264 points by vzhou842 8 days ago
Unfortunately the reference to the actual minifier, SVGO (https://github.com/svg/svgo), is buried in the middle of the post.
SVGO is an incredible project driven by a single volunteer in spare time (originally Kir Belevich https://twitter.com/deepsweet , and for the last few years Lev Solntsev https://twitter.com/ruGreLI). It's the type of project you probably use regularly behind another tool like Sketch or SVGOMG (https://jakearchibald.github.io/svgomg/). They accept donations (https://www.paypal.me/deepsweet) and you should definitely consider contributing if you find it useful.
Another svg optimizer (link contains comparison to svgo): https://github.com/RazrFalcon/svgcleaner
Pro tip for people with SVGs that have an image inside. You can pull that base64 image out of the svg, convert to image, compress it (tinypng, pngcrush etc) then convert back to base64 and out back in the SVG. We automated this with a slackbot but it’s often overlooked with all SVG optimization scripts.
Somewhat related: a while ago realized that the SVGs that draw.io generates (exports) have an interesting property. If you include an SVG inside such a drawing and you export it to SVG, that SVG is actually embedded as base64. This is unfortunately not supported by some programs (e.g., Microsoft Office). A quick fix is to actually expand the SVG into the big SVG. I wrote a quick tool for this that does this for you (I should consider open sourcing this maybe).
Anyway, I bet doing this and then minimizing the SVG could result in more savings?
Interesting. Care to share one of these images? I’d love to check it out.
I use it every time I'm using SVG for the web.
My only problem is that when pasting SVG directly from Illustrator it finds an invalid character at the end and refuses to use it. I have to paste the SVG into an editor, remove some invisible character at the end, and then paste to svgomg.
I reported this and offered to PR but never got an answer: https://github.com/jakearchibald/svgomg/issues/145
My protocol for dealing with no-PR's has been (especially in this case where you can just host the tool on github.io)
1. Just fork it and use your own version until the PR is accepted
2. Directly email the maintainer if you think your PR is worthwhile
3. If you have a bunch of PR's that are pretty useful to others, consider making it a real distributable (and not just a personal) fork if the maintainer still doesn't respond
Is it legal to have a NUL byte in an SVG? Might want to also contact Illustrator as well if it's not.
I doubt it, but I also doubt Adobe would pay attention...
It might be disallowed by XML already.
Why ask, just write the PR and see if they'll take it.
I've done that in the past and some PRs are still there collecting dust.
Their loss. On the flip side, if anyone ever comes back or takes over the project, they'll have PR's ready to go.
Hardly a day goes by that I don't use svgomg. If you need a one-off SVG minified, this is the tool for you.
I used this tool just the other day; very comprehensive, and very fantastic.
Be aware that svgo is a lossy operation. One thing it does is round the digits in numbers to a certain precision. When I last worked with it, it also did some things I did not like such as mess with viewboxes. I found that it messed up the sizing of SVGs embedded with <img> tags in certain browsers. It has been a while since I used it though.
I was curious about this. It looks like it has numerous individual plugins that provide the actual functionality. Do you know if some of the plugins are lossless, while others lossy? It may be possible to pick and choose the lossless ones.
It has a boat load of plugins. I looked at a few and after puzzling over which were default enabled and default disabled I decided I didn't want to use a program with unsafe defaults.
I absolutely love ImageOptim for manually optimizing SVGs, JPGs, PNGs, and GIFs. It's a simple GUI that wraps some excellent compression libs (including SVGO), which I find much faster to use for one-off image optimization than the CLI. It does a clever job of trying multiple compressors and picking whichever one creates the smallest file for each image. You can run it in a lossless or lossy mode, and tune various compression options. It's free (gratis), OSS, and runs on Mac, Windows, and Linux.
I make it a habit of running images through ImageOptim before sending them via Messages or Slack, as a courtesy for people on slow connections.
https://imageoptim.com (by HN user pornel)
Bonus round: When it comes to optimizing SVGs, nothing beats hand-editing the SVG source. I've done a fair bit of that optimizing the artwork for https://www.lunchboxsessions.com and it's been totally worth the effort. Taking a detailed 10k SVG down to 500 bytes with no perceptual difference is a huge win when you have hundreds of those SVGs on the page and you serve people who still use dialup.
ImageOptim also strips out a lot of (all?) the image metadata which is often a place people leak data unintentionally.
I've been using ImageOptim for a while now. Saves me so much trouble when I get junky SVG's as well as exporting my own.
Yet another alternative that has benchmarks against svgo and scour is svgcleaner: https://github.com/RazrFalcon/svgcleaner.
It is written in Rust and has been my go to for SVG optimization due to its speed.
Would be more compelling with the sizes of that neutral net in the different passes. As it is, this is like comparing two algorithms on a small dataset without accounting for growth in data on behavior.
There's also https://vecta.io/nano, which I've found to sometimes give better results than SVGOMG
A related (but less automated) advantage is if you're building react components that render SVG, this minified output will be much shorter and hopefully easier to reach through and figure out where you want to insert your own colours/child components.
Illustrator has an export SVG function (as opposed to just saving as an SVG, which I was doing and then using a minifer for a long time) that will minify it for you.
That said, it currently appears unable to reopen those minified SVGs (go figure).
FYI you can copy any vector in Illustrator and when you paste it somewhere it is actually SVG code.
Which is why I fairly frequently end up with amusingly-long pastes of raw svg markup being posted into email fields... .
I love seeing efforts to make software more efficient, so I enjoyed this post, but at the same time your own blog has a 434kb 512x512 png favicon. I'm not at all a web developer and so I don't know if there is a technical reason for that, but it seems a bit absurd in juxtaposition to your goal to reduce a 2kb svg to 100 bytes.
I've seen this on a number of other blogs, which seem to focus on minimal design. Can anyone explain what the need is for these enormous favicons?
512x512 is specifically for browser loading splash screens when the site is added as a launcher on a phone. It's required for Chrome to have an "Add to Home Screen" prompt and requires a 192x192 and 512x512 icon. I imagine most minimalist sites you browse are expecting people to add the website as an icon/app launcher on their phones. Not sure how common an experience that is for users, I sure as hell wouldn't but maybe other people might.
Was intended for web apps I believe but is usable by sites that aren't web apps.
Read more here: https://developers.google.com/web/fundamentals/web-app-manif...
Don’t forget to also use GZip or another transport compression on them as well. Text tends to compress well, and can net additional benefits on top of minification.
I ran the examples in the article through gzip. The long one is 789 bytes compressed, and the short one is 266 bytes compressed.
The reality is that while text compresses well, gzip doesn't know what data is relevant. It keeps it all. So for this example, the original long svg contains things like inkscape configuration (the window height, the current layer name, etc.) while the shorter one omits all that. So not emitting that information will always be more efficient than emitting that information and compressing it.
He means in addition to minifying the SVG, you should compress it for transport (which is likely happening already depending on how you serve it)
Of course, but even the minified versions should net some benefits with compression. Most SVGs probably fit in 7-bit ASCII, which on itself should lead to improvements via (Huffman?) compression. Importantly, as was pointed out, I say “also” not “instead.”
This is neat, but I wonder: are there path optimization tools? For instance: flattening, removal of entirely occluded paths, and combination of paths.
Not exactly that but comes close (and perhaps could be used for that if you rasterized the input vector): https://vectormagic.com/
I haven’t used their desktop edition but on the web it worked really well for cleaning up images. I imagine similar algorithms could be applied directly to SVGs to simplify them.
Worth noting is the manual step in their workflow - I think there’s only so much you can do before you need a human to decide whether any important meaning has been lost.
SVG optimization seems like a good choice for delivery, but oftentimes I find myself using SVG in a development context. Having human readable, annotated SVG files is useful when looking through previous revisions. In terms of compressed filesize: there's no serious practical gain.
It's analogous to JS minification -- Clients on the production site should get a bare minimum and the VCS should have the readable form with all the digits of precision and whatever comments, non-functional groups acting as layers in the editing software, etc.
Aren't garden variety SVG files ~75% smaller than other lossless compression methods (e.g. PNG). As such, is minification necessary or a best practice?
In short, is this an intellectual exercise or something that should be in an SVG producer's toolkit?
Raster formats have a worst-case file size that is linked to their visual size.
SVG has no such limit; for a given visual size it can have arbitrarily many objects. If you’re not careful, you can quite easily end up with multi-megabytes SVGs where a PNG would have been <100KB.
SVG is therefore a dangerous format to use if you don’t know what you’re doing, because its performance characteristics are quite unlike raster formats.
This is an excellent post!
It does a fantastic job of defining the problem and walking the reader through the process of solving it. The language is so succinct and clear that even if the reader had never heard of SVGs before, they would understand this post.
Well done, Victor!
Thanks, appreciate that!
Nice work, svgo is awesome. I'd love to adapt this to a non-gatsby post-build tool that also runs image compression on any jpg/png files.
Big advantage to running post-build besides convenience is that sometimes Inkscape doesn't render optimized svg's correctly. So I have to keep originals elsewhere.
We provide hosted service for optimising SVG. If anyone is interested for image optimisation as a service, take a look here: https://www.gumlet.com
Just another shameless plug here. I'm developing an SVG editor which focuses on generating clean and readable SVG files (e.g. no redundant namespaces, metadata or deeply nested structures).
You can try it online on https://boxy-svg.com, there are also desktop apps for macOS, Windows, Linux and Chrome OS.
There is a typo on this page:
I think you mean "images" not "imags"
Also the link to the docs doesn't work.
For inlining, you can even go further and remove that xmlns attribute to end with something around 50 bytes. I'd recommend keeping the viewbox though, since it's required for proper resizing.
Most of the time, automated optimizers for SVG are relatively bad, though, and rewriting the SVG by hand can get a 2KiB svgo result down to 200 bytes or less.
Don't SVGs have to have an XML declaration? I think that magic will fail to identify the content type of an SVG if it's not a valid XML file.
No, the declaration is optional in many cases.
Very neat! Now just need to figure out how to plug svgo into my jsx rendering pipeline, at build time.
I've also noticed the cruft in Inkscape SVG's.
The TXR logo SVG is cleaned with a TXR script:
The original looks like this:
The cleaned one is not in the repo. It's obtained by "txr cleansvg.txr txr.svg":
<?xml version="1.0" encoding="UTF-8" standalone="no"?>
<!-- TXR Logo
Copyright ... -->
viewBox="0 0 277.4375 297.4375"
<path d="m 476.96875,651.28125 c -7.29171,7e-5 -14.03911,1.37247 -20.21875,4.125 -6.17972,2.75267 -11.36983,6.39069 -15.5625,10.875 -4.33856,4.5938 -7.66668,9.81776 -10,15.6875 -2.33334,5.86983 -3.50001,11.92191 -3.5,18.15625 -10e-6,7.29167 1.28645,13.62241 3.875,19 2.58853,5.3776 6.02082,9.89063 10.25,13.5 4.26559,3.60937 9.05205,6.29426 14.375,8.0625 5.32287,1.76822 10.74475,2.65623 16.25,2.65625 6.1614,-2e-5 11.73692,-0.987 16.75,-2.9375 5.01294,-1.95054 9.35409,-4.51564 13,-7.6875 l -1.71875,-2.78125 c -3.60945,2.98957 -7.66675,5.27603 -12.1875,6.84375 -4.5209,1.5677 -9.67194,2.34374 -15.46875,2.34375 -5.14067,-10e-6 -9.93234,-0.82814 -14.34375,-2.46875 -4.41149,-1.64064 -8.224,-4.09376 -11.46875,-7.375 -3.31773,-3.39062 -5.89586,-7.61198 -7.71875,-12.625 -1.82294,-5.013 -2.71877,-10.80728 -2.71875,-17.40625 -2e-5,-6.85413 1.16665,-13.00517 3.5,-18.4375 2.33331,-5.43223 5.36976,-9.94787 9.125,-13.59375 3.86454,-3.75515 8.14059,-6.59889 12.84375,-8.53125 4.70307,-1.93222 9.54162,-2.90618 14.5,-2.90625 5.21348,7e-5 9.82806,0.80215 13.875,2.40625 4.04679,1.60423 7.40096,3.79174 10.0625,6.5625 2.77074,2.80735 4.83846,6.02349 6.1875,9.6875 1.34887,3.6641 2.03116,7.66672 2.03125,11.96875 -0.1377,8.31795 -2.4429,16.899 -11.71184,23.81796 -4.24484,3.16863 -9.29313,4.01343 -14.18104,4.49201 -2.5623,0.25088 -7.11725,0.30793 -9.74156,0.25794 -1.81112,-0.0345 -3.02482,-0.27866 -3.74043,-0.81795 -0.71564,-0.53929 -1.06251,-1.22605 -1.0625,-2.0625 -1e-5,-0.90247 0.57819,-3.28219 1.71875,-7.15625 l 7.90625,-28.15625 3.59375,0 c 3.15321,4e-5 5.11461,0.28996 5.875,0.90625 1.00632,0.81447 1.49996,2.04944 1.5,3.65625 -4e-5,1.36476 -0.20014,2.91736 -0.625,4.65625 l 2.78622,0 4.09375,-13.875 -37.14277,0 -3.53125,13.87504 2.29405,0 c 0.62617,-2.3112 1.42225,-4.1203 2.40625,-5.375 0.98398,-1.25462 2.27599,-2.21462 3.875,-2.875 1.59897,-0.66031 3.69596,-0.96871 6.3125,-0.96875 l 2.71875,0 -8.34375,29.125 c -0.93928,3.16968 -1.69323,5.23924 -2.21875,6.21875 -0.52556,0.97952 -1.10147,1.65706 -1.75,2.03125 -0.93928,0.55029 -2.28692,0.8125 -4.03125,0.8125 l -0.875,0 -0.96875,3.46875 19.90942,0.12627 c 4.65825,0.0295 14.28759,-0.73275 17.58443,-1.82102 6.80584,-2.24657 10.97411,-5.63785 13.61133,-8.47036 1.87071,-2.00924 7.5173,-9.18725 7.82441,-20.78032 0.12261,-4.62853 0.078,-9.57025 -1.39847,-13.83592 -1.47666,-4.26557 -3.67197,-8.03119 -6.625,-11.3125 -2.95321,-3.31763 -6.74488,-6.00774 -11.375,-8.03125 -4.63028,-2.02336 -10.09903,-3.03118 -16.40625,-3.03125 z"/>
Are you talking about inkscape SVG which also save many details for inkscape or raw/small svg saved by inkscape ? Those if needed could use a patch if theyre too big and there are low hanging fruits
Why? Why minify everything? It all gets gzipped at the HTTP layer anyway! I truly cannot understand this need to minify everything (JS included)
I started by downvoting your comment, and I realized I was choosing the cowardly way of reacting. I am saying this just to illustrate my own shallow reading of your comment at first and how I was avoiding actually being helpful.
The short answer is that 99% of the time no purpose is being served by that cruft except to be bloat. Minifiers remove that cruft and optimize for the intended audience, which is machines.
The longer answer is that in optimizing for machines, you're also optimizing for humans in the long run. Anyone stuck with a 2G connection, dial-up, or Bluetooth tethered device, or a new ultra low powered device on a different CPU architecture, will appreciate not having to operate solely in the world of gigabit+ connections, 8GB mobile devices, or 8 core hyperfast CPUs. Realize that such use cases still exist, and for very legitimate, not truly edge cases.
Think of pages with hundreds or thousands of vector images. Not everyone can afford to run a Blink-powered browser or have CPU cycles for days. Every little bit helps.
Edit to fix typo and add further clarity.
Fun fact! SVG is one of those things that is often easy to minify down below your nginx/apache/cdn minimum gzip length.
If your SVG files are < ~1kb, it's usually faster to just not gzip them.
If they're < ~150b, gzipping will probably make both slower _and_ bigger.
Because when the browser un-gzips it, it takes a lot more memory than the minified version. And everyone seems to complain about browser bloat.
It doesn't always get gzipped. That is highly dependent on the server.
Minification usually reduces code size, and some minifiers will also implement various optimizations.
Did you read the article? SVGs are super crufty and extremely easy to optimize.
In general I'd recommend the opposite: don't minify anything unless necessary, prettify it instead. Every plaintext-based format file should be optimized for human readability unless you actually need to make it hard to read (e.g. you are developing a proprietary product not meant to be open-source) or your bandwidth is limited so severely that every single byte matters.
Nevertheless the kind of minification demonstrated in the example (removing Inkscape bloat) feels really great and actually makes the file more human-readable. This reminds me of HTML files generated by MS Word and other WYSIWG editors which included tons of bloat code that actually harmed rendering (needless to say human readability).
I completely disagree. It's trivial enough to run any minified file through a prettifier.
Even in the first world, on a spotty mobile connection every single byte matters, and some of your users are almost always going to be in that situation some of the time.
Think of minifying as "compiling" for the web. A good dev makes their site performant, and doesn't care about generating readable source code for end-users which 99.9% of them don't care about.
If you want to expose the site's source code to the world in a readable format, then just publish it on GitHub as well, which is the audience that might actually care.
Readable plaintext files are for developers on your team. Not for end users of your product.
I'm not sure I agree in the usual case. This is sort of what sourcemaps are for. Provide a link in a comment in the resource source to a prettified or at least non-minified version at a separate URL. The typical user won't need to download the extra bytes, but you can still access the original if you want.