L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Design et création · SKILL.md

hz-unity-meta-quest-ui

hz-unity-meta-quest-ui

Configures Unity UI for Meta VR and Horizon OS VR development — world-space canvases, TextMesh Pro setup, comfortable sizing, viewing distances, and interaction readiness.

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: hz-unity-meta-quest-ui license: Apache-2.0 description: Configures Unity UI for Meta VR and Horizon OS VR development — world-space canvases, TextMesh Pro setup, comfortable sizing, viewing distances, and interaction readiness.

Meta VR UI Setup

When to use this skill

Use this skill automatically when:

  • Setting up a Canvas for VR
  • Creating UI text with TextMesh Pro in a VR project
  • Adding buttons, sliders, or other interactive UI in VR
  • User reports pink/magenta text, unclickable buttons, or UI sizing issues in VR
  • Configuring VR interaction (ray or poke) on a Canvas

Driving the Editor

Build the UI against an open Editor with unity-cli rather than hand-editing scene YAML — the Editor applies changes to the actual active scene.

unity status --format json          # look for state "ready"
unity command run_script --file AgentScripts/BuildUI.cs --entry BuildUI.Run --format json

Canvas construction is multi-component work, so it belongs in a run_script .cs file (kept outside Assets/, path relative to the project root) rather than a chain of single commands. Add --dry_run true to compile-check a script before it mutates the scene — but read the verdict from data.result.success: a compile failure still returns outer success: true and exit code 0, with the errors in data.result.diagnostics. Connecting, --project-path, and Safe Mode recovery are in the unity-cli skill; MCP equivalents are in references/unity-mcp-fallback.md.

Single-property reads and writes are cheaper as direct commands than as a script:

unity command get_component_properties --target MenuUI --type Canvas --format json
unity command set_transform --target MenuUI --position "[0,1.5,2]" --format json
unity command find_gameobjects --name MenuUI --format json

Array parameters take a JSON array in one argument (--position "[0,1.5,2]"). Space- or comma-separated components are rejected with INVALID_COMMAND_ARGS. Note set_transform writes local position — see the unity-placement skill for parented canvases.

Prerequisite: TMP Essential Resources

Before creating ANY VR UI, verify TMP resources are imported:

unity command find_assets --name "LiberationSans SDF" --format json

A data.result.count of 0 means they are missing — use the tmp-resources skill before proceeding.

Step 1: Create World Space Canvas

// AgentScripts/BuildUI.cs
using UnityEditor;
using UnityEditor.SceneManagement;
using UnityEngine;
using UnityEngine.UI;

public static class BuildUI
{
    public static string Run()
    {
        // Adapt the name to match your canvas (e.g., "MainMenu", "SettingsUI")
        var go = new GameObject("MenuUI");
        var canvas = go.AddComponent<Canvas>();
        canvas.renderMode = RenderMode.WorldSpace;

        go.AddComponent<GraphicRaycaster>();

        // Remove CanvasScaler — not appropriate for VR
        var scaler = go.GetComponent<CanvasScaler>();
        if (scaler != null)
            Object.DestroyImmediate(scaler);

        var rt = go.GetComponent<RectTransform>();
        rt.localScale = new Vector3(0.001f, 0.001f, 0.001f);
        rt.sizeDelta = new Vector2(1920f, 1080f);
        rt.position = new Vector3(0f, 1.5f, 2f);

        Undo.RegisterCreatedObjectUndo(go, "Create VR Canvas");
        EditorSceneManager.MarkSceneDirty(EditorSceneManager.GetActiveScene());
        return quot;Created VR Canvas '{go.name}'. Scale: {rt.localScale}, Size: {rt.sizeDelta}, Position: {rt.position}";
    }
}

Persist it when the shape is right:

unity command save_scene --format json

Canvas rules

  • Render Mode: Always World Space. Screen Space modes break stereo rendering.
  • Scale: 0.001 on all axes (1 unit in canvas = 1mm in world).
  • CanvasScaler: Remove it. Physical size is controlled by world scale, not screen adaptation.
  • Distance: Place 1.5-3m from user. Never closer than 0.5m. Max 5m for readable text.
  • Physical size formula: Canvas sizeDelta * scale = meters. Example: 1920 * 0.001 = 1.92m wide.

Step 2: Create child UI elements

All child elements (panels, buttons, text) must follow these rules:

  • localScale: Always [1, 1, 1]. Never scale children to compensate for canvas scale.
  • localPosition.z: Always 0. Children must sit on the canvas plane.
  • Size control: Use RectTransform.sizeDelta and anchors, never scale.
// AgentScripts/BuildPanel.cs
using TMPro;
using UnityEditor;
using UnityEditor.SceneManagement;
using UnityEngine;
using UnityEngine.UI;

public static class BuildPanel
{
    public static string Run()
    {
        // Replace "MenuUI" with the actual canvas name used in Step 1
        var canvas = GameObject.Find("MenuUI");
        if (canvas == null) return "ERROR: Canvas 'MenuUI' not found.";

        // Panel
        var panel = new GameObject("ButtonPanel");
        panel.transform.SetParent(canvas.transform, false);
        var panelRT = panel.AddComponent<RectTransform>();
        panelRT.localScale = Vector3.one;
        panelRT.sizeDelta = new Vector2(800f, 600f);
        var panelImg = panel.AddComponent<Image>();
        panelImg.color = new Color(0.1f, 0.1f, 0.1f, 0.95f);
        panelImg.raycastTarget = false;

        var layout = panel.AddComponent<VerticalLayoutGroup>();
        layout.spacing = 50f;
        layout.padding = new RectOffset(80, 80, 100, 100);
        layout.childAlignment = TextAnchor.MiddleCenter;

        // Button
        var btnGO = new GameObject("StartButton");
        btnGO.transform.SetParent(panel.transform, false);
        var btnRT = btnGO.AddComponent<RectTransform>();
        btnRT.localScale = Vector3.one;
        btnRT.sizeDelta = new Vector2(400f, 120f);
        var btnImg = btnGO.AddComponent<Image>();
        btnImg.color = new Color(0.2f, 0.6f, 1f, 1f);
        btnGO.AddComponent<Button>();

        // Button text
        var textGO = new GameObject("Text");
        textGO.transform.SetParent(btnGO.transform, false);
        var textRT = textGO.AddComponent<RectTransform>();
        textRT.localScale = Vector3.one;
        textRT.anchorMin = Vector2.zero;
        textRT.anchorMax = Vector2.one;
        textRT.sizeDelta = Vector2.zero;
        var tmp = textGO.AddComponent<TextMeshProUGUI>();
        tmp.text = "Start";
        tmp.fontSize = 48f;
        tmp.color = new Color(0.95f, 0.95f, 0.95f, 1f);
        tmp.alignment = TextAlignmentOptions.Center;
        tmp.raycastTarget = false;

        Undo.RegisterCreatedObjectUndo(panel, "Create VR Panel");
        EditorSceneManager.MarkSceneDirty(EditorSceneManager.GetActiveScene());
        return quot;Created panel with button. Panel scale: {panelRT.localScale}, Button scale: {btnRT.localScale}";
    }
}

Step 3: Validate created UI

Spot-check the canvas itself with a direct command:

unity command get_component_properties --target MenuUI --type Canvas --format json
unity command get_scene_hierarchy --format json                # whole active scene, incl. children

get_scene_hierarchy --path takes an open scene path, not a GameObject — --path MenuUI fails with "Scene 'MenuUI' is not open." Omit it and read the MenuUI subtree out of the active scene.

Editor checks only go so far — deploy the build and pull metavr capture screenshot from the headset to see the UI the way users will.

To sweep every child and auto-fix, use a script:

// AgentScripts/ValidateUI.cs
using System.Text;
using UnityEngine;

public static class ValidateUI
{
    public static string Run()
    {
        // Replace "MenuUI" with the actual canvas name
        var canvas = GameObject.Find("MenuUI");
        if (canvas == null) return "ERROR: Canvas not found.";

        var crt = canvas.GetComponent<RectTransform>();
        var c = canvas.GetComponent<Canvas>();
        var sb = new StringBuilder();

        if (c.renderMode != RenderMode.WorldSpace)
            sb.AppendLine("FAIL: Canvas renderMode is not World Space.");

        if (Mathf.Abs(crt.localScale.x - 0.001f) > 0.0001f)
            sb.AppendLine(quot;FAIL: Canvas scale is {crt.localScale}, expected 0.001.");

        foreach (Transform child in canvas.GetComponentsInChildren<Transform>(true))
        {
            if (child == canvas.transform) continue;
            var rt = child.GetComponent<RectTransform>();
            if (rt == null) continue;

            if (rt.localScale != Vector3.one)
            {
                sb.AppendLine(quot;FIXED: '{child.name}' had localScale {rt.localScale}, expected (1,1,1).");
                rt.localScale = Vector3.one;
            }
            if (Mathf.Abs(rt.localPosition.z) > 0.01f)
            {
                sb.AppendLine(quot;FIXED: '{child.name}' had localPosition.z={rt.localPosition.z}, expected 0.");
                rt.localPosition = new Vector3(rt.localPosition.x, rt.localPosition.y, 0f);
            }
        }

        return sb.Length == 0 ? "Validation PASSED. All properties correct." : sb.ToString();
    }
}

Common issues this catches:

  • Canvas scale stuck at 1,1,1 instead of 0.001
  • Child elements auto-scaled to compensate for parent (e.g., 1000x)
  • Child elements offset on Z axis (e.g., localPosition.z = -2500)

Step 4: Add VR interaction (interactive UI only)

Only add interaction when your Canvas has interactive elements (buttons, sliders, dropdowns, toggles, input fields). Skip this for display-only UI (HUD, score, labels).

Decision tree

Does your Canvas have buttons, dropdowns, sliders, toggles, or input fields?
  YES -> Add VR interaction. Without it, interactive elements won't work in VR.
  NO  -> Skip. Display-only UI doesn't need interaction overhead.

Wiring it — ISDK Quick Actions

The supported entry point is the ISDK Quick Actions API, called through run_script. See the hz-unity-meta-interaction-sdk skill for signatures and the behavior these calls bake in.

// AgentScripts/AddCanvasInteraction.cs
using Oculus.Interaction.Editor.QuickActions;   // Editor-only assembly
using UnityEngine;

public static class AddCanvasInteraction
{
    public static string Run()
    {
        var target = GameObject.Find("MenuUI");
        QuickActionsAPI.AddRayCanvasInteraction(target);    // or AddPokeCanvasInteraction
        return "ok";
    }
}

These calls take the canvas's GameObject, not a Canvas component — passing Canvas is a compile error (cannot convert from 'UnityEngine.Canvas' to 'UnityEngine.GameObject'). Verified signatures:

InteractionSignatureWhen to use
Ray (default)AddRayCanvasInteraction(GameObject target, bool fixPointableCanvasModule = true)Menus, settings panels, any UI beyond arm's reach
Poke (close range)AddPokeCanvasInteraction(GameObject target, bool fixPointableCanvasModule = true)Control panels on surfaces, virtual keyboards, diegetic UI on props (< 0.8m)
GazeAddGazeCanvasInteraction(GameObject target, bool enableRayFallback = true, bool addToChildCanvases = true, bool fixPointableCanvasModule = true)Eye-tracked selection
Both (advanced)Ray first, then pokeHybrid UIs where users can point OR touch

Ray adds RayInteractable, PointableCanvas, an ISDK_RayCanvasInteraction child, and a scene-level Pointable Canvas Module. Poke adds PokeInteractable, PointableCanvas, and close-range collision detection.

Pass fixPointableCanvasModule: false when the PointableCanvasModule comes from a prefab or an unloaded additive scene, or you get a second one in the open scene.

If a Unity MCP server with the Meta Unity extension is what you have instead, the equivalent tools are meta_add_canvas_interaction_ray / meta_add_canvas_interaction_poke — see references/unity-mcp-fallback.md.

Verify interaction was added

unity command get_component_properties --target MenuUI --type PointableCanvas --format json
unity command find_gameobjects --name "Pointable Canvas Module" --format json

get_component_properties needs --type whenever --target is a GameObject (without it the call fails: "'type' is required when 'target' is a GameObject."), and it addresses one component rather than listing them — so name the component you are asserting. Success means the component is there; a failure naming the type means it is not.

PointableCanvas on the canvas plus a Pointable Canvas Module in the scene is the bar. Note that PointableElement wiring is assigned in Awake(), so on a prefab asset it always reads null — check the serialized field instead:

unity command get_serialized_fields --target <interactable> --field _pointableElement --format json

If buttons work in Editor but not on Meta VR device

The scene needs an interaction rig at runtime:

using Oculus.Interaction.OVR.Editor.QuickActions;
OVRQuickActionsAPI.AddOVRInteractionRig();      // also adds an OVRCameraRig if the scene has none

This provides all hand/controller interactors. See hz-unity-meta-interaction-sdk for the rig it builds and the "interactable authored without a rig present" caveat.

VR UI sizing reference

All sizes assume canvas scale of 0.001 (1 unit = 1mm).

Element sizes (minimum)

ElementsizeDelta (units)Physical size
Button (small)250 x 10025cm x 10cm
Button (standard)300 x 12030cm x 12cm
Button (large)400 x 15040cm x 15cm
Dropdown500 x 14050cm x 14cm
Slider600 x 8060cm x 8cm
Toggle120 x 12012cm x 12cm
Input Field800 x 14080cm x 14cm
Panel/Menu1200-2000 x 800-14001.2-2m x 0.8-1.4m

Button spacing

Minimum 50 units (5cm) between interactive elements to prevent mis-clicks.

Font sizes (at 0.001 canvas scale)

DistanceMinimumComfortableLarge/Title
1.5m32pt40-48pt60-72pt
2.0m36pt48-56pt72-84pt
2.5m40pt52-64pt84-96pt
3.0m48pt64-72pt96-120pt

Never use auto-sizing in VR. Never use legacy Text components — always TextMeshPro.

Colors

  • Text: off-white (0.9, 0.9, 0.9) or dark gray (0.1, 0.1, 0.1). Avoid pure white/black.
  • Backgrounds: dark (0.1, 0.1, 0.1, 0.95) or light (0.9, 0.9, 0.9, 0.95).
  • Contrast ratio: minimum 4.5:1, prefer 7:1+.
  • Prefer opaque backgrounds over transparent (cheaper to render).

Performance tips

  • Disable raycastTarget on non-interactive elements (labels, backgrounds, decorative images).
  • Split static and dynamic content onto separate canvases to minimize rebuilds.
  • Canvas rebuild cost should be < 1-2ms to maintain 72/90Hz.
  • Share one SDF font asset per font family across all text elements.
  • Disable canvases or GameObjects when not visible.

Troubleshooting

SymptomCauseFix
Pink/magenta textTMP resources not importedUse tmp-resources skill
Buttons not clickable in VRMissing VR interaction on CanvasQuickActionsAPI.AddRayCanvasInteraction(go)
Buttons work in Editor, not on deviceMissing interaction rigOVRQuickActionsAPI.AddOVRInteractionRig()
UI too small/largeWrong canvas scaleSet to 0.001, verify with the validation script
Children offset in ZAuto-positioning bugSet localPosition.z = 0 on all children
Children scaled to 1000xAuto-scale compensationSet localScale = Vector3.one on all children
Text blurryLow atlas resolution or small fontUse SDF 2048x2048+, font size 48+
Frame dropsCanvas rebuilding too oftenSplit into static/dynamic canvases
unity command won't connectEditor in Safe Mode from compile errorsunity pipeline list, fix the errors, restart Unity

Checklists

Interactive UI (buttons, sliders, etc.)

[ ] TMP Essential Resources imported
[ ] Canvas: renderMode = World Space
[ ] Canvas: localScale = [0.001, 0.001, 0.001]
[ ] Canvas: positioned 1.5-3m from user
[ ] All children: localScale = [1, 1, 1]
[ ] All children: localPosition.z = 0
[ ] Sizes via sizeDelta, not scale
[ ] AddRayCanvasInteraction called on Canvas
[ ] PointableCanvas component verified on Canvas
[ ] Pointable Canvas Module exists in scene
[ ] Interaction rig present (AddOVRInteractionRig)
[ ] Buttons >= 250x100 units
[ ] Button spacing >= 50 units
[ ] Text >= 48pt, raycastTarget = false on non-interactive text
[ ] Validation script run and passed
[ ] Scene saved (unity command save_scene)

Display-only UI (HUD, score, labels)

[ ] TMP Essential Resources imported
[ ] Canvas: renderMode = World Space
[ ] Canvas: localScale = [0.001, 0.001, 0.001]
[ ] All children: localScale = [1, 1, 1]
[ ] All children: localPosition.z = 0
[ ] Text >= 48pt
[ ] raycastTarget = false on all elements
[ ] NO interaction components needed
[ ] Scene saved (unity command save_scene)

Integration with other skills

  • tmp-resources: Use first to import TMP Essential Resources before creating any UI.
  • hz-unity-meta-interaction-sdk: Owns the Quick Actions APIs used in Step 4.
  • unity-placement: Use for positioning canvases relative to other scene objects.
  • unity-cli: Connecting to an Editor, --project-path, Safe Mode recovery.
Skills enregistrées