diff --git a/comfy/cli_args.py b/comfy/cli_args.py
index 3947e62a8..2f30b72d2 100644
--- a/comfy/cli_args.py
+++ b/comfy/cli_args.py
@@ -145,10 +145,11 @@ class PerformanceFeature(enum.Enum):
     Fp8MatrixMultiplication = "fp8_matrix_mult"
     CublasOps = "cublas_ops"
     AutoTune = "autotune"
-    PinnedMem = "pinned_memory"
 
 parser.add_argument("--fast", nargs="*", type=PerformanceFeature, help="Enable some untested and potentially quality deteriorating optimizations. This is used to test new features so using it might crash your comfyui. --fast with no arguments enables everything. You can pass a list specific optimizations if you only want to enable specific ones. Current valid optimizations: {}".format(" ".join(map(lambda c: c.value, PerformanceFeature))))
 
+parser.add_argument("--disable-pinned-memory", action="store_true", help="Disable pinned memory use.")
+
 parser.add_argument("--mmap-torch-files", action="store_true", help="Use mmap when loading ckpt/pt files.")
 parser.add_argument("--disable-mmap", action="store_true", help="Don't use mmap when loading safetensors.")
 
diff --git a/comfy/ldm/qwen_image/controlnet.py b/comfy/ldm/qwen_image/controlnet.py
index 92ac3cf0a..a6d408104 100644
--- a/comfy/ldm/qwen_image/controlnet.py
+++ b/comfy/ldm/qwen_image/controlnet.py
@@ -44,7 +44,7 @@ class QwenImageControlNetModel(QwenImageTransformer2DModel):
         txt_start = round(max(((x.shape[-1] + (self.patch_size // 2)) // self.patch_size) // 2, ((x.shape[-2] + (self.patch_size // 2)) // self.patch_size) // 2))
         txt_ids = torch.arange(txt_start, txt_start + context.shape[1], device=x.device).reshape(1, -1, 1).repeat(x.shape[0], 1, 3)
         ids = torch.cat((txt_ids, img_ids), dim=1)
-        image_rotary_emb = self.pe_embedder(ids).squeeze(1).unsqueeze(2).to(x.dtype)
+        image_rotary_emb = self.pe_embedder(ids).to(x.dtype).contiguous()
         del ids, txt_ids, img_ids
 
         hidden_states = self.img_in(hidden_states) + self.controlnet_x_embedder(hint)
diff --git a/comfy/model_management.py b/comfy/model_management.py
index 56a16be33..26f4e269f 100644
--- a/comfy/model_management.py
+++ b/comfy/model_management.py
@@ -1086,22 +1086,21 @@ def cast_to_device(tensor, device, dtype, copy=False):
 
 PINNED_MEMORY = {}
 TOTAL_PINNED_MEMORY = 0
-if PerformanceFeature.PinnedMem in args.fast:
-    if WINDOWS:
-        MAX_PINNED_MEMORY = get_total_memory(torch.device("cpu")) * 0.45  # Windows limit is apparently 50%
-    else:
-        MAX_PINNED_MEMORY = get_total_memory(torch.device("cpu")) * 0.95
-else:
-    MAX_PINNED_MEMORY = -1
+MAX_PINNED_MEMORY = -1
+if not args.disable_pinned_memory:
+    if is_nvidia() or is_amd():
+        if WINDOWS:
+            MAX_PINNED_MEMORY = get_total_memory(torch.device("cpu")) * 0.45  # Windows limit is apparently 50%
+        else:
+            MAX_PINNED_MEMORY = get_total_memory(torch.device("cpu")) * 0.95
+        logging.info("Enabled pinned memory {}".format(MAX_PINNED_MEMORY // (1024 * 1024)))
+
 
 def pin_memory(tensor):
     global TOTAL_PINNED_MEMORY
     if MAX_PINNED_MEMORY <= 0:
         return False
 
-    if not is_nvidia():
-        return False
-
     if not is_device_cpu(tensor.device):
         return False
 
@@ -1122,9 +1121,6 @@ def unpin_memory(tensor):
     if MAX_PINNED_MEMORY <= 0:
         return False
 
-    if not is_nvidia():
-        return False
-
     if not is_device_cpu(tensor.device):
         return False