diff --git a/backend/src/packages/chaiNNer_onnx/__init__.py b/backend/src/packages/chaiNNer_onnx/__init__.py
index e99d17407..a57ca7b83 100644
--- a/backend/src/packages/chaiNNer_onnx/__init__.py
+++ b/backend/src/packages/chaiNNer_onnx/__init__.py
@@ -2,7 +2,7 @@
 
 from api import KB, MB, Dependency, add_package
 from gpu import nvidia
-from system import is_arm_mac
+from system import is_arm_mac, is_windows
 
 general = "ONNX uses .onnx models to upscale images."
 conversion = "It also helps to convert between PyTorch and NCNN."
@@ -15,7 +15,7 @@
         f"{general} {conversion} It is fastest when CUDA is supported. If TensorRT is"
         " installed on the system, it can also be configured to use that."
     )
-    inst_hint = f"{general} It does not support AMD GPUs."
+    inst_hint = f"{general} It does not support AMD GPUs, in linux."
 
 
 def get_onnx_runtime():
@@ -28,6 +28,13 @@ def get_onnx_runtime():
             import_name="onnxruntime",
             extra_index_url="https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-12/pypi/simple/",
         )
+    elif is_windows:
+        return Dependency(
+            display_name="ONNX Runtime (DirectMl)",
+            pypi_name="onnxruntime-directml",
+            version="1.17.1",
+            size_estimate=15 * MB,
+        )
     else:
         return Dependency(
             display_name="ONNX Runtime",