
A link to Apple web site

For this study, the Apple touch icons of the 5,000 most visited web sites were studied. More precisely, this analysis focuses on the apple-touch-icon.png file located in the root directory. Although this file is not a strict requirement, it is so emblematic that studying it alone makes sense. More about this in the methodology below.

First things first: how many apple-touch-icon.png were found and probed? 804. This 16% score sounds low, but again, this file is not required for the Apple touch icon to work. A previous study concludes to a 60% support rate. Therefore, this figure should not be taken too seriously. What matters is what lies in these 804 icons.

The right answer was…

152×152. 180×180, thanks to iOS 8 (updated on September 26th, 2014).

Apple defines 4 icon sizes, from 60×60 up to 152×152. What the specifications forget to mention is that these sizes are for iOS 7. To support iOS 6 and prior, you need 4 other dimensions.

iPhone iPad
non-Retina Retina non-Retina Retina
iOS 6 and prior 57×57 114×114 72×72 144×144
iOS 7 60×60 120×120 76×76 152×152

The specifications do not explicitly define the size of the apple-touch-icon.png picture, but  a good practice is to provide a 152×152 picture. An iPad with a Retina screen running iOS 7 will use it as is, while the other devices will scale it as needed (unless they find a more suitable icon).

How many web sites follow the 152×152 recommendation? 33. A good 4%. Oh. Among them, Apple.com.

Nearly as many sites have a 60×60 icon, which is the other size you might pick after reading the Apple specs. Some others resolutions are more popular, like 114×114 (79 sites) and 144×144 (72 sites). But the winner is clearly 57×57 with 281 sites. This resolution is outdated, but it is still the reference. Google for “apple touch icon size”: at the time of writing, the first result is from StackOverflow and the first answer documents 57×57.

An apple-touch-icon.png with an iOS 6 size is the sign that it has not been updated for a while. Yet, it does not prove that the Apple touch icon won’t work on modern devices. Take Bing.com, one of the 57×57 supporters: it also defines apple-touch-icon-152x152.png. You shiny iPad will use this picture.

made with ChartBoot

This leaves us with 68% of apple-touch-icon.png with a resolution recommended (or used to be recommended) by Apple. And 32% of… something else.

Apple touch icon bloopers

Creative resolutions

18 sites have a 100×100 icon, like imgur. Stackoverflow, TheGuardian and 15 others use a 158×158 picture. NewRelic’s icon is 80×80.

Some sites anticipate the Super Retina HD screen: Tumblr exhibits a 300×300 picture. BuildWith and 6 others come with a 512×512 icon. ClipHunter (not safe for work…) takes this very seriously with a 1024×1024 picture.

Although these sizes are unexpected, iPhone and iPad are able to process them.

Square icons are so mainstream

Sure, the 175×175 icon of Flickr is not standard. But at least, this square picture can be scaled as needed by iOS. On the other hand, the 155×45 picture of Kioskea cannot fit the iOS UI. Apple’s mobile platform does its best but it makes no miracle.

The 155x45 apple-touch-icon.png of Kioskea
The 155×45 apple-touch-icon.png of Kioskea
Kioskea bookmark
Kioskea bookmark

The 1194×687 apple-touch-icon.png of 1111.com.tw is another interesting case: this is a white picture.

1111.com.tw, 1194x687 pixels of zen and fullness
1111.com.tw, 1194×687 pixels of zen and fullness

12 sites are impacted by this issue and 3 others were probably tricked during the creation of the icon. There is a difference of 1 pixel between their width and height. For example, the picture of HTCMania is 56×57. So close.

Small is beautiful. Sometimes.

Some sites have really big icons and some others take the opposite path. 7 sites have a 32×32 picture, among them Media.net and ADSLGate. Zillow and 4 others are 16×16. Duowan.com’s is only 8×8.

Pixel art maybe?
Pixel art maybe?

No less than 26 sites, such as AOL and NBCSports, choose a radical approach: the 1×1 icon. It reminds me of the old times, when complex HTML designs were made of stretched transparent 1×1 GIF pics. Except that such setting does not make any sense for apple-touch-icon.png. Note that some of these sites are a bit tricky. AOL, for example, also declares valid Apple touch icons. As a consequence, iOS visitors do not hear about the strange 1×1 picture.

4G recommended

Resolution is one thing, file size is another. The studied icons weigh 8.5KB on average, but some pictures are way larger. Yelp’s icon is 91KB, which is a lot when you consider that it is 57×57.

91KB, 57x57 Yelp icon
91KB, 57×57 Yelp icon

Addic7ed doubles this figure with its fancy 198KB, 1341×1609 icon. Very strange design by the way.

Addic7ed's... icon
Addic7ed’s… icon
Addic7ed's icon with iOS best effort
Addic7ed’s icon with iOS best effort

The winner in this category is ClipHunter with its 631KB icon! Sure, the drawing deserves all its bits (not safe for work, really).

But the most disappointing icon is probably Apple’s. With 96KB, the message is clear: buy an iPhone 5 and upgrade your plan to 4G.

Conclusion

Among the apple-touch-icon.png pictures that follow the dimensions of Apple specifications, 85% of them stick to the old versions of iOS. Only 15% follow iOS 7, released 6 months ago. Apparently, webmasters are not in a hurry to update them.

made with ChartBoot

The overall results are positive, with only 32% of picture with undocumented dimensions. Even if these dimensions are sometimes surprising, most of them are processed correctly by iOS. Yet some sites should definitely update their icons.

made with ChartBoot

Maybe the most striking figure is the amount of different sizes encountered during this study: 60 different resolutions were found, from the popular 57×57 to the unique 110×110 of the NCAA. This number reveals how fragmented the information is. 57×57 is still broadly advertised. 114×114 has its fans. 129×129 was popular for a while…

When creating an Apple touch icon, you should use an up-to-date favicon generator (and now let’s see if auto-promotion is efficient). You probably don’t bother checking Apple specifications every so often. Fortunately, some people do it for you. To make sure your favicon stays relevant, you can follow us (Twitter, Google+ or Facebook) to be notified when the generated pictures and code are updated.

Favicon update is still unpractical. You can expect more in the next few weeks… stay tuned!

Methodology

The apple-touch-icon.png picture is famous but not required to enable the Apple touch icon. There are actually four ways to display this icon:

  • HTML declaration. For example, <link rel="apple-touch-icon" sizes="76x76" href="touch-icon-ipad.png">
  • Dimension-specific pictures, such as apple-touch-icon-76x76.png
  • Precomposed touch icons (eg. apple-touch-icon-precomposed.png)
  • apple-touch-icon.png

As a consequence, when a web site as no apple-touch-icon.png or this picture is somewhat wrong, it may still support the Apple touch icon successfully. For example, by providing alternative pictures declared in the HTML code.

To collect data, some portions of the favicon checker were reused. The script parsed the Alexa top web sites listing and requested each site using an iPad Mini user agent to make sure it gets the mobile version of the site. Then, it tried to download apple-touch-icon.png and get its dimensions. All failures were ignored. For example, when the file is not present, some sites do not return a plain 404 error but an error page. The script failed at parsing such “picture”, yet this case was not distinguished from genuine corrupted pictures.

The 129×129 dimension, used by 49 sites, is a bit special. Apparently, it has been the dimension of the apple-touch-icon.png of Apple for a while, making it somehow “official”. Yet, I chose to consider it as non-official since there is apparently no reason for such resolution.

Oh, and in case you wonder, yes, I cleared my browser history at the end of this study. Thank you for reminding me.

What is the best scaling algorithm? You choose!

Generating a favicon for all platforms is a lot a matter of resizing a big picture to smaller sizes. Submit a 1000×1000 picture to RealFaviconGenerator and it will scale it to 16×16, 32×32, 152×152… and many other dimensions.

At first, I was pleased by the Mitchell algorithm. This is the default algorithm of ImageMagick, the tool used by RFG to perform image manipulation. A few weeks ago, Mazyad Alabduljaleel, an iOS developer, submitted his own picture and the result was not that great.

The Mitchell algorithm tends to blur the original image in order to smooth the edges. This is usually a good thing and thus makes Mitchell a good default choice. However, Mazyad’s picture was a pixel art design, where angles are a desired feature. And RFG does its best to lessen them. Oops!

Nearest Neighbor was the right solution for this picture. Different images, different scaling algorithms.

What is the best scaling algorithm? It depends.
What is the best scaling algorithm? It depends.

RFG does not pretend to know the right answer anymore. Although it still offers Mitchell by default, you can now pick the algorithm that fits your picture best. That was issue31.

You picture, your choice
You picture, your choice

ImageMagick offers something like 40 algorithms. I did by best to select the most relevant ones in order to not overwhelm the regular user with too many choices. However, I may have overlooked important ones. Please drop a comment whenever you feel something is missing.

What’s next? Vector graphics support and maybe vectorization of bitmap pictures for better scaling results.

40% of the most visited websites have no favicon for mobile platforms

If you play golf, your swing may be just okay. But when Tiger Woods is on the green, you expect the greatest performance.  If you can run a 100 meters dash in less than 12 seconds, you are already good. Enters Usan Bolt and you expect less than 10 seconds. And a gold medal. Do you know how much time it takes for your car to accelerate from 0 to 60 mph? Probably not. Buy a $100,000+ sport car and this figure will obsess you.

The same phenomenon applies to websites. Whatever your expectations for your own work are, when it comes to the greatest, most visited websites, you look forward for the best design. The highest security. The greatest user experience. The smallest latency. Zero downtime. Of course, these sites must do well on all modern platforms: desktop, smartphones and tablets. And, yes, so should do their favicons.

This survey targets the first 100 top sites listed by Alexa. In this list, you find the classic gang: Google, Facebook, Youtube, Wikipedia, Amazon… Some Chinese and Russian giants: Baidu, Yandex… And many others.

The top websites are mobile-ready

No surprise. In 2014, nearly all major websites are designed with the mobile in mind. Most of them come with a “m.” version, while some others are responsive. Whatever the technical solution and the attractiveness of the result, this means one thing: the big players take the mobile seriously.

What “most” means exactly? 88%. What about the others 12%? After all, at this level of audience, we could aim for 100%. Well, a few intruders slip into the list. Take t.co, the Twitter URL shortener. While this service is extensively used across the Web, it is not a website by itself. Visit its homepage: this is just a few lines of content and a minimalist design.

All in all, there is no surprise at this level.

The favicon is stuck in the desktop era

Favicon starts with the favicon.ico file introduced by Internet Explorer. It is well-known and creating it is part of any web projects nowadays. In this respect, top websites are doing well: 98% of them exhibit a favicon for desktop browsers. This figure makes sense: favicon.ico was here before the iPhone, so even the outdated sites benefit from this old practice.

What about the mobile platforms?

The support rate for the iOS touch icon is 60%. If you bookmark or add to your home screen one of the others 40% websites, you will end up with a default icon or a “thumbnail” icon. Definitely not something you want on your Retina screen.

Would you rather click Ask.com or the calculator?
Would you rather click Ask.com or the calculator?
Buy a book? Nah, let's spend some time on Reddit.
Buy a book? Nah, let’s spend some time on Reddit.

Unlike iOS, Chrome for Android has no equivalent for the Touch icon, except two 196×196 and 128×128 pictures with very little audience for the moment. It sticks to the iOS rules to get the job done, except that is does not handle as many cases. As a consequence, Android score is mechanically lower. The score found here is 29%, which is quite low. However, this figure should be considered carefully, see the methodology below. Suffice it to says that Android support is lower than iOS.

Paypal did the job for iOS, but not for Android
Paypal did the job for iOS, but not for Android

The last platform is Windows 8. As a challenger compared to iOS and Android, it cannot pretend to a huge support. No surprise, only 14% of the top websites define Windows 8 tiles. When Windows 8 cannot find a dedicated picture, it arranges something with the classic favicon, which is way too small to produce something good.

 

WordPress defines a tile picture. Not Blogger.
WordPress defines a tile picture. Not Blogger.

Let’s sum up:

Mobile support (responsive site or dedicated mobile site)88%
Desktop favicon support98%
iOS TOuch Icon support60%
Android icon support29%
Windows 8 tile support14%

Conclusion

These results are surprising and disappointing.

Sure, there are explanations. For example, add Google.com to the home screen of your Android device. Oops! Google has no icon for its own mobile platform! Is it so important? After all, Google is so integrated to Android that it makes this use case irrelevant. Better save a few bytes in the most visited web page in the world. Some sites, like Instagram, redirect you to their native apps, making their mobile websites almost useless. No need for a mobile favicon when the visitors are invited… to never come back.

Yet, why a site like Amazon has no Touch Icon? Can the existence of an iOS app justify this? As a website, Amazon.com is doing well on an iPad or an Android phone. A clear sign that Amazon does not bet everything on its native apps. A dedicated favicon would be the natural conclusion of the investment in web design: a pleasant experience for the loyal customers who want to see a big “A” each time they turn on their device.

Maybe the big editors ignore the mobile favicon because it costs more than it’s worth. After all, in many cases, supporting mobile devices adds a few bytes to the HTML code and leads to additional picture downloads. The browsing experience may suffer because of this. But it doesn’t seem to be the way the major players think. Let’s consider the Touch Icon. 23% of the studied web sites have a apple-touch-icon.png in their root directory. This file is supposed to be a 152×152 picture. Why 152×152? Because this is the resolution needed by the iOS device with the highest screen density, a Retina iPad with iOS 7. Only two websites do this. You guessed right, one of them is Apple.com. The resolution of the pictures of the other sites is beyond the scope of this post. But you will sometimes find 57×57 images, often documented here and there on the Web as the reference for the iOS Touch Icon. Unfortunately, this information is outdated: 57×57 is the size for non-Retina iPhone running iOS 6 or prior.

In the end, top websites give the feeling to neglect their favicon. In many cases, the absence of support for iOS just doesn’t make sense. The outdated picture resolution reinforces this hypothesis.

On mobile platforms, pictures and buttons are preferred to text and typing. The favicon, used as bookmark or home screen icon, is more exposed than on desktop, where the tiny picture is mostly here to spot a browser’s tab in a blink. It is a bit awkward to note that so many major websites don’t pay enough attention to this aspect, whereas so many time and money are spent in the sites themselves.

What should top websites do? They should review AppleMicrosoft and (optionally) Google specifications. These three pages contain everything you need to know to support most iOS, Android and Windows 8 devices and browsers. Keep in mind that Android follows iOS rules, this is why Google’s recommendations can be skipped. Then, top players should investigate their requirements in term of support. Declare all iOS icons for a better support or declare only the biggest one to shorten the HTML code? Oh, they could also use a great favicon generator to fix all their issues at once. A little bit of self-promotion never hurts.

Methodology

To perform this study, I hacked the favicon checker to iterate over the top 100 list and dump the data I needed. If you tried the checker before, you noticed that it checks a lot of things and it is not very forgiving. Unless you generated your code with RealFaviconGenerator, you will probably get a warning or two. Obviously, I was not so inflexible for this study. For example, if a site had just apple-touch-icon.png in its root directory, the test was positive. Even if Apple recommends a lot of other pictures.

The checker looks for some known pictures and/or code to decide if a kind of icon is supported. For example, if apple-touch-icon.png is present, the test passes. Else, if apple-touch-icon-precomposed.png is present, the test also passes. Etc. If no rule succeeds, the test fails. This means that, if the checker lacks a rule, if might wrongly fail. In other words, although the checker should not generate wrong positives (“site abc.com support Windows 8 tiles”… whereas abc.com does not support Windows 8 tiles), it can generate wrong negatives (“site abc.com does not support iOS touch icon”… whereas abc.com supports iOS touch icon).

In order to mitigate the risk of massive error, I manually cross-checked some results I found suspicious. Apparently the checker did quite well.

Android Chrome adds some complexity to this guess-work. It mainly re-uses the iOS touch icon but it often fails at looking for conventional pictures (ie. trying to get /apple-touch-icon.png even when it is not declared in the HTML page). But not always. Therefore, it is hard to predict if a favicon will do well on Android or not. Android Chrome also looks for PNG favicon. Because these pictures are still rarely used, we can state this simple rule: Android Chrome support < iOS Safari support. For these reasons, the Android figure above should be considered “okay” but not very reliable.

Last but not least: out of the 100 top websites, only 89 were actually studied. Some “top sites” are actually “top domains”, such as akamaihd.net. Accessing http://akamaihd.net/ just doesn’t work. In addition, some of the homepages made the HTML parser failed, like Dailymail.co.uk. Such site was simply excluded. So when this post announces that 40% of the sites have no mobile favicon, it means 35 sites out of 89.