diff --git a/adafruit_is31fl3731/__init__.py b/adafruit_is31fl3731/__init__.py
index 2d65f4e..55b53df 100644
--- a/adafruit_is31fl3731/__init__.py
+++ b/adafruit_is31fl3731/__init__.py
@@ -92,11 +92,11 @@ class IS31FL3731:
     width = 16
     height = 9
 
-    def __init__(self, i2c, address=0x74):
+    def __init__(self, i2c, address=0x74, frames=None):
         self.i2c = i2c
         self.address = address
         self._frame = None
-        self._init()
+        self._init(frames=frames)
 
     def _i2c_read_reg(self, reg, result):
         # Read a buffer of data from the specified 8-bit I2C register address.
@@ -111,19 +111,21 @@ def _i2c_read_reg(self, reg, result):
             self.i2c.unlock()
         return None
 
-    def _i2c_write_reg(self, reg, data):
-        # Write a buffer of data (byte array) to the specified I2C register
-        # address.
+    def _i2c_write_block(self, data):
+        # Writes a contiguous block of data (bytearray) where the first byte
+        # is the starting I2C register address (register is not an argument).
         while not self.i2c.try_lock():
             pass
         try:
-            buf = bytearray(1)
-            buf[0] = reg
-            buf.extend(data)
-            self.i2c.writeto(self.address, buf)
+            self.i2c.writeto(self.address, data)
         finally:
             self.i2c.unlock()
 
+    def _i2c_write_reg(self, reg, data):
+        # Write a contiguous block of data (bytearray) starting at the
+        # specified I2C register address (register passed as argument).
+        self._i2c_write_block(bytes([reg]) + data)
+
     def _bank(self, bank=None):
         if bank is None:
             result = bytearray(1)
@@ -142,16 +144,21 @@ def _register(self, bank, register, value=None):
     def _mode(self, mode=None):
         return self._register(_CONFIG_BANK, _MODE_REGISTER, mode)
 
-    def _init(self):
+    def _init(self, frames=None):
         self.sleep(True)
-        time.sleep(0.01)  # 10 MS pause to reset.
-        self._mode(_PICTURE_MODE)
-        self.frame(0)
-        for frame in range(8):
-            self.fill(0, False, frame=frame)
-            for col in range(18):
-                self._register(frame, _ENABLE_OFFSET + col, 0xFF)
-        self.audio_sync(False)
+        # Clear config; sets to Picture Mode, no audio sync, maintains sleep
+        self._bank(_CONFIG_BANK)
+        self._i2c_write_block(bytes([0] * 14))
+        enable_data = bytes([_ENABLE_OFFSET] + [255] * 18)
+        fill_data = bytearray([0] * 25)
+        # Initialize requested frames, or all 8 if unspecified
+        for frame in frames if frames else range(8):
+            self._bank(frame)
+            self._i2c_write_block(enable_data)  # Set all enable bits
+            for row in range(6):  # Barebones quick fill() w/0
+                fill_data[0] = _COLOR_OFFSET + row * 24
+                self._i2c_write_block(fill_data)
+        self._frame = 0  # To match config bytes above
         self.sleep(False)
 
     def reset(self):
diff --git a/adafruit_is31fl3731/matrix.py b/adafruit_is31fl3731/matrix.py
index 94a2c9e..93b3034 100644
--- a/adafruit_is31fl3731/matrix.py
+++ b/adafruit_is31fl3731/matrix.py
@@ -40,3 +40,40 @@ class Matrix(IS31FL3731):
     def pixel_addr(x, y):
         """Calulate the offset into the device array for x,y pixel"""
         return x + y * 16
+
+    # This takes precedence over image() in __init__ and is tuned for the
+    # Matrix class. Some shortcuts can be taken because matrix layout is
+    # very straightforward, and a few large write operations are used
+    # rather than pixel-by-pixel writes, yielding significant speed gains
+    # for animation. Buffering the full matrix for a quick write is not a
+    # memory concern here, as by definition this method is used with PIL
+    # images; we're not running on a RAM-constrained microcontroller.
+    def image(self, img, blink=None, frame=None):
+        """Set buffer to value of Python Imaging Library image.
+        The image should be in 8-bit mode (L) and a size equal to the
+        display size.
+
+        :param img: Python Imaging Library image
+        :param blink: True to blink
+        :param frame: the frame to set the image
+        """
+        if img.mode != "L":
+            raise ValueError("Image must be in mode L.")
+        if img.size[0] != self.width or img.size[1] != self.height:
+            raise ValueError(
+                "Image must be same dimensions as display ({0}x{1}).".format(
+                    self.width, self.height
+                )
+            )
+
+        # Frame-select and then write pixel data in one big operation
+        if frame is not None:
+            self._bank(frame)
+        # We can safely reduce the image to a "flat" byte sequence because
+        # the matrix layout is known linear; no need to go through a 2D
+        # pixel array or invoke pixel_addr().
+        self._i2c_write_block(bytes([0x24]) + img.tobytes())
+        # Set or clear blink state if requested, for all pixels at once
+        if blink is not None:
+            # 0x12 is _BLINK_OFFSET in __init__.py
+            self._i2c_write_block(bytes([0x12] + [1 if blink else 0] * 18))