diff --git a/sdk/android/api/org/webrtc/VideoFrame.java b/sdk/android/api/org/webrtc/VideoFrame.java
index 3e8a1e4..3b73543 100644
--- a/sdk/android/api/org/webrtc/VideoFrame.java
+++ b/sdk/android/api/org/webrtc/VideoFrame.java
@@ -177,6 +177,7 @@
     buffer.retain();
   }
 
+  @CalledByNative
   public void release() {
     buffer.release();
   }
diff --git a/sdk/android/native_api/video/wrapper.cc b/sdk/android/native_api/video/wrapper.cc
index 0f1bcaf..cc626c8 100644
--- a/sdk/android/native_api/video/wrapper.cc
+++ b/sdk/android/native_api/video/wrapper.cc
@@ -26,7 +26,7 @@
 
 ScopedJavaLocalRef<jobject> NativeToJavaVideoFrame(JNIEnv* jni,
                                                    const VideoFrame& frame) {
-  return jni::NativeToJavaFrame(jni, frame);
+  return jni::NativeToJavaVideoFrame(jni, frame);
 }
 
 }  // namespace webrtc
diff --git a/sdk/android/native_api/video/wrapper.h b/sdk/android/native_api/video/wrapper.h
index 35c305c..fc670ea 100644
--- a/sdk/android/native_api/video/wrapper.h
+++ b/sdk/android/native_api/video/wrapper.h
@@ -26,7 +26,8 @@
     JNIEnv* jni,
     jobject video_sink);
 
-// Creates a Java VideoFrame object from a native VideoFrame.
+// Creates a Java VideoFrame object from a native VideoFrame. The returned
+// object has to be released by calling release.
 ScopedJavaLocalRef<jobject> NativeToJavaVideoFrame(JNIEnv* jni,
                                                    const VideoFrame& frame);
 
diff --git a/sdk/android/src/jni/androidmediaencoder.cc b/sdk/android/src/jni/androidmediaencoder.cc
index f71b508..7f835f7 100644
--- a/sdk/android/src/jni/androidmediaencoder.cc
+++ b/sdk/android/src/jni/androidmediaencoder.cc
@@ -728,11 +728,14 @@
       case AndroidVideoFrameBuffer::AndroidType::kTextureBuffer:
         encode_status = EncodeTexture(jni, key_frame, input_frame);
         break;
-      case AndroidVideoFrameBuffer::AndroidType::kJavaBuffer:
+      case AndroidVideoFrameBuffer::AndroidType::kJavaBuffer: {
+        ScopedJavaLocalRef<jobject> j_frame =
+            NativeToJavaVideoFrame(jni, frame);
         encode_status =
-            EncodeJavaFrame(jni, key_frame, NativeToJavaFrame(jni, input_frame),
-                            j_input_buffer_index);
+            EncodeJavaFrame(jni, key_frame, j_frame, j_input_buffer_index);
+        ReleaseJavaVideoFrame(jni, j_frame);
         break;
+      }
       default:
         RTC_NOTREACHED();
         return WEBRTC_VIDEO_CODEC_ERROR;
diff --git a/sdk/android/src/jni/videoencoderwrapper.cc b/sdk/android/src/jni/videoencoderwrapper.cc
index 5ef30c1..ef0cdb4 100644
--- a/sdk/android/src/jni/videoencoderwrapper.cc
+++ b/sdk/android/src/jni/videoencoderwrapper.cc
@@ -128,8 +128,10 @@
   info.timestamp_rtp = frame.timestamp();
   frame_extra_infos_.push_back(info);
 
-  ScopedJavaLocalRef<jobject> ret = Java_VideoEncoder_encode(
-      jni, encoder_, NativeToJavaFrame(jni, frame), encode_info);
+  ScopedJavaLocalRef<jobject> j_frame = NativeToJavaVideoFrame(jni, frame);
+  ScopedJavaLocalRef<jobject> ret =
+      Java_VideoEncoder_encode(jni, encoder_, j_frame, encode_info);
+  ReleaseJavaVideoFrame(jni, j_frame);
   return HandleReturnCode(jni, ret);
 }
 
diff --git a/sdk/android/src/jni/videoframe.cc b/sdk/android/src/jni/videoframe.cc
index f638e19..22867e4 100644
--- a/sdk/android/src/jni/videoframe.cc
+++ b/sdk/android/src/jni/videoframe.cc
@@ -383,8 +383,8 @@
          AndroidVideoFrameBuffer::AndroidType::kJavaBuffer;
 }
 
-ScopedJavaLocalRef<jobject> NativeToJavaFrame(JNIEnv* jni,
-                                              const VideoFrame& frame) {
+ScopedJavaLocalRef<jobject> NativeToJavaVideoFrame(JNIEnv* jni,
+                                                   const VideoFrame& frame) {
   rtc::scoped_refptr<VideoFrameBuffer> buffer = frame.video_frame_buffer();
 
   if (IsJavaVideoBuffer(buffer)) {
@@ -396,9 +396,11 @@
     AndroidVideoBuffer* android_video_buffer =
         static_cast<AndroidVideoBuffer*>(android_buffer);
 
+    ScopedJavaLocalRef<jobject> j_video_frame_buffer(
+        jni, android_video_buffer->video_frame_buffer());
+    Java_Buffer_retain(jni, j_video_frame_buffer);
     return Java_VideoFrame_Constructor(
-        jni, android_video_buffer->video_frame_buffer(),
-        static_cast<jint>(frame.rotation()),
+        jni, j_video_frame_buffer, static_cast<jint>(frame.rotation()),
         static_cast<jlong>(frame.timestamp_us() *
                            rtc::kNumNanosecsPerMicrosec));
   } else {
@@ -410,6 +412,10 @@
   }
 }
 
+void ReleaseJavaVideoFrame(JNIEnv* jni, const JavaRef<jobject>& j_video_frame) {
+  Java_VideoFrame_release(jni, j_video_frame);
+}
+
 static void JNI_VideoFrame_CropAndScaleI420(
     JNIEnv* jni,
     const JavaParamRef<jclass>&,
diff --git a/sdk/android/src/jni/videoframe.h b/sdk/android/src/jni/videoframe.h
index 5cb7a2d..6f5720d 100644
--- a/sdk/android/src/jni/videoframe.h
+++ b/sdk/android/src/jni/videoframe.h
@@ -155,8 +155,11 @@
                              const JavaRef<jobject>& j_video_frame,
                              uint32_t timestamp_rtp);
 
-ScopedJavaLocalRef<jobject> NativeToJavaFrame(JNIEnv* jni,
-                                              const VideoFrame& frame);
+// NOTE: Returns a new video frame that has to be released by calling
+// ReleaseJavaVideoFrame.
+ScopedJavaLocalRef<jobject> NativeToJavaVideoFrame(JNIEnv* jni,
+                                                   const VideoFrame& frame);
+void ReleaseJavaVideoFrame(JNIEnv* jni, const JavaRef<jobject>& j_video_frame);
 
 int64_t GetJavaVideoFrameTimestampNs(JNIEnv* jni,
                                      const JavaRef<jobject>& j_video_frame);
diff --git a/sdk/android/src/jni/videosink.cc b/sdk/android/src/jni/videosink.cc
index 89d6d2e..bbdda31 100644
--- a/sdk/android/src/jni/videosink.cc
+++ b/sdk/android/src/jni/videosink.cc
@@ -23,7 +23,9 @@
 
 void VideoSinkWrapper::OnFrame(const VideoFrame& frame) {
   JNIEnv* jni = AttachCurrentThreadIfNeeded();
-  Java_VideoSink_onFrame(jni, j_sink_, NativeToJavaFrame(jni, frame));
+  ScopedJavaLocalRef<jobject> j_frame = NativeToJavaVideoFrame(jni, frame);
+  Java_VideoSink_onFrame(jni, j_sink_, j_frame);
+  ReleaseJavaVideoFrame(jni, j_frame);
 }
 
 }  // namespace jni