diff --git a/Common/ColorConv.cpp b/Common/ColorConv.cpp
index bf4276f1e3..6206847d2e 100644
--- a/Common/ColorConv.cpp
+++ b/Common/ColorConv.cpp
@@ -308,7 +308,7 @@ void ConvertRGBA565ToRGBA8888(u32 *dst32, const u16 *src, u32 numPixels) {
 		b = _mm_or_si128(_mm_slli_epi16(b, 3), _mm_srli_epi16(b, 2));
 		b = _mm_and_si128(b, mask8);
 
-		// Always set to 00FF 00FF.
+		// Always set alpha to 00FF 00FF.
 		__m128i a = _mm_slli_epi16(mask8, 8);
 
 		// Now combine them, RRGG RRGG and BBAA BBAA, and then interleave.
@@ -472,36 +472,36 @@ void ConvertRGBA4444ToBGRA8888(u32 *dst32, const u16 *src, u32 numPixels) {
 	u8 *dst = (u8 *)dst32;
 	for (u32 x = 0; x < numPixels; x++) {
 		u16 c = src[x];
-		u32 r = c & 0x000f;
-		u32 g = (c >> 4) & 0x000f;
-		u32 b = (c >> 8) & 0x000f;
-		u32 a = (c >> 12) & 0x000f;
+		u32 r = Convert4To8(c & 0x000f);
+		u32 g = Convert4To8((c >> 4) & 0x000f);
+		u32 b = Convert4To8((c >> 8) & 0x000f);
+		u32 a = Convert4To8((c >> 12) & 0x000f);
 
-		dst[x] = (r << (16 + 4)) | (g << (8 + 4)) | (b << 4) | (a << (24 + 4));
+		dst[x] = (a << 24) | (r << 16) | (g << 8) | b;
 	}
 }
 
 void ConvertRGBA5551ToBGRA8888(u32 *dst, const u16 *src, u32 numPixels) {
 	for (u32 x = 0; x < numPixels; x++) {
 		u16 c = src[x];
-		u32 r = c & 0x001f;
-		u32 g = (c >> 5) & 0x001f;
-		u32 b = (c >> 10) & 0x001f;
+		u32 r = Convert5To8(c & 0x001f);
+		u32 g = Convert6To8((c >> 5) & 0x001f);
+		u32 b = Convert5To8((c >> 10) & 0x001f);
 		// We force an arithmetic shift to get the sign bits/
 		u32 a = ((s32)(s16)c) & 0xff000000;
 
-		dst[x] = (r << (16 + 3)) | (g << (8 + 3)) | (b << 3) | a;
+		dst[x] = a | (r << 16) | (g << 8) | b;
 	}
 }
 
 void ConvertRGB565ToBGRA8888(u32 *dst, const u16 *src, u32 numPixels) {
 	for (u32 x = 0; x < numPixels; x++) {
 		u16 c = src[x];
-		u32 r = c & 0x001f;
-		u32 g = (c >> 5) & 0x003f;
-		u32 b = (c >> 11) & 0x001f;
+		u32 r = Convert5To8(c & 0x001f);
+		u32 g = Convert6To8((c >> 5) & 0x003f);
+		u32 b = Convert5To8((c >> 11) & 0x001f);
 
-		dst[x] = (r << (16 + 3)) | (g << (8 + 2)) | (b << 3) | 0xFF000000;
+		dst[x] = 0xFF000000 | (r << 16) | (g << 8) | b;
 	}
 }
 
diff --git a/GPU/Common/FramebufferCommon.cpp b/GPU/Common/FramebufferCommon.cpp
index 8853b2a262..0bd3f03985 100644
--- a/GPU/Common/FramebufferCommon.cpp
+++ b/GPU/Common/FramebufferCommon.cpp
@@ -1212,7 +1212,7 @@ void FramebufferManagerCommon::ResizeFramebufFBO(VirtualFramebuffer *vfb, int w,
 	}
 }
 
-// This is called from detected memcopies only. Not block transfers.
+// This is called from detected memcopies and framebuffer initialization from VRAM. Not block transfers.
 // MotoGP goes this path so we need to catch those copies here.
 bool FramebufferManagerCommon::NotifyFramebufferCopy(u32 src, u32 dst, int size, bool isMemset, u32 skipDrawReason) {
 	if (size == 0) {
diff --git a/GPU/D3D11/TextureCacheD3D11.cpp b/GPU/D3D11/TextureCacheD3D11.cpp
index cb593d7873..568343bdd8 100644
--- a/GPU/D3D11/TextureCacheD3D11.cpp
+++ b/GPU/D3D11/TextureCacheD3D11.cpp
@@ -399,6 +399,8 @@ void TextureCacheD3D11::ApplyTextureFramebuffer(TexCacheEntry *entry, VirtualFra
 		shaderApply.ApplyBounds(gstate_c.vertBounds, gstate_c.curTextureXOffset, gstate_c.curTextureYOffset, xoff, yoff);
 		shaderApply.Use(depalShaderCache_->GetDepalettizeVertexShader(), depalShaderCache_->GetInputLayout());
 
+		ID3D11ShaderResourceView *nullTexture = nullptr;
+		context_->PSSetShaderResources(0, 1, &nullTexture);  // In case the target was used in the last draw call. Happens in Sega Rally.
 		draw_->BindFramebufferAsRenderTarget(depalFBO, { Draw::RPAction::DONT_CARE, Draw::RPAction::DONT_CARE, Draw::RPAction::DONT_CARE });
 		context_->PSSetShaderResources(3, 1, &clutTexture);
 		context_->PSSetSamplers(3, 1, &stockD3D11.samplerPoint2DWrap);
diff --git a/ext/native/thin3d/VulkanQueueRunner.cpp b/ext/native/thin3d/VulkanQueueRunner.cpp
index 77c1c603f5..df2d386255 100644
--- a/ext/native/thin3d/VulkanQueueRunner.cpp
+++ b/ext/native/thin3d/VulkanQueueRunner.cpp
@@ -776,7 +776,10 @@ void VulkanQueueRunner::ApplyRenderPassMerge(std::vector<VKRStep *> &steps) {
 					if (steps[j]->dependencies.contains(touchedFramebuffers)) {
 						goto done_fb;
 					}
-					if (steps[j]->render.framebuffer == fb) {
+					if (steps[j]->render.framebuffer == fb &&
+						steps[j]->render.color != VKRRenderPassAction::CLEAR &&
+						steps[j]->render.depth != VKRRenderPassAction::CLEAR &&
+						steps[j]->render.stencil != VKRRenderPassAction::CLEAR) {
 						// ok. Now, if it's a render, slurp up all the commands
 						// and kill the step.
 						// Also slurp up any pretransitions.