diff --git a/image.go b/image.go
index 1557e33d2..cd3f412f9 100644
--- a/image.go
+++ b/image.go
@@ -308,7 +308,7 @@ func (i *Image) drawImage(img *Image, options *DrawImageOptions) {
 		} else {
 			s = shareable.NewImage(w2, h2)
 		}
-		vs := src.QuadVertices(0, 0, w, h, 0.5, 0, 0, 0.5, 0, 0)
+		vs := src.QuadVertices(0, 0, w, h, 0.5, 0, 0, 0.5, 0, 0, 1, 1, 1, 1)
 		is := graphicsutil.QuadIndices()
 		s.DrawImage(src, vs, is, options.ColorM.impl, opengl.CompositeModeCopy, graphics.FilterLinear)
 		img.shareableImages = append(img.shareableImages, s)
@@ -318,9 +318,22 @@ func (i *Image) drawImage(img *Image, options *DrawImageOptions) {
 
 	if level < len(img.shareableImages) {
 		src := img.shareableImages[level]
-		vs := src.QuadVertices(sx0, sy0, sx1, sy1, a, b, c, d, tx, ty)
+		colorm := options.ColorM.impl
+		cr, cg, cb, ca := float32(1), float32(1), float32(1), float32(1)
+		if colorm.ScaleOnly() {
+			body, _ := colorm.UnsafeElements()
+			cr = body[0]
+			cg = body[5]
+			cb = body[10]
+			ca = body[15]
+		}
+		vs := src.QuadVertices(sx0, sy0, sx1, sy1, a, b, c, d, tx, ty, cr, cg, cb, ca)
 		is := graphicsutil.QuadIndices()
-		i.shareableImages[0].DrawImage(src, vs, is, options.ColorM.impl, mode, filter)
+
+		if colorm.ScaleOnly() {
+			colorm = nil
+		}
+		i.shareableImages[0].DrawImage(src, vs, is, colorm, mode, filter)
 	}
 	i.disposeMipmaps()
 }
diff --git a/internal/affine/colorm.go b/internal/affine/colorm.go
index 81d506d74..479affcea 100644
--- a/internal/affine/colorm.go
+++ b/internal/affine/colorm.go
@@ -63,6 +63,30 @@ func (c *ColorM) isInited() bool {
 	return c != nil && (c.body != nil || c.translate != nil)
 }
 
+func (c *ColorM) ScaleOnly() bool {
+	if c == nil {
+		return true
+	}
+	if c.body != nil {
+		for i, e := range c.body {
+			if i == 0 || i == 5 || i == 10 || i == 15 {
+				continue
+			}
+			if e != 0 {
+				return false
+			}
+		}
+	}
+	if c.translate != nil {
+		for _, e := range c.translate {
+			if e != 0 {
+				return false
+			}
+		}
+	}
+	return true
+}
+
 func (c *ColorM) Apply(clr color.Color) color.Color {
 	if !c.isInited() {
 		return clr
diff --git a/internal/graphics/program.go b/internal/graphics/program.go
index 330d2c5e6..61f999e68 100644
--- a/internal/graphics/program.go
+++ b/internal/graphics/program.go
@@ -94,6 +94,11 @@ var (
 				dataType: opengl.Float,
 				num:      4,
 			},
+			{
+				name:     "color_scale",
+				dataType: opengl.Float,
+				num:      4,
+			},
 		},
 	}
 )
diff --git a/internal/graphics/shader.go b/internal/graphics/shader.go
index 2fd0efa4a..d6543d848 100644
--- a/internal/graphics/shader.go
+++ b/internal/graphics/shader.go
@@ -50,14 +50,17 @@ const (
 uniform mat4 projection_matrix;
 attribute vec2 vertex;
 attribute vec4 tex_coord;
+attribute vec4 color_scale;
 varying vec2 varying_tex_coord;
 varying vec2 varying_tex_coord_min;
 varying vec2 varying_tex_coord_max;
+varying vec4 varying_color_scale;
 
 void main(void) {
   varying_tex_coord = vec2(tex_coord[0], tex_coord[1]);
   varying_tex_coord_min = vec2(min(tex_coord[0], tex_coord[2]), min(tex_coord[1], tex_coord[3]));
   varying_tex_coord_max = vec2(max(tex_coord[0], tex_coord[2]), max(tex_coord[1], tex_coord[3]));
+  varying_color_scale = color_scale;
   gl_Position = projection_matrix * vec4(vertex, 0, 1);
 }
 `
@@ -85,6 +88,7 @@ uniform highp float scale;
 varying highp vec2 varying_tex_coord;
 varying highp vec2 varying_tex_coord_min;
 varying highp vec2 varying_tex_coord_max;
+varying highp vec4 varying_color_scale;
 
 void main(void) {
   highp vec2 pos = varying_tex_coord;
@@ -147,8 +151,9 @@ void main(void) {
   if (0.0 < color.a) {
     color.rgb /= color.a;
   }
-  // Apply the color matrix
+  // Apply the color matrix or scale.
   color = (color_matrix_body * color) + color_matrix_translation;
+  color *= varying_color_scale;
   color = clamp(color, 0.0, 1.0);
   // Premultiply alpha
   color.rgb *= color.a;
diff --git a/internal/graphicsutil/vertices.go b/internal/graphicsutil/vertices.go
index 9eb8e5744..f57895212 100644
--- a/internal/graphicsutil/vertices.go
+++ b/internal/graphicsutil/vertices.go
@@ -47,7 +47,7 @@ func isPowerOf2(x int) bool {
 	return (x & (x - 1)) == 0
 }
 
-func QuadVertices(width, height int, sx0, sy0, sx1, sy1 int, a, b, c, d, tx, ty float32) []float32 {
+func QuadVertices(width, height int, sx0, sy0, sx1, sy1 int, a, b, c, d, tx, ty float32, cr, cg, cb, ca float32) []float32 {
 	if !isPowerOf2(width) {
 		panic("not reached")
 	}
@@ -65,13 +65,13 @@ func QuadVertices(width, height int, sx0, sy0, sx1, sy1 int, a, b, c, d, tx, ty
 	wf := float32(width)
 	hf := float32(height)
 	u0, v0, u1, v1 := float32(sx0)/wf, float32(sy0)/hf, float32(sx1)/wf, float32(sy1)/hf
-	return quadVerticesImpl(float32(sx1-sx0), float32(sy1-sy0), u0, v0, u1, v1, a, b, c, d, tx, ty)
+	return quadVerticesImpl(float32(sx1-sx0), float32(sy1-sy0), u0, v0, u1, v1, a, b, c, d, tx, ty, cr, cg, cb, ca)
 }
 
-func quadVerticesImpl(x, y, u0, v0, u1, v1, a, b, c, d, tx, ty float32) []float32 {
+func quadVerticesImpl(x, y, u0, v0, u1, v1, a, b, c, d, tx, ty, cr, cg, cb, ca float32) []float32 {
 	// Specifying a range explicitly here is redundant but this helps optimization
 	// to eliminate boundry checks.
-	vs := theVerticesBackend.sliceForOneQuad()[0:24]
+	vs := theVerticesBackend.sliceForOneQuad()[0:40]
 
 	ax, by, cx, dy := a*x, b*y, c*x, d*y
 
@@ -86,28 +86,44 @@ func quadVerticesImpl(x, y, u0, v0, u1, v1, a, b, c, d, tx, ty float32) []float3
 	vs[3] = v0
 	vs[4] = u1
 	vs[5] = v1
+	vs[6] = cr
+	vs[7] = cg
+	vs[8] = cb
+	vs[9] = ca
 
 	// and the same for the other three coordinates
-	vs[6] = ax + tx
-	vs[7] = cx + ty
-	vs[8] = u1
-	vs[9] = v0
-	vs[10] = u0
-	vs[11] = v1
-
-	vs[12] = by + tx
-	vs[13] = dy + ty
+	vs[10] = ax + tx
+	vs[11] = cx + ty
+	vs[12] = u1
+	vs[13] = v0
 	vs[14] = u0
 	vs[15] = v1
-	vs[16] = u1
-	vs[17] = v0
+	vs[16] = cr
+	vs[17] = cg
+	vs[18] = cb
+	vs[19] = ca
 
-	vs[18] = ax + by + tx
-	vs[19] = cx + dy + ty
-	vs[20] = u1
-	vs[21] = v1
+	vs[20] = by + tx
+	vs[21] = dy + ty
 	vs[22] = u0
-	vs[23] = v0
+	vs[23] = v1
+	vs[24] = u1
+	vs[25] = v0
+	vs[26] = cr
+	vs[27] = cg
+	vs[28] = cb
+	vs[29] = ca
+
+	vs[30] = ax + by + tx
+	vs[31] = cx + dy + ty
+	vs[32] = u1
+	vs[33] = v1
+	vs[34] = u0
+	vs[35] = v0
+	vs[36] = cr
+	vs[37] = cg
+	vs[38] = cb
+	vs[39] = ca
 
 	return vs
 }
diff --git a/internal/restorable/image.go b/internal/restorable/image.go
index 63651d431..d0b24870f 100644
--- a/internal/restorable/image.go
+++ b/internal/restorable/image.go
@@ -163,7 +163,8 @@ func (i *Image) ReplacePixels(pixels []byte, x, y, width, height int) {
 		colorm := (*affine.ColorM)(nil).Scale(0, 0, 0, 0)
 		vs := graphicsutil.QuadVertices(w, h, 0, 0, w, h,
 			float32(width)/float32(w), 0, 0, float32(height)/float32(h),
-			float32(x), float32(y))
+			float32(x), float32(y),
+			1, 1, 1, 1)
 		is := graphicsutil.QuadIndices()
 		i.image.DrawImage(dummyImage.image, vs, is, colorm, opengl.CompositeModeCopy, graphics.FilterNearest)
 	}
diff --git a/internal/restorable/images_test.go b/internal/restorable/images_test.go
index 4aefae02c..40fbb49e9 100644
--- a/internal/restorable/images_test.go
+++ b/internal/restorable/images_test.go
@@ -116,7 +116,7 @@ func TestRestoreChain(t *testing.T) {
 	fill(imgs[0], clr.R, clr.G, clr.B, clr.A)
 	for i := 0; i < num-1; i++ {
 		w, h := imgs[i].Size()
-		vs := graphicsutil.QuadVertices(w, h, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0)
+		vs := graphicsutil.QuadVertices(w, h, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 		is := graphicsutil.QuadIndices()
 		imgs[i+1].DrawImage(imgs[i], vs, is, nil, opengl.CompositeModeCopy, graphics.FilterNearest)
 	}
@@ -158,7 +158,7 @@ func TestRestoreChain2(t *testing.T) {
 	clr8 := color.RGBA{0x00, 0x00, 0xff, 0xff}
 	fill(imgs[8], clr8.R, clr8.G, clr8.B, clr8.A)
 
-	vs := graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 0, 0)
+	vs := graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	is := graphicsutil.QuadIndices()
 	imgs[8].DrawImage(imgs[7], vs, is, nil, opengl.CompositeModeCopy, graphics.FilterNearest)
 	imgs[9].DrawImage(imgs[8], vs, is, nil, opengl.CompositeModeCopy, graphics.FilterNearest)
@@ -204,7 +204,7 @@ func TestRestoreOverrideSource(t *testing.T) {
 	clr0 := color.RGBA{0x00, 0x00, 0x00, 0xff}
 	clr1 := color.RGBA{0x00, 0x00, 0x01, 0xff}
 	fill(img1, clr0.R, clr0.G, clr0.B, clr0.A)
-	vs := graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 0, 0)
+	vs := graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	is := graphicsutil.QuadIndices()
 	img2.DrawImage(img1, vs, is, nil, opengl.CompositeModeSourceOver, graphics.FilterNearest)
 	img3.DrawImage(img2, vs, is, nil, opengl.CompositeModeSourceOver, graphics.FilterNearest)
@@ -289,24 +289,24 @@ func TestRestoreComplexGraph(t *testing.T) {
 		img1.Dispose()
 		img0.Dispose()
 	}()
-	vs := graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 0, 0)
+	vs := graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	is := graphicsutil.QuadIndices()
 	img3.DrawImage(img0, vs, is, nil, opengl.CompositeModeSourceOver, graphics.FilterNearest)
-	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 1, 0)
+	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 1, 0, 1, 1, 1, 1)
 	img3.DrawImage(img1, vs, is, nil, opengl.CompositeModeSourceOver, graphics.FilterNearest)
-	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 1, 0)
+	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 1, 0, 1, 1, 1, 1)
 	img4.DrawImage(img1, vs, is, nil, opengl.CompositeModeSourceOver, graphics.FilterNearest)
-	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 2, 0)
+	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 2, 0, 1, 1, 1, 1)
 	img4.DrawImage(img2, vs, is, nil, opengl.CompositeModeSourceOver, graphics.FilterNearest)
-	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 0, 0)
+	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	img5.DrawImage(img3, vs, is, nil, opengl.CompositeModeSourceOver, graphics.FilterNearest)
-	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 0, 0)
+	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	img6.DrawImage(img3, vs, is, nil, opengl.CompositeModeSourceOver, graphics.FilterNearest)
-	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 1, 0)
+	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 1, 0, 1, 1, 1, 1)
 	img6.DrawImage(img4, vs, is, nil, opengl.CompositeModeSourceOver, graphics.FilterNearest)
-	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 0, 0)
+	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	img7.DrawImage(img2, vs, is, nil, opengl.CompositeModeSourceOver, graphics.FilterNearest)
-	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 2, 0)
+	vs = graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 2, 0, 1, 1, 1, 1)
 	img7.DrawImage(img3, vs, is, nil, opengl.CompositeModeSourceOver, graphics.FilterNearest)
 	ResolveStaleImages()
 	if err := Restore(); err != nil {
@@ -397,7 +397,7 @@ func TestRestoreRecursive(t *testing.T) {
 		img1.Dispose()
 		img0.Dispose()
 	}()
-	vs := graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 1, 0)
+	vs := graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 1, 0, 1, 1, 1, 1)
 	is := graphicsutil.QuadIndices()
 	img1.DrawImage(img0, vs, is, nil, opengl.CompositeModeSourceOver, graphics.FilterNearest)
 	img0.DrawImage(img1, vs, is, nil, opengl.CompositeModeSourceOver, graphics.FilterNearest)
@@ -485,7 +485,7 @@ func TestDrawImageAndReplacePixels(t *testing.T) {
 	img1 := NewImage(2, 1, false)
 	defer img1.Dispose()
 
-	vs := graphicsutil.QuadVertices(1, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0)
+	vs := graphicsutil.QuadVertices(1, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	is := graphicsutil.QuadIndices()
 	img1.DrawImage(img0, vs, is, nil, opengl.CompositeModeCopy, graphics.FilterNearest)
 	img1.ReplacePixels([]byte{0xff, 0xff, 0xff, 0xff}, 1, 0, 1, 1)
@@ -517,7 +517,7 @@ func TestDispose(t *testing.T) {
 	img2 := newImageFromImage(base2)
 	defer img2.Dispose()
 
-	vs := graphicsutil.QuadVertices(1, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0)
+	vs := graphicsutil.QuadVertices(1, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	is := graphicsutil.QuadIndices()
 	img1.DrawImage(img2, vs, is, nil, opengl.CompositeModeCopy, graphics.FilterNearest)
 	img0.DrawImage(img1, vs, is, nil, opengl.CompositeModeCopy, graphics.FilterNearest)
@@ -545,7 +545,7 @@ func TestDoubleResolve(t *testing.T) {
 	base.Pix[3] = 0xff
 	img1 := newImageFromImage(base)
 
-	vs := graphicsutil.QuadVertices(1, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0)
+	vs := graphicsutil.QuadVertices(1, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	is := graphicsutil.QuadIndices()
 	img0.DrawImage(img1, vs, is, nil, opengl.CompositeModeCopy, graphics.FilterNearest)
 	img0.ReplacePixels([]uint8{0x00, 0xff, 0x00, 0xff}, 1, 1, 1, 1)
diff --git a/internal/shareable/shareable.go b/internal/shareable/shareable.go
index d116b8c8e..03db8324e 100644
--- a/internal/shareable/shareable.go
+++ b/internal/shareable/shareable.go
@@ -63,7 +63,7 @@ func (b *backend) TryAlloc(width, height int) (*packing.Node, bool) {
 	newImg := restorable.NewImage(s, s, false)
 	oldImg := b.restorable
 	w, h := oldImg.Size()
-	vs := graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 0, 0)
+	vs := graphicsutil.QuadVertices(w, h, 0, 0, w, h, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	is := graphicsutil.QuadIndices()
 	newImg.DrawImage(oldImg, vs, is, nil, opengl.CompositeModeCopy, graphics.FilterNearest)
 	oldImg.Dispose()
@@ -129,7 +129,7 @@ func (i *Image) ensureNotShared() {
 	x, y, w, h := i.region()
 	newImg := restorable.NewImage(w, h, false)
 	vw, vh := i.backend.restorable.Size()
-	vs := graphicsutil.QuadVertices(vw, vh, x, y, x+w, y+h, 1, 0, 0, 1, 0, 0)
+	vs := graphicsutil.QuadVertices(vw, vh, x, y, x+w, y+h, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	is := graphicsutil.QuadIndices()
 	newImg.DrawImage(i.backend.restorable, vs, is, nil, opengl.CompositeModeCopy, graphics.FilterNearest)
 
@@ -184,13 +184,13 @@ func (i *Image) Size() (width, height int) {
 	return i.width, i.height
 }
 
-func (i *Image) QuadVertices(sx0, sy0, sx1, sy1 int, a, b, c, d, tx, ty float32) []float32 {
+func (i *Image) QuadVertices(sx0, sy0, sx1, sy1 int, a, b, c, d, tx, ty float32, cr, cg, cb, ca float32) []float32 {
 	if i.backend == nil {
 		i.allocate(true)
 	}
 	ox, oy, _, _ := i.region()
 	w, h := i.backend.restorable.SizePowerOf2()
-	return graphicsutil.QuadVertices(w, h, sx0+ox, sy0+oy, sx1+ox, sy1+oy, a, b, c, d, tx, ty)
+	return graphicsutil.QuadVertices(w, h, sx0+ox, sy0+oy, sx1+ox, sy1+oy, a, b, c, d, tx, ty, cr, cg, cb, ca)
 }
 
 const MaxCountForShare = 10
diff --git a/internal/shareable/shareable_test.go b/internal/shareable/shareable_test.go
index 5f6ae44da..b7e19b308 100644
--- a/internal/shareable/shareable_test.go
+++ b/internal/shareable/shareable_test.go
@@ -86,7 +86,7 @@ func TestEnsureNotShared(t *testing.T) {
 		dy1 = size * 3 / 4
 	)
 	// img4.ensureNotShared() should be called.
-	vs := img3.QuadVertices(0, 0, size/2, size/2, 1, 0, 0, 1, size/4, size/4)
+	vs := img3.QuadVertices(0, 0, size/2, size/2, 1, 0, 0, 1, size/4, size/4, 1, 1, 1, 1)
 	is := graphicsutil.QuadIndices()
 	img4.DrawImage(img3, vs, is, nil, opengl.CompositeModeCopy, graphics.FilterNearest)
 	want := false
@@ -150,7 +150,7 @@ func Disabled_TestReshared(t *testing.T) {
 	}
 
 	// Use img1 as a render target.
-	vs := img2.QuadVertices(0, 0, size, size, 1, 0, 0, 1, 0, 0)
+	vs := img2.QuadVertices(0, 0, size, size, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1)
 	is := graphicsutil.QuadIndices()
 	img1.DrawImage(img2, vs, is, nil, opengl.CompositeModeCopy, graphics.FilterNearest)
 	want = false