diff --git a/bgfilter/pipeline.py b/bgfilter/pipeline.py
index 1c30a13..4f02fee 100644
--- a/bgfilter/pipeline.py
+++ b/bgfilter/pipeline.py
@@ -28,13 +28,36 @@ class MattingPipeline:
         self.settings = settings
         self._runner: ViTMatteRunner | None = None
         self._segmenter = None
+        self._fill_segmenter = None
 
     def _segment(self, rgb: np.ndarray) -> np.ndarray:
         if self._segmenter is None:
             from .segmentation import make_segmenter
 
             self._segmenter = make_segmenter(self.settings.segmentation)
-        return self._segmenter.mask(rgb)
+        mask = self._segmenter.mask(rgb)
+        seg = self.settings.segmentation
+        if seg.sharpen > 0:
+            from .segmentation import unsharp_mask
+
+            mask = unsharp_mask(mask, seg.sharpen, seg.sharpen_radius)
+        return mask
+
+    def _segment_fill(self, rgb: np.ndarray, default_mask: np.ndarray) -> np.ndarray:
+        """Mask for the dropout-fill; may use a different (higher-recall) backend
+        than the trimap, since the fill only adds and cannot break anything."""
+        seg = self.settings.segmentation
+        if not seg.fill_backend or seg.fill_backend == seg.backend:
+            return default_mask
+        if self._fill_segmenter is None:
+            from dataclasses import replace
+
+            from .segmentation import make_segmenter
+
+            self._fill_segmenter = make_segmenter(
+                replace(seg, backend=seg.fill_backend, model_name=seg.fill_model_name)
+            )
+        return self._fill_segmenter.mask(rgb)
 
     def _predict_alpha(self, rgb: np.ndarray, trimap: np.ndarray, bg_confidence: np.ndarray) -> tuple[np.ndarray, str]:
         if self.settings.model.matting_method == "chroma":
@@ -93,7 +116,8 @@ def _run_image(
     alpha = suppress_alpha_by_chroma(alpha, bg_confidence, trimap, settings.alpha_post)
     alpha = clean_alpha(alpha, trimap, settings.alpha_post)
     if seg_mask is not None:
-        alpha = fill_seg_dropouts(alpha, seg_mask, settings.alpha_post)
+        fill_mask = pipeline._segment_fill(rgb, seg_mask)
+        alpha = fill_seg_dropouts(alpha, fill_mask, settings.alpha_post)
     foreground = estimate_foreground_rgb(
         rgb, alpha, bg_confidence, model, settings.foreground
     )
diff --git a/bgfilter/segmentation.py b/bgfilter/segmentation.py
index e6269bf..fbfcae5 100644
--- a/bgfilter/segmentation.py
+++ b/bgfilter/segmentation.py
@@ -111,6 +111,19 @@ class AnimeSegSegmenter:
         return np.asarray(mask_img, dtype=np.float32) / 255.0
 
 
+def unsharp_mask(mask: np.ndarray, amount: float, radius: int) -> np.ndarray:
+    """Sharpen a soft mask: raise thin-strand peaks and crisp the blur so strands
+    reach the sure-foreground threshold instead of dissolving into the unknown band."""
+    if amount <= 0:
+        return mask
+    from .deps import require_cv2
+
+    cv2 = require_cv2()
+    k = max(1, radius) * 2 + 1
+    blur = cv2.GaussianBlur(mask, (k, k), 0)
+    return np.clip(mask + amount * (mask - blur), 0.0, 1.0).astype(np.float32)
+
+
 def make_segmenter(settings: SegmentationSettings):
     if settings.backend == "birefnet":
         return BiRefNetSegmenter(settings)
diff --git a/bgfilter/settings.py b/bgfilter/settings.py
index 6c27e6e..03988d5 100644
--- a/bgfilter/settings.py
+++ b/bgfilter/settings.py
@@ -106,8 +106,12 @@ class SegmentationSettings:
     enabled: bool = True
     backend: str = "birefnet"  # "birefnet" or "anime-seg"
     model_name: str = "ZhengPeng7/BiRefNet"
+    fill_backend: str = ""     # "" = reuse `backend` for dropout-fill; else e.g. "anime-seg"
+    fill_model_name: str = ""
     device: str = "cuda"
     input_size: int = 1024
+    sharpen: float = 0.0       # unsharp-mask amount on the mask (0 = off)
+    sharpen_radius: int = 4
 
 
 @dataclass(frozen=True)
diff --git a/configs/default.yaml b/configs/default.yaml
index abae22e..3fc21e9 100644
--- a/configs/default.yaml
+++ b/configs/default.yaml
@@ -8,6 +8,8 @@ segmentation:
   enabled: true
   backend: birefnet  # birefnet | anime-seg (anime-seg needs model_name: skytnt/anime-seg)
   model_name: ZhengPeng7/BiRefNet
+  fill_backend: anime-seg          # dropout-fill uses anime-seg (higher wisp recall; add-only, harmless)
+  fill_model_name: skytnt/anime-seg
   device: cuda
   input_size: 1024