Detect touching duplicate sprite subjects

app.py

@@ -1774,7 +1774,7 @@ def primary_diffusion_png(spec: AssetSpec, index: int, run_id: int) -> tuple[byt
             if is_background:
                 return content, None
 
-            last_component_count = len(significant_foreground_component_areas(content))
+            last_component_count = estimated_foreground_subject_count(content)
             if last_component_count == 1:
                 return content, None
 
@@ -1928,6 +1928,56 @@ def significant_foreground_component_areas(content: bytes) -> list[int]:
     return sorted((area for area in areas if area >= minimum), reverse=True)
 
 
+def has_deep_vertical_foreground_valley(content: bytes) -> bool:
+    """Detect two large side-by-side silhouettes joined by a thin prop, tail, or shadow."""
+    image = Image.open(io.BytesIO(content)).convert("RGBA")
+    width, height = image.size
+    alpha = image.getchannel("A")
+    foreground = [[alpha.getpixel((x, y)) >= 64 for x in range(width)] for y in range(height)]
+    active_x = [x for x in range(width) if any(foreground[y][x] for y in range(height))]
+    active_y = [y for y in range(height) if any(foreground[y][x] for x in range(width))]
+    if not active_x or not active_y:
+        return False
+
+    x0, x1 = min(active_x), max(active_x)
+    y0, y1 = min(active_y), max(active_y)
+    bbox_width = x1 - x0 + 1
+    bbox_height = y1 - y0 + 1
+    if bbox_width < 12 or bbox_height < 12:
+        return False
+
+    columns = [sum(foreground[y][x] for y in range(y0, y1 + 1)) for x in range(x0, x1 + 1)]
+    total = sum(columns)
+    minimum_side_area = total * 0.28
+    minimum_side_width = max(4, int(bbox_width * 0.22))
+    minimum_side_peak = bbox_height * 0.55
+    maximum_valley = bbox_height * 0.25
+    prefix = 0
+
+    for index in range(1, len(columns) - 1):
+        prefix += columns[index - 1]
+        suffix = total - prefix - columns[index]
+        if prefix < minimum_side_area or suffix < minimum_side_area:
+            continue
+        if index < minimum_side_width or len(columns) - index - 1 < minimum_side_width:
+            continue
+        valley = sum(columns[index - 1 : index + 2]) / 3
+        if valley > maximum_valley:
+            continue
+        if max(columns[:index]) < minimum_side_peak or max(columns[index + 1 :]) < minimum_side_peak:
+            continue
+        return True
+    return False
+
+
+def estimated_foreground_subject_count(content: bytes) -> int:
+    """Estimate material subjects, including silhouettes that touch through a narrow bridge."""
+    component_count = len(significant_foreground_component_areas(content))
+    if component_count != 1:
+        return component_count
+    return 2 if has_deep_vertical_foreground_valley(content) else 1
+
+
 def free_diffusion_png(spec: AssetSpec, index: int, run_id: int) -> tuple[bytes | None, str | None]:
     global FREE_DIFFUSION_PIPE, FREE_DIFFUSION_ERROR
     if FREE_DIFFUSION_ERROR:
@@ -2173,11 +2223,11 @@ def validate_asset_png(content: bytes, spec: AssetSpec) -> list[str]:
         )
         if any(value > 16 for value in corners):
             warnings.append("sprite has opaque corner pixels")
-        component_areas = significant_foreground_component_areas(content)
-        if not component_areas:
+        subject_count = estimated_foreground_subject_count(content)
+        if not subject_count:
             warnings.append("sprite has no significant foreground subject")
-        elif len(component_areas) > 1:
-            warnings.append(f"sprite contains {len(component_areas)} significant foreground subjects")
+        elif subject_count > 1:
+            warnings.append(f"sprite contains {subject_count} significant foreground subjects")
     return warnings