Holistic Landmarker

The MediaPipe Holistic Landmarker task lets you combine components of the face, hand, and pose landmarkers to detect a total of 553 human body landmarks in real time. You can use this task to analyze full-body gestures, poses, and actions across a continuous stream of images or video. The task outputs a combination of normalized landmarks, world landmarks, blendshapes (optional), and segmentation masks (optional).

Interactive demo Holistic Landmarker Simultaneously track 543 face, pose, and hand landmarks in a single pipeline. Launch interactive demo Open in Web DemosEdit in StackBlitz Runs in your browser with WebAssembly.

Open in Web DemosEdit in StackBlitz

Quickstart

Install the SDK for your platform, prepare input data, choose a running mode, and run inference.

Web

npmnpm install @mediapipe/tasks-visionWeb demo·Edit in StackBlitz·Setup guide

Image

IMAGEStateless · single frameSynchronous call on independent still images. Resets temporal tracking state on every call.

import { FilesetResolver, HolisticLandmarker } from "@mediapipe/tasks-vision";

const vision = await FilesetResolver.forVisionTasks(
  "https://cdn.jsdelivr.net/npm/@mediapipe/tasks-vision@latest/wasm"
);
const holisticLandmarker = await HolisticLandmarker.createFromOptions(vision, {
  baseOptions: { modelAssetPath: "holistic_landmarker.task", delegate: "GPU" },
  runningMode: "IMAGE",
  minFaceDetectionConfidence: 0.5,
  minPoseDetectionConfidence: 0.5,
});

// Synchronous single-frame call (stateless across images)
const image = document.getElementById("image") as HTMLImageElement;
const result = holisticLandmarker.detect(image);

Video

VIDEOEvery frame · no dropsSynchronous tracking across decoded video frames with monotonic timestamps. Processes every frame in order and never drops frames.

import { FilesetResolver, HolisticLandmarker } from "@mediapipe/tasks-vision";

const vision = await FilesetResolver.forVisionTasks(
  "https://cdn.jsdelivr.net/npm/@mediapipe/tasks-vision@latest/wasm"
);
const holisticLandmarker = await HolisticLandmarker.createFromOptions(vision, {
  baseOptions: { modelAssetPath: "holistic_landmarker.task", delegate: "GPU" },
  runningMode: "VIDEO",
  minFaceDetectionConfidence: 0.5,
  minPoseDetectionConfidence: 0.5,
});

// Synchronous across decoded frames with monotonic timestamps (never drops frames)
for (const { frame, timestampMs } of decodedVideoFrames) {
  const result = holisticLandmarker.detectForVideo(frame, timestampMs);
}

Livestream

VIDEO + rAFLive camera · latest frameWeb tasks run synchronously—pair runningMode: "VIDEO" with requestAnimationFrame so each paint inspects only the latest camera frame (skipping intermediate frames when busy).

import { FilesetResolver, HolisticLandmarker } from "@mediapipe/tasks-vision";

const vision = await FilesetResolver.forVisionTasks(
  "https://cdn.jsdelivr.net/npm/@mediapipe/tasks-vision@latest/wasm"
);
// On Web, live camera loops pair runningMode: "VIDEO" with requestAnimationFrame
const holisticLandmarker = await HolisticLandmarker.createFromOptions(vision, {
  baseOptions: { modelAssetPath: "holistic_landmarker.task", delegate: "GPU" },
  runningMode: "VIDEO",
  minFaceDetectionConfidence: 0.5,
  minPoseDetectionConfidence: 0.5,
});

const video = document.getElementById("webcam") as HTMLVideoElement;
video.srcObject = await navigator.mediaDevices.getUserMedia({ video: true });
await video.play();
let lastVideoTime = -1;

function renderLoop() {
  if (video.readyState >= 2 && video.currentTime !== lastVideoTime) {
    lastVideoTime = video.currentTime;
    const result = holisticLandmarker.detectForVideo(video, performance.now());
    console.log(result.poseLandmarks, result.faceLandmarks);
  }
  requestAnimationFrame(renderLoop);
}
renderLoop();

Android (Kotlin)

Gradleimplementation("com.google.mediapipe:tasks-vision:latest.release")Sample code·Setup guide

Image

IMAGEStateless · single frameSynchronous call on independent still images. Resets temporal tracking state on every call.

import com.google.mediapipe.framework.image.BitmapImageBuilder
import com.google.mediapipe.framework.image.MPImage
import com.google.mediapipe.tasks.core.BaseOptions
import com.google.mediapipe.tasks.vision.core.RunningMode
import com.google.mediapipe.tasks.vision.holisticlandmarker.HolisticLandmarker

val baseOptions = BaseOptions.builder()
    .setModelAssetPath("holistic_landmarker.task")
    .build()
val options = HolisticLandmarker.HolisticLandmarkerOptions.builder()
    .setBaseOptions(baseOptions)
    .setRunningMode(RunningMode.IMAGE)
    .setMinFaceDetectionConfidence(0.5f)
    .setMinPoseDetectionConfidence(0.5f)
    .build()

val holisticLandmarker = HolisticLandmarker.createFromOptions(context, options)

// Synchronous single-image call (no temporal tracking across calls)
val mpImage = BitmapImageBuilder(bitmap).build()
val result = holisticLandmarker.detect(mpImage)

Video

VIDEOEvery frame · no dropsSynchronous tracking across decoded video frames with monotonic timestamps. Processes every frame in order and never drops frames.

import com.google.mediapipe.framework.image.BitmapImageBuilder
import com.google.mediapipe.framework.image.MPImage
import com.google.mediapipe.tasks.core.BaseOptions
import com.google.mediapipe.tasks.vision.core.RunningMode
import com.google.mediapipe.tasks.vision.holisticlandmarker.HolisticLandmarker

val baseOptions = BaseOptions.builder()
    .setModelAssetPath("holistic_landmarker.task")
    .build()
val options = HolisticLandmarker.HolisticLandmarkerOptions.builder()
    .setBaseOptions(baseOptions)
    .setRunningMode(RunningMode.VIDEO)
    .setMinFaceDetectionConfidence(0.5f)
    .setMinPoseDetectionConfidence(0.5f)
    .build()

val holisticLandmarker = HolisticLandmarker.createFromOptions(context, options)

// Synchronous: processes every frame in order with monotonic timestamps (no frame drops)
val mpImage = BitmapImageBuilder(frameBitmap).build()
val result = holisticLandmarker.detectForVideo(mpImage, frameTimestampMs)

Livestream

LIVE_STREAMAsync · drops frames when busyAsynchronous camera pipeline with a result callback. Automatically drops incoming frames when busy so results stay current with low latency.

import android.os.SystemClock
import android.util.Log
import com.google.mediapipe.framework.image.BitmapImageBuilder
import com.google.mediapipe.framework.image.MPImage
import com.google.mediapipe.tasks.core.BaseOptions
import com.google.mediapipe.tasks.vision.core.RunningMode
import com.google.mediapipe.tasks.vision.holisticlandmarker.HolisticLandmarker

val baseOptions = BaseOptions.builder()
    .setModelAssetPath("holistic_landmarker.task")
    .build()
val options = HolisticLandmarker.HolisticLandmarkerOptions.builder()
    .setBaseOptions(baseOptions)
    .setRunningMode(RunningMode.LIVE_STREAM)
    .setMinFaceDetectionConfidence(0.5f)
    .setMinPoseDetectionConfidence(0.5f)
    .setResultListener { result, _ -> val pose = result.poseLandmarks() }
    .setErrorListener { error -> Log.e("MediaPipe", error.message.orEmpty()) }
    .build()

val holisticLandmarker = HolisticLandmarker.createFromOptions(context, options)

// Async: automatically drops incoming camera frames when busy so results stay current
holisticLandmarker.detectAsync(mpImage, SystemClock.uptimeMillis())

Android (Java)

Gradleimplementation 'com.google.mediapipe:tasks-vision:latest.release'Sample code·Setup guide

Image

IMAGEStateless · single frameSynchronous call on independent still images. Resets temporal tracking state on every call.

import com.google.mediapipe.framework.image.BitmapImageBuilder;
import com.google.mediapipe.framework.image.MPImage;
import com.google.mediapipe.tasks.core.BaseOptions;
import com.google.mediapipe.tasks.vision.core.RunningMode;
import com.google.mediapipe.tasks.vision.holisticlandmarker.HolisticLandmarker;
import com.google.mediapipe.tasks.vision.holisticlandmarker.HolisticLandmarkerResult;

BaseOptions baseOptions = BaseOptions.builder()
    .setModelAssetPath("holistic_landmarker.task")
    .build();
HolisticLandmarker.HolisticLandmarkerOptions options = HolisticLandmarker.HolisticLandmarkerOptions.builder()
    .setBaseOptions(baseOptions)
    .setRunningMode(RunningMode.IMAGE)
    .setMinFaceDetectionConfidence(0.5f)
    .setMinPoseDetectionConfidence(0.5f)
    .build();

HolisticLandmarker holisticLandmarker = HolisticLandmarker.createFromOptions(context, options);

// Synchronous single-image call (no temporal tracking across calls)
MPImage mpImage = new BitmapImageBuilder(bitmap).build();
HolisticLandmarkerResult result = holisticLandmarker.detect(mpImage);

Video

VIDEOEvery frame · no dropsSynchronous tracking across decoded video frames with monotonic timestamps. Processes every frame in order and never drops frames.

import com.google.mediapipe.framework.image.BitmapImageBuilder;
import com.google.mediapipe.framework.image.MPImage;
import com.google.mediapipe.tasks.core.BaseOptions;
import com.google.mediapipe.tasks.vision.core.RunningMode;
import com.google.mediapipe.tasks.vision.holisticlandmarker.HolisticLandmarker;
import com.google.mediapipe.tasks.vision.holisticlandmarker.HolisticLandmarkerResult;

BaseOptions baseOptions = BaseOptions.builder()
    .setModelAssetPath("holistic_landmarker.task")
    .build();
HolisticLandmarker.HolisticLandmarkerOptions options = HolisticLandmarker.HolisticLandmarkerOptions.builder()
    .setBaseOptions(baseOptions)
    .setRunningMode(RunningMode.VIDEO)
    .setMinFaceDetectionConfidence(0.5f)
    .setMinPoseDetectionConfidence(0.5f)
    .build();

HolisticLandmarker holisticLandmarker = HolisticLandmarker.createFromOptions(context, options);

// Synchronous: processes every frame in order with monotonic timestamps (no frame drops)
MPImage mpImage = new BitmapImageBuilder(frameBitmap).build();
HolisticLandmarkerResult result = holisticLandmarker.detectForVideo(mpImage, frameTimestampMs);

Livestream

LIVE_STREAMAsync · drops frames when busyAsynchronous camera pipeline with a result callback. Automatically drops incoming frames when busy so results stay current with low latency.

import android.os.SystemClock;
import android.util.Log;
import com.google.mediapipe.framework.image.BitmapImageBuilder;
import com.google.mediapipe.framework.image.MPImage;
import com.google.mediapipe.tasks.core.BaseOptions;
import com.google.mediapipe.tasks.vision.core.RunningMode;
import com.google.mediapipe.tasks.vision.holisticlandmarker.HolisticLandmarker;

BaseOptions baseOptions = BaseOptions.builder()
    .setModelAssetPath("holistic_landmarker.task")
    .build();
HolisticLandmarker.HolisticLandmarkerOptions options = HolisticLandmarker.HolisticLandmarkerOptions.builder()
    .setBaseOptions(baseOptions)
    .setRunningMode(RunningMode.LIVE_STREAM)
    .setMinFaceDetectionConfidence(0.5f)
    .setMinPoseDetectionConfidence(0.5f)
    .setResultListener((result, inputImage) -> { var pose = result.poseLandmarks(); })
    .setErrorListener(error -> Log.e("MediaPipe", String.valueOf(error.getMessage())))
    .build();

HolisticLandmarker holisticLandmarker = HolisticLandmarker.createFromOptions(context, options);

// Async: automatically drops incoming camera frames when busy so results stay current
holisticLandmarker.detectAsync(mpImage, SystemClock.uptimeMillis());

iOS

SwiftPMhttps://github.com/google-ai-edge/mediapipe → MediaPipeTasksVisionPackage.swift·Sample code·Setup guide

Image

IMAGEStateless · single frameSynchronous call on independent still images. Resets temporal tracking state on every call.

import Foundation
import MediaPipeTasksVision

guard let modelPath = Bundle.main.path(forResource: "holistic_landmarker", ofType: "task") else { return }

let options = HolisticLandmarkerOptions()
options.baseOptions.modelAssetPath = modelPath
options.runningMode = .image
options.minFaceDetectionConfidence = 0.5
options.minPoseDetectionConfidence = 0.5

let holisticLandmarker = try HolisticLandmarker(options: options)

// Synchronous single-image call (stateless across images)
let mpImage = try MPImage(uiImage: image)
let result = try holisticLandmarker.detect(image: mpImage)

Video

VIDEOEvery frame · no dropsSynchronous tracking across decoded video frames with monotonic timestamps. Processes every frame in order and never drops frames.

import Foundation
import MediaPipeTasksVision

guard let modelPath = Bundle.main.path(forResource: "holistic_landmarker", ofType: "task") else { return }

let options = HolisticLandmarkerOptions()
options.baseOptions.modelAssetPath = modelPath
options.runningMode = .video
options.minFaceDetectionConfidence = 0.5
options.minPoseDetectionConfidence = 0.5

let holisticLandmarker = try HolisticLandmarker(options: options)

// Synchronous: processes every frame in order with monotonic timestamps (no frame drops)
let mpImage = try MPImage(pixelBuffer: pixelBuffer)
let result = try holisticLandmarker.detect(videoFrame: mpImage, timestampInMilliseconds: timestampMs)

Livestream

LIVE_STREAMAsync · drops frames when busyAsynchronous camera pipeline with a result callback. Automatically drops incoming frames when busy so results stay current with low latency.

import Foundation
import MediaPipeTasksVision

final class HolisticLandmarkerStreamService: NSObject {
    private var holisticLandmarker: HolisticLandmarker?

    func setUp() throws {
        guard let modelPath = Bundle.main.path(
            forResource: "holistic_landmarker",
            ofType: "task"
        ) else { return }

        let options = HolisticLandmarkerOptions()
        options.baseOptions.modelAssetPath = modelPath
        options.runningMode = .liveStream
        options.minFaceDetectionConfidence = 0.5
        options.minPoseDetectionConfidence = 0.5
        options.holisticLandmarkerLiveStreamDelegate = self

        holisticLandmarker = try HolisticLandmarker(options: options)
    }

    func detectAsync(mpImage: MPImage, timestampMs: Int) throws {
        // Async: automatically drops incoming camera frames when busy so results stay current
        try holisticLandmarker?.detectAsync(image: mpImage, timestampInMilliseconds: timestampMs)
    }
}

extension HolisticLandmarkerStreamService: HolisticLandmarkerLiveStreamDelegate {
    func holisticLandmarker(
        _ holisticLandmarker: HolisticLandmarker,
        didFinishDetection result: HolisticLandmarkerResult?,
        timestampInMilliseconds: Int,
        error: (any Error)?
    ) {
        guard let result else { return }
        DispatchQueue.main.async {
            let pose = result.poseLandmarks
        }
    }
}

Python

pippip install mediapipeSample code·Setup guide

Image

IMAGEStateless · single frameSynchronous call on independent still images. Resets temporal tracking state on every call.

import mediapipe as mp

BaseOptions = mp.tasks.BaseOptions
HolisticLandmarker = mp.tasks.vision.HolisticLandmarker
HolisticLandmarkerOptions = mp.tasks.vision.HolisticLandmarkerOptions
VisionRunningMode = mp.tasks.vision.RunningMode

options = HolisticLandmarkerOptions(
    base_options=BaseOptions(model_asset_path="holistic_landmarker.task"),
    running_mode=VisionRunningMode.IMAGE,
    min_face_detection_confidence=0.5,
    min_pose_detection_confidence=0.5,
)

with HolisticLandmarker.create_from_options(options) as landmarker:
    # Synchronous single-image call (stateless across images)
    mp_image = mp.Image.create_from_file("input.jpg")
    result = landmarker.detect(mp_image)

Video

VIDEOEvery frame · no dropsSynchronous tracking across decoded video frames with monotonic timestamps. Processes every frame in order and never drops frames.

import mediapipe as mp

BaseOptions = mp.tasks.BaseOptions
HolisticLandmarker = mp.tasks.vision.HolisticLandmarker
HolisticLandmarkerOptions = mp.tasks.vision.HolisticLandmarkerOptions
VisionRunningMode = mp.tasks.vision.RunningMode

options = HolisticLandmarkerOptions(
    base_options=BaseOptions(model_asset_path="holistic_landmarker.task"),
    running_mode=VisionRunningMode.VIDEO,
    min_face_detection_confidence=0.5,
    min_pose_detection_confidence=0.5,
)

with HolisticLandmarker.create_from_options(options) as landmarker:
    # Synchronous: processes every frame in order with monotonic timestamps (no frame drops)
    for mp_image, timestamp_ms in decoded_video_frames:
        result = landmarker.detect_for_video(mp_image, timestamp_ms)

Livestream

LIVE_STREAMAsync · drops frames when busyAsynchronous camera pipeline with a result callback. Automatically drops incoming frames when busy so results stay current with low latency.

import mediapipe as mp

BaseOptions = mp.tasks.BaseOptions
HolisticLandmarker = mp.tasks.vision.HolisticLandmarker
HolisticLandmarkerOptions = mp.tasks.vision.HolisticLandmarkerOptions
VisionRunningMode = mp.tasks.vision.RunningMode

def on_result(result, output_image: mp.Image, timestamp_ms: int):
    print(f"Pose landmarks at {timestamp_ms} ms: {len(result.pose_landmarks)}")

options = HolisticLandmarkerOptions(
    base_options=BaseOptions(model_asset_path="holistic_landmarker.task"),
    running_mode=VisionRunningMode.LIVE_STREAM,
    min_face_detection_confidence=0.5,
    min_pose_detection_confidence=0.5,
    result_callback=on_result,
)

with HolisticLandmarker.create_from_options(options) as landmarker:
    # Async: automatically drops incoming camera frames when busy so results stay current
    mp_image = mp.Image(image_format=mp.ImageFormat.SRGB, data=rgb_frame)
    landmarker.detect_async(mp_image, timestamp_ms)

On Web, you can use DrawingUtils(ctx).drawConnectors(...) with PoseLandmarker.POSE_CONNECTIONS, HandLandmarker.HAND_CONNECTIONS, and FaceLandmarker.FACE_LANDMARKS_TESSELATION from @mediapipe/tasks-vision to render holistic overlays on an HTML <canvas>.

Task details

This section describes the capabilities, inputs, outputs, and configuration options of this task.

Features

• Input image processing - Processing includes image rotation, resizing, normalization, and color space conversion.
• Score thresholds - Filter results based on prediction and confidence scores.
• Optional outputs - Optional output of face blendshapes and pose segmentation masks.

Inputs

The Holistic Landmarker accepts an input of one of the following data types:

• Still images
• Decoded video frames
• Live video feed

See Prepare vision input data for MPImage, mp.Image, and browser media examples.

Outputs

The Holistic Landmarker outputs the following results:

• Face landmarks in normalized image coordinates (478 landmarks)
• Pose landmarks in normalized image coordinates and 3D world coordinates (33 landmarks)
• Left hand landmarks in normalized image coordinates and 3D world coordinates (21 landmarks)
• Right hand landmarks in normalized image coordinates and 3D world coordinates (21 landmarks)
• Optional: Face blendshapes scores
• Optional: A segmentation mask for the pose region

Configuration options

This task has the following configuration options:

Web

Option Name	Description	Value Range	Default Value
runningMode	Sets the running mode for the task: IMAGE: Single image inputs. VIDEO: Decoded video frames and live camera streams.	{IMAGE, VIDEO}	IMAGE
minFaceDetectionConfidence	The minimum confidence score for face detection to be considered successful.	Float [0.0, 1.0]	0.5
minFaceSuppressionThreshold	The minimum non-maximum-suppression threshold for overlapping face detections.	Float [0.0, 1.0]	0.3 (Python: 0.5)
minFacePresenceConfidence	The minimum confidence score for face presence in face landmark detection (min_face_landmarks_confidence in Python).	Float [0.0, 1.0]	0.5
minPoseDetectionConfidence	The minimum confidence score for pose detection to be considered successful.	Float [0.0, 1.0]	0.5
minPoseSuppressionThreshold	The minimum non-maximum-suppression threshold for overlapping pose detections.	Float [0.0, 1.0]	0.3 (Python: 0.5)
minPosePresenceConfidence	The minimum confidence score for pose presence in pose landmark detection (min_pose_landmarks_confidence in Python).	Float [0.0, 1.0]	0.5
minHandLandmarksConfidence	The minimum confidence score for hand presence in hand landmark detection.	Float [0.0, 1.0]	0.5
outputFaceBlendshapes	Whether to output face blendshape classifications for 3D facial expression rendering.	Boolean	false
outputPoseSegmentationMasks	Whether to output segmentation masks for the detected pose (output_segmentation_mask in Python).	Boolean	false

Android

Option Name	Description	Value Range	Default Value
runningMode	Sets the running mode for the task: IMAGE: Single image inputs. VIDEO: Decoded frames of a video. LIVE_STREAM: Livestream of camera input data (requires asynchronous listener/delegate).	{IMAGE, VIDEO, LIVE_STREAM}	IMAGE
minFaceDetectionConfidence	The minimum confidence score for face detection to be considered successful.	Float [0.0, 1.0]	0.5
minFaceSuppressionThreshold	The minimum non-maximum-suppression threshold for overlapping face detections.	Float [0.0, 1.0]	0.3 (Python: 0.5)
minFacePresenceConfidence	The minimum confidence score for face presence in face landmark detection (min_face_landmarks_confidence in Python).	Float [0.0, 1.0]	0.5
minPoseDetectionConfidence	The minimum confidence score for pose detection to be considered successful.	Float [0.0, 1.0]	0.5
minPoseSuppressionThreshold	The minimum non-maximum-suppression threshold for overlapping pose detections.	Float [0.0, 1.0]	0.3 (Python: 0.5)
minPosePresenceConfidence	The minimum confidence score for pose presence in pose landmark detection (min_pose_landmarks_confidence in Python).	Float [0.0, 1.0]	0.5
minHandLandmarksConfidence	The minimum confidence score for hand presence in hand landmark detection.	Float [0.0, 1.0]	0.5
outputFaceBlendshapes	Whether to output face blendshape classifications for 3D facial expression rendering.	Boolean	false
outputPoseSegmentationMasks	Whether to output segmentation masks for the detected pose (output_segmentation_mask in Python).	Boolean	false
resultListener	Asynchronous callback for LIVE_STREAM mode results.	OutputHandler.ResultListener<HolisticLandmarkerResult, MPImage>	Not set
errorListener	Optional error listener for LIVE_STREAM mode.	ErrorListener	Not set

iOS

Option Name	Description	Value Range	Default Value
runningMode	Sets the running mode for the task: IMAGE: Single image inputs. VIDEO: Decoded frames of a video. LIVE_STREAM: Livestream of camera input data (requires asynchronous listener/delegate).	{IMAGE, VIDEO, LIVE_STREAM}	IMAGE
minFaceDetectionConfidence	The minimum confidence score for face detection to be considered successful.	Float [0.0, 1.0]	0.5
minFaceSuppressionThreshold	The minimum non-maximum-suppression threshold for overlapping face detections.	Float [0.0, 1.0]	0.3 (Python: 0.5)
minFacePresenceConfidence	The minimum confidence score for face presence in face landmark detection (min_face_landmarks_confidence in Python).	Float [0.0, 1.0]	0.5
minPoseDetectionConfidence	The minimum confidence score for pose detection to be considered successful.	Float [0.0, 1.0]	0.5
minPoseSuppressionThreshold	The minimum non-maximum-suppression threshold for overlapping pose detections.	Float [0.0, 1.0]	0.3 (Python: 0.5)
minPosePresenceConfidence	The minimum confidence score for pose presence in pose landmark detection (min_pose_landmarks_confidence in Python).	Float [0.0, 1.0]	0.5
minHandLandmarksConfidence	The minimum confidence score for hand presence in hand landmark detection.	Float [0.0, 1.0]	0.5
outputFaceBlendshapes	Whether to output face blendshape classifications for 3D facial expression rendering.	Boolean	false
outputPoseSegmentationMasks	Whether to output segmentation masks for the detected pose (output_segmentation_mask in Python).	Boolean	false
holisticLandmarkerLiveStreamDelegate	Asynchronous delegate for LIVE_STREAM mode results (invoked on a serial background queue).	HolisticLandmarkerLiveStreamDelegate	Not set

Python

Option Name	Description	Value Range	Default Value
running_mode	Sets the running mode for the task: IMAGE: Single image inputs. VIDEO: Decoded frames of a video. LIVE_STREAM: Livestream of camera input data (requires result_callback).	{IMAGE, VIDEO, LIVE_STREAM}	IMAGE
min_face_detection_confidence	The minimum confidence score for face detection to be considered successful.	Float [0.0, 1.0]	0.5
min_face_suppression_threshold	The minimum non-maximum-suppression threshold for overlapping face detections.	Float [0.0, 1.0]	0.5
min_face_landmarks_confidence	The minimum confidence score for face presence in face landmark detection.	Float [0.0, 1.0]	0.5
min_pose_detection_confidence	The minimum confidence score for pose detection to be considered successful.	Float [0.0, 1.0]	0.5
min_pose_suppression_threshold	The minimum non-maximum-suppression threshold for overlapping pose detections.	Float [0.0, 1.0]	0.5
min_pose_landmarks_confidence	The minimum confidence score for pose presence in pose landmark detection.	Float [0.0, 1.0]	0.5
min_hand_landmarks_confidence	The minimum confidence score for hand presence in hand landmark detection.	Float [0.0, 1.0]	0.5
output_face_blendshapes	Whether to output face blendshape classifications for 3D facial expression rendering.	Boolean	False
output_segmentation_mask	Whether to output a segmentation mask for the detected pose.	Boolean	False
result_callback	Asynchronous callback function for LIVE_STREAM mode results.	Callable	Not set

Task output

Each HolisticLandmarkerResult returns synchronized face, pose, and left/right hand landmarks from a single unified pipeline:

Field	Type / Coordinate Space	Description
faceLandmarks	NormalizedLandmark[] (x, y in [0.0, 1.0])	478 3D facial landmarks normalized by image width and height, with z relative to the center of the head (face_landmarks in Python).
poseLandmarks	NormalizedLandmark[] (x, y in [0.0, 1.0])	33 pose landmarks normalized by image width and height, with z relative to the midpoint of the hips (pose_landmarks in Python).
poseWorldLandmarks	Landmark[] (meters)	33 3D pose landmarks in real-world meters with the origin at the midpoint of the hips (pose_world_landmarks in Python).
leftHandLandmarks	NormalizedLandmark[] (x, y in [0.0, 1.0])	21 left-hand 3D landmarks normalized by image width and height (left_hand_landmarks in Python).
leftHandWorldLandmarks	Landmark[] (meters)	21 left-hand 3D landmarks in real-world meters (left_hand_world_landmarks in Python).
rightHandLandmarks	NormalizedLandmark[] (x, y in [0.0, 1.0])	21 right-hand 3D landmarks normalized by image width and height (right_hand_landmarks in Python).
rightHandWorldLandmarks	Landmark[] (meters)	21 right-hand 3D landmarks in real-world meters (right_hand_world_landmarks in Python).
faceBlendshapes	Classifications (score in [0.0, 1.0])	Optional 52 facial expression coefficients when outputFaceBlendshapes is enabled (face_blendshapes in Python).
segmentationMask	MPMask (Float32 in [0.0, 1.0])	Optional single-channel float pose confidence mask when outputPoseSegmentationMasks is enabled (segmentation_mask in Python).

Example HolisticLandmarkerResult payload

The following example shows synchronized face, pose, and hand landmarks for a single subject:

{
  faceLandmarks: [
    { x: 0.4980, y: 0.3120, z: -0.0240 }, // #0..#477 (478 normalized face landmarks)
  ],
  poseLandmarks: [
    { x: 0.5010, y: 0.2640, z: -0.4120, visibility: 0.9980, presence: 0.9990 }, // #0 nose (33 pose landmarks)
  ],
  poseWorldLandmarks: [
    { x: 0.0120, y: -0.5480, z: -0.2850, visibility: 0.9980, presence: 0.9990 }, // #0 nose (3D meters)
  ],
  leftHandLandmarks: [
    { x: 0.6840, y: 0.6120, z: 0.0 }, // #0 WRIST (21 left hand landmarks)
  ],
  rightHandLandmarks: [
    { x: 0.3180, y: 0.6050, z: 0.0 }, // #0 WRIST (21 right hand landmarks)
  ]
}

Models

The Holistic Landmarker uses a series of packaged models to perform holistic landmarks detection. These models include face detection, facial landmarks, pose detection, pose landmarker, palm detection, and hand landmark models.

The following models are packaged together into a downloadable model bundle:

• Pose detection and landmark model: tracks 33 body pose coordinates.
• Face detection and landmark model: detects and tracks 478 3D facial landmarks, including 10 iris landmarks with optional 52 blendshape coefficients.
• Palm detection and hand landmark model: tracks 21 knuckle coordinates per hand.

Model bundle	Data type	Model Cards	Versions
Holistic landmarker	float 16	BlazePose FaceLandmarker-V2 HandLandmarker	Latest

Landmark coordinates

The Holistic Landmarker outputs a total of 553 landmarks that represent the full body:

• Pose landmarks: 33 landmarks representing key body coordinate points (same as the Pose Landmarker).
• Face landmarks: 478 landmarks representing the facial structure including 10 iris landmarks (same as the Face Landmarker).
• Hand landmarks: 21 landmarks per hand (42 total) representing knuckle coordinates (same as the Hand Landmarker).