diff --git a/bgfilter/segmentation.py b/bgfilter/segmentation.py
index f7fa882..5792dcd 100644
--- a/bgfilter/segmentation.py
+++ b/bgfilter/segmentation.py
@@ -38,10 +38,11 @@ class BiRefNetSegmenter:
 
         self.torch = torch
         self.model = AutoModelForImageSegmentation.from_pretrained(
-            resolve_model_source(settings.model_name), trust_remote_code=True
+            resolve_model_source(settings.model_name),
+            trust_remote_code=True,
+            torch_dtype=torch.float16,  # checkpoint ships fp16; run it as-is, inputs cast to match
         )
         self.model.eval()
-        self.model.float()  # checkpoint ships as fp16; force fp32 to match inputs
         self.device = self._resolve_device(settings.device)
         self.model.to(self.device)
 
@@ -64,8 +65,8 @@ class BiRefNetSegmenter:
     def mask(self, rgb: np.ndarray) -> np.ndarray:
         """Return a soft foreground mask, float32 0..1, at the input resolution."""
         image = Image.fromarray(rgb.astype(np.uint8), mode="RGB")
-        tensor = self.transform(image).unsqueeze(0).to(self.device)
-        with self.torch.no_grad():
+        tensor = self.transform(image).unsqueeze(0).to(self.device, dtype=self.torch.float16)
+        with self.torch.inference_mode():
             pred = self.model(tensor)[-1].sigmoid()
         pred = pred[0, 0].detach().float().cpu().numpy()
         if pred.shape != rgb.shape[:2]:
diff --git a/bgfilter/vitmatte_infer.py b/bgfilter/vitmatte_infer.py
index 9884f4f..3d5f953 100644
--- a/bgfilter/vitmatte_infer.py
+++ b/bgfilter/vitmatte_infer.py
@@ -21,7 +21,7 @@ class ViTMatteRunner:
         self.torch = torch
         source = resolve_model_source(settings.model_name)
         self.processor = VitMatteImageProcessor.from_pretrained(source)
-        self.model = VitMatteForImageMatting.from_pretrained(source)
+        self.model = VitMatteForImageMatting.from_pretrained(source, torch_dtype=torch.float16)
         self.device = self._resolve_device(settings.device)
         self.model.to(self.device)
         self.model.eval()
@@ -37,8 +37,13 @@ class ViTMatteRunner:
         image = Image.fromarray(rgb.astype(np.uint8), mode="RGB")
         trimap_image = Image.fromarray(trimap.astype(np.uint8), mode="L")
         inputs = self.processor(images=image, trimaps=trimap_image, return_tensors="pt")
-        inputs = {key: value.to(self.device) for key, value in inputs.items()}
-        with self.torch.no_grad():
+        inputs = {
+            key: value.to(self.device, dtype=self.torch.float16)
+            if value.is_floating_point()
+            else value.to(self.device)
+            for key, value in inputs.items()
+        }
+        with self.torch.inference_mode():
             outputs = self.model(**inputs)
 
         alpha = outputs.alphas[0, 0].detach().float().cpu().numpy()