| name | qt-ui-design |
| description | Design or audit UI for Qt/QML, Qt projects, web, or embedded MPU or MCU targets. Use when creating screens, layouts, navigation, or auditing UX. |
| license | LicenseRef-Qt-Commercial OR BSD-3-Clause |
| compatibility | Designed for Claude Code, GitHub Copilot, and similar agents. |
| disable-model-invocation | false |
| metadata | {"author":"qt-ai-skills","version":"1.0","qt-version":"6.x","category":"conceptual","changelog":"Initial release"} |
Qt UI Design
Before producing UI output, confirm you know: target platform, screen geometry, design system, content priority, viewing distance, locale, and input methods. Run the seven items below as a check against the conversation and the project state; ask only the items that are genuinely missing. When the user cannot answer an item, choose a sensible Qt default and name it in your response so the user can correct it.
Small edits to an existing design — for example "move the OK button to the right", "change this label", "make this red" — do not trigger the checklist. Apply section 1 silently and verify section 2 (contrast, hit-target) where relevant.
0. Context check (before designing)
Use the seven items below to decide what is already known and what to ask. If the conversation or repository has already answered an item, do not re-ask.
- Target platform — Desktop, web browser, mobile, or specific hardware (MCU, Raspberry Pi, other embedded board)?
- If a specific board: ask whether a board-specific skill exists for it and load it if so.
- Screen shape — Rectangle (default), Square, or Circle?
- Resolution and DPI — Do you know the screen resolution and DPI? (Approximate is fine.)
- Design system — Check whether the project already uses a design system or Qt Quick Controls style. If so, follow it and reuse its tokens. If not, recommend a Qt Quick Controls style: Basic, Fusion, Imagine, Material, Universal, iOS, or FluentWinUI3 (the iOS and FluentWinUI3 styles require Qt 6.7 or later). Where the project follows a third-party design language (Material Design 3, Apple Human Interface Guidelines, Fluent 2), map its tokens to the corresponding Qt Quick Controls style rather than introducing a parallel token vocabulary.
- Content priority — What information is most important (primary), secondary, and tertiary on this screen?
- Viewing distance — How far will users be from the screen? (e.g. handheld ~30 cm, desk ~60 cm, panel ~1.5 m, wall ~3 m)
- Locale and input — What is the primary locale/language? Is RTL (Arabic, Hebrew, Farsi, Urdu) support required? What input methods must be supported (touch, keyboard, mouse/pointer, hardware buttons, voice)?
If the target is an embedded or MCU device, also read section 4 in full before any design decisions — it overrides several desktop defaults.
If the user is requesting an audit of an existing design, skip to section 5 (Audit).
1. Design principles to apply (all targets)
Apply these while designing. Do not ask about each one — use them to inform decisions silently.
Content and layout:
- Golden Ratio + Rule of Thirds: Place primary elements at visual intersections.
- Progressive Disclosure: Show only what is needed at the current step.
- Inverted Pyramid: Critical information first, elaboration after.
- Modularity: Divide complex flows into smaller, self-contained screens.
- Ockham's Razor: When two designs are equivalent, choose the simpler one.
- Performance Load: Fewer steps = higher task completion.
- Five Hat Racks: Organise by category, time, location, alphabet, or continuum.
Perception and interaction:
- Jakob's Law: Match patterns users already know.
- Affordance: Controls should look like what they do.
- Hick's Law: More choices = slower decisions. Limit options per screen.
- Miller's Law: Working memory holds ~7 items. Chunk accordingly.
- Recognition Over Recall: Show options; don't require memorisation.
- Proximity + Similarity: Group related elements visually.
- Uniform Connectedness: Shared border or color = same group.
- von Restorff Effect: One visually distinct element draws attention — use sparingly.
- Peak-End Rule: Users remember the peak moment and the ending. Design completion states (e.g. installer finish screens) to feel rewarding, not abrupt.
- Doherty Threshold: System feedback within 400 ms, or show a progress indicator.
- Aesthetic-Usability Effect: Polished design is perceived as more usable.
- Wayfinding: Users must always know where they are, where they've been, where they can go.
Reading patterns (use to guide information placement):
- F-shaped: Text-heavy content — top bar, shorter secondary bar, left-edge scan.
- Z-shaped: Sparse content — top-left → top-right → diagonal → bottom-right.
- Layer-cake: Users scan headings and skip body text.
- Spotted: Users jump to landmarks — links, capitals, list items.
Buttons and CTAs:
- Limit CTA buttons per group. OK + Cancel = one group; additional actions must be visually secondary.
- Use Proximity and Similarity to distinguish primary, secondary, and tertiary controls.
1.1 Motion and animation (desktop/web — embedded: see section 4)
Motion communicates state, relationship, and causality. Every animation must be functional, not decorative.
- Enter animations: Use deceleration easing (fast start, slow end). Elements should appear to arrive, not just pop.
- Exit animations: Use acceleration easing (slow start, fast end). Elements should appear to leave, not vanish.
- Direct manipulation feedback: Respond within 100 ms. Operations taking > 1 s must show a progress indicator; > 10 s must show estimated time.
- Duration budgets: Small elements (icons, badges): 100–150 ms. Medium elements (cards, panels): 200–300 ms. Full-screen transitions: 300–400 ms. Never exceed 500 ms for any UI animation — slower feels broken.
- Limit simultaneous animations to one or two elements. Animating the whole screen at once disorients.
- Animate only
transform and opacity in QML/CSS — these are GPU-composited. Animating geometry (width, height, anchors) triggers layout recalculation and causes jank.
- Honour user preference for reduced motion. On the web, gate non-essential animation behind the
prefers-reduced-motion CSS media query. Qt 6.x has no built-in equivalent: expose a project-level setting (for example a singleton property bound to QSettings or to a runtime accessibility option) and gate animations on it. When the user opts out, replace animation with instant transitions — do not simply slow them down.
- Spatial consistency: Elements that move between screens should animate in the direction that matches their destination (forward = slide left, back = slide right for LTR layouts).
1.2 Typography scale (desktop/web)
For embedded typography, see section 4.4. This subsection governs desktop and web targets only.
Use a modular scale to derive all type sizes. A modular scale is a sequence of numbers related by a fixed ratio — every size is mathematically proportional to every other, producing a scale that feels harmonious rather than arbitrary. The ratios are listed below; see also https://www.modularscale.com/ for an interactive generator.
How to build the scale
-
Choose a base — the size at which your body text looks best at the target viewing distance. For desktop at ~60 cm, 16 px is a reliable starting point. The base is your ms(0).
-
Choose a ratio — multiply the base by this ratio to get the next step up, divide to get the next step down. Pick based on the character of the product:
| Ratio | Name | Factor | Character |
|---|---|---|---|
| 8:9 | Major second | 1.125 | Compact, dense — good for data-heavy UIs, installer flows |
| 5:6 | Minor third | 1.200 | Moderate — good for general desktop apps |
| 4:5 | Major third | 1.250 | Open, comfortable — good for marketing, onboarding |
| 3:4 | Perfect fourth | 1.333 | Strong contrast — good for dashboards, bold hierarchy |
| 1:1.618 | Golden section | 1.618 | High contrast — use sparingly; large gaps between steps |
-
Map scale steps to roles — assign a step number to each typographic role. Never add roles not on the scale; if a size is needed, pick the nearest step.
Worked example (base 16 px, Perfect Fourth 1.333)
| ms() | Value (px, rounded) | Role |
|---|
| ms(3) | 37.9 → 38 px | Display / hero text |
| ms(2) | 28.4 → 28 px | Page title / H1 |
| ms(1) | 21.3 → 21 px | Section heading / H2 |
| ms(0) | 16 px | Body — base |
| ms(-1) | 12.0 → 12 px | Caption / label / metadata |
Use a maximum of three to four of these steps per screen. More than four active sizes creates visual noise.
Rules that apply to all modular scales
- Body (ms(0)) minimum is 16 px at desktop viewing distance (~60 cm). Never use ms(-1) or smaller for primary reading content.
- Line height: 1.4–1.6× for body. 1.1–1.2× for headings (they need less leading).
- Line length: 45–75 characters per line for comfortable reading. Use
max-width on text containers — do not let prose span the full viewport.
- Weight pairs: Use Regular (400) for body and captions; Medium (500) for headings. Never use Bold (700) inside body text.
- System font first. Use the OS/platform system font by default (Segoe UI Variable on Windows, SF Pro on macOS, Roboto on Android/Linux). Only introduce a custom font when there is a brand requirement and a Figma token exists for it.
- Verify at Large system font size. Do not hardcode pixel values that ignore the OS font scale. In Qt, prefer
font.pointSize (which respects the platform DPI scale) over font.pixelSize for body text, or derive sizes from a singleton driven by Screen.pixelDensity. On the web, use relative units (rem).
Qt/QML implementation
In QML, define the scale as a singleton (e.g. TypeScale.qml) so sizes are referenced by role name, not hardcoded values:
// TypeScale.qml — generated from modular scale, base 16, ratio 1.333
pragma Singleton
import QtQuick
QtObject {
readonly property real base: 16 // ms(0)
readonly property real h2: 21 // ms(1)
readonly property real h1: 28 // ms(2)
readonly property real display: 38 // ms(3)
readonly property real caption: 12 // ms(-1)
}
Reference via TypeScale.h1 in components. Recalculate the whole singleton when the base or ratio changes — never patch individual values.
Register the singleton in your QML module so it can be imported. In a qmldir file:
module MyApp.Theme
singleton TypeScale 1.0 TypeScale.qml
In CMake with qt_add_qml_module:
qt_add_qml_module(myapp_theme
URI MyApp.Theme
VERSION 1.0
QML_FILES TypeScale.qml
)
set_source_files_properties(TypeScale.qml PROPERTIES QT_QML_SINGLETON_TYPE TRUE)
1.3 Multi-input and keyboard navigation (desktop/web)
Every desktop and web interface must support multiple input methods. Do not design for pointer alone.
- Full keyboard navigability is required. Tab order must follow the visual reading order (top-left to bottom-right for LTR). Always render a visible focus indicator — in QML, drive it from
activeFocus (for example a Rectangle border or scale bound to activeFocus) or use the style-specific focus visuals on Control. On the web, do not remove the focus ring without a styled replacement.
- No keyboard traps. Users navigating by keyboard must always be able to exit any modal, popover, or overlay using Escape or a keyboard-accessible close control.
- Every interactive element must be reachable by pointer, keyboard, and — where Qt's accessibility APIs expose it — screen reader. This is a requirement, not a recommendation.
- Hover states are an enhancement, not the primary disclosure mechanism. Any information shown on hover must also be accessible without a pointer (e.g. a visible label, a dedicated info button, or keyboard-triggered tooltip).
- Touch and pointer coexist. On touch devices that also support a stylus or pointer, do not remove touch targets when a pointer is detected.
1.4 Semantic colour
Colour should communicate meaning consistently across the entire interface. Avoid arbitrary colour choices.
- Use role-based tokens, not raw hex values. Token names should describe the role a colour plays (interactive, surface, on-surface, error, outline), not its appearance (blue, dark-blue, grey-2). The role vocabulary above mirrors Material Design 3; when the project's design language is Fluent 2 or Apple Human Interface Guidelines, map to the equivalent role names from those systems rather than mixing vocabularies. If the project ships with a Figma token set, copy the tokens manually into a singleton until tooling is available to extract them.
- Distinguish interactive colour from decorative colour. A colour used on a button to signal "this is tappable" must not also be used as a background accent that doesn't invite interaction. Reusing interactive colour decoratively trains users to tap things that aren't tappable.
- Colour is never the sole carrier of state. Already covered in section 2 (WCAG), but applies equally to non-disabled states: success, warning, and error must always pair colour with an icon or text label.
- Dark mode: Design for both light and dark themes from the start. Token-based colour systems handle this automatically — if hardcoded colours are used, a dark variant must be explicitly defined.
- Culturally variable colour meanings (see also §1.5): Red = danger in Western contexts but good fortune in some East Asian contexts. White = purity in Western contexts but mourning in some East Asian contexts. For safety-critical or internationally shipped products, do not rely on colour alone to convey critical meaning — always pair with text or universal iconography.
1.5 Localisation and RTL layout
Ask question 7 in the intake. Act on the answer here.
- Date, time, and number formats must be locale-aware. Never hardcode format strings like
DD/MM/YYYY. Use Qt's QLocale or the platform locale API.
- RTL layout mirroring: For Arabic, Hebrew, Farsi, and Urdu, the entire layout mirrors horizontally. Navigation moves from right to left. Back buttons point right. Icons that imply direction (arrows, chevrons, playback controls) must flip. Enable
LayoutMirroring.enabled with LayoutMirroring.childrenInherit: true near the root of your scene; it handles anchors and Qt Quick Layouts. The following items still need explicit attention even when LayoutMirroring is enabled: Text.horizontalAlignment defaults, Image content (use mirror: true on directional icons), custom Canvas painting, and any manual x positions.
- Text expansion: Translated strings are typically 30–40% longer than English source. Design containers and buttons to accommodate expansion — avoid fixed-width containers for any user-visible string.
- Icons: Avoid icons that are culturally specific or that carry variable meaning across regions (e.g. a thumbs-up, certain hand gestures). Prefer universal symbols (checkmark for success, X for close, magnifying glass for search).
- Avoid embedding text in images or icons. Localisation requires all text to be in string resources — text baked into assets cannot be translated.
2. Accessibility (WCAG 2.2)
Check every design against these before delivering:
- Perceivable: All information has a non-visual equivalent (alt text, labels). Contrast ≥ 4.5:1 for text, ≥ 3:1 for large text and UI components.
- Operable: Full keyboard navigation, no focus traps. All interactive targets reachable without a pointer.
- Understandable: Labels, instructions, and error messages are plain language.
- Robust: Markup and structure work with screen readers and assistive tools.
- Never rely on color alone to communicate state — always pair with shape, icon, or text.
- Test with OS font size set to Large — verify that layout tolerates text scaling without truncation or overflow.
- Test with a colour-blindness simulation (Deuteranopia is the most common) — confirm that all state changes are distinguishable without colour.
- Validate focus order — tab through every interactive element and confirm the sequence is logical. Focus must be visible at all times — see §1.3 for the Qt-side focus indicator pattern.
3. AI-specific UX
When designing screens that involve AI features:
- User Control: Users must be able to start, stop, and modify AI actions. Always include Undo or Regenerate.
- Transparency: Show why the AI made a decision when it affects the user.
- Graceful Failure: When AI fails, provide a clear manual fallback path.
- Value over Novelty: AI features must solve a real problem — not exist for demonstration.
- Uncertainty Communication: AI output is probabilistic. When confidence is low or the result may be incorrect, surface that — use hedging language ("suggested", "approximately", "review before using"), visual confidence indicators, or explicit labelling. Never present probabilistic output as authoritative fact. For Qt AI Assistant and similar features, provide a "show source" or "why this?" affordance wherever the answer affects consequential decisions.
- Latency patterns: AI operations typically take 1–15 seconds. Do not show a generic spinner — users cannot estimate duration from a spinner and become anxious after ~3 s.
- Streaming output (text generation): begin rendering partial results immediately. Show a blinking cursor or pulsing indicator at the insertion point while generation continues.
- Long operations (code generation, image processing): show a skeleton screen or placeholder that matches the shape of the expected result. This anchors attention and sets expectations.
- Background operations (indexing, analysis): surface as a subtle persistent status indicator (progress bar in a status bar, animated icon in a toolbar), not a blocking modal.
- Never blank the entire screen while waiting for an AI response.
- Consent before action: If an AI feature will read, send, or modify user data (files, code, settings), make this explicit before the action executes. A one-line confirmation ("This will read your project files to generate a suggestion") is sufficient — it does not need to be a modal dialog.
4. Embedded and MCU targets
This section overrides desktop defaults. Read it fully before designing for any embedded, MCU, or hardware-constrained target.
Hardware facts that drive every decision (collect these during the intake step above if not already answered):
- Physical pixel dimensions and DPI
- GPU present? If unknown, assume none.
- Available RAM and flash for UI assets
- Input method: touchscreen, hardware buttons, rotary encoder, pointer
- Whether a board-specific skill exists
4.1 Rendering without a GPU
Software rasterisation means every visual effect costs CPU time.
| Allowed | Not allowed |
|---|
| Flat solid fills | Gradients |
| 1 px solid strokes | Drop shadows |
| Bitmap (PNG) icons and fonts | Blur or transparency layers |
| Opacity-only transitions | Simultaneous move + fade animations |
| Fixed pixel layout | Flexbox or fluid layout |
Prefer bitmap icons over vector paths — raster cost is paid at compile time, not runtime. Relax these constraints only when a GPU is confirmed.
Exceptions. The constraints above apply to runtime software rasterisation. Qt Quick Ultralite supports compile-time-baked gradients via its static rendering pipeline. Some MCUs ship with a 2D blitter or a small GPU that relaxes the no-blur and no-transparency constraints. Verify the hardware data sheet and the Qt toolchain in use before treating the table as a hard prohibition.
4.2 Layout
- Fixed pixel layout. No fluid grids, no breakpoints. Design to the exact screen dimensions.
- One task per screen. No overlapping panels.
- Flat navigation: move between screens, not within them. Max 2 levels deep.
- System state (connected, error, active mode) must be permanently visible — no tooltips or hover states.
4.3 Interaction
- No hover states. Replace hover-triggered disclosure with explicit buttons or dedicated screens.
- Touch minimum: 48 px is the default. Some controlled environments (fixed-grip medical instruments, secured industrial panels) may justify smaller targets after explicit safety review. Gloved-hand environments (industrial, outdoor): 60–72 px.
- Every action must have a hardware button fallback. No touch-only actions.
- No drag, scroll inertia, or multi-touch unless hardware confirms support.
- Confirm before any action that controls a physical actuator.
4.4 Typography
Embedded fonts are bitmaps — size is fixed at build time. Get it right from the start.
| Viewing distance | Minimum size |
|---|
| ~50 cm (handheld, appliance) | 14 px min, 16 px preferred |
| ~1.0–1.5 m (HMI, instrument cluster) | 20 px min |
| ~2–3 m (wall panel, public display) | 28 px min |
Contrast ≥ 4.5:1 for all text. Embedded screens often have poor viewing angles and bright ambient light — higher contrast is always better.
4.5 Error and safety
- Confirm before irreversible physical actuator commands — this is a safety requirement.
- Error state indicators must be persistent: driven by application state, not transient UI state. They must survive a screen repaint.
- Alarms require three independent cues: color + shape + text.
- Define a safe default screen the UI returns to if the application crashes or loses communication.
- No silent failures — unacknowledged hardware commands must be surfaced visibly.
4.6 Desktop → MCU quick reference
| Rule | Desktop / web | MCU / embedded |
|---|
| Responsiveness | Flexbox, fluid | Fixed pixel layout |
| Animation | Easing curves, ≤ 400 ms | Opacity-only, < 200 ms |
| Progressive disclosure | Tooltips, hover | Explicit button only |
| Touch targets | 44 px recommended | 48 px default (relax only with explicit safety review) |
| Font rendering | Vector, OS-scalable | Bitmap, fixed at build time |
| Status communication | Color + token-based | Color + shape + text |
| Error recovery | Undo | Confirm before action |
| State visibility | Status bar / tooltip | Always on screen |
| Navigation depth | Unlimited | Max 2 levels |
| Localisation | QLocale, RTL mirroring | QLocale only (RTL rarely applicable) |
5. Audit instructions
When auditing an existing design, categorize every finding:
- Critical — Violates a core UX law or WCAG accessibility rule. Must fix.
- Warning — Potential friction or elevated cognitive load. Should fix.
- Opportunity — Enhancement or AI-driven improvement. Consider.
Apply section 4 constraints first if the target is embedded.
Extended audit checklist for desktop/web targets:
6. References
