NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI
PyPI, #2623 most downloaded on PyPI
MediaPipe is the simplest way for researchers and developers to build world-class ML solutions and applications for mobile, edge, cloud and the web. See the privacy notice at https://goo.gle/mediapipe-privacy.
Latest stable
1.2.2
Last release 19 days ago
03 Nov 2026
Ships unpredictably
gaps range from 8 days to 10 months
Most releases are documented
notes for 20 of 23 stable releases
Nothing withdrawn
no release was ever pulled
3 years old
23 releases · first in 2023
Drag across the ruler to compare, or tap a release to open it.
Fix the tensor shape for column-major matrices in ConvertMatrixToTensorOnCpu.
'anonymous' instead of 'Anonymous' for HTMLImageElement.crossOrigin.HandLandmarkerResult and GestureRecognizerResult documentation across C, C++, iOS, Java, Python, and Web that Left and Right handedness correspond to the subject's hands rather than the viewer's, and that passing a horizontally mirrored image inverts these outputs.TaskRunnerOptions::memory_manager to register a shared MemoryManager as kMemoryManagerService on the task's CalculatorGraph before initialization.This section should highlight the changes that are done specifically for any platform and don't propagate to
other platforms.
Your coding agent can read these notes before it upgrades. Set up the MCP server →
Bump MediaPipe version to 1.2.0
This section should highlight the changes that are done specifically for any platform and don't propagate to
other platforms.
PrivacyInfo.xcprivacy) to MediaPipe Tasks iOS Swift Package Manager package.Bump MediaPipe version to 1.1.1
This section should highlight the changes that are done specifically for any platform and don't propagate to
other platforms.
Fix deprecated CreateSourceTexture usage in desktop GPU example.
resources.bzl in mediapipe/metal.bzl, using the metal rule provided by Bazel Apple rules.WORKSPACE.kissfft in build configuration.xla llvm download URL by using mediapipe-assets mirror.unistd.h header imports.TaskPart and EmbeddingProvider interface.TextPart, AudioPart, and ImagePart into tasks.core and unify duplicates.TextEmbedder and ImageEmbedder to use executor abstractions (TextEmbedderExecutor, ImageEmbedderExecutor).HasFile, GetBundleManifest, and GetBundleId helpers to ModelAssetBundleResources.noexcept to move ctor/operator and add swap to Packet.ABSL_ASSIGN_OR_RETURN and ABSL_RETURN_IF_ERROR in MediaPipe.GeckoPreprocessorCalculator to SentencePiecePreprocessorCalculator.encoded_media_bytes in UnpackMediaSequenceCalculator.use_litert option in MediaPipe BaseOptions.LiteRtSystemHandles to Environment in LiteRT Inference Calculator.MTLCommandQueue directly instead of DrishtiMetalHelper in InferenceRunnerLiteRt.InferenceRunnerLiteRt.TensorsToDetectionsCalculator.GpuResources.CreateSourceTexture usage in desktop GPU example.This section highlights the changes that are done specifically for each platform and don't propagate to other platforms.
AHardwareBuffer (AHWB) output from the GPU.AHardwareBuffer synchronization when converting between write and read views.AHardwareBuffer (kRGBA32).MediaPipeException instead of crashing in surface_output_jni on Android EGL surface creation failure.PathToResourceAsFile when TEST_SRCDIR is not set on Android.LITERT delegate flag from Android API.MediaPipeTasksRetrieval).MPPAudioRecord.h to MediaPipeTasksRetrieval XCFramework headers._Mp*) symbols from the iOS symbol rename map.inference_calculator_litert dependencies to iOS Tasks targets.NSError instead of crashing when universal embedder model path is invalid.UniversalEmbedder, SemanticRetriever).SemanticRetriever.DecisionMaker web API usable from the minified decision_js bundle.UniversalEmbedder on Web.UniversalEmbedder on Web.ImageFrame and YUVImage conversions.await calls to UniversalEmbedder example in web retrieval documentation.AudioEmbedder.cosine_similarity utility method to UniversalEmbedder.cache_dir option and default to :nocache.HolisticLandmarker crash when face detections are missing.HolisticLandmarker Any Builder return type and empty packet MediaPipeExceptions.ImageEmbedder.close() method.numpy_view() SIGABRT on non-contiguous float32 masks.HAND_PALM_CONNECTIONS.LandmarksSmoothingCalculator (OneEuroFilter) when stream resolution dynamically changes between preview and still capture.audio_decoder_calculator_test.WORKSPACE.Nothing published for this version
Replace ITUIOSTestUtil with ITUTestRunEnvironment in iOS tests.
vImageExtractChannelAndConvertToFloat function.This section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
Add histogram information to Tensor::DebugString.
This section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
tasks/core deps[mediapipe] mark mediapipe/gpu:gl_context incompatible on Windows
This section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
Enables ml drift metal delegate as inference calculator backend.
fp16 activations in tests.xnnpack version.xnn_reduce_mean_squared reduction for the RMSNorm.ImageProcessingOptions support to FaceLandmarker C API.ImageProcessingOptions support to Object Detector C API.ImageProcessingOptions support to MediaPipe PoseLandmarker C API.ImageProcessingOptions support to Image Segmenter C APIHandAssociationCalculatorTestMpStatus.has_confidence_masks field from ImageSegmenterResult.MpStatus.This section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
MatrixC ctypes structure for MediaPipe C APINothing published for this version
Nothing published for this version
Update tensors_to_image_calculator.cc
This section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
Add comments to explain how to configure OpenCV in the opencv_macos.BUILD file.
shadow_copy parameter to PathToResourceAsFileThis section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
toBuilder method to the LlmInferenceOptions object.Remove deprecated TfLiteModelLoader::LoadFromPath.
third_party/emscriptenReturnType helper, since it's part of the standard now.This section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
numOfSupportedLoraRanks parameter.AVAudioPCMBuffer+TestUtils, few basic iOS audio classifier tests, options tests to iOS audio classifier, utils for AVAudioFile, test for score threshold to MPPAudioClassifierTests, constants in MPPAudioClassifierTests, close method to iOS audio classifier, iOS MPPAudioData test utils, stream mode tests for iOS audio classifier, iOS audio classifier to cocoapods build, audio record creation tests to MPPAudioClassifierTests, close method to MPPAudioEmbedder, iOS audio embedder tests, more utility methods to MPPAudioEmbedderTests, streams mode tests for iOS audio embedder, iOS audio embedder to cocoapods build, comments to MPPAudioClassifierTests, iOS audio embedder header and implementation, iOS audio classifier implementation file, method for getting interleaved float32 pcm buffer from audio file, refactored iOS LlmTaskRunner, iOS LlmSessionRunner, more errors to GenAiInferenceError, refactored LlmInference, iOS session runner to build files, extra safeguards for response context in LlmSessionRunner, LlmInference+Session.swift and documentation regarding session and inference life times to iOS LLM Inference.LlmInference.Session.clone() method.mock with unittest.mock in model_maker tests.Remove deprecated TfLiteOperatorCreateWithData function
graph_service headers.PointToForeign with an optional cleanup object.BeginLoopCalculator for move-only types (e.g. Tensor) without Packet::Consume usage and copyable types without copying unless it's a fundamental type.Packet::Share() method in replacement of SharedPtrWithPacket() function.PacketGetter.Pair public.expected_seq_len param to ComputeLogits()This section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
Update C API documentation to make it clear that the callback is invoked multiple times
This section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
Deprecated ImageSource in favor of standard TexImageSource.
text_embedder_cosing_similarity signature to use Embedding pointersallow_cropping_outside_frame is setIsConnected() to graph builder SideOutThis section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
Updated genai C package visibility
addRawDataSpanToInputStream binding functions.This section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
LlmInferenceOptionsvisibility field in landmark.d.tsvisibility supportDeprecate ImageFrame::ByteDepth
This section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
Add libtext and libvision build rules
This section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
Allow Python to be build on Mac with GPU support
This section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
Fix win32 build break in mediapipe.
This section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
MPImage with source type UIImage from a C++ image.Fixed Pose Landmarker jittering issue
Add export_model_with_tokenizer to Text Classifier API.
Update WASM files for 0.10.7 release
Add a new from_image API to create face stylizer dataset from a single image. Also deprecate the from_folder API since we only support one-shot use ca…
This section should highlight the changes that are done specifically for any platform and don't propagate to other platforms.
Add tensorflow-addons to model_maker requirements.txt
Change to add the w_avg latent code to style encoding before layer swapping. This is a bug in the previous code. Also set training=True for encoder since this affect the encoding performance.
add metadata writer into face stylizer.
Refactor text_classifier preprocessor to move away from using classifier_data_lib
Import image_util for using it in mediapipe face stylizer open sourcing.
Fix image_util shortcut import line
Change supported_ops to a Tuple instead of List to match the API definition.
Add a new from_image API to create face stylizer dataset from a single image. Also deprecate the from_folder API since we only support one-shot use case now.
Add an API to run inference with face stylizer TF model.
Check if the image contains valid face that can be aligned for stylization. If not, throw an exception for invalid input image. This is applied to both input stylized face and raw face.
Add allow_custom_ops to model_util.convert_to_tflite and enable custom ops for face stylizer.
Update WASM files for 10.5 release
One column per quarter.