Liquid Glass Web logo

Liquid Glass Web

Community
gentpan
liquid-glass-web

Build refractive "liquid glass" UI for the web (Apple Liquid Glass style) that renders the same in Chrome, Safari/iOS and Firefox — SVG feDisplacementMap applied through CSS filter to the content itself, runtime displacement maps, chromatic fringe and rim highlights, press/squish springs, WebKit-specific fixes, WebGL for video. Ships the liquidglass.dev web components (switch, slider, segmented, button, panel, navbar, menu bar, toast, dock, lens), the zero-dependency engine, a reference page and a three-engine screenshot script. Use whenever a web or front-end task (HTML/CSS/React/Vue/Svelte) asks for liquid glass, glass refraction, a glass lens or magnifier, glassmorphism that should actually bend light, or Apple-style glass switches, sliders, segmented controls, tab indicators, cards, navbars, menus, toasts or docks — and when a backdrop-filter glass effect does not show up in Safari or Firefox. Not for native SwiftUI/UIKit apps (use the system glassEffect there).

Overview

Publishergentpan
Repositoryliquidglass
Skill nameliquid-glass-web
Stars
16
Forks
0
Bundled files
13
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 13 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by gentpan on GitHub. Read the source before you install it.

Installation

Install the Liquid Glass Web AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/gentpan/liquidglass/tree/main/skill/liquid-glass-web
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/gentpan/liquidglass.git /tmp/liquidglass
mkdir -p .claude/skills
cp -r /tmp/liquidglass/skill/liquid-glass-web .claude/skills/liquid-glass-web
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Liquid Glass Web in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Liquid Glass Web on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Liquid Glass Web is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Liquid glass for the web

Goal: glass that really bends light, like Apple's Liquid Glass, rather than a blur — and looks the same in Chromium, WebKit (Safari and every iOS browser) and Gecko (Firefox). The approach follows Aave Labs' Building Glass for the Web; the bundled code is an independent implementation, verified screenshot-by-screenshot in all three engines. Live site and docs: https://liquidglass.dev.

How it works

  1. Refraction goes on filter: url(#id), not backdrop-filter. backdrop-filter: url() only renders in Chromium. A plain filter referencing an SVG filter works everywhere, but it bends the element's own pixels — so whatever should be bent must be inside the glass element: the real content, or a decorative copy of it.
  2. The displacement map is generated at runtime from the lens shape. Red/green = horizontal/vertical bend, blue = rim highlight, alpha = lens coverage, neutral grey 128 outside the lens. Pixels outside the lens pass through untouched, so text stays selectable and links clickable.
  3. Moving is cheap, reshaping regenerates. Moving a lens only changes filter sub-region coordinates; the map is redrawn only when shape or optics change (cached, only a quarter computed).
  4. Surfaces without live DOM use WebGL with the same map. Safari never runs SVG filters on a playing <video>; canvas content has no DOM either.

Step 1 — pick the path

Fastest: the ready-made web components. If the project can load an ES module, use assets/components.js (copy it next to assets/liquid-glass.js, or load https://liquidglass.dev/lib/components.js). Elements: lg-switch, lg-slider, lg-segmented, lg-button, lg-panel, lg-navbar, lg-menubar + lg-menu, toast(), lg-dock, lg-lens. Usage, attributes and theming variables: references/components.md.

Custom glass: use the engine (createGlass) with one of these patterns:

SituationPatternHow
A lens moving over content you control (magnifier, hero, indicator on a card)A wrapPut the content inside the element; createGlass(el, { x, y, width, height, radius })
Switch, slider, segmented control, tab indicatorB overlay copyNative control stays underneath; an aria-hidden copy on top with clip: true
Glass card/panel over a known backdrop (image, illustration, CSS background)C backdrop copyAn aligned copy of the backdrop inside the card; fit: true
Glass over arbitrary scrolling DOM (fixed navbar)D backdropmode: 'backdrop': real refraction in Chromium, frosted blur elsewhere
<video>, canvas, WebGL scenesE WebGLcreateGlassRenderer(canvas, source)

D is the only case that cannot refract in all three engines — tell the user instead of passing a blur off as refraction. If the backdrop is actually known (a fixed background image), use C. Full code for each pattern plus React/Vue/Svelte wiring: references/patterns.md. A runnable reference page: assets/demo.html.

Step 2 — use the bundled engine

  • Copy assets/liquid-glass.js into the project (e.g. src/lib/); for pattern E also assets/glass-webgl.js. Both are dependency-free ES modules. Search for an existing createGlass first and reuse it.
  • Do not rewrite the filter from memory. The file encodes fixes that were found by measurement (filter units, per-engine scale, WebKit caching). A rewrite almost always renders as a blank element in Safari. Modify it if needed, then re-verify with Step 5.
js
import { createGlass, createSpring } from './lib/liquid-glass.js';

const glass = createGlass(surface, { x: 40, y: 40, width: 180, height: 120, radius: 60, refraction: 28 });
glass.update({ x: 120 });        // move: coordinates only, cheap
glass.update({ width: 200 });    // reshape: new map (cached)
glass.update({ hidden: true });  // remove the lens without destroying it
glass.destroy();                 // always on unmount

Options: x y width height radius (CSS px, relative to the element's border box), refraction (max rim bend in px), bezel (curved-rim width as a share of the half size), curvature, ior, invert, chroma, blur, specular, specularWidth, lightAngle, mapScale; plus fit (lens follows the element's size), clip (clip the element to the lens), mode: 'backdrop', fallback, sync, hidden. Meaning and ranges: references/optics.md.

In frameworks: create after mount, destroy() on unmount; drive per-frame changes (drag, springs) by calling glass.update() directly, not through component state (that re-renders every frame). Create on the client only.

Step 3 — make it read as liquid glass

  • Bend at the rim, flat in the middle. bezel sets the width of the curved band; the flat centre keeps content readable. Glass over text (segmented, slider) uses a small bezel and small refraction; strong bends are for decorative highlights. A thin bar (menu bar) reads best as a glass rod: bezel: 1, curvature: 2.5, so the backdrop lenses across the whole height.
  • Light chroma (0.1–0.25), only as a fringe at the rim.
  • Directional highlight + shadow. specular draws a rim that is brightest facing the light and dimly reflects on the far side (default light from the upper left); add a decorative shadow element (pointer-events: none). On photos, tint the glass: a white tint (8–20 %) under dark text, a dark tint (~30 %) under white text — contrast has to come from the tint, not luck.
  • Motion makes it liquid. Press → grow and squish (wider, shorter); drag → stretch with velocity; release → spring back (createSpring, which sub-steps so slow frames don't slow the animation). Switch and slider thumbs are solid white at rest and turn into glass while pressed, as on iOS.
  • The glass layer needs an opaque background (chroma sums channels in premultiplied space and over-counts alpha on transparent pixels), and room to grow: filter output is clipped to the element's border box.

Starting values (from assets/demo.html and liquidglass.dev, checked in all three engines):

Userefractionbezelcurvaturechromaspecularnotes
Lens over content / magnifier24–320.640.150.5180×120, radius 60
Switch thumb100.930.20.736×24 at rest → 54×36 pressed, specularWidth 1.5
Slider thumb60.930.150.7gentle: the fill must stay readable
Segmented indicator50.3540.20.6rim bends, labels stay legible
Menu bar (glass rod)2012.50.20.944 px tall capsule, specularWidth 2.5
Glass card over a photo24–260.3–0.3540.10.5tint + level-2 shadow, blur 2
Video controls (WebGL)140.830.250.8—

When the project has a design system, shadows, tints and radii come from its tokens; the table only covers optics.

Step 4 — rules that break an engine when ignored

The bundled engine already does all of this. Keep it when writing or changing filters:

  1. primitiveUnits="objectBoundingBox". WebKit renders any HTML element whose filter uses userSpaceOnUse primitives as completely blank.
  2. In bounding-box units, feDisplacementMap's scale is converted differently by each engine; compensate per engine (bendScale).
  3. color-interpolation-filters="sRGB", or 128 stops meaning "no bend" and everything shifts.
  4. WebKit caches filter output per filter id: new id on every update; decode the map before swapping it in; re-issue the id shortly after a swap.
  5. The <svg> holding the filters must not be display: none.
  6. Keep filtered surfaces small; never filter the whole page. WebKit caps a filter's input size and breaks large areas into tiles.
  7. <video> and canvas in Safari → WebGL.

Symptom → cause → fix, with the measurements: references/browser-quirks.md. Read it first whenever glass is blank, frozen, shifted or too strong on one axis in some browser.

Accessibility and fallbacks

  • Real controls stay native, focusable and labelled; copies and decoration are aria-hidden="true" with pointer-events: none; draw focus rings on the visible thumb (input:focus-visible ~ .thumb).
  • prefers-reduced-motion: createSpring jumps straight to the target; don't add squish on top.
  • prefers-reduced-transparency / prefers-contrast: more: stronger tint or a solid surface.
  • Text on glass: at least 4.5:1, guaranteed by the tint layer.
  • Without JavaScript the content renders normally — the filter is only attached once its map is ready.

Step 5 — verify before handing over

  1. Serve the page (the project's dev server, or python3 scripts/serve.py <dir>). Avoid browser-blocked ports (e.g. 4190, 6000, 6665–6669, 10080): WebKit silently refuses them and the page stays about:blank.
  2. On macOS run scripts/shoot.sh "<url>?hold=2500" <out-dir> → Chrome (2x), WebKit (2x — the system WebKit is Safari's engine) and Firefox screenshots. Crop and compare at full resolution; don't judge from thumbnails. Check: the glass is present in all three, rims really bend, text outside the lens is crisp, controls line up with their copies.
    • hold only works with serve.py: firefox --screenshot fires on the load event, before the glass exists, unless the load event is held back.
    • Firefox headless screenshots omit WebGL canvases; read pixels back in the page to check WebGL there.
  3. Check interactive states in real time with node scripts/chrome-shot.mjs <url> out.png --actions '[{"type":"move","x":300,"y":200}]' --eval "js". Chrome's --virtual-time-budget screenshots can freeze springs mid-animation and show states no user will ever see.
  4. Motion: two WebKit snapshots at different delays (5th argument of snap-webkit) — the lens must render in its new position and the filter id must change.
  5. When possible, look at it on a real iPhone.
  6. When unsure how something should behave, open assets/demo.html through serve.py: ?pressed freezes pressed states, ?auto animates everything.

Files

  • assets/liquid-glass.js — engine: map generation, filter graph, controller, springs
  • assets/components.js — the liquidglass.dev web components (needs liquid-glass.js beside it)
  • assets/glass-webgl.js — WebGL renderer (video, canvas)
  • assets/demo.html — reference implementations of every pattern with a parameter panel
  • scripts/serve.py — static server with ?hold=<ms>
  • scripts/shoot.sh — screenshots in Chrome, WebKit and Firefox
  • scripts/snap-webkit.swift — WebKit snapshots via WKWebView
  • scripts/chrome-shot.mjs — real-time Chrome driver for hover/press states
  • references/components.md — the web components: markup, attributes, events, theming
  • references/patterns.md — code for each pattern and framework wiring
  • references/browser-quirks.md — compatibility, performance, measurements
  • references/optics.md — map encoding, filter graph, parameters, writing a filter by hand

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Liquid Glass Web AI skill do?

Build refractive "liquid glass" UI for the web (Apple Liquid Glass style) that renders the same in Chrome, Safari/iOS and Firefox — SVG feDisplacementMap applied through CSS filter to the content itself, runtime displacement maps, chromatic fringe and rim highlights, press/squish springs, WebKit-specific fixes, WebGL for video. Ships the liquidglass.dev web components (switch, slider, segmented, button, panel, navbar, menu bar, toast, dock, lens), the zero-dependency engine, a reference page and a three-engine screenshot script. Use whenever a web or front-end task (HTML/CSS/React/Vue/Svelt...

Why use Liquid Glass Web on TypingMind?

Because you install it once and use it with any model. Liquid Glass Web is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Liquid Glass Web in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/gentpan/liquidglass/tree/main/skill/liquid-glass-web. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Liquid Glass Web?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Liquid Glass Web?

As many as you like. As long as a model supports skills, you can use Liquid Glass Web with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Liquid Glass Web AI skill free?

Yes. It is published on GitHub by gentpan under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇