diff --git a/src/enc/picture_tools_enc.c b/src/enc/picture_tools_enc.c
index be292d43..d0e8a495 100644
--- a/src/enc/picture_tools_enc.c
+++ b/src/enc/picture_tools_enc.c
@@ -16,10 +16,6 @@
 #include "src/enc/vp8i_enc.h"
 #include "src/dsp/yuv.h"
 
-static WEBP_INLINE uint32_t MakeARGB32(int r, int g, int b) {
-  return (0xff000000u | (r << 16) | (g << 8) | b);
-}
-
 //------------------------------------------------------------------------------
 // Helper: clean up fully transparent area to help compressibility.
 
@@ -195,6 +191,10 @@ void WebPCleanupTransparentAreaLossless(WebPPicture* const pic) {
 #define BLEND_10BIT(V0, V1, ALPHA) \
     ((((V0) * (1020 - (ALPHA)) + (V1) * (ALPHA)) * 0x101 + 1024) >> 18)
 
+static WEBP_INLINE uint32_t MakeARGB32(int r, int g, int b) {
+  return (0xff000000u | (r << 16) | (g << 8) | b);
+}
+
 void WebPBlendAlpha(WebPPicture* pic, uint32_t background_rgb) {
   const int red = (background_rgb >> 16) & 0xff;
   const int green = (background_rgb >> 8) & 0xff;
@@ -208,39 +208,44 @@ void WebPBlendAlpha(WebPPicture* pic, uint32_t background_rgb) {
     const int U0 = VP8RGBToU(4 * red, 4 * green, 4 * blue, 4 * YUV_HALF);
     const int V0 = VP8RGBToV(4 * red, 4 * green, 4 * blue, 4 * YUV_HALF);
     const int has_alpha = pic->colorspace & WEBP_CSP_ALPHA_BIT;
-    if (!has_alpha || pic->a == NULL) return;    // nothing to do
+    uint8_t* y_ptr = pic->y;
+    uint8_t* u_ptr = pic->u;
+    uint8_t* v_ptr = pic->v;
+    uint8_t* a_ptr = pic->a;
+    if (!has_alpha || a_ptr == NULL) return;    // nothing to do
     for (y = 0; y < pic->height; ++y) {
       // Luma blending
-      uint8_t* const y_ptr = pic->y + y * pic->y_stride;
-      uint8_t* const a_ptr = pic->a + y * pic->a_stride;
       for (x = 0; x < pic->width; ++x) {
-        const int alpha = a_ptr[x];
+        const uint8_t alpha = a_ptr[x];
         if (alpha < 0xff) {
-          y_ptr[x] = BLEND(Y0, y_ptr[x], a_ptr[x]);
+          y_ptr[x] = BLEND(Y0, y_ptr[x], alpha);
         }
       }
       // Chroma blending every even line
       if ((y & 1) == 0) {
-        uint8_t* const u = pic->u + (y >> 1) * pic->uv_stride;
-        uint8_t* const v = pic->v + (y >> 1) * pic->uv_stride;
         uint8_t* const a_ptr2 =
             (y + 1 == pic->height) ? a_ptr : a_ptr + pic->a_stride;
         for (x = 0; x < uv_width; ++x) {
           // Average four alpha values into a single blending weight.
           // TODO(skal): might lead to visible contouring. Can we do better?
-          const int alpha =
+          const uint32_t alpha =
               a_ptr[2 * x + 0] + a_ptr[2 * x + 1] +
               a_ptr2[2 * x + 0] + a_ptr2[2 * x + 1];
-          u[x] = BLEND_10BIT(U0, u[x], alpha);
-          v[x] = BLEND_10BIT(V0, v[x], alpha);
+          u_ptr[x] = BLEND_10BIT(U0, u_ptr[x], alpha);
+          v_ptr[x] = BLEND_10BIT(V0, v_ptr[x], alpha);
         }
         if (pic->width & 1) {   // rightmost pixel
-          const int alpha = 2 * (a_ptr[2 * x + 0] + a_ptr2[2 * x + 0]);
-          u[x] = BLEND_10BIT(U0, u[x], alpha);
-          v[x] = BLEND_10BIT(V0, v[x], alpha);
+          const uint32_t alpha = 2 * (a_ptr[2 * x + 0] + a_ptr2[2 * x + 0]);
+          u_ptr[x] = BLEND_10BIT(U0, u_ptr[x], alpha);
+          v_ptr[x] = BLEND_10BIT(V0, v_ptr[x], alpha);
         }
+      } else {
+        u_ptr += pic->uv_stride;
+        v_ptr += pic->uv_stride;
       }
-      memset(a_ptr, 0xff, pic->width);
+      memset(a_ptr, 0xff, pic->width);  // reset alpha value to opaque
+      a_ptr += pic->a_stride;
+      y_ptr += pic->y_stride;
     }
   } else {
     uint32_t* argb = pic->argb;