vision_ai_flutter

Pre-built Flutter UI overlay widgets for vision_ai. Camera preview with hand skeleton, face bounding boxes, contour mesh, gesture labels, and emotion indicators — one widget does it all.

pub package

Hand skeleton Face detection

Installation

dependencies:
  vision_ai: ^0.1.0
  vision_ai_flutter: ^0.1.0

Quick Start

import 'package:vision_ai/vision_ai.dart';
import 'package:vision_ai_flutter/vision_ai_flutter.dart';

final vision = VisionAi(hand: HandConfig(), face: FaceConfig());
final textureId = await vision.start();

// One widget handles everything:
VisionAiCameraView(
  controller: vision,
  textureId: textureId,
)

That's it — hand skeleton, face box, gesture label, and emotion indicator all render automatically with sensible defaults.

VisionAiCameraView

The main composite widget. Stacks a Texture preview with configurable overlay layers, all driven by a single StreamBuilder so every layer stays frame-synced.

All Parameters

VisionAiCameraView(
  // Required
  controller: vision,             // VisionAi instance
  textureId: textureId,           // from vision.start()
  
  // Toggle overlays (all have sensible defaults)
  showHandLandmarks: true,        // 21-point skeleton with bone connections
  showHandBoundingBox: false,     // yellow rectangle around detected hand
  showFaceBoundingBox: true,      // cyan rectangle around detected face
  showFaceContours: false,        // 15-type face mesh with dots and polylines
  showGestureLabel: true,         // gesture name + confidence at top center
  showEmotionLabel: true,         // emotion name + emoji at bottom center
  
  // Advanced
  mirrorLandmarks: false,         // flip skeleton horizontally (for non-mirrored preview)
  style: const OverlayStyle(),    // colors, sizes, fonts for all overlays
  overlayBuilder: null,           // custom widget on top of everything (gets VisionResult)
)

Overlay Toggle Guide

Scenario Recommended toggles
Hand gesture app showHandLandmarks: true, showGestureLabel: true
Emotion detector showFaceBoundingBox: true, showEmotionLabel: true
Face mesh / AR showFaceContours: true, showFaceBoundingBox: false
Debug / testing Everything on
Clean camera only Everything off, use overlayBuilder for custom UI
Performance-sensitive showHandLandmarks: false (painters cost ~1-2ms per frame)

Custom Overlay Builder

For anything beyond the built-in painters, overlayBuilder gives you full VisionResult access:

VisionAiCameraView(
  controller: vision,
  textureId: textureId,
  showHandLandmarks: false,  // disable built-in, draw your own
  overlayBuilder: (context, result) {
    if (!result.hasHands) return const SizedBox.shrink();
    final hand = result.primaryHand!;
    
    return Stack(
      children: [
        // Show extended fingers
        Positioned(
          top: 20, left: 20,
          child: Text(
            hand.fingerStates.entries
              .where((e) => e.value == FingerState.extended)
              .map((e) => e.key.name)
              .join(', '),
            style: const TextStyle(color: Colors.white, fontSize: 18),
          ),
        ),
        // Show world-coordinate measurements
        if (hand.worldLandmarks.length >= 21)
          Positioned(
            bottom: 20, left: 20,
            child: Text(
              'Span: ${(hand.worldLandmarks[HandLandmarkIndex.thumbTip].distanceTo(hand.worldLandmarks[HandLandmarkIndex.pinkyTip]) * 100).toStringAsFixed(1)}cm',
              style: const TextStyle(color: Colors.yellow, fontSize: 16),
            ),
          ),
      ],
    );
  },
)

The overlayBuilder renders on top of all built-in overlays. Return SizedBox.shrink() when you have nothing to show.

Painters

Use these CustomPainter classes directly if you need custom layout or want to compose your own camera view instead of using VisionAiCameraView.

HandLandmarkPainter

Draws 21 red landmark dots connected by 23 green bone lines.

CustomPaint(
  painter: HandLandmarkPainter(
    hands: result.hands,        // required
    imageSize: result.imageSize, // required: for coordinate mapping
    style: LandmarkStyle(
      dotColor: Colors.red,     // landmark dot color (default: red)
      lineColor: Colors.green,  // bone connection color (default: green)
      dotRadius: 4.0,           // dot radius in logical pixels
      lineWidth: 2.0,           // line width in logical pixels
    ),
    mirrored: false,            // flip horizontally for non-mirrored previews
  ),
)

Coordinate space: Landmarks are normalized 0,1. The painter maps directly to canvas size since Flutter's Texture stretches to fill its container.

HandBoundingBoxPainter

Draws a rectangle computed from the min/max of all 21 landmarks.

CustomPaint(
  painter: HandBoundingBoxPainter(
    hands: result.hands,
    boxColor: Colors.yellow,    // stroke color (default: yellow)
    boxWidth: 2.0,              // stroke width
    mirrored: false,            // mirrors the box for flipped previews
  ),
)

FaceOverlayPainter

Draws a rectangle around each detected face.

CustomPaint(
  painter: FaceOverlayPainter(
    faces: result.faces,
    imageSize: result.imageSize,  // required: face bbox is in pixel coords
    boxColor: Colors.cyan,       // stroke color (default: cyan)
    boxWidth: 2.0,               // stroke width
  ),
)

Coordinate space: Face bounding boxes are in pixel coordinates relative to imageSize. The painter scales to canvas size.

FaceContourPainter

Draws 15 contour polylines with dots for each face. Polylines are open (not closed paths) — ML Kit doesn't guarantee closed contours.

CustomPaint(
  painter: FaceContourPainter(
    faces: result.faces,
    imageSize: result.imageSize,
    dotColor: Colors.greenAccent,  // contour point color
    lineColor: Colors.greenAccent, // polyline color
    dotRadius: 2.0,
    lineWidth: 1.5,
  ),
)

Requires FaceConfig(detectContours: true). Renders: face outline, left/right eyebrow (top + bottom), left/right eye, upper/lower lip (top + bottom), nose bridge, nose bottom, left/right cheek.

Label Widgets

GestureLabel

Shows the gesture name and confidence on a rounded dark background.

GestureLabel(
  hand: result.primaryHand!,
  style: LabelStyle(
    textColor: Colors.white,
    backgroundColor: Colors.black87,
    fontSize: 20.0,
    fontWeight: FontWeight.bold,
    borderRadius: 12.0,
    padding: EdgeInsets.symmetric(horizontal: 16, vertical: 8),
  ),
)

Returns SizedBox.shrink() when gesture == Gesture.none.

EmotionIndicator

Shows the emotion name with an emoji and optional confidence percentage.

EmotionIndicator(
  face: result.primaryFace!,
  style: LabelStyle(backgroundColor: Colors.blue),
  showConfidence: true,  // "HAPPY 😊 98%" vs just "HAPPY 😊"
)

Emoji mapping: happy=😊, sad=😢, angry=😠, surprised=😮, disgusted=🤢, fearful=😨, neutral=😐.

ConfidenceBar

Horizontal progress bar showing a 0-100% score.

ConfidenceBar(
  value: face.emotionConfidence,  // [0.0, 1.0]
  color: Colors.green,
  backgroundColor: Colors.grey[800]!,
  height: 8.0,
  width: 120.0,
)

Full Styling Reference

OverlayStyle

Controls all visual properties when using VisionAiCameraView:

OverlayStyle(
  // Hand skeleton
  handLandmark: LandmarkStyle(
    dotColor: Colors.red,
    lineColor: Colors.green,
    dotRadius: 4.0,
    lineWidth: 2.0,
  ),
  
  // Hand bounding box
  handBoundingBoxColor: Colors.yellow,
  handBoundingBoxWidth: 2.0,
  
  // Face bounding box  
  faceBoundingBoxColor: Colors.cyan,
  faceBoundingBoxWidth: 2.0,
  
  // Gesture label
  gestureLabel: LabelStyle(
    textColor: Colors.white,
    backgroundColor: Colors.black87,
    fontSize: 20.0,
    fontWeight: FontWeight.bold,
    borderRadius: 12.0,
    padding: EdgeInsets.symmetric(horizontal: 16, vertical: 8),
  ),
  
  // Emotion label
  emotionLabel: LabelStyle(backgroundColor: Colors.blue),
  showEmotionConfidence: true,
)

Theme Ideas

Minimal white:

OverlayStyle(
  handLandmark: LandmarkStyle(dotColor: Colors.white, lineColor: Colors.white70, dotRadius: 3, lineWidth: 1),
  faceBoundingBoxColor: Colors.white54,
)

Neon cyberpunk:

OverlayStyle(
  handLandmark: LandmarkStyle(dotColor: Color(0xFFFF00FF), lineColor: Color(0xFF00FFFF), dotRadius: 5, lineWidth: 3),
  faceBoundingBoxColor: Color(0xFFFF00FF),
  handBoundingBoxColor: Color(0xFF00FFFF),
)

Subtle debug:

OverlayStyle(
  handLandmark: LandmarkStyle(dotColor: Colors.blue.withOpacity(0.5), lineColor: Colors.blue.withOpacity(0.3)),
  faceBoundingBoxColor: Colors.grey.withOpacity(0.3),
)

Example App

The vision_ai example app is a full working demo of every widget and painter in this package. Run it to see all overlays in action before writing code:

git clone https://github.com/OttomanDeveloper/vision_ai.git
cd vision_ai/packages/vision_ai/example
flutter run

What you can toggle in the Overlays card:

  • Hand skeleton (21-point landmark connections)
  • Hand bounding box (yellow rectangle)
  • Face bounding box (cyan rectangle)
  • Face contours (15-type green mesh)
  • Gesture label (top center)
  • Emotion label (bottom center)
  • World coordinates (pinch/span in cm)
  • Stats overlay (inference time, finger states, attention score, etc.)

Each overlay toggle applies instantly — no restart needed. The settings panel also groups related options into cards so hand overlays disappear when hand detection is off, and face overlays disappear when face detection is off.

The example also shows how to use VisionAiCameraView with ValueNotifier-driven settings, so you can see how overlay flags flow from your state into the widget without setState.

Tips

  • Performance: If you only need gesture labels (no skeleton), set showHandLandmarks: false — saves ~1-2ms per frame of painter work
  • Contours vs landmarks: Contours give you detailed face mesh (15 types, hundreds of points). Landmarks give you 10 key points. Contours disable face tracking — use landmarks if you need both tracking and facial points
  • Custom overlay: Use overlayBuilder for game logic, AR effects, or any UI that depends on detection results. It runs on every frame inside the StreamBuilder
  • Mirror: Set mirrorLandmarks: true only if your Texture is NOT already mirrored. The front camera output is mirrored by default on both Android and iOS

License

Apache 2.0 — see LICENSE and NOTICE.

Libraries

vision_ai_flutter