diff --git a/oci/src/proxmenux_oci/installer.py b/oci/src/proxmenux_oci/installer.py
index 72f0c284..96d47bd6 100644
--- a/oci/src/proxmenux_oci/installer.py
+++ b/oci/src/proxmenux_oci/installer.py
@@ -962,15 +962,14 @@ def _ask_immich_acceleration(ui, rootfs_storage: str | None = None) -> tuple[str
         return software
     # Recognition on an AMD GPU depends on ROCm: it is offered when the host
     # can run it, marked as experimental on a GPU ROCm does not support
-    # officially, and left out, with the reason, when it cannot.
+    # officially, and left out when it cannot. A GPU of a generation ROCm does
+    # not run is simply not offered; what the host can fix is explained.
     amd_blocker = host.rocm_blocker(rootfs_storage) if found["amd"] else None
     amd_experimental = (found["amd"] and not amd_blocker
                         and host.rocm_support(found.get("amd_gfx_target")) == "experimental")
-    if amd_blocker:
+    if amd_blocker in ("kfd", "space"):
         reason = (translate("The AMD driver does not offer its compute interface (/dev/kfd) on this host.")
                   if amd_blocker == "kfd" else
-                  f"{translate('The ROCm image has no support for the AMD GPU of this host:')} {host.amd_gpu_name()}."
-                  if amd_blocker == "generation" else
                   translate("The storage has less than 40 GB free for the ROCm image."))
         real.message(f"{reason}\n\n{translate('Recognition is not offered on the AMD GPU; video transcoding is.')}")
     mark = f" — {translate('experimental on this GPU')}" if amd_experimental else ""
@@ -1916,11 +1915,6 @@ def configure_acceleration(installer_profile, environment, unprivileged, ui, mod
             found.get("amd_gfx_target"), tuple(item["amd_gfx_targets"])) == "experimental"}
         options = [(item["id"], item["label"]) for item in usable]
         asked = advanced or not installer_profile.get("selkies")
-        if asked and found["amd"] and any(item.get("amd_gfx_targets") and item not in usable and host.rocm_support(
-                found.get("amd_gfx_target"), tuple(item["amd_gfx_targets"])) is None for item in profiles):
-            # Its owner may expect the GPU to be offered: say why it is not.
-            ui.message(f"{translate('The ROCm image of this application has no support for the AMD GPU of this host:')} "
-                       f"{host.amd_gpu_name()}. {translate('Its ROCm profile is not offered.')}")
         if asked and len(usable) < len(profiles) and len(usable) == 1:
             # Nothing but the CPU is left: say why there is nothing to choose.
             ui.message(translate("No usable GPU was found on this host. The application will be installed "
diff --git a/oci/tests/test_frigate_amd_profile.py b/oci/tests/test_frigate_amd_profile.py
index ae121d10..447eae02 100644
--- a/oci/tests/test_frigate_amd_profile.py
+++ b/oci/tests/test_frigate_amd_profile.py
@@ -191,8 +191,10 @@ class FrigateExperimentalRocmTests(unittest.TestCase):
         self.assertEqual(plan["hardware_profile"], "vaapi")
         self.assertEqual(self.override(plan), [])
 
-    def test_a_gpu_the_image_cannot_use_is_not_offered_and_the_reason_is_said(self, *_):
+    def test_a_gpu_the_image_cannot_use_is_simply_not_offered(self, *_):
+        # The menu offers what works on the GPU; it does not explain ROCm.
         with patch("proxmenux_oci.installer.host.amd_gpu_name", return_value="Lucienne (gfx90c)"):
             ui, plan = self.build(90012, {})
         self.assertNotIn("rocm", ui.labels[PROMPT])
-        self.assertTrue(any("Lucienne (gfx90c)" in message and "not offered" in message for message in ui.messages))
+        self.assertIn("vaapi", ui.labels[PROMPT])
+        self.assertFalse(any("ROCm" in message for message in ui.messages))
diff --git a/oci/tests/test_immich_acceleration.py b/oci/tests/test_immich_acceleration.py
index 4a52f6e9..586dca5b 100644
--- a/oci/tests/test_immich_acceleration.py
+++ b/oci/tests/test_immich_acceleration.py
@@ -92,16 +92,21 @@ class ImmichAccelerationTests(unittest.TestCase):
             self.assertIn("No usable GPU", ui.messages[0])
             self.assertEqual(result, ("cpu", "cpu"), mode)
 
-    def test_an_amd_host_that_cannot_run_rocm_offers_video_only_and_says_why(self, *_):
-        for blocker, text in (("kfd", "/dev/kfd"), ("space", "40 GB"), ("generation", "no support for the AMD GPU")):
+    def test_an_amd_host_that_cannot_run_rocm_offers_video_only(self, *_):
+        # What the host can fix is said; a GPU generation ROCm does not run is not.
+        for blocker, text in (("kfd", "/dev/kfd"), ("space", "40 GB"), ("generation", None)):
             with patch("proxmenux_oci.installer.host.rocm_blocker", return_value=blocker), \
                     patch("proxmenux_oci.installer.host.amd_gpu_name", return_value="Lucienne (gfx90c)"):
                 ui, result, _ = self.plan(AMD, DEFAULT_MODE)
             self.assertEqual(ui.options[PROMPT], ["cpu", "amd-video"], blocker)
             self.assertEqual(ui.defaults[PROMPT], "amd-video", blocker)
             self.assertEqual(result, ("vaapi", "cpu"), blocker)
-            self.assertTrue(any(text in message and "Recognition is not offered" in message
-                                for message in ui.messages), blocker)
+            if text:
+                self.assertTrue(any(text in message and "Recognition is not offered" in message
+                                    for message in ui.messages), blocker)
+            else:
+                self.assertFalse(any("ROCm" in message or "Recognition" in message
+                                     for message in ui.messages), blocker)
 
     def test_a_gpu_rocm_supports_is_proposed_whole(self, *_):
         for target in (100300, 110501, None):