diff --git a/image_test.go b/image_test.go
index 5151fa11b..8e7e97862 100644
--- a/image_test.go
+++ b/image_test.go
@@ -504,7 +504,7 @@ func TestImageFill(t *testing.T) {
 	}
 }
 
-// Issue #317
+// Issue #317, #558
 func TestImageEdge(t *testing.T) {
 	const (
 		img0Width  = 16
@@ -536,15 +536,23 @@ func TestImageEdge(t *testing.T) {
 	red := color.RGBA{0xff, 0, 0, 0xff}
 	transparent := color.RGBA{0, 0, 0, 0}
 
+	angles := []float64{}
+	for a := 0; a < 360; a++ {
+		angles = append(angles, float64(a)/360*2*math.Pi)
+	}
+	for a := 0; a < 256; a++ {
+		angles = append(angles, float64(a)/256*2*math.Pi)
+	}
+
 	for _, f := range []Filter{FilterNearest, FilterLinear} {
-		for a := 0; a < 360; a += 5 {
+		for _, a := range angles {
 			img1.Clear()
 			op := &DrawImageOptions{}
 			w, h := img0.Size()
 			r := image.Rect(0, 0, w, h/2)
 			op.SourceRect = &r
 			op.GeoM.Translate(-float64(img0Width)/2, -float64(img0Height)/2)
-			op.GeoM.Rotate(float64(a) * math.Pi / 180)
+			op.GeoM.Rotate(a)
 			op.GeoM.Translate(img1Width/2, img1Height/2)
 			op.Filter = f
 			img1.DrawImage(img0, op)
@@ -565,7 +573,7 @@ func TestImageEdge(t *testing.T) {
 							continue
 						}
 					}
-					t.Errorf("img1.At(%d, %d) (filter: %d, angle: %d) want: red or transparent, got: %v", i, j, f, a, c)
+					t.Errorf("img1.At(%d, %d) (filter: %d, angle: %f) want: red or transparent, got: %v", i, j, f, a, c)
 				}
 			}
 		}
diff --git a/internal/graphics/program.go b/internal/graphics/program.go
index 73dec8933..833277668 100644
--- a/internal/graphics/program.go
+++ b/internal/graphics/program.go
@@ -321,17 +321,16 @@ func (s *openGLState) useProgram(proj []float32, texture opengl.Texture, dst, sr
 		s.lastColorMatrixTranslation = esTranslate
 	}
 
-	if program == s.programLinear || program == s.programScreen {
-		sw, sh := src.Size()
-		sw = emath.NextPowerOf2Int(sw)
-		sh = emath.NextPowerOf2Int(sh)
+	sw, sh := src.Size()
+	sw = emath.NextPowerOf2Int(sw)
+	sh = emath.NextPowerOf2Int(sh)
 
-		if s.lastSourceWidth != sw || s.lastSourceHeight != sh {
-			c.UniformFloats(program, "source_size", []float32{float32(sw), float32(sh)})
-			s.lastSourceWidth = sw
-			s.lastSourceHeight = sh
-		}
+	if s.lastSourceWidth != sw || s.lastSourceHeight != sh {
+		c.UniformFloats(program, "source_size", []float32{float32(sw), float32(sh)})
+		s.lastSourceWidth = sw
+		s.lastSourceHeight = sh
 	}
+
 	if program == s.programScreen {
 		sw, _ := src.Size()
 		dw, _ := dst.Size()
diff --git a/internal/graphics/shader.go b/internal/graphics/shader.go
index e9b521ba5..40a7c9786 100644
--- a/internal/graphics/shader.go
+++ b/internal/graphics/shader.go
@@ -76,9 +76,7 @@ uniform sampler2D texture;
 uniform mat4 color_matrix;
 uniform vec4 color_matrix_translation;
 
-#if defined(FILTER_LINEAR) || defined(FILTER_SCREEN)
 uniform highp vec2 source_size;
-#endif
 
 #if defined(FILTER_SCREEN)
 uniform highp float scale;
@@ -104,51 +102,60 @@ void main(void) {
   // roundTexel adjusts pos by rounding it (#315, #558).
   pos = roundTexel(pos);
 
-#if defined(FILTER_NEAREST)
-  vec4 color = texture2D(texture, pos);
-  if (pos.x < varying_tex_coord_min.x ||
-    pos.y < varying_tex_coord_min.y ||
-    varying_tex_coord_max.x <= pos.x ||
-    varying_tex_coord_max.y <= pos.y) {
-    color = vec4(0, 0, 0, 0);
-  }
-#endif
-
-#if defined(FILTER_LINEAR)
   highp vec2 texel_size = 1.0 / source_size;
 
+#if defined(FILTER_NEAREST) || defined(FILTER_LINEAR)
+  // Even with the nearest filter, it is better to calculate the color with around 4 texels (#558).
+  
   highp vec2 p0 = pos - texel_size / 2.0;
   highp vec2 p1 = pos + texel_size / 2.0;
   vec4 c0 = texture2D(texture, p0);
   vec4 c1 = texture2D(texture, vec2(p1.x, p0.y));
   vec4 c2 = texture2D(texture, vec2(p0.x, p1.y));
   vec4 c3 = texture2D(texture, p1);
-  if (p0.x < varying_tex_coord_min.x) {
+  if (p0.x < (varying_tex_coord_min.x)) {
     c0 = vec4(0, 0, 0, 0);
     c2 = vec4(0, 0, 0, 0);
   }
-  if (p0.y < varying_tex_coord_min.y) {
+  if (p0.y < (varying_tex_coord_min.y)) {
     c0 = vec4(0, 0, 0, 0);
     c1 = vec4(0, 0, 0, 0);
   }
-  if (varying_tex_coord_max.x <= p1.x) {
+  if ((varying_tex_coord_max.x - texel_size.x / 256.0) <= p1.x) {
     c1 = vec4(0, 0, 0, 0);
     c3 = vec4(0, 0, 0, 0);
   }
-  if (varying_tex_coord_max.y <= p1.y) {
+  if ((varying_tex_coord_max.y - texel_size.y / 256.0) <= p1.y) {
     c2 = vec4(0, 0, 0, 0);
     c3 = vec4(0, 0, 0, 0);
   }
 
   vec2 rate = fract(p0 * source_size);
+
+#if defined(FILTER_NEAREST)
+  vec4 color;
+  if (rate.x < 0.5) {
+    if (rate.y < 0.5) {
+      color = c0;
+    } else {
+      color = c2;
+    }
+  } else {
+    if (rate.y < 0.5) {
+      color = c1;
+    } else {
+      color = c3;
+    }
+  }
+#elif defined(FILTER_LINEAR)
   vec4 color = mix(mix(c0, c1, rate.x), mix(c2, c3, rate.x), rate.y);
 #endif
 
+#endif
+
 #if defined(FILTER_SCREEN)
-  highp vec2 texel_size = 1.0 / source_size;
   highp vec2 p0 = pos - texel_size / 2.0 / scale;
   highp vec2 p1 = pos + texel_size / 2.0 / scale;
-
   vec4 c0 = texture2D(texture, p0);
   vec4 c1 = texture2D(texture, vec2(p1.x, p0.y));
   vec4 c2 = texture2D(texture, vec2(p0.x, p1.y));