General note: This site’s appearance, configuration, hosting, and other basic considerations will change over time. As a result, certain content on this page could be at variance with what you’re currently seeing on the site, but the two were consistent when this post originally appeared.
Who needs heroes? Not this site, I’ve decided.
Of course, I’m referring to a hero image — a massive photo or other graphic element that constitutes much if not all of the “above-the-fold” content on the typical website these days.
If you’re among my few (but beloved, I assure you) regular readers and this is your latest visit since before I first issued this post, you’re probably wondering, “Hey, where’s the usual great, big, honkin’ stock photo he always puts up here underneath the post title and all that stuff?” For example, it’d usually look something like this (taken from a pre-today incarnation of this post):
Not today, though. In fact, you might want to open another browser tab or window and check around the rest of the site, except for the home page. You’ll see that this isn’t an aberration.
It’s a new look.
And it’s here because of stuff I learned — and a decision I made.
As part of the conversion of this site to an Eleventy/webpack combo, I determined to do a better job where providing responsive images is concerned. In the beginning, I accomplished this through use of responsive-loader, but that gave me only different sizes of JPGs and PNGs to serve visitors. There was no provision for the Google-created WebP format which, although it’s been around for nearly a decade and is much more efficient than either JPG or PNG in most cases, has only recently gained support among a sufficient number of non-Chrome browsers to make it worth providing, in my opinion.1
I decided to do something about that. So, since responsive-loader doesn’t support WebP as an output format, I ended up wiring the npm package imagemin-webp-webpack-plugin, among others, into the site’s webpack config.
A weekend’s worth of false starts later, I had the sequence just right and, on build, the setup converted each JPG or PNG image in
/src/images/ to up to four different sizes in the image’s original format plus WebP. Then, finally, I retooled an Eleventy shortcode I’d written so that it would produce, rather than just an
<img> with a
data-srcset (to conform to the needs of the lazysizes library), a
<picture> which would give each browser its choice of sizes and formats. Browsers which couldn’t handle WebP files would fall back to each file’s original format, whether JPG or PNG.
It seemed simple enough. But I gave you a hint above as to why it came a-cropper — “Webp . . . is much more efficient than either JPG or PNG in most cases” (emphasis added this time around).
Obviously, “in most cases” != “all the time”; and therein lay the problem.
When I saw that some of the files actually were larger in WebP format — especially when I ran the site’s
testbuild script (see the site repo’s README file for more details) — I wondered how that could be.
It didn’t take long to find out.
I won’t get into all the details, but the skinny on why the WebP images weren’t always skinnier than their counterparts is that, simply put, you can’t run an already-lossy JPG or PNG file through the conversion to an also-lossy WebP file and not get a fatter file. This was especially true for images with a lot of specular highlights2; sometimes they really ballooned, because the conversion process went ape on those tiny details, adding many more pixels.
More maddeningly, it was just hit-or-miss enough that I couldn’t reliably configure all the settings to prevent it. So, rather than “rewarding” WebP-compatible browsers with smaller-size graphics, in many cases I would be “punishing” them with bigger ones. Definitely not okay.
While figuring out what to do, I tried Google’s versatile
cwebp (WebP encoder) tool. Although it, understandably, produced better results than my webpack config, I soon saw that the only reliable setup which always produced a WebP file that was smaller than the original JPG or PNG was when I had the truly original JPG or PNG. That is, I couldn’t use a processed version of the original, or I’d almost always get a larger WebP file.
Only problem with that was, in many cases, I had no unprocessed version of the JPG or PNG file.
This was particularly true for nearly all of the stock images. For each, I had—in the interests of keeping file sizes lower, ironically enough—always downloaded a smaller image size than the maximum that the vendor offered. Now, it appeared, I’d have to go back and re-download each stock image at its maximum size, then use
cwebp to resize it.
One. At. A. Time.
— and, often, with different settings for each. (By the way, I had no way of knowing that even the largest image offered by the vendor hadn’t already been processed somehow. After all, these were JPG or PNG files, not raw files.)
It got worse. Often, I also had to convert the WebP version of a stock image at only twenty or thirty percent quality to achieve a smaller file size. To the credit of the WebP format, that ugly process didn’t result in the horrible, artifact-laden mess you’d get if you reduced other formats’ files to so low a quality. Still, it was visibly not as clear as the JPG or PNG, especially when I compared the two on a high-resolution display.3
So I had to decide what to do.
I first figured I had two choices:
Keep the process and images I already had, accepting that I’d sometimes be delivering “fatter” WebP files.
Go back and re-do nearly all the stock images, converting each one separately (and usually at low quality) to WebP.
However, in the early hours of yesterday morning, I revisited something I’d played with only in separate branches of my repo but never yet committed to the remotes: doing without hero images altogether.
I’d thought about it before because, frankly, I had tired of always having to find some stock photo to illustrate each and every post, particularly since I had to stick to free stock photos — yeah, right, the same free stock photos everyone else uses. And it’s not as if I could always find an image that both fit the “hole” I had to give it and had at least some relation to the subject of the post.
Besides, as I kept telling myself: “Dude, what few people you have visiting here from time to time don’t come here looking for the Picture of the Day. They come here looking for information. For words. If you want to stick images inside certain posts to illustrate points, that’s a totally different matter.4 But this ain’t Life Magazine.”
As a result, I created a new branch, retooled the site accordingly, and today published the redesigned, hero-image-less site you now see.
It’s ironic that one aspect of image processing has to do with reducing so-called noise; because the “noise” of having the hero images (even the cute ones) became an end unto itself. In the end, that noise — and the growing technical debt, and growing expenditure of my already limited time, which now was going to come with it — had become unacceptable.
So, it’s gone.
Will I miss the hero images? Perhaps. Maybe some of you will, too. And I did think they added a nice feel to things at times. Nonetheless, I hope this decision, like others5 I’ve made over the brief history of this site, will ensure that each visitor’s browsing experience will be the best possible.
After all, you folks are the real heroes. You keep coming back. If that’s not heroic, I don’t know what is.
Of course, between Google Chrome’s already massive market share and the adoption by the Windows-default Microsoft Edge of the Chromium engine, that provided even more incentive to serve WebP-format images. ↩︎
Examples of images that were especially problematic for this reason: road surfaces (because little bits of light gravel and other shiny crud appeared); and sunlit beach sand (because you could see individual sand granules, pebbles, and the like). ↩︎
A WebP converted at such low quality looked passable on the old low-resolution monitor at work, but on the double- or triple-quality displays commonly used by today’s smartphones, much less my iMac’s 5K display — not so much. ↩︎
As of the day this was first posted, those posts with included images are: “The client is too smart for you”; “Readable web type, pretty please”; “A stacked deck”; “Blox sux”; “iA for IO?”; “Why I left Ulysses”; “Ahoy, ‘Mate”; “Independence”; and “Roger, Copy.” ↩︎
Commenting by giscus.