diff --git a/Common/Data/Encoding/Utf8.cpp b/Common/Data/Encoding/Utf8.cpp
index 09abf6090d..4e7a2c7a63 100644
--- a/Common/Data/Encoding/Utf8.cpp
+++ b/Common/Data/Encoding/Utf8.cpp
@@ -24,6 +24,7 @@
 #include <cstdint>
 #include <algorithm>
 #include <string>
+#include <string_view>
 
 #include "Common/Data/Encoding/Utf8.h"
 #include "Common/Data/Encoding/Utf16.h"
@@ -373,6 +374,46 @@ void ConvertUTF8ToJavaModifiedUTF8(std::string *output, std::string_view input)
 	_dbg_assert_(output->size() >= input.size());
 }
 
+std::string NormalizeForSearch(std::string_view input) {
+	std::string result;
+	// Pre-allocating input size is a good heuristic, though the
+	// result could be slightly smaller after normalization.
+	result.reserve(input.size());
+
+	int index = 0;
+	int size = static_cast<int>(input.size());
+	char buffer[4]; // Temporary buffer for UTF-8 encoding
+
+	while (index < size) {
+		uint32_t codepoint = u8_nextchar(input.data(), &index, size);
+
+		// 1. Convert Fullwidth Roman/Numbers to ASCII
+		// Range: U+FF01 (！) to U+FF5E (～)
+		if (codepoint >= 0xFF01 && codepoint <= 0xFF5E) {
+			codepoint -= 0xFEE0;
+		}
+		// Convert Fullwidth Space (U+3000) to standard space
+		else if (codepoint == 0x3000) {
+			codepoint = 0x20;
+		}
+
+		// 2. Lowercase (Basic Latin range)
+		// We do this after the wide-to-ascii conversion to catch characters
+		// that were originally wide uppercase (e.g., 'Ａ' -> 'A' -> 'a').
+		if (codepoint >= 'A' && codepoint <= 'Z') {
+			codepoint += ('a' - 'A');
+		}
+
+		// 3. Re-encode back to UTF-8
+		int bytes_written = u8_wc_toutf8(buffer, codepoint);
+		if (bytes_written > 0) {
+			result.append(buffer, bytes_written);
+		}
+	}
+
+	return result;
+}
+
 #ifndef _WIN32
 
 // Replacements for the Win32 wstring functions. Not to be used from emulation code!
diff --git a/Common/Data/Encoding/Utf8.h b/Common/Data/Encoding/Utf8.h
index ba72eeb9fe..9a752f8c70 100644
--- a/Common/Data/Encoding/Utf8.h
+++ b/Common/Data/Encoding/Utf8.h
@@ -129,3 +129,6 @@ std::u16string ConvertUTF8ToUCS2(std::string_view source);
 // Java needs 4-byte UTF-8 to be converted to surrogate pairs, each component of which get
 // encoded into 3-byte UTF-8.
 void ConvertUTF8ToJavaModifiedUTF8(std::string *output, std::string_view input);
+
+// Applies "tolower" and also rewrites full-with japanese characters.
+std::string NormalizeForSearch(std::string_view input);
diff --git a/UI/GameBrowser.cpp b/UI/GameBrowser.cpp
index 96e2c3fee5..6e3841f77f 100644
--- a/UI/GameBrowser.cpp
+++ b/UI/GameBrowser.cpp
@@ -563,6 +563,8 @@ bool GameBrowser::Key(const KeyInput &input) {
 					// TODO: Restore focus state here.
 					UI::EnableFocusMovement(false);
 				}
+			} else {
+				// Empty search filter. Navigate upwards on backspace?
 			}
 		} else if (!searchFilter_.empty() && input.keyCode == NKCODE_ESCAPE) {
 			searchFilter_.clear();
@@ -594,8 +596,7 @@ void GameBrowser::ApplySearchFilter(bool setKeyboardFocus) {
 		return;
 	}
 
-	std::string filter = searchFilter_;
-	std::transform(filter.begin(), filter.end(), filter.begin(), tolower);
+	std::string filter = NormalizeForSearch(searchFilter_);
 
 	searchPending_ = false;
 	// By default, everything is matching.
@@ -632,7 +633,7 @@ void GameBrowser::ApplySearchFilter(bool setKeyboardFocus) {
 			continue;
 		}
 
-		std::transform(label.begin(), label.end(), label.begin(), tolower);
+		label = NormalizeForSearch(label);
 		bool match = v->CanBeFocused() && label.find(filter) != label.npos;
 		if (match && !firstMatch) {
 			firstMatch = v;
@@ -1152,6 +1153,9 @@ void GameBrowser::NavigateClick(UI::EventParams &e) {
 		path_.Navigate(text.ToString());
 	}
 	g_Config.currentDirectory = path_.GetPath();
+	// Clear the search filter. This allow for smooth directory navigation using search
+	// (although there's no good way of going up...)
+	SetSearchFilter("", false);
 	Refresh();
 }