diff --git a/piet-gpu-types/src/annotated.rs b/piet-gpu-types/src/annotated.rs
index 0e36f62..681a7ec 100644
--- a/piet-gpu-types/src/annotated.rs
+++ b/piet-gpu-types/src/annotated.rs
@@ -7,6 +7,10 @@ piet_gpu! {
             rgba_color: u32,
             bbox: [f32; 4],
         }
+        struct AnnoFillMask {
+            mask: f32,
+            bbox: [f32; 4],
+        }
         struct AnnoStroke {
             rgba_color: u32,
             bbox: [f32; 4],
@@ -18,6 +22,8 @@ piet_gpu! {
             Nop,
             Stroke(AnnoStroke),
             Fill(AnnoFill),
+            FillMask(AnnoFillMask),
+            FillMaskInv(AnnoFillMask),
         }
     }
 }
diff --git a/piet-gpu-types/src/ptcl.rs b/piet-gpu-types/src/ptcl.rs
index 1deac38..95dcdc6 100644
--- a/piet-gpu-types/src/ptcl.rs
+++ b/piet-gpu-types/src/ptcl.rs
@@ -25,9 +25,17 @@ piet_gpu! {
             backdrop: i32,
             rgba_color: u32,
         }
+        struct CmdFillMask {
+            tile_ref: u32,
+            backdrop: i32,
+            mask: f32,
+        }
         struct CmdSolid {
             rgba_color: u32,
         }
+        struct CmdSolidMask {
+            mask: f32,
+        }
         struct CmdJump {
             new_ref: u32,
         }
@@ -36,8 +44,11 @@ piet_gpu! {
             Circle(CmdCircle),
             Line(CmdLine),
             Fill(CmdFill),
+            FillMask(CmdFillMask),
+            FillMaskInv(CmdFillMask),
             Stroke(CmdStroke),
             Solid(CmdSolid),
+            SolidMask(CmdSolidMask),
             Jump(CmdJump),
         }
 
diff --git a/piet-gpu-types/src/scene.rs b/piet-gpu-types/src/scene.rs
index 0d72650..1359c1b 100644
--- a/piet-gpu-types/src/scene.rs
+++ b/piet-gpu-types/src/scene.rs
@@ -23,6 +23,9 @@ piet_gpu! {
         struct Fill {
             rgba_color: u32,
         }
+        struct FillMask {
+            mask: f32,
+        }
         struct Stroke {
             rgba_color: u32,
         }
@@ -50,6 +53,8 @@ piet_gpu! {
             Fill(Fill),
             SetLineWidth(SetLineWidth),
             Transform(Transform),
+            FillMask(FillMask),
+            FillMaskInv(FillMask),
         }
     }
 }
diff --git a/piet-gpu/shader/annotated.h b/piet-gpu/shader/annotated.h
index 5dcb4ad..847ca06 100644
--- a/piet-gpu/shader/annotated.h
+++ b/piet-gpu/shader/annotated.h
@@ -4,6 +4,10 @@ struct AnnoFillRef {
     uint offset;
 };
 
+struct AnnoFillMaskRef {
+    uint offset;
+};
+
 struct AnnoStrokeRef {
     uint offset;
 };
@@ -23,6 +27,17 @@ AnnoFillRef AnnoFill_index(AnnoFillRef ref, uint index) {
     return AnnoFillRef(ref.offset + index * AnnoFill_size);
 }
 
+struct AnnoFillMask {
+    float mask;
+    vec4 bbox;
+};
+
+#define AnnoFillMask_size 20
+
+AnnoFillMaskRef AnnoFillMask_index(AnnoFillMaskRef ref, uint index) {
+    return AnnoFillMaskRef(ref.offset + index * AnnoFillMask_size);
+}
+
 struct AnnoStroke {
     uint rgba_color;
     vec4 bbox;
@@ -38,6 +53,8 @@ AnnoStrokeRef AnnoStroke_index(AnnoStrokeRef ref, uint index) {
 #define Annotated_Nop 0
 #define Annotated_Stroke 1
 #define Annotated_Fill 2
+#define Annotated_FillMask 3
+#define Annotated_FillMaskInv 4
 #define Annotated_size 28
 
 AnnotatedRef Annotated_index(AnnotatedRef ref, uint index) {
@@ -66,6 +83,28 @@ void AnnoFill_write(AnnoFillRef ref, AnnoFill s) {
     annotated[ix + 4] = floatBitsToUint(s.bbox.w);
 }
 
+AnnoFillMask AnnoFillMask_read(AnnoFillMaskRef ref) {
+    uint ix = ref.offset >> 2;
+    uint raw0 = annotated[ix + 0];
+    uint raw1 = annotated[ix + 1];
+    uint raw2 = annotated[ix + 2];
+    uint raw3 = annotated[ix + 3];
+    uint raw4 = annotated[ix + 4];
+    AnnoFillMask s;
+    s.mask = uintBitsToFloat(raw0);
+    s.bbox = vec4(uintBitsToFloat(raw1), uintBitsToFloat(raw2), uintBitsToFloat(raw3), uintBitsToFloat(raw4));
+    return s;
+}
+
+void AnnoFillMask_write(AnnoFillMaskRef ref, AnnoFillMask s) {
+    uint ix = ref.offset >> 2;
+    annotated[ix + 0] = floatBitsToUint(s.mask);
+    annotated[ix + 1] = floatBitsToUint(s.bbox.x);
+    annotated[ix + 2] = floatBitsToUint(s.bbox.y);
+    annotated[ix + 3] = floatBitsToUint(s.bbox.z);
+    annotated[ix + 4] = floatBitsToUint(s.bbox.w);
+}
+
 AnnoStroke AnnoStroke_read(AnnoStrokeRef ref) {
     uint ix = ref.offset >> 2;
     uint raw0 = annotated[ix + 0];
@@ -103,6 +142,14 @@ AnnoFill Annotated_Fill_read(AnnotatedRef ref) {
     return AnnoFill_read(AnnoFillRef(ref.offset + 4));
 }
 
+AnnoFillMask Annotated_FillMask_read(AnnotatedRef ref) {
+    return AnnoFillMask_read(AnnoFillMaskRef(ref.offset + 4));
+}
+
+AnnoFillMask Annotated_FillMaskInv_read(AnnotatedRef ref) {
+    return AnnoFillMask_read(AnnoFillMaskRef(ref.offset + 4));
+}
+
 void Annotated_Nop_write(AnnotatedRef ref) {
     annotated[ref.offset >> 2] = Annotated_Nop;
 }
@@ -117,3 +164,13 @@ void Annotated_Fill_write(AnnotatedRef ref, AnnoFill s) {
     AnnoFill_write(AnnoFillRef(ref.offset + 4), s);
 }
 
+void Annotated_FillMask_write(AnnotatedRef ref, AnnoFillMask s) {
+    annotated[ref.offset >> 2] = Annotated_FillMask;
+    AnnoFillMask_write(AnnoFillMaskRef(ref.offset + 4), s);
+}
+
+void Annotated_FillMaskInv_write(AnnotatedRef ref, AnnoFillMask s) {
+    annotated[ref.offset >> 2] = Annotated_FillMaskInv;
+    AnnoFillMask_write(AnnoFillMaskRef(ref.offset + 4), s);
+}
+
diff --git a/piet-gpu/shader/backdrop.comp b/piet-gpu/shader/backdrop.comp
index beba683..20c6ce9 100644
--- a/piet-gpu/shader/backdrop.comp
+++ b/piet-gpu/shader/backdrop.comp
@@ -53,7 +53,10 @@ void main() {
     uint row_count = 0;
     if (element_ix < n_elements) {
         uint tag = Annotated_tag(ref);
-        if (tag == Annotated_Fill) {
+        switch (tag) {
+        case Annotated_Fill:
+        case Annotated_FillMask:
+        case Annotated_FillMaskInv:
             PathRef path_ref = PathRef(element_ix * Path_size);
             Path path = Path_read(path_ref);
             sh_row_width[th_ix] = path.bbox.z - path.bbox.x;
diff --git a/piet-gpu/shader/backdrop.spv b/piet-gpu/shader/backdrop.spv
index fdbb4f1..e2093d9 100644
Binary files a/piet-gpu/shader/backdrop.spv and b/piet-gpu/shader/backdrop.spv differ
diff --git a/piet-gpu/shader/binning.comp b/piet-gpu/shader/binning.comp
index 046c4fb..ee3301b 100644
--- a/piet-gpu/shader/binning.comp
+++ b/piet-gpu/shader/binning.comp
@@ -60,6 +60,8 @@ void main() {
     float my_right_edge = INFINITY;
     switch (tag) {
     case Annotated_Fill:
+    case Annotated_FillMask:
+    case Annotated_FillMaskInv:
     case Annotated_Stroke:
         // Note: we take advantage of the fact that fills and strokes
         // have compatible layout.
diff --git a/piet-gpu/shader/binning.spv b/piet-gpu/shader/binning.spv
index d1c469c..50070a1 100644
Binary files a/piet-gpu/shader/binning.spv and b/piet-gpu/shader/binning.spv differ
diff --git a/piet-gpu/shader/coarse.comp b/piet-gpu/shader/coarse.comp
index 57cbc8b..16573e8 100644
--- a/piet-gpu/shader/coarse.comp
+++ b/piet-gpu/shader/coarse.comp
@@ -173,6 +173,8 @@ void main() {
         uint tile_count;
         switch (tag) {
         case Annotated_Fill:
+        case Annotated_FillMask:
+        case Annotated_FillMaskInv:
         case Annotated_Stroke:
             // Because the only elements we're processing right now are
             // paths, we can just use the element index as the path index.
@@ -222,12 +224,20 @@ void main() {
                     el_ix = probe;
                 }
             }
+            AnnotatedRef ref = AnnotatedRef(el_ix * Annotated_size);
+            uint tag = Annotated_tag(ref);
             uint seq_ix = ix - (el_ix > 0 ? sh_tile_count[el_ix - 1] : 0);
             uint width = sh_tile_width[el_ix];
             uint x = sh_tile_x0[el_ix] + seq_ix % width;
             uint y = sh_tile_y0[el_ix] + seq_ix / width;
             Tile tile = Tile_read(TileRef(sh_tile_base[el_ix] + (sh_tile_stride[el_ix] * y + x) * Tile_size));
-            if (tile.tile.offset != 0 || tile.backdrop != 0) {
+            // Include the path in the tile if
+            // - the tile contains at least a segment (tile offset non-zero)
+            // - the tile is completely covered (backdrop non-zero)
+            // - the tile is not covered and we're filling everything outside the path (backdrop zero, inverse fills).
+            bool inside = tile.backdrop != 0;
+            bool fill = tag != Annotated_FillMaskInv;
+            if (tile.tile.offset != 0 || inside == fill) {
                 uint el_slice = el_ix / 32;
                 uint el_mask = 1 << (el_ix & 31);
                 atomicOr(sh_bitmaps[el_slice][y * N_TILE_X + x], el_mask);
@@ -281,6 +291,27 @@ void main() {
                 }
                 cmd_ref.offset += Cmd_size;
                 break;
+            case Annotated_FillMask:
+            case Annotated_FillMaskInv:
+                tile = Tile_read(TileRef(sh_tile_base[element_ref_ix]
+                    + (sh_tile_stride[element_ref_ix] * tile_y + tile_x) * Tile_size));
+                AnnoFillMask fill_mask = Annotated_FillMask_read(ref);
+                alloc_cmd(cmd_ref, cmd_limit);
+                if (tile.tile.offset != 0) {
+                    CmdFillMask cmd_fill;
+                    cmd_fill.tile_ref = tile.tile.offset;
+                    cmd_fill.backdrop = tile.backdrop;
+                    cmd_fill.mask = fill_mask.mask;
+                    if (tag == Annotated_FillMask) {
+                        Cmd_FillMask_write(cmd_ref, cmd_fill);
+                    } else {
+                        Cmd_FillMaskInv_write(cmd_ref, cmd_fill);
+                    }
+                } else {
+                    Cmd_SolidMask_write(cmd_ref, CmdSolidMask(fill_mask.mask));
+                }
+                cmd_ref.offset += Cmd_size;
+                break;
             case Annotated_Stroke:
                 tile = Tile_read(TileRef(sh_tile_base[element_ref_ix]
                     + (sh_tile_stride[element_ref_ix] * tile_y + tile_x) * Tile_size));
diff --git a/piet-gpu/shader/coarse.spv b/piet-gpu/shader/coarse.spv
index 334e1f5..7758b96 100644
Binary files a/piet-gpu/shader/coarse.spv and b/piet-gpu/shader/coarse.spv differ
diff --git a/piet-gpu/shader/elements.comp b/piet-gpu/shader/elements.comp
index 3c38f38..b0e4779 100644
--- a/piet-gpu/shader/elements.comp
+++ b/piet-gpu/shader/elements.comp
@@ -130,6 +130,8 @@ State map_element(ElementRef ref, inout bool is_fill) {
         c.pathseg_count = 1;
         break;
     case Element_Fill:
+    case Element_FillMask:
+    case Element_FillMaskInv:
         is_fill = true;
         // fall-through
     case Element_Stroke:
@@ -381,6 +383,23 @@ void main() {
             out_ref = AnnotatedRef((st.path_count - 1) * Annotated_size);
             Annotated_Fill_write(out_ref, anno_fill);
             break;
+        case Element_FillMask:
+            FillMask fill_mask = Element_FillMask_read(this_ref);
+            AnnoFillMask anno_fill_mask;
+            anno_fill_mask.mask = fill_mask.mask;
+            anno_fill_mask.bbox = st.bbox;
+            out_ref = AnnotatedRef((st.path_count - 1) * Annotated_size);
+            Annotated_FillMask_write(out_ref, anno_fill_mask);
+            break;
+        case Element_FillMaskInv:
+            fill_mask = Element_FillMaskInv_read(this_ref);
+            anno_fill_mask.mask = fill_mask.mask;
+            // The inverse fill conceptually takes up the entire screen.
+            // TODO: Tighten bounds to contain only affected paths.
+            anno_fill_mask.bbox = vec4(0, 0, 1e9, 1e9);
+            out_ref = AnnotatedRef((st.path_count - 1) * Annotated_size);
+            Annotated_FillMaskInv_write(out_ref, anno_fill_mask);
+            break;
         }
     }
 }
diff --git a/piet-gpu/shader/elements.spv b/piet-gpu/shader/elements.spv
index 4b7a1c2..599465a 100644
Binary files a/piet-gpu/shader/elements.spv and b/piet-gpu/shader/elements.spv differ
diff --git a/piet-gpu/shader/kernel4.comp b/piet-gpu/shader/kernel4.comp
index 70c879f..d21aa87 100644
--- a/piet-gpu/shader/kernel4.comp
+++ b/piet-gpu/shader/kernel4.comp
@@ -27,6 +27,40 @@ layout(rgba8, set = 0, binding = 2) uniform writeonly image2D image;
 #include "ptcl.h"
 #include "tile.h"
 
+// Calculate coverage based on backdrop + coverage of each line segment
+float[CHUNK] computeArea(vec2 xy, int backdrop, uint tile_ref) {
+    // Probably better to store as float, but conversion is no doubt cheap.
+    float area[CHUNK];
+    for (uint k = 0; k < CHUNK; k++) area[k] = float(backdrop);
+    TileSegRef tile_seg_ref = TileSegRef(tile_ref);
+    do {
+        TileSeg seg = TileSeg_read(tile_seg_ref);
+        for (uint k = 0; k < CHUNK; k++) {
+            vec2 my_xy = vec2(xy.x, xy.y + float(k * CHUNK_DY));
+            vec2 start = seg.start - my_xy;
+            vec2 end = seg.end - my_xy;
+            vec2 window = clamp(vec2(start.y, end.y), 0.0, 1.0);
+            if (window.x != window.y) {
+                vec2 t = (window - start.y) / (end.y - start.y);
+                vec2 xs = vec2(mix(start.x, end.x, t.x), mix(start.x, end.x, t.y));
+                float xmin = min(min(xs.x, xs.y), 1.0) - 1e-6;
+                float xmax = max(xs.x, xs.y);
+                float b = min(xmax, 1.0);
+                float c = max(b, 0.0);
+                float d = max(xmin, 0.0);
+                float a = (b + 0.5 * (d * d - c * c) - xmin) / (xmax - xmin);
+                area[k] += a * (window.x - window.y);
+            }
+            area[k] += sign(end.x - start.x) * clamp(my_xy.y - seg.y_edge + 1.0, 0.0, 1.0);
+        }
+        tile_seg_ref = seg.next;
+    } while (tile_seg_ref.offset != 0);
+    for (uint k = 0; k < CHUNK; k++) {
+        area[k] = min(abs(area[k]), 1.0);
+    }
+    return area;
+}
+
 void main() {
     uint tile_ix = gl_WorkGroupID.y * WIDTH_IN_TILES + gl_WorkGroupID.x;
     CmdRef cmd_ref = CmdRef(tile_ix * PTCL_INITIAL_ALLOC);
@@ -34,8 +68,10 @@ void main() {
     uvec2 xy_uint = uvec2(gl_GlobalInvocationID.x, gl_LocalInvocationID.y + TILE_HEIGHT_PX * gl_WorkGroupID.y);
     vec2 xy = vec2(xy_uint);
     vec3 rgb[CHUNK];
+    float mask[CHUNK];
     for (uint i = 0; i < CHUNK; i++) {
         rgb[i] = vec3(0.5);
+        mask[i] = 1.0;
     }
 
     while (true) {
@@ -52,7 +88,7 @@ void main() {
                 float r = length(vec2(xy.x, xy.y + dy) + vec2(0.5, 0.5) - circle.center.xy);
                 float alpha = clamp(0.5 + circle.radius - r, 0.0, 1.0);
                 // TODO: sRGB
-                rgb[i] = mix(rgb[i], fg_rgba.rgb, alpha * fg_rgba.a);
+                rgb[i] = mix(rgb[i], fg_rgba.rgb, mask[i] * alpha * fg_rgba.a);
             }
             break;
         case Cmd_Stroke:
@@ -75,49 +111,43 @@ void main() {
             fg_rgba = unpackUnorm4x8(stroke.rgba_color).wzyx;
             for (uint k = 0; k < CHUNK; k++) {
                 float alpha = clamp(stroke.half_width + 0.5 - df[k], 0.0, 1.0);
-                rgb[k] = mix(rgb[k], fg_rgba.rgb, alpha * fg_rgba.a);
+                rgb[k] = mix(rgb[k], fg_rgba.rgb, mask[k] * alpha * fg_rgba.a);
             }
             break;
         case Cmd_Fill:
-            // Calculate coverage based on backdrop + coverage of each line segment
             CmdFill fill = Cmd_Fill_read(cmd_ref);
-            // Probably better to store as float, but conversion is no doubt cheap.
             float area[CHUNK];
-            for (uint k = 0; k < CHUNK; k++) area[k] = float(fill.backdrop);
-            tile_seg_ref = TileSegRef(fill.tile_ref);
-            do {
-                TileSeg seg = TileSeg_read(tile_seg_ref);
-                for (uint k = 0; k < CHUNK; k++) {
-                    vec2 my_xy = vec2(xy.x, xy.y + float(k * CHUNK_DY));
-                    vec2 start = seg.start - my_xy;
-                    vec2 end = seg.end - my_xy;
-                    vec2 window = clamp(vec2(start.y, end.y), 0.0, 1.0);
-                    if (window.x != window.y) {
-                        vec2 t = (window - start.y) / (end.y - start.y);
-                        vec2 xs = vec2(mix(start.x, end.x, t.x), mix(start.x, end.x, t.y));
-                        float xmin = min(min(xs.x, xs.y), 1.0) - 1e-6;
-                        float xmax = max(xs.x, xs.y);
-                        float b = min(xmax, 1.0);
-                        float c = max(b, 0.0);
-                        float d = max(xmin, 0.0);
-                        float a = (b + 0.5 * (d * d - c * c) - xmin) / (xmax - xmin);
-                        area[k] += a * (window.x - window.y);
-                    }
-                    area[k] += sign(end.x - start.x) * clamp(my_xy.y - seg.y_edge + 1.0, 0.0, 1.0);
-                }
-                tile_seg_ref = seg.next;
-            } while (tile_seg_ref.offset != 0);
+            area = computeArea(xy, fill.backdrop, fill.tile_ref);
             fg_rgba = unpackUnorm4x8(fill.rgba_color).wzyx;
             for (uint k = 0; k < CHUNK; k++) {
-                float alpha = min(abs(area[k]), 1.0);
-                rgb[k] = mix(rgb[k], fg_rgba.rgb, alpha * fg_rgba.a);
+                rgb[k] = mix(rgb[k], fg_rgba.rgb, mask[k] * area[k] * fg_rgba.a);
+            }
+            break;
+        case Cmd_FillMask:
+            CmdFillMask fill_mask = Cmd_FillMask_read(cmd_ref);
+            area = computeArea(xy, fill_mask.backdrop, fill_mask.tile_ref);
+            for (uint k = 0; k < CHUNK; k++) {
+                mask[k] = mix(mask[k], fill_mask.mask, area[k]);
+            }
+            break;
+        case Cmd_FillMaskInv:
+            fill_mask = Cmd_FillMask_read(cmd_ref);
+            area = computeArea(xy, fill_mask.backdrop, fill_mask.tile_ref);
+            for (uint k = 0; k < CHUNK; k++) {
+                mask[k] = mix(mask[k], fill_mask.mask, 1.0 - area[k]);
             }
             break;
         case Cmd_Solid:
             CmdSolid solid = Cmd_Solid_read(cmd_ref);
             fg_rgba = unpackUnorm4x8(solid.rgba_color).wzyx;
             for (uint k = 0; k < CHUNK; k++) {
-                rgb[k] = mix(rgb[k], fg_rgba.rgb, fg_rgba.a);
+                rgb[k] = mix(rgb[k], fg_rgba.rgb, mask[k] * fg_rgba.a);
+            }
+            break;
+        case Cmd_SolidMask:
+            CmdSolidMask solid_mask = Cmd_SolidMask_read(cmd_ref);
+            for (uint k = 0; k < CHUNK; k++) {
+                mask[k] = solid_mask.mask;
             }
             break;
         case Cmd_Jump:
diff --git a/piet-gpu/shader/kernel4.spv b/piet-gpu/shader/kernel4.spv
index 84919e0..d159fe5 100644
Binary files a/piet-gpu/shader/kernel4.spv and b/piet-gpu/shader/kernel4.spv differ
diff --git a/piet-gpu/shader/ptcl.h b/piet-gpu/shader/ptcl.h
index c9be302..d1cc83d 100644
--- a/piet-gpu/shader/ptcl.h
+++ b/piet-gpu/shader/ptcl.h
@@ -16,10 +16,18 @@ struct CmdFillRef {
     uint offset;
 };
 
+struct CmdFillMaskRef {
+    uint offset;
+};
+
 struct CmdSolidRef {
     uint offset;
 };
 
+struct CmdSolidMaskRef {
+    uint offset;
+};
+
 struct CmdJumpRef {
     uint offset;
 };
@@ -83,6 +91,18 @@ CmdFillRef CmdFill_index(CmdFillRef ref, uint index) {
     return CmdFillRef(ref.offset + index * CmdFill_size);
 }
 
+struct CmdFillMask {
+    uint tile_ref;
+    int backdrop;
+    float mask;
+};
+
+#define CmdFillMask_size 12
+
+CmdFillMaskRef CmdFillMask_index(CmdFillMaskRef ref, uint index) {
+    return CmdFillMaskRef(ref.offset + index * CmdFillMask_size);
+}
+
 struct CmdSolid {
     uint rgba_color;
 };
@@ -93,6 +113,16 @@ CmdSolidRef CmdSolid_index(CmdSolidRef ref, uint index) {
     return CmdSolidRef(ref.offset + index * CmdSolid_size);
 }
 
+struct CmdSolidMask {
+    float mask;
+};
+
+#define CmdSolidMask_size 4
+
+CmdSolidMaskRef CmdSolidMask_index(CmdSolidMaskRef ref, uint index) {
+    return CmdSolidMaskRef(ref.offset + index * CmdSolidMask_size);
+}
+
 struct CmdJump {
     uint new_ref;
 };
@@ -107,9 +137,12 @@ CmdJumpRef CmdJump_index(CmdJumpRef ref, uint index) {
 #define Cmd_Circle 1
 #define Cmd_Line 2
 #define Cmd_Fill 3
-#define Cmd_Stroke 4
-#define Cmd_Solid 5
-#define Cmd_Jump 6
+#define Cmd_FillMask 4
+#define Cmd_FillMaskInv 5
+#define Cmd_Stroke 6
+#define Cmd_Solid 7
+#define Cmd_SolidMask 8
+#define Cmd_Jump 9
 #define Cmd_size 20
 
 CmdRef Cmd_index(CmdRef ref, uint index) {
@@ -219,6 +252,25 @@ void CmdFill_write(CmdFillRef ref, CmdFill s) {
     ptcl[ix + 2] = s.rgba_color;
 }
 
+CmdFillMask CmdFillMask_read(CmdFillMaskRef ref) {
+    uint ix = ref.offset >> 2;
+    uint raw0 = ptcl[ix + 0];
+    uint raw1 = ptcl[ix + 1];
+    uint raw2 = ptcl[ix + 2];
+    CmdFillMask s;
+    s.tile_ref = raw0;
+    s.backdrop = int(raw1);
+    s.mask = uintBitsToFloat(raw2);
+    return s;
+}
+
+void CmdFillMask_write(CmdFillMaskRef ref, CmdFillMask s) {
+    uint ix = ref.offset >> 2;
+    ptcl[ix + 0] = s.tile_ref;
+    ptcl[ix + 1] = uint(s.backdrop);
+    ptcl[ix + 2] = floatBitsToUint(s.mask);
+}
+
 CmdSolid CmdSolid_read(CmdSolidRef ref) {
     uint ix = ref.offset >> 2;
     uint raw0 = ptcl[ix + 0];
@@ -232,6 +284,19 @@ void CmdSolid_write(CmdSolidRef ref, CmdSolid s) {
     ptcl[ix + 0] = s.rgba_color;
 }
 
+CmdSolidMask CmdSolidMask_read(CmdSolidMaskRef ref) {
+    uint ix = ref.offset >> 2;
+    uint raw0 = ptcl[ix + 0];
+    CmdSolidMask s;
+    s.mask = uintBitsToFloat(raw0);
+    return s;
+}
+
+void CmdSolidMask_write(CmdSolidMaskRef ref, CmdSolidMask s) {
+    uint ix = ref.offset >> 2;
+    ptcl[ix + 0] = floatBitsToUint(s.mask);
+}
+
 CmdJump CmdJump_read(CmdJumpRef ref) {
     uint ix = ref.offset >> 2;
     uint raw0 = ptcl[ix + 0];
@@ -261,6 +326,14 @@ CmdFill Cmd_Fill_read(CmdRef ref) {
     return CmdFill_read(CmdFillRef(ref.offset + 4));
 }
 
+CmdFillMask Cmd_FillMask_read(CmdRef ref) {
+    return CmdFillMask_read(CmdFillMaskRef(ref.offset + 4));
+}
+
+CmdFillMask Cmd_FillMaskInv_read(CmdRef ref) {
+    return CmdFillMask_read(CmdFillMaskRef(ref.offset + 4));
+}
+
 CmdStroke Cmd_Stroke_read(CmdRef ref) {
     return CmdStroke_read(CmdStrokeRef(ref.offset + 4));
 }
@@ -269,6 +342,10 @@ CmdSolid Cmd_Solid_read(CmdRef ref) {
     return CmdSolid_read(CmdSolidRef(ref.offset + 4));
 }
 
+CmdSolidMask Cmd_SolidMask_read(CmdRef ref) {
+    return CmdSolidMask_read(CmdSolidMaskRef(ref.offset + 4));
+}
+
 CmdJump Cmd_Jump_read(CmdRef ref) {
     return CmdJump_read(CmdJumpRef(ref.offset + 4));
 }
@@ -292,6 +369,16 @@ void Cmd_Fill_write(CmdRef ref, CmdFill s) {
     CmdFill_write(CmdFillRef(ref.offset + 4), s);
 }
 
+void Cmd_FillMask_write(CmdRef ref, CmdFillMask s) {
+    ptcl[ref.offset >> 2] = Cmd_FillMask;
+    CmdFillMask_write(CmdFillMaskRef(ref.offset + 4), s);
+}
+
+void Cmd_FillMaskInv_write(CmdRef ref, CmdFillMask s) {
+    ptcl[ref.offset >> 2] = Cmd_FillMaskInv;
+    CmdFillMask_write(CmdFillMaskRef(ref.offset + 4), s);
+}
+
 void Cmd_Stroke_write(CmdRef ref, CmdStroke s) {
     ptcl[ref.offset >> 2] = Cmd_Stroke;
     CmdStroke_write(CmdStrokeRef(ref.offset + 4), s);
@@ -302,6 +389,11 @@ void Cmd_Solid_write(CmdRef ref, CmdSolid s) {
     CmdSolid_write(CmdSolidRef(ref.offset + 4), s);
 }
 
+void Cmd_SolidMask_write(CmdRef ref, CmdSolidMask s) {
+    ptcl[ref.offset >> 2] = Cmd_SolidMask;
+    CmdSolidMask_write(CmdSolidMaskRef(ref.offset + 4), s);
+}
+
 void Cmd_Jump_write(CmdRef ref, CmdJump s) {
     ptcl[ref.offset >> 2] = Cmd_Jump;
     CmdJump_write(CmdJumpRef(ref.offset + 4), s);
diff --git a/piet-gpu/shader/scene.h b/piet-gpu/shader/scene.h
index eee8075..6823fe6 100644
--- a/piet-gpu/shader/scene.h
+++ b/piet-gpu/shader/scene.h
@@ -16,6 +16,10 @@ struct FillRef {
     uint offset;
 };
 
+struct FillMaskRef {
+    uint offset;
+};
+
 struct StrokeRef {
     uint offset;
 };
@@ -78,6 +82,16 @@ FillRef Fill_index(FillRef ref, uint index) {
     return FillRef(ref.offset + index * Fill_size);
 }
 
+struct FillMask {
+    float mask;
+};
+
+#define FillMask_size 4
+
+FillMaskRef FillMask_index(FillMaskRef ref, uint index) {
+    return FillMaskRef(ref.offset + index * FillMask_size);
+}
+
 struct Stroke {
     uint rgba_color;
 };
@@ -120,6 +134,8 @@ TransformRef Transform_index(TransformRef ref, uint index) {
 #define Element_Fill 8
 #define Element_SetLineWidth 9
 #define Element_Transform 10
+#define Element_FillMask 11
+#define Element_FillMaskInv 12
 #define Element_size 36
 
 ElementRef Element_index(ElementRef ref, uint index) {
@@ -179,6 +195,14 @@ Fill Fill_read(FillRef ref) {
     return s;
 }
 
+FillMask FillMask_read(FillMaskRef ref) {
+    uint ix = ref.offset >> 2;
+    uint raw0 = scene[ix + 0];
+    FillMask s;
+    s.mask = uintBitsToFloat(raw0);
+    return s;
+}
+
 Stroke Stroke_read(StrokeRef ref) {
     uint ix = ref.offset >> 2;
     uint raw0 = scene[ix + 0];
@@ -253,3 +277,11 @@ Transform Element_Transform_read(ElementRef ref) {
     return Transform_read(TransformRef(ref.offset + 4));
 }
 
+FillMask Element_FillMask_read(ElementRef ref) {
+    return FillMask_read(FillMaskRef(ref.offset + 4));
+}
+
+FillMask Element_FillMaskInv_read(ElementRef ref) {
+    return FillMask_read(FillMaskRef(ref.offset + 4));
+}
+
diff --git a/piet-gpu/shader/tile_alloc.comp b/piet-gpu/shader/tile_alloc.comp
index b0c99a5..3e1e52f 100644
--- a/piet-gpu/shader/tile_alloc.comp
+++ b/piet-gpu/shader/tile_alloc.comp
@@ -47,6 +47,8 @@ void main() {
     int x0 = 0, y0 = 0, x1 = 0, y1 = 0;
     switch (tag) {
     case Annotated_Fill:
+    case Annotated_FillMask:
+    case Annotated_FillMaskInv:
     case Annotated_Stroke:
         // Note: we take advantage of the fact that fills and strokes
         // have compatible layout.
diff --git a/piet-gpu/shader/tile_alloc.spv b/piet-gpu/shader/tile_alloc.spv
index e012b7c..af52665 100644
Binary files a/piet-gpu/shader/tile_alloc.spv and b/piet-gpu/shader/tile_alloc.spv differ