Do you have a sense of which aspects are the most resource-intensive? Naively I would guess it's the backdrop-filter.

But missing out on the refractive aspect of glass takes away the strong visual separation of layers that IMO is Liquid Glass's biggest contribution.

Material has these wonderful designer resources showing how the app ought to be built of consistent moving layers, shown in 3D from the side. It's clear that there's these layers. But once you go 2d, put it all together, its incredibly hard for me to find all the features. The number of times that there's an action button in some corner that folks don't see massive. Motion sort of helps highlight the chrome vs content, but there's just so little thats visually defining.

Liquid Glass's biggest strength IMO is the edge distortion. That when the content moves, there's this non-linear motion that the human visual sense immediately picks up on. Static screenshots don't always look great, but in motion there's so much more distinction of elements than anything else I've seen.

And that key refractive element, that takes such huge advantage of human's motion sensing, is missing here.

I'd seen one other great web web demo, but am struggling to find it right now. I did find this one, which I don't think looks as good, and the performance is bad for me here on mobile. But it nicely breaks down steps and shows how complex a task this can be. Getting these edge distortions in is hard, and relying on SVG filters to do it is a brutal path to hack together. https://atlaspuplabs.com/blog/liquid-glass-but-in-css#breaki...

There is a collection of attempts (CodePen Spark) at Liquid Glass that I just found. Second link is one I thought did a pretty nice job, via very specifically pre defined / hard coded SVG. https://codepen.io/spark/453 https://codepen.io/lucasromerodb/pen/vEOWpYM

I think the most immediate difference is how light has no interaction with the bevels. I also expect some light to shine back into the glass and affect the lighting and coloring. It's not enough to just throw a blur in there.

Also, glass can have its own shadow with some interesting caustics (not sure even Apple does this). I see the shadow here, and it feels like a simple drop shadow. It makes the box feel like a flat card more than a 3d physical object which I think is part of the new trend.

Either way, This will not be easy to emulate with just css, it's probably more suitable to be a shader running in a gpu.

others

- https://ui.glass/generator/ Get started with this free CSS generator based on the glassmorphism design specifications to quickly design and customize the style properties.

- frosted glass sticky header https://www.joshwcomeau.com/css/backdrop-filter/

- glassy glassmorphism [codepen](https://codepen.io/a-trost/pen/dypQzwq), [in context](https://codepen.io/TurkAysenur/pen/ZEpxeYm)

  ```css
  .blur-and-rotate {
    border-radius: 20px;
    backdrop-filter: blur(20px) hue-rotate(120deg);
    -webkit-backdrop-filter: hue-rotate(120deg);
  }
  ```

Not a lot of web sites have made my Macbook M3 show signs of stress ;)

Thus, the odds you're doing glass-morphism via a Gaussian blur and drop shadow in CSS, are exactly 0. They are assuredly at abstraction levels far below that.

(disclaimer: worked on Pixel, did the color extraction + color system stuff for Material You)

  filter: blur(16px);

It shows, this is awesome, especially that rice paper effect!

Especially with respect to the abuse of box shadow.

Putting aside for the moment that I personally find Apple’s iOS 26 design objectionable[1], I don’t understand why `backdrop-filter: blur` is the focus of recent implementations jumping on the Liquid Glass hype train.

Using background blur to create UI layer separation (often in combination with darkening/saturation or other contrast enhancements) has been around for over a decade on iOS and almost as long on the web. So what’s new here?

Adding to my confusion, I’m a bit surprised folks here think this is so challenging in CSS. A few commenters have pointed to great implementations elsewhere, but I think they’re underselling them.

Plainly: the solution that involves `backdrop-filter: filter(#filter)` where `#filter` references an inline SVG embedded with `<feImage/>` and `<feDisplacementMap/>` works very convincingly well.

Check out this demo for example:

https://codepen.io/Mikhail-Bespalov/pen/MYwrMNy

This implementation choose to hard-code `<feImage href="data:image/png;base64,..."/>`. But if glass3d.dev chose to implement 3d highlights in a similar fashion (and wanted to support e.g. a dynamic `border-radius`) this image could easily be rendered in the browser runtime in canvas, and dumped into the CSS using `toDataURL()`. Similarly, a component library with a CSS pre-processor could generate these for any static shape at build-time.

The closest thing I’ve seen to this is:

https://ruri.design/glass

These implementations are out there.

Coming back to the why/should for a second though.

[1] In its current realization, Liquid Glass cannot effectively replace what blur accomplishes. Because Liquid Glass layers are more transparent, the contrast issues are real, and the spectral highlights distract the eye as much as (or more than) they help make up for that lack of contrast. It draws attention to the eye where blur would relax it. It’s a UI meant for an XR system (where it arguably solves a real problem, much like a HUD does in a video game) hacked into devices where it makes no sense, all in the name of a “unified OS” design language.

If any aspect of Liquid Glass is successful it will be when it’s used sparingly to delight in places that are low stakes and where other affordances are present (like a small circular button floating in the corner of the screen with hardware concentricity). A circle shape’s refractions would be smaller, softer, more symmetrical, and therefore arguably less noisy/distracting—in a way resembling a softer blurry background.

Which brings me full circle back to two wrongs.

This website doesn’t do anything new, but that’s why it’s good. Because the truth is, Apple failed to deliver a Siri-based LLM on a schedule it announced and is now trying to distract us with some shiny new thing. Damn, it worked.