used a proper normal matrix to transform normals

src/blender-addon/leaf/export.py

@@ -178,7 +178,7 @@ def export_texture(tex):
 
 def export_image(img):
     options = {
-        "cuda": True,
+        "cuda": False,
         "linear": img.colorspace_settings.name == "Linear",
         "normal_map": img.leaf.is_normal_map
     }

src/engine/Renderer.cpp

@@ -67,8 +67,7 @@ struct SceneState
 struct InstanceData
 {
     glm::mat4 modelMatrix;
-    glm::mat3 normalMatrix;
-    float _padding[3];
+    glm::mat3x4 normalMatrix; // use 3x4 to match cbuffer packing rules
 };
 #pragma pack(pop)
 
@@ -434,7 +433,7 @@ void Renderer::render(const Scene *scene, int width, int height, bool overrideCa
             CHECK_HRESULT(res);
             InstanceData *instanceData = (InstanceData *)mappedResource.pData;
             instanceData->modelMatrix = job.transform;
-            instanceData->normalMatrix = glm::mat3(glm::inverseTranspose(job.transform));
+            instanceData->normalMatrix = glm::mat3x4(glm::inverseTranspose(glm::mat3(job.transform)));
             Device::context->Unmap(this->cbInstance, 0);
 
             Device::context->DrawIndexed(currentMesh->getIndexCount(), 0, 0);

src/engine/shaders/instance.h

@@ -1,4 +1,5 @@
 cbuffer InstanceData : register(b2)
 {
     float4x4 modelMatrix;
+    float3x3 normalMatrix;
 };

src/engine/shaders/standard.vs.hlsl

@@ -22,8 +22,8 @@ STANDARD_PS_INPUT main(VS_INPUT input)
     output.position.z = (output.position.z + output.position.w) * 0.5;
 
     output.worldPosition = worldPosition.xyz;
-    output.normal = mul((float3x3)modelMatrix, input.normal);
-    output.tangent = float4(mul((float3x3)modelMatrix, input.tangent.xyz), input.tangent.w);
+    output.normal = mul(normalMatrix, input.normal);
+    output.tangent = float4(mul(normalMatrix, input.tangent.xyz), input.tangent.w);
     output.uv = float2(input.uv.x, 1.0 - input.uv.y);
 
     return output;