diff --git a/MobileGL/MG_Test/Texture/TextureTest.cpp b/MobileGL/MG_Test/Texture/TextureTest.cpp
index 40534417..08339cf3 100644
--- a/MobileGL/MG_Test/Texture/TextureTest.cpp
+++ b/MobileGL/MG_Test/Texture/TextureTest.cpp
@@ -4033,6 +4033,8 @@ TEST_F(TextureTest, ThreeChannelWideningRetargetsInternalFormatAndTransferPairTo
     const Flags<PixelFormatNormalizeOptionBit> widenNoSnorm16 =
         PixelFormatNormalizeOptionBit::NoThreeChannelRenderTarget |
         PixelFormatNormalizeOptionBit::NoSnorm16RenderTarget;
+    const Flags<PixelFormatNormalizeOptionBit> widenNoNorm16 =
+        PixelFormatNormalizeOptionBit::NoThreeChannelRenderTarget | PixelFormatNormalizeOptionBit::NoNorm16;
 
     const Case cases[] = {
         // Complementary's colortex1 and colortex2. The transfer pair used to stay three-channel
@@ -4047,10 +4049,16 @@ TEST_F(TextureTest, ThreeChannelWideningRetargetsInternalFormatAndTransferPairTo
         // cannot render to the encoding gets the 32-bit float rather than the half.
         {GL_RGB16_SNORM, widen, GL_RGBA16_SNORM, GL_RGBA, GL_SHORT},
         {GL_RGB16_SNORM, widenNoSnorm16, GL_RGBA32F, GL_RGBA, GL_FLOAT},
-        // 16-bit UNORM and the legacy 10/12-bit formats stored as RGB16.
-        {GL_RGB16, widen, GL_RGBA32F, GL_RGBA, GL_FLOAT},
-        {GL_RGB10, widen, GL_RGBA32F, GL_RGBA, GL_FLOAT},
-        {GL_RGB12, widen, GL_RGBA32F, GL_RGBA, GL_FLOAT},
+        // 16-bit UNORM and the legacy 10/12-bit formats stored as RGB16. The same-width sibling
+        // whenever the driver has EXT_texture_norm16 - which is what keeps the whole 48-bit
+        // ARB_texture_view class on one ES view class, so a GL_RGB16 texture can be viewed as
+        // GL_RGB16UI - and the 32-bit float only when it does not.
+        {GL_RGB16, widen, GL_RGBA16, GL_RGBA, GL_UNSIGNED_SHORT},
+        {GL_RGB10, widen, GL_RGBA16, GL_RGBA, GL_UNSIGNED_SHORT},
+        {GL_RGB12, widen, GL_RGBA16, GL_RGBA, GL_UNSIGNED_SHORT},
+        {GL_RGB16, widenNoNorm16, GL_RGBA32F, GL_RGBA, GL_FLOAT},
+        {GL_RGB10, widenNoNorm16, GL_RGBA32F, GL_RGBA, GL_FLOAT},
+        {GL_RGB12, widenNoNorm16, GL_RGBA32F, GL_RGBA, GL_FLOAT},
         // sRGB and the integer formats: the base format has to move to the four-channel one of the
         // right class, GL_RGBA_INTEGER included.
         {GL_SRGB8, widen, GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE},
diff --git a/MobileGL/MG_Util/Texture/TextureFormatProcessor.cpp b/MobileGL/MG_Util/Texture/TextureFormatProcessor.cpp
index 7ae3886d..df504d4e 100644
--- a/MobileGL/MG_Util/Texture/TextureFormatProcessor.cpp
+++ b/MobileGL/MG_Util/Texture/TextureFormatProcessor.cpp
@@ -134,7 +134,21 @@ namespace MobileGL::MG_Util::TextureFormatProcessor {
             case GL_RGB16:
             case GL_RGB10:
             case GL_RGB12:
-                return {GL_RGBA32F, GL_RGBA, GL_FLOAT};
+                // Same reasoning as GL_RGB16_SNORM above, and the same shape: keep the
+                // unsigned-normalized encoding whenever the driver has it, because
+                // GL_RGBA16 is the SAME-WIDTH four-channel sibling and GL_RGBA32F is not.
+                // That matters beyond storage size. ARB_texture_view puts all five 48-bit
+                // formats in one view class, so a GL_RGB16 texture viewed as GL_RGB16UI has
+                // to alias storage the ES driver also considers compatible; against a
+                // GL_RGBA32F carrier the view is a different class and glTextureView is
+                // refused outright (KHR-GL4x.texture_view.view_classes). Against GL_RGBA16
+                // the whole class lands on ES's 64-bit class and every channel reinterprets
+                // bit-exactly. EXT_texture_norm16 - the absence of which is what NoNorm16
+                // means - is also what makes GL_RGBA16 colour-renderable, so the two
+                // questions have one answer.
+                return (options & PixelFormatNormalizeOptionBit::NoNorm16)
+                           ? ThreeChannelWidening{GL_RGBA32F, GL_RGBA, GL_FLOAT}
+                           : ThreeChannelWidening{GL_RGBA16, GL_RGBA, GL_UNSIGNED_SHORT};
             // Floating point.
             case GL_RGB16F:
                 return {GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT};