diff --git a/GPU/Common/SoftwareTransformCommon.cpp b/GPU/Common/SoftwareTransformCommon.cpp
index a29c987c11..93385d8aa3 100644
--- a/GPU/Common/SoftwareTransformCommon.cpp
+++ b/GPU/Common/SoftwareTransformCommon.cpp
@@ -125,6 +125,33 @@ static bool IsReallyAClear(const TransformedVertex *transformed, int numVerts, f
 	return true;
 }
 
+static int ColorIndexOffset(int prim, GEShadeMode shadeMode, bool clearMode) {
+	if (shadeMode != GE_SHADE_FLAT || clearMode) {
+		return 0;
+	}
+
+	switch (prim) {
+	case GE_PRIM_LINES:
+	case GE_PRIM_LINE_STRIP:
+		return 1;
+
+	case GE_PRIM_TRIANGLES:
+	case GE_PRIM_TRIANGLE_STRIP:
+		return 2;
+
+	case GE_PRIM_TRIANGLE_FAN:
+		return 1;
+
+	case GE_PRIM_RECTANGLES:
+		// We already use BR color when expanding, so no need to offset.
+		return 0;
+
+	default:
+		break;
+	}
+	return 0;
+}
+
 void SoftwareTransform(
 	int prim, int vertexCount, u32 vertType, u16 *&inds, int indexType,
 	const DecVtxFormat &decVtxFormat, int &maxIndex, TransformedVertex *&drawBuffer, int &numTrans, bool &drawIndexed, const SoftwareTransformParams *params, SoftwareTransformResult *result) {
@@ -168,6 +195,11 @@ void SoftwareTransform(
 		fog_slope = 1.0f;
 	}
 
+	int colorIndOffset = 0;
+	if (params->provokeFlatFirst) {
+		colorIndOffset = ColorIndexOffset(prim, gstate.getShadeMode(), gstate.isModeClear());
+	}
+
 	VertexReader reader(decoded, decVtxFormat, vertType);
 	if (throughmode) {
 		for (int index = 0; index < maxIndex; index++) {
@@ -178,7 +210,13 @@ void SoftwareTransform(
 			reader.ReadPos(vert.pos);
 
 			if (reader.hasColor0()) {
-				reader.ReadColor0_8888(vert.color0);
+				if (colorIndOffset != 0 && index + colorIndOffset < maxIndex) {
+					reader.Goto(index + colorIndOffset);
+					reader.ReadColor0_8888(vert.color0);
+					reader.Goto(index);
+				} else {
+					reader.ReadColor0_8888(vert.color0);
+				}
 			} else {
 				vert.color0_32 = gstate.getMaterialAmbientRGBA();
 			}
@@ -260,7 +298,13 @@ void SoftwareTransform(
 			// Perform lighting here if enabled. don't need to check through, it's checked above.
 			Vec4f unlitColor = Vec4f(1, 1, 1, 1);
 			if (reader.hasColor0()) {
-				reader.ReadColor0(&unlitColor.x);
+				if (colorIndOffset != 0 && index + colorIndOffset < maxIndex) {
+					reader.Goto(index + colorIndOffset);
+					reader.ReadColor0(&unlitColor.x);
+					reader.Goto(index);
+				} else {
+					reader.ReadColor0(&unlitColor.x);
+				}
 			} else {
 				unlitColor = Vec4f::FromRGBA(gstate.getMaterialAmbientRGBA());
 			}
diff --git a/GPU/Common/SoftwareTransformCommon.h b/GPU/Common/SoftwareTransformCommon.h
index 9bc2e4da49..a7efc396c9 100644
--- a/GPU/Common/SoftwareTransformCommon.h
+++ b/GPU/Common/SoftwareTransformCommon.h
@@ -46,6 +46,7 @@ struct SoftwareTransformParams {
 	TextureCacheCommon *texCache;
 	bool allowClear;
 	bool allowSeparateAlphaClear;
+	bool provokeFlatFirst;
 };
 
 void SoftwareTransform(int prim, int vertexCount, u32 vertexType, u16 *&inds, int indexType, const DecVtxFormat &decVtxFormat, int &maxIndex, TransformedVertex *&drawBuffer,
diff --git a/GPU/D3D11/DrawEngineD3D11.cpp b/GPU/D3D11/DrawEngineD3D11.cpp
index 2b49ae2d6e..4651c38290 100644
--- a/GPU/D3D11/DrawEngineD3D11.cpp
+++ b/GPU/D3D11/DrawEngineD3D11.cpp
@@ -587,6 +587,7 @@ rotateVBO:
 		params.texCache = textureCache_;
 		params.allowClear = true;
 		params.allowSeparateAlphaClear = false;  // D3D11 doesn't support separate alpha clears
+		params.provokeFlatFirst = true;
 
 		int maxIndex = indexGen.MaxIndex();
 		SoftwareTransform(
diff --git a/GPU/Directx9/DrawEngineDX9.cpp b/GPU/Directx9/DrawEngineDX9.cpp
index c72d256727..de1a85d093 100644
--- a/GPU/Directx9/DrawEngineDX9.cpp
+++ b/GPU/Directx9/DrawEngineDX9.cpp
@@ -545,6 +545,7 @@ rotateVBO:
 		params.texCache = textureCache_;
 		params.allowClear = true;
 		params.allowSeparateAlphaClear = true;
+		params.provokeFlatFirst = true;
 
 		int maxIndex = indexGen.MaxIndex();
 		SoftwareTransform(
diff --git a/GPU/GLES/DrawEngineGLES.cpp b/GPU/GLES/DrawEngineGLES.cpp
index d79ee4efd4..79589a113b 100644
--- a/GPU/GLES/DrawEngineGLES.cpp
+++ b/GPU/GLES/DrawEngineGLES.cpp
@@ -553,6 +553,7 @@ rotateVBO:
 		params.texCache = textureCache_;
 		params.allowClear = true;
 		params.allowSeparateAlphaClear = true;
+		params.provokeFlatFirst = false;
 
 		int maxIndex = indexGen.MaxIndex();
 		int vertexCount = indexGen.VertexCount();
diff --git a/GPU/Vulkan/DrawEngineVulkan.cpp b/GPU/Vulkan/DrawEngineVulkan.cpp
index 76643d6066..f68f989e7f 100644
--- a/GPU/Vulkan/DrawEngineVulkan.cpp
+++ b/GPU/Vulkan/DrawEngineVulkan.cpp
@@ -860,6 +860,7 @@ void DrawEngineVulkan::DoFlush() {
 		// do not respect scissor rects.
 		params.allowClear = g_Config.iRenderingMode != 0;
 		params.allowSeparateAlphaClear = false;
+		params.provokeFlatFirst = true;
 
 		int maxIndex = indexGen.MaxIndex();
 		SoftwareTransform(