diff --git a/codex-rs/app-server-protocol/schema/json/ClientRequest.json b/codex-rs/app-server-protocol/schema/json/ClientRequest.json
index d2a9de926a..d05de8638d 100644
--- a/codex-rs/app-server-protocol/schema/json/ClientRequest.json
+++ b/codex-rs/app-server-protocol/schema/json/ClientRequest.json
@@ -660,6 +660,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -671,9 +685,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -683,7 +694,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageContentItem",
@@ -1420,6 +1430,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -1431,9 +1455,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -1443,7 +1464,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/ServerNotification.json b/codex-rs/app-server-protocol/schema/json/ServerNotification.json
index 83aa38fdcf..6d7bc110d5 100644
--- a/codex-rs/app-server-protocol/schema/json/ServerNotification.json
+++ b/codex-rs/app-server-protocol/schema/json/ServerNotification.json
@@ -1759,6 +1759,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -1770,9 +1784,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -1782,7 +1793,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.schemas.json b/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.schemas.json
index ac76ffbb2d..813df4d7f3 100644
--- a/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.schemas.json
+++ b/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.schemas.json
@@ -10472,6 +10472,20 @@
             "type": "object"
           },
           {
+            "anyOf": [
+              {
+                "properties": {
+                  "image_url": {
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "image_url"
+                ],
+                "title": "ImageUrlContentItem",
+                "type": "object"
+              }
+            ],
             "properties": {
               "detail": {
                 "anyOf": [
@@ -10483,9 +10497,6 @@
                   }
                 ]
               },
-              "image_url": {
-                "type": "string"
-              },
               "type": {
                 "enum": [
                   "input_image"
@@ -10495,7 +10506,6 @@
               }
             },
             "required": [
-              "image_url",
               "type"
             ],
             "title": "InputImageContentItem",
@@ -12333,6 +12343,20 @@
             "type": "object"
           },
           {
+            "anyOf": [
+              {
+                "properties": {
+                  "image_url": {
+                    "type": "string"
+                  }
+                },
+                "required": [
+                  "image_url"
+                ],
+                "title": "ImageUrlFunctionCallOutputContentItem",
+                "type": "object"
+              }
+            ],
             "properties": {
               "detail": {
                 "anyOf": [
@@ -12344,9 +12368,6 @@
                   }
                 ]
               },
-              "image_url": {
-                "type": "string"
-              },
               "type": {
                 "enum": [
                   "input_image"
@@ -12356,7 +12377,6 @@
               }
             },
             "required": [
-              "image_url",
               "type"
             ],
             "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.v2.schemas.json b/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.v2.schemas.json
index 7a95a06a2a..0e4c577b0c 100644
--- a/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.v2.schemas.json
+++ b/codex-rs/app-server-protocol/schema/json/codex_app_server_protocol.v2.schemas.json
@@ -6352,6 +6352,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -6363,9 +6377,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -6375,7 +6386,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageContentItem",
@@ -8213,6 +8223,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -8224,9 +8248,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -8236,7 +8257,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ItemCompletedNotification.json b/codex-rs/app-server-protocol/schema/json/v2/ItemCompletedNotification.json
index f3caf84e96..dc2d0dc619 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ItemCompletedNotification.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ItemCompletedNotification.json
@@ -359,6 +359,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -370,9 +384,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -382,7 +393,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ItemStartedNotification.json b/codex-rs/app-server-protocol/schema/json/v2/ItemStartedNotification.json
index 8432eb5437..c153e128b3 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ItemStartedNotification.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ItemStartedNotification.json
@@ -359,6 +359,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -370,9 +384,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -382,7 +393,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/RawResponseItemCompletedNotification.json b/codex-rs/app-server-protocol/schema/json/v2/RawResponseItemCompletedNotification.json
index c6d8c66fe1..c05b788d8a 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/RawResponseItemCompletedNotification.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/RawResponseItemCompletedNotification.json
@@ -80,6 +80,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -91,9 +105,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -103,7 +114,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageContentItem",
@@ -188,6 +198,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -199,9 +223,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -211,7 +232,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ReviewStartResponse.json b/codex-rs/app-server-protocol/schema/json/v2/ReviewStartResponse.json
index 7da4ff34d8..cddc22f59f 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ReviewStartResponse.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ReviewStartResponse.json
@@ -499,6 +499,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -510,9 +524,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -522,7 +533,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ThreadForkResponse.json b/codex-rs/app-server-protocol/schema/json/v2/ThreadForkResponse.json
index 8a7fa76e53..a10d95cf00 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ThreadForkResponse.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ThreadForkResponse.json
@@ -580,6 +580,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -591,9 +605,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -603,7 +614,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ThreadItemsListResponse.json b/codex-rs/app-server-protocol/schema/json/v2/ThreadItemsListResponse.json
index d11f47942a..f467944e67 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ThreadItemsListResponse.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ThreadItemsListResponse.json
@@ -359,6 +359,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -370,9 +384,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -382,7 +393,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ThreadListResponse.json b/codex-rs/app-server-protocol/schema/json/v2/ThreadListResponse.json
index 32101efd89..537af065c3 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ThreadListResponse.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ThreadListResponse.json
@@ -502,6 +502,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -513,9 +527,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -525,7 +536,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ThreadMetadataUpdateResponse.json b/codex-rs/app-server-protocol/schema/json/v2/ThreadMetadataUpdateResponse.json
index f411af6d60..c76c1d2491 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ThreadMetadataUpdateResponse.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ThreadMetadataUpdateResponse.json
@@ -502,6 +502,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -513,9 +527,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -525,7 +536,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ThreadReadResponse.json b/codex-rs/app-server-protocol/schema/json/v2/ThreadReadResponse.json
index a639e719b8..510c26c2ee 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ThreadReadResponse.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ThreadReadResponse.json
@@ -502,6 +502,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -513,9 +527,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -525,7 +536,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ThreadResumeParams.json b/codex-rs/app-server-protocol/schema/json/v2/ThreadResumeParams.json
index 5c35287d0e..753f062836 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ThreadResumeParams.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ThreadResumeParams.json
@@ -142,6 +142,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -153,9 +167,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -165,7 +176,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageContentItem",
@@ -250,6 +260,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -261,9 +285,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -273,7 +294,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ThreadResumeResponse.json b/codex-rs/app-server-protocol/schema/json/v2/ThreadResumeResponse.json
index 2019868698..3a6fb8bdaa 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ThreadResumeResponse.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ThreadResumeResponse.json
@@ -596,6 +596,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -607,9 +621,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -619,7 +630,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ThreadRevertResponse.json b/codex-rs/app-server-protocol/schema/json/v2/ThreadRevertResponse.json
index 8dd0da9b6a..8d8a69f51d 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ThreadRevertResponse.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ThreadRevertResponse.json
@@ -502,6 +502,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -513,9 +527,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -525,7 +536,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ThreadStartResponse.json b/codex-rs/app-server-protocol/schema/json/v2/ThreadStartResponse.json
index 3f193b0de8..da04bf0a27 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ThreadStartResponse.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ThreadStartResponse.json
@@ -580,6 +580,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -591,9 +605,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -603,7 +614,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ThreadStartedNotification.json b/codex-rs/app-server-protocol/schema/json/v2/ThreadStartedNotification.json
index e7eb265dcb..084cc6522d 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ThreadStartedNotification.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ThreadStartedNotification.json
@@ -502,6 +502,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -513,9 +527,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -525,7 +536,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ThreadTurnsListResponse.json b/codex-rs/app-server-protocol/schema/json/v2/ThreadTurnsListResponse.json
index c8d1dd0354..3e06a271fa 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ThreadTurnsListResponse.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ThreadTurnsListResponse.json
@@ -499,6 +499,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -510,9 +524,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -522,7 +533,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/ThreadUnarchiveResponse.json b/codex-rs/app-server-protocol/schema/json/v2/ThreadUnarchiveResponse.json
index 7c0f053e20..d396062af1 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/ThreadUnarchiveResponse.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/ThreadUnarchiveResponse.json
@@ -502,6 +502,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -513,9 +527,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -525,7 +536,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/TurnCompletedNotification.json b/codex-rs/app-server-protocol/schema/json/v2/TurnCompletedNotification.json
index 7db8c18c82..9ca8c3b087 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/TurnCompletedNotification.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/TurnCompletedNotification.json
@@ -499,6 +499,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -510,9 +524,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -522,7 +533,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/TurnStartParams.json b/codex-rs/app-server-protocol/schema/json/v2/TurnStartParams.json
index 9f8637ea17..68f4b3dff9 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/TurnStartParams.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/TurnStartParams.json
@@ -166,6 +166,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -177,9 +191,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -189,7 +200,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/TurnStartResponse.json b/codex-rs/app-server-protocol/schema/json/v2/TurnStartResponse.json
index 9552369480..f2d2251e03 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/TurnStartResponse.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/TurnStartResponse.json
@@ -499,6 +499,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -510,9 +524,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -522,7 +533,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/json/v2/TurnStartedNotification.json b/codex-rs/app-server-protocol/schema/json/v2/TurnStartedNotification.json
index 57052afd9f..bdf800c52a 100644
--- a/codex-rs/app-server-protocol/schema/json/v2/TurnStartedNotification.json
+++ b/codex-rs/app-server-protocol/schema/json/v2/TurnStartedNotification.json
@@ -499,6 +499,20 @@
           "type": "object"
         },
         {
+          "anyOf": [
+            {
+              "properties": {
+                "image_url": {
+                  "type": "string"
+                }
+              },
+              "required": [
+                "image_url"
+              ],
+              "title": "ImageUrlFunctionCallOutputContentItem",
+              "type": "object"
+            }
+          ],
           "properties": {
             "detail": {
               "anyOf": [
@@ -510,9 +524,6 @@
                 }
               ]
             },
-            "image_url": {
-              "type": "string"
-            },
             "type": {
               "enum": [
                 "input_image"
@@ -522,7 +533,6 @@
             }
           },
           "required": [
-            "image_url",
             "type"
           ],
           "title": "InputImageFunctionCallOutputContentItem",
diff --git a/codex-rs/app-server-protocol/schema/precomputed/app-server-exports-experimental.json.zst b/codex-rs/app-server-protocol/schema/precomputed/app-server-exports-experimental.json.zst
index b4f9045ce0..79d6a3dfb1 100644
Binary files a/codex-rs/app-server-protocol/schema/precomputed/app-server-exports-experimental.json.zst and b/codex-rs/app-server-protocol/schema/precomputed/app-server-exports-experimental.json.zst differ
diff --git a/codex-rs/app-server-protocol/schema/precomputed/app-server-exports-stable.json.zst b/codex-rs/app-server-protocol/schema/precomputed/app-server-exports-stable.json.zst
index 8222eefe7f..14bdda9860 100644
Binary files a/codex-rs/app-server-protocol/schema/precomputed/app-server-exports-stable.json.zst and b/codex-rs/app-server-protocol/schema/precomputed/app-server-exports-stable.json.zst differ
diff --git a/codex-rs/app-server-protocol/schema/typescript/ContentItem.ts b/codex-rs/app-server-protocol/schema/typescript/ContentItem.ts
index 9e53b5fc25..dd5b0a1929 100644
--- a/codex-rs/app-server-protocol/schema/typescript/ContentItem.ts
+++ b/codex-rs/app-server-protocol/schema/typescript/ContentItem.ts
@@ -3,4 +3,4 @@
 // This file was generated by [ts-rs](https://github.com/Aleph-Alpha/ts-rs). Do not edit this file manually.
 import type { ImageDetail } from "./ImageDetail";
 
-export type ContentItem = { "type": "input_text", text: string, } | { "type": "input_image", image_url: string, detail?: ImageDetail, } | { "type": "input_audio", audio_url: string, } | { "type": "output_text", text: string, };
+export type ContentItem = { "type": "input_text", text: string, } | { "type": "input_image", detail?: ImageDetail, } & ({ image_url: string, }) | { "type": "input_audio", audio_url: string, } | { "type": "output_text", text: string, };
diff --git a/codex-rs/app-server-protocol/schema/typescript/FunctionCallOutputContentItem.ts b/codex-rs/app-server-protocol/schema/typescript/FunctionCallOutputContentItem.ts
index 6c2ab2afa1..8ebefc596e 100644
--- a/codex-rs/app-server-protocol/schema/typescript/FunctionCallOutputContentItem.ts
+++ b/codex-rs/app-server-protocol/schema/typescript/FunctionCallOutputContentItem.ts
@@ -7,4 +7,4 @@ import type { ImageDetail } from "./ImageDetail";
  * Responses API compatible content items that can be returned by a tool call.
  * This is a subset of ContentItem with the types we support as function call outputs.
  */
-export type FunctionCallOutputContentItem = { "type": "input_text", text: string, } | { "type": "input_image", image_url: string, detail?: ImageDetail, } | { "type": "input_audio", audio_url: string, } | { "type": "encrypted_content", encrypted_content: string, };
+export type FunctionCallOutputContentItem = { "type": "input_text", text: string, } | { "type": "input_image", detail?: ImageDetail, } & ({ image_url: string, }) | { "type": "input_audio", audio_url: string, } | { "type": "encrypted_content", encrypted_content: string, };
diff --git a/codex-rs/app-server/src/request_processors/turn_processor.rs b/codex-rs/app-server/src/request_processors/turn_processor.rs
index 95e1d62f5c..a051c867db 100644
--- a/codex-rs/app-server/src/request_processors/turn_processor.rs
+++ b/codex-rs/app-server/src/request_processors/turn_processor.rs
@@ -8,6 +8,7 @@ use codex_protocol::models::ContentItem;
 use codex_protocol::models::FunctionCallOutputBody;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
+use codex_protocol::models::ImageReference;
 use codex_protocol::protocol::AdditionalContextEntry as CoreAdditionalContextEntry;
 use codex_protocol::protocol::AdditionalContextKind as CoreAdditionalContextKind;
 use codex_protocol::protocol::TurnSettingsUpdate;
@@ -36,7 +37,7 @@ fn validate_response_item_image_urls(items: &[ResponseItem]) -> Result<(), JSONR
         ResponseItem::Message { content, .. } => content.iter().any(|item| {
             matches!(
                 item,
-                ContentItem::InputImage { image_url, .. } if is_remote_image_url(image_url)
+                ContentItem::InputImage { image: ImageReference::Inline { image_url }, .. } if is_remote_image_url(image_url)
             )
         }),
         ResponseItem::FunctionCallOutput { output, .. }
@@ -45,7 +46,7 @@ fn validate_response_item_image_urls(items: &[ResponseItem]) -> Result<(), JSONR
                 content.iter().any(|item| {
                     matches!(
                         item,
-                        FunctionCallOutputContentItem::InputImage { image_url, .. }
+                        FunctionCallOutputContentItem::InputImage { image: ImageReference::Inline { image_url }, .. }
                             if is_remote_image_url(image_url)
                     )
                 })
diff --git a/codex-rs/app-server/tests/suite/v2/dynamic_tools.rs b/codex-rs/app-server/tests/suite/v2/dynamic_tools.rs
index 38cad9c430..38f34330ba 100644
--- a/codex-rs/app-server/tests/suite/v2/dynamic_tools.rs
+++ b/codex-rs/app-server/tests/suite/v2/dynamic_tools.rs
@@ -30,6 +30,7 @@ use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
 use codex_protocol::models::FunctionCallOutputBody;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
+use codex_protocol::models::ImageReference;
 use codex_protocol::openai_models::InputModality;
 use core_test_support::load_default_config_for_test;
 use core_test_support::responses;
@@ -715,7 +716,9 @@ async fn dynamic_tool_call_round_trip_handles_content_items() -> Result<()> {
             text: "dynamic-ok".to_string(),
         },
         FunctionCallOutputContentItem::InputImage {
-            image_url: TINY_PNG_DATA_URL.to_string(),
+            image: ImageReference::Inline {
+                image_url: TINY_PNG_DATA_URL.to_string(),
+            },
             detail: Some(DEFAULT_IMAGE_DETAIL),
         },
         FunctionCallOutputContentItem::InputAudio {
diff --git a/codex-rs/app-server/tests/suite/v2/request_validation.rs b/codex-rs/app-server/tests/suite/v2/request_validation.rs
index 1b5abe2f17..042236ed8e 100644
--- a/codex-rs/app-server/tests/suite/v2/request_validation.rs
+++ b/codex-rs/app-server/tests/suite/v2/request_validation.rs
@@ -14,6 +14,7 @@ use codex_app_server_protocol::ThreadStartResponse;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseItem;
 use pretty_assertions::assert_eq;
 use serde_json::json;
@@ -173,7 +174,9 @@ async fn request_handlers_reject_remote_image_urls() -> Result<()> {
         namespace: None,
         output: FunctionCallOutputPayload::from_content_items(vec![
             FunctionCallOutputContentItem::InputImage {
-                image_url: "https://example.com/tool.png".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "https://example.com/tool.png".to_string(),
+                },
                 detail: Some(ImageDetail::High),
             },
         ]),
diff --git a/codex-rs/codex-api/src/endpoint/search.rs b/codex-rs/codex-api/src/endpoint/search.rs
index 131a335e26..6b2450e649 100644
--- a/codex-rs/codex-api/src/endpoint/search.rs
+++ b/codex-rs/codex-api/src/endpoint/search.rs
@@ -72,6 +72,7 @@ mod tests {
     use codex_client::TransportError;
     use codex_protocol::ResponseItemId;
     use codex_protocol::models::ContentItem;
+    use codex_protocol::models::ImageReference;
     use codex_protocol::models::ResponseItem;
     use http::StatusCode;
     use pretty_assertions::assert_eq;
@@ -164,7 +165,9 @@ mod tests {
                                 text: "find this".to_string(),
                             },
                             ContentItem::InputImage {
-                                image_url: "https://example.com/image.png".to_string(),
+                                image: ImageReference::Inline {
+                                    image_url: "https://example.com/image.png".to_string(),
+                                },
                                 detail: None,
                             },
                         ],
diff --git a/codex-rs/core/src/client_common_tests.rs b/codex-rs/core/src/client_common_tests.rs
index e8b030e6a1..1cda4ef13d 100644
--- a/codex-rs/core/src/client_common_tests.rs
+++ b/codex-rs/core/src/client_common_tests.rs
@@ -6,6 +6,7 @@ use codex_models_manager::model_info::model_info_from_slug;
 use codex_protocol::config_types::ServiceTier;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use pretty_assertions::assert_eq;
 use serde_json::value::RawValue;
 use std::sync::Arc;
@@ -27,7 +28,9 @@ fn prompt_with_image_outputs(detail: Option<ImageDetail>) -> Prompt {
                         text: "Describe this image.".to_string(),
                     },
                     ContentItem::InputImage {
-                        image_url: "https://example.com/image.png".to_string(),
+                        image: ImageReference::Inline {
+                            image_url: "https://example.com/image.png".to_string(),
+                        },
                         detail,
                     },
                 ],
@@ -41,7 +44,9 @@ fn prompt_with_image_outputs(detail: Option<ImageDetail>) -> Prompt {
                 namespace: None,
                 output: FunctionCallOutputPayload::from_content_items(vec![
                     FunctionCallOutputContentItem::InputImage {
-                        image_url: "data:image/png;base64,function".to_string(),
+                        image: ImageReference::Inline {
+                            image_url: "data:image/png;base64,function".to_string(),
+                        },
                         detail,
                     },
                 ]),
@@ -53,7 +58,9 @@ fn prompt_with_image_outputs(detail: Option<ImageDetail>) -> Prompt {
                 name: None,
                 output: FunctionCallOutputPayload::from_content_items(vec![
                     FunctionCallOutputContentItem::InputImage {
-                        image_url: "data:image/png;base64,custom".to_string(),
+                        image: ImageReference::Inline {
+                            image_url: "data:image/png;base64,custom".to_string(),
+                        },
                         detail,
                     },
                 ]),
diff --git a/codex-rs/core/src/compact_remote_v2.rs b/codex-rs/core/src/compact_remote_v2.rs
index 7a5d3ad416..b367ba4bea 100644
--- a/codex-rs/core/src/compact_remote_v2.rs
+++ b/codex-rs/core/src/compact_remote_v2.rs
@@ -45,6 +45,8 @@ use codex_protocol::items::ContextCompactionItem;
 use codex_protocol::items::TurnItem;
 use codex_protocol::models::AgentMessageInputContent;
 use codex_protocol::models::ContentItem;
+#[cfg(test)]
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::protocol::EventMsg;
 use codex_protocol::protocol::TokenUsage;
@@ -981,11 +983,15 @@ mod tests {
                     text: "user".to_string(),
                 },
                 ContentItem::InputImage {
-                    image_url: "data:image/png;base64,abc".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,abc".to_string(),
+                    },
                     detail: None,
                 },
                 ContentItem::InputImage {
-                    image_url: "data:image/png;base64,def".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,def".to_string(),
+                    },
                     detail: None,
                 },
             ],
@@ -1047,7 +1053,9 @@ mod tests {
                     text: "abcdef".to_string(),
                 },
                 ContentItem::InputImage {
-                    image_url: "data:image/png;base64,abc".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,abc".to_string(),
+                    },
                     detail: None,
                 },
                 ContentItem::OutputText {
@@ -1057,7 +1065,9 @@ mod tests {
                     text: "discarded after the text budget is exhausted".to_string(),
                 },
                 ContentItem::InputImage {
-                    image_url: "data:image/png;base64,def".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,def".to_string(),
+                    },
                     detail: None,
                 },
             ],
@@ -1089,14 +1099,18 @@ mod tests {
                         text: "abcdef".to_string(),
                     },
                     ContentItem::InputImage {
-                        image_url: "data:image/png;base64,abc".to_string(),
+                        image: ImageReference::Inline {
+                            image_url: "data:image/png;base64,abc".to_string()
+                        },
                         detail: None,
                     },
                     ContentItem::OutputText {
                         text: "uv…1 tokens truncated…yz".to_string(),
                     },
                     ContentItem::InputImage {
-                        image_url: "data:image/png;base64,def".to_string(),
+                        image: ImageReference::Inline {
+                            image_url: "data:image/png;base64,def".to_string()
+                        },
                         detail: None,
                     },
                 ],
@@ -1123,7 +1137,9 @@ mod tests {
             id: None,
             role: "user".to_string(),
             content: vec![ContentItem::InputImage {
-                image_url: "data:image/png;base64,abc".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,abc".to_string(),
+                },
                 detail: None,
             }],
             phase: None,
@@ -1147,7 +1163,9 @@ mod tests {
             id: None,
             role: "user".to_string(),
             content: vec![ContentItem::InputImage {
-                image_url: "data:image/png;base64,abc".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,abc".to_string(),
+                },
                 detail: None,
             }],
             phase: None,
diff --git a/codex-rs/core/src/compact_remote_v2_image_budget_tests.rs b/codex-rs/core/src/compact_remote_v2_image_budget_tests.rs
index ffc3a6f85a..7c7d94ac8b 100644
--- a/codex-rs/core/src/compact_remote_v2_image_budget_tests.rs
+++ b/codex-rs/core/src/compact_remote_v2_image_budget_tests.rs
@@ -1,5 +1,6 @@
 use super::*;
 use codex_protocol::models::ContentItemKind;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::InternalChatMessageMetadataPassthrough;
 use codex_protocol::models::image_close_tag_text;
 use codex_protocol::models::local_image_open_tag_text_with_path;
@@ -17,7 +18,9 @@ fn message(content: Vec<ContentItem>) -> ResponseItem {
 
 fn image() -> ContentItem {
     ContentItem::InputImage {
-        image_url: "data:image/png;base64,abc".to_string(),
+        image: ImageReference::Inline {
+            image_url: "data:image/png;base64,abc".to_string(),
+        },
         detail: None,
     }
 }
diff --git a/codex-rs/core/src/compact_remote_v2_images.rs b/codex-rs/core/src/compact_remote_v2_images.rs
index d2fecb03a2..0575e07f01 100644
--- a/codex-rs/core/src/compact_remote_v2_images.rs
+++ b/codex-rs/core/src/compact_remote_v2_images.rs
@@ -4,6 +4,7 @@ use codex_context_fragments::set_annotated_content;
 use codex_context_fragments::to_annotated_content;
 use codex_history::ResponseItemEnvelope;
 use codex_protocol::models::ContentItem;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::is_image_close_tag_text;
 use codex_protocol::models::is_image_open_tag_text;
 use codex_protocol::models::is_local_image_open_tag_text;
@@ -17,9 +18,12 @@ pub(super) fn content_item_token_count(item: &ContentItem) -> usize {
         ContentItem::InputText { text } | ContentItem::OutputText { text } => {
             approx_token_count(text)
         }
-        ContentItem::InputImage { image_url, detail } => usize::try_from(
-            approx_tokens_from_byte_count_i64(estimate_image_bytes(image_url, *detail)),
-        )
+        ContentItem::InputImage {
+            image: ImageReference::Inline { image_url },
+            detail,
+        } => usize::try_from(approx_tokens_from_byte_count_i64(estimate_image_bytes(
+            image_url, *detail,
+        )))
         .unwrap_or(usize::MAX),
         ContentItem::InputAudio { .. } => 0,
     }
diff --git a/codex-rs/core/src/compact_tests.rs b/codex-rs/core/src/compact_tests.rs
index b319e75ddb..531d093061 100644
--- a/codex-rs/core/src/compact_tests.rs
+++ b/codex-rs/core/src/compact_tests.rs
@@ -10,6 +10,7 @@ use codex_protocol::models::ContentItemKind;
 use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
 use codex_protocol::models::ExecutedToolCall;
 use codex_protocol::models::FunctionCallOutputPayload;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::InternalChatMessageMetadataPassthrough;
 use core_test_support::responses;
 use core_test_support::skip_if_no_network;
@@ -207,7 +208,9 @@ fn content_items_to_text_joins_non_empty_segments() {
 #[test]
 fn content_items_to_text_ignores_image_only_content() {
     let items = vec![ContentItem::InputImage {
-        image_url: "file://image.png".to_string(),
+        image: ImageReference::Inline {
+            image_url: "file://image.png".to_string(),
+        },
         detail: Some(DEFAULT_IMAGE_DETAIL),
     }];
 
diff --git a/codex-rs/core/src/context/node_repl_review_evidence.rs b/codex-rs/core/src/context/node_repl_review_evidence.rs
index fdc2027da2..16cf5f62d1 100644
--- a/codex-rs/core/src/context/node_repl_review_evidence.rs
+++ b/codex-rs/core/src/context/node_repl_review_evidence.rs
@@ -12,6 +12,7 @@ use codex_guardian_context::NodeReplContext;
 use codex_guardian_context::NodeReplResponse;
 pub use codex_guardian_context::NodeReplReviewEvidenceMode;
 use codex_protocol::models::ContentItem;
+use codex_protocol::models::ImageReference;
 use codex_protocol::user_input::UserInput;
 use codex_protocol::user_input::UserInput::Image;
 use codex_protocol::user_input::UserInput::Text;
@@ -117,7 +118,9 @@ impl NodeReplReviewEvidence {
             .filter_map(|item| match item {
                 Image { image_url, detail } if seen_images.insert(image_url) => {
                     Some(ContentItem::InputImage {
-                        image_url: image_url.clone(),
+                        image: ImageReference::Inline {
+                            image_url: image_url.clone(),
+                        },
                         detail: *detail,
                     })
                 }
diff --git a/codex-rs/core/src/context_manager/history.rs b/codex-rs/core/src/context_manager/history.rs
index 3695fc7b77..4617e462ab 100644
--- a/codex-rs/core/src/context_manager/history.rs
+++ b/codex-rs/core/src/context_manager/history.rs
@@ -43,6 +43,7 @@ use codex_protocol::models::ContentItem;
 use codex_protocol::models::FunctionCallOutputBody;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::openai_models::InputModality;
 use codex_protocol::protocol::InterAgentCommunication;
@@ -850,9 +851,10 @@ fn estimate_response_item_model_visible_bytes(item: &ResponseItem) -> i64 {
                 ContentItem::InputText { text } | ContentItem::OutputText { text } => {
                     text_bytes(text)
                 }
-                ContentItem::InputImage { image_url, detail } => {
-                    estimate_image_bytes(image_url, *detail)
-                }
+                ContentItem::InputImage {
+                    image: ImageReference::Inline { image_url },
+                    detail,
+                } => estimate_image_bytes(image_url, *detail),
                 ContentItem::InputAudio { audio_url } => estimate_audio_bytes(audio_url),
             })
             .fold(0i64, i64::saturating_add),
@@ -1055,9 +1057,10 @@ fn estimate_function_output_bytes(output: &FunctionCallOutputBody) -> i64 {
             .iter()
             .map(|part| match part {
                 FunctionCallOutputContentItem::InputText { text } => text_bytes(text),
-                FunctionCallOutputContentItem::InputImage { image_url, detail } => {
-                    estimate_image_bytes(image_url, *detail)
-                }
+                FunctionCallOutputContentItem::InputImage {
+                    image: ImageReference::Inline { image_url },
+                    detail,
+                } => estimate_image_bytes(image_url, *detail),
                 FunctionCallOutputContentItem::InputAudio { audio_url } => {
                     estimate_audio_bytes(audio_url)
                 }
diff --git a/codex-rs/core/src/context_manager/history_tests.rs b/codex-rs/core/src/context_manager/history_tests.rs
index c0f35668a9..e0844511bd 100644
--- a/codex-rs/core/src/context_manager/history_tests.rs
+++ b/codex-rs/core/src/context_manager/history_tests.rs
@@ -18,6 +18,7 @@ use codex_protocol::models::FunctionCallOutputBody;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::InternalChatMessageMetadataPassthrough;
 use codex_protocol::models::LocalShellAction;
 use codex_protocol::models::LocalShellExecAction;
@@ -715,7 +716,9 @@ fn for_prompt_annotated_preserves_metadata_while_normalizing_item() {
                     text: "keep".to_string(),
                 },
                 ContentItem::InputImage {
-                    image_url: "data:image/png;base64,abc".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,abc".to_string(),
+                    },
                     detail: None,
                 },
             ],
@@ -796,7 +799,9 @@ fn for_prompt_strips_media_when_model_does_not_support_it() {
                     text: "look at this".to_string(),
                 },
                 ContentItem::InputImage {
-                    image_url: "https://example.com/img.png".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "https://example.com/img.png".to_string(),
+                    },
                     detail: Some(DEFAULT_IMAGE_DETAIL),
                 },
                 ContentItem::InputAudio {
@@ -838,7 +843,9 @@ fn for_prompt_strips_media_when_model_does_not_support_it() {
                     text: "image result".to_string(),
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "https://example.com/result.png".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "https://example.com/result.png".to_string(),
+                    },
                     detail: Some(DEFAULT_IMAGE_DETAIL),
                 },
                 FunctionCallOutputContentItem::InputAudio {
@@ -865,7 +872,9 @@ fn for_prompt_strips_media_when_model_does_not_support_it() {
                     text: "js repl result".to_string(),
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "https://example.com/js-repl-result.png".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "https://example.com/js-repl-result.png".to_string(),
+                    },
                     detail: Some(DEFAULT_IMAGE_DETAIL),
                 },
                 FunctionCallOutputContentItem::InputAudio {
@@ -991,7 +1000,9 @@ fn for_prompt_strips_media_when_model_does_not_support_it() {
                 text: "look".to_string(),
             },
             ContentItem::InputImage {
-                image_url: "https://example.com/img.png".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "https://example.com/img.png".to_string(),
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
         ],
@@ -2414,7 +2425,7 @@ fn image_data_url_payload_does_not_dominate_message_estimate() {
                 text: "Here is the screenshot".to_string(),
             },
             ContentItem::InputImage {
-                image_url,
+                image: ImageReference::Inline { image_url },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
         ],
@@ -2455,7 +2466,7 @@ fn image_data_url_payload_does_not_dominate_function_call_output_estimate() {
                 text: "Screenshot captured".to_string(),
             },
             FunctionCallOutputContentItem::InputImage {
-                image_url,
+                image: ImageReference::Inline { image_url },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
         ]),
@@ -2484,7 +2495,7 @@ fn image_data_url_payload_does_not_dominate_custom_tool_call_output_estimate() {
                 text: "Screenshot captured".to_string(),
             },
             FunctionCallOutputContentItem::InputImage {
-                image_url,
+                image: ImageReference::Inline { image_url },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
         ]),
@@ -2628,7 +2639,9 @@ fn non_base64_image_urls_use_image_estimates() {
         id: None,
         role: "user".to_string(),
         content: vec![ContentItem::InputImage {
-            image_url: "https://example.com/foo.png".to_string(),
+            image: ImageReference::Inline {
+                image_url: "https://example.com/foo.png".to_string(),
+            },
             detail: Some(DEFAULT_IMAGE_DETAIL),
         }],
         phase: None,
@@ -2641,7 +2654,9 @@ fn non_base64_image_urls_use_image_estimates() {
         namespace: None,
         output: FunctionCallOutputPayload::from_content_items(vec![
             FunctionCallOutputContentItem::InputImage {
-                image_url: "file:///tmp/foo.png".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "file:///tmp/foo.png".to_string(),
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
         ]),
@@ -2746,7 +2761,10 @@ fn data_url_without_base64_marker_uses_image_estimate() {
         id: None,
         role: "user".to_string(),
         content: vec![ContentItem::InputImage {
-            image_url: "data:image/svg+xml,<svg xmlns='http://www.w3.org/2000/svg'/>".to_string(),
+            image: ImageReference::Inline {
+                image_url: "data:image/svg+xml,<svg xmlns='http://www.w3.org/2000/svg'/>"
+                    .to_string(),
+            },
             detail: Some(DEFAULT_IMAGE_DETAIL),
         }],
         phase: None,
@@ -2771,7 +2789,7 @@ fn non_image_base64_data_url_uses_image_estimate() {
         namespace: None,
         output: FunctionCallOutputPayload::from_content_items(vec![
             FunctionCallOutputContentItem::InputImage {
-                image_url,
+                image: ImageReference::Inline { image_url },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
         ]),
@@ -2791,7 +2809,7 @@ fn mixed_case_data_url_markers_are_adjusted() {
         id: None,
         role: "user".to_string(),
         content: vec![ContentItem::InputImage {
-            image_url,
+            image: ImageReference::Inline { image_url },
             detail: Some(DEFAULT_IMAGE_DETAIL),
         }],
         phase: None,
@@ -2818,11 +2836,15 @@ fn multiple_inline_images_apply_multiple_fixed_costs() {
                 text: "images".to_string(),
             },
             ContentItem::InputImage {
-                image_url: image_url_one,
+                image: ImageReference::Inline {
+                    image_url: image_url_one,
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
             ContentItem::InputImage {
-                image_url: image_url_two,
+                image: ImageReference::Inline {
+                    image_url: image_url_two,
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
         ],
@@ -2858,7 +2880,7 @@ fn original_detail_images_scale_with_dimensions() {
         namespace: None,
         output: FunctionCallOutputPayload::from_content_items(vec![
             FunctionCallOutputContentItem::InputImage {
-                image_url,
+                image: ImageReference::Inline { image_url },
                 detail: Some(ImageDetail::Original),
             },
         ]),
@@ -2891,7 +2913,7 @@ fn original_detail_images_are_capped_at_max_patch_count() {
         namespace: None,
         output: FunctionCallOutputPayload::from_content_items(vec![
             FunctionCallOutputContentItem::InputImage {
-                image_url,
+                image: ImageReference::Inline { image_url },
                 detail: Some(ImageDetail::Original),
             },
         ]),
@@ -2927,7 +2949,7 @@ fn original_detail_webp_images_scale_with_dimensions() {
         namespace: None,
         output: FunctionCallOutputPayload::from_content_items(vec![
             FunctionCallOutputContentItem::InputImage {
-                image_url,
+                image: ImageReference::Inline { image_url },
                 detail: Some(ImageDetail::Original),
             },
         ]),
diff --git a/codex-rs/core/src/event_mapping.rs b/codex-rs/core/src/event_mapping.rs
index d386b08b5d..ed983fe644 100644
--- a/codex-rs/core/src/event_mapping.rs
+++ b/codex-rs/core/src/event_mapping.rs
@@ -5,6 +5,7 @@ use codex_protocol::items::TurnItem;
 use codex_protocol::items::UserMessageItem;
 use codex_protocol::items::WebSearchItem;
 use codex_protocol::models::ContentItem;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::MessagePhase;
 use codex_protocol::models::ReasoningItemContent;
 use codex_protocol::models::ReasoningItemReasoningSummary;
@@ -126,7 +127,10 @@ fn parse_user_message(message: &[ContentItem]) -> Option<UserMessageItem> {
                     text_elements: Vec::new(),
                 });
             }
-            ContentItem::InputImage { image_url, detail } => {
+            ContentItem::InputImage {
+                image: ImageReference::Inline { image_url },
+                detail,
+            } => {
                 content.push(UserInput::Image {
                     image_url: image_url.clone(),
                     detail: *detail,
diff --git a/codex-rs/core/src/event_mapping_tests.rs b/codex-rs/core/src/event_mapping_tests.rs
index 972df5e400..b6ac81c9da 100644
--- a/codex-rs/core/src/event_mapping_tests.rs
+++ b/codex-rs/core/src/event_mapping_tests.rs
@@ -12,6 +12,7 @@ use codex_protocol::items::WebSearchItem;
 use codex_protocol::items::build_hook_prompt_message;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ReasoningItemContent;
 use codex_protocol::models::ReasoningItemReasoningSummary;
 use codex_protocol::models::ResponseItem;
@@ -83,11 +84,15 @@ fn parses_user_message_with_text_and_two_images() {
                 text: "Hello world".to_string(),
             },
             ContentItem::InputImage {
-                image_url: img1.clone(),
+                image: ImageReference::Inline {
+                    image_url: img1.clone(),
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
             ContentItem::InputImage {
-                image_url: img2.clone(),
+                image: ImageReference::Inline {
+                    image_url: img2.clone(),
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
         ],
@@ -131,7 +136,9 @@ fn skips_local_image_label_text() {
         content: vec![
             ContentItem::InputText { text: label },
             ContentItem::InputImage {
-                image_url: image_url.clone(),
+                image: ImageReference::Inline {
+                    image_url: image_url.clone(),
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
             ContentItem::InputText {
@@ -258,7 +265,9 @@ fn skips_unnamed_image_label_text() {
         content: vec![
             ContentItem::InputText { text: label },
             ContentItem::InputImage {
-                image_url: image_url.clone(),
+                image: ImageReference::Inline {
+                    image_url: image_url.clone(),
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
             ContentItem::InputText {
diff --git a/codex-rs/core/src/guardian/review_session.rs b/codex-rs/core/src/guardian/review_session.rs
index 23e756fcf4..397f1eb635 100644
--- a/codex-rs/core/src/guardian/review_session.rs
+++ b/codex-rs/core/src/guardian/review_session.rs
@@ -35,6 +35,7 @@ use codex_protocol::items::TurnItem;
 use codex_protocol::mcp::is_node_repl_backed_server;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseInputItem;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::openai_models::InputModality;
@@ -502,7 +503,7 @@ async fn run_review_on_session(
                         _ => &[],
                     })
                     .filter_map(|item| match item {
-                        ContentItem::InputImage { image_url, .. } => Some(image_url.as_str()),
+                        ContentItem::InputImage { image: ImageReference::Inline { image_url }, .. } => Some(image_url.as_str()),
                         _ => None,
                     })
                     .collect::<HashSet<_>>();
@@ -550,7 +551,7 @@ async fn run_review_on_session(
                             return false;
                         };
                         content.iter().any(|item| {
-                            matches!(item, ContentItem::InputImage { image_url, .. }
+                            matches!(item, ContentItem::InputImage { image: ImageReference::Inline { image_url }, .. }
                                 if !reviewer_image_urls.contains(image_url.as_str()))
                         })
                     });
diff --git a/codex-rs/core/src/guardian/tests.rs b/codex-rs/core/src/guardian/tests.rs
index 777b77c54a..cf191851bf 100644
--- a/codex-rs/core/src/guardian/tests.rs
+++ b/codex-rs/core/src/guardian/tests.rs
@@ -45,6 +45,7 @@ use codex_protocol::config_types::ApprovalsReviewer;
 use codex_protocol::config_types::ReasoningSummary;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::FunctionCallOutputContentItem;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::PermissionProfile;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::models::SandboxPermissions;
@@ -1130,7 +1131,9 @@ fn collect_guardian_transcript_entries_preserves_named_unpaired_tool_sources() {
     if let ResponseItem::FunctionCallOutput { output, .. } = &mut items[0] {
         *output = codex_protocol::models::FunctionCallOutputPayload::from_content_items(vec![
             FunctionCallOutputContentItem::InputImage {
-                image_url: "data:image/png;base64,image".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,image".to_string(),
+                },
                 detail: None,
             },
         ]);
diff --git a/codex-rs/core/src/image_preparation.rs b/codex-rs/core/src/image_preparation.rs
index 78170db319..adc886f7c1 100644
--- a/codex-rs/core/src/image_preparation.rs
+++ b/codex-rs/core/src/image_preparation.rs
@@ -15,6 +15,7 @@ use codex_protocol::models::ContentItemKind;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::openai_models::ModelInfo;
 use codex_utils_image::ImageProcessingError;
@@ -185,7 +186,11 @@ fn prepare_message_content(
     let mut image_number = 0;
     let mut resized_images = Vec::new();
     for item in items {
-        if let ContentItem::InputImage { image_url, detail } = item.content_mut() {
+        if let ContentItem::InputImage {
+            image: ImageReference::Inline { image_url },
+            detail,
+        } = item.content_mut()
+        {
             image_number += 1;
             match prepare_image(image_url, detail, origin, metadata, mode) {
                 Ok(Some(resize)) if resize_notice_mode == ImageResizeNoticeMode::Enabled => {
@@ -226,7 +231,11 @@ fn prepare_tool_output_content(
     let mut image_number = 0;
     let mut resized_images = Vec::new();
     for item in items {
-        if let FunctionCallOutputContentItem::InputImage { image_url, detail } = item {
+        if let FunctionCallOutputContentItem::InputImage {
+            image: ImageReference::Inline { image_url },
+            detail,
+        } = item
+        {
             image_number += 1;
             match prepare_image(image_url, detail, origin, metadata, mode) {
                 Ok(Some(resize)) if resize_notice_mode == ImageResizeNoticeMode::Enabled => {
diff --git a/codex-rs/core/src/image_preparation_tests.rs b/codex-rs/core/src/image_preparation_tests.rs
index 67b1f1d4df..a0cfbf5479 100644
--- a/codex-rs/core/src/image_preparation_tests.rs
+++ b/codex-rs/core/src/image_preparation_tests.rs
@@ -5,6 +5,7 @@ use base64::engine::general_purpose::STANDARD as BASE64_STANDARD;
 use codex_protocol::models::ContentItemKind;
 use codex_protocol::models::FunctionCallOutputBody;
 use codex_protocol::models::FunctionCallOutputPayload;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::InternalChatMessageMetadataPassthrough;
 use codex_utils_image::data_url_from_bytes;
 use image::DynamicImage;
@@ -41,11 +42,15 @@ fn preparation_preserves_small_image_bytes_and_replaces_remote_urls() {
         role: "user".to_string(),
         content: vec![
             ContentItem::InputImage {
-                image_url: data_url,
+                image: ImageReference::Inline {
+                    image_url: data_url,
+                },
                 detail: Some(ImageDetail::High),
             },
             ContentItem::InputImage {
-                image_url: "https://example.com/image.png".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "https://example.com/image.png".to_string(),
+                },
                 detail: Some(ImageDetail::Low),
             },
         ],
@@ -56,7 +61,9 @@ fn preparation_preserves_small_image_bytes_and_replaces_remote_urls() {
         id: None,
         role: "developer".to_string(),
         content: vec![ContentItem::InputImage {
-            image_url: "https://example.com/developer-image.png".to_string(),
+            image: ImageReference::Inline {
+                image_url: "https://example.com/developer-image.png".to_string(),
+            },
             detail: Some(ImageDetail::High),
         }],
         phase: None,
@@ -73,7 +80,10 @@ fn preparation_preserves_small_image_bytes_and_replaces_remote_urls() {
         panic!("expected message");
     };
     let [
-        ContentItem::InputImage { image_url, .. },
+        ContentItem::InputImage {
+            image: ImageReference::Inline { image_url },
+            ..
+        },
         ContentItem::InputText { text },
     ] = content.as_slice()
     else {
@@ -135,7 +145,10 @@ fn detail_policies_apply_the_expected_budgets() {
         let mut items = vec![ResponseItem::Message {
             id: None,
             role: "user".to_string(),
-            content: vec![ContentItem::InputImage { image_url, detail }],
+            content: vec![ContentItem::InputImage {
+                image: ImageReference::Inline { image_url },
+                detail,
+            }],
             phase: None,
             internal_chat_message_metadata_passthrough: None,
         }];
@@ -149,7 +162,13 @@ fn detail_policies_apply_the_expected_budgets() {
         let ResponseItem::Message { content, .. } = &items[0] else {
             panic!("expected message");
         };
-        let [ContentItem::InputImage { image_url, .. }] = content.as_slice() else {
+        let [
+            ContentItem::InputImage {
+                image: ImageReference::Inline { image_url },
+                ..
+            },
+        ] = content.as_slice()
+        else {
             panic!("expected image");
         };
         assert_eq!(decoded_image(image_url).1.dimensions(), expected_dimensions);
@@ -179,7 +198,7 @@ fn preparation_reports_tool_output_item_id() {
         namespace: None,
         output: FunctionCallOutputPayload::from_content_items(vec![
             FunctionCallOutputContentItem::InputImage {
-                image_url,
+                image: ImageReference::Inline { image_url },
                 detail: Some(ImageDetail::High),
             },
         ]),
@@ -215,15 +234,21 @@ fn resize_notices_preserve_original_image_positions_and_skip_failed_images() {
             role: "user".to_string(),
             content: vec![
                 ContentItem::InputImage {
-                    image_url: small_image_url,
+                    image: ImageReference::Inline {
+                        image_url: small_image_url,
+                    },
                     detail: Some(ImageDetail::High),
                 },
                 ContentItem::InputImage {
-                    image_url: "data:image/png;base64,%%%".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,%%%".to_string(),
+                    },
                     detail: Some(ImageDetail::High),
                 },
                 ContentItem::InputImage {
-                    image_url: large_image_url.clone(),
+                    image: ImageReference::Inline {
+                        image_url: large_image_url.clone(),
+                    },
                     detail: Some(ImageDetail::High),
                 },
             ],
@@ -246,11 +271,15 @@ fn resize_notices_preserve_original_image_positions_and_skip_failed_images() {
             namespace: None,
             output: FunctionCallOutputPayload::from_content_items(vec![
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "data:image/png;base64,%%%".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,%%%".to_string(),
+                    },
                     detail: Some(ImageDetail::High),
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: large_image_url,
+                    image: ImageReference::Inline {
+                        image_url: large_image_url,
+                    },
                     detail: Some(ImageDetail::High),
                 },
             ]),
@@ -290,14 +319,20 @@ fn resize_notices_preserve_original_image_positions_and_skip_failed_images() {
     );
     let [
         ContentItem::InputImage {
-            image_url: small_message_image_url,
+            image:
+                ImageReference::Inline {
+                    image_url: small_message_image_url,
+                },
             ..
         },
         ContentItem::InputText {
             text: failed_message_image,
         },
         ContentItem::InputImage {
-            image_url: resized_message_image_url,
+            image:
+                ImageReference::Inline {
+                    image_url: resized_message_image_url,
+                },
             ..
         },
     ] = content.as_slice()
@@ -342,7 +377,10 @@ fn resize_notices_preserve_original_image_positions_and_skip_failed_images() {
             text: failed_tool_image,
         },
         FunctionCallOutputContentItem::InputImage {
-            image_url: resized_tool_image_url,
+            image:
+                ImageReference::Inline {
+                    image_url: resized_tool_image_url,
+                },
             ..
         },
     ] = output.content_items().expect("tool output content items")
@@ -394,19 +432,27 @@ fn preparation_replaces_only_failed_tool_images_and_preserves_metadata() {
                     text: "before".to_string(),
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "data:image/png;base64,%%%".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,%%%".to_string(),
+                    },
                     detail: Some(ImageDetail::High),
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: data_url_from_bytes("image/png", b"not an image"),
+                    image: ImageReference::Inline {
+                        image_url: data_url_from_bytes("image/png", b"not an image"),
+                    },
                     detail: Some(ImageDetail::High),
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: valid_image_url.clone(),
+                    image: ImageReference::Inline {
+                        image_url: valid_image_url.clone(),
+                    },
                     detail: Some(ImageDetail::Low),
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: valid_image_url,
+                    image: ImageReference::Inline {
+                        image_url: valid_image_url,
+                    },
                     detail: Some(ImageDetail::High),
                 },
             ]),
@@ -442,7 +488,9 @@ fn preparation_replaces_only_failed_tool_images_and_preserves_metadata() {
                         text: UNSUPPORTED_LOW_DETAIL_PLACEHOLDER.to_string(),
                     },
                     FunctionCallOutputContentItem::InputImage {
-                        image_url: expected_valid_image_url,
+                        image: ImageReference::Inline {
+                            image_url: expected_valid_image_url
+                        },
                         detail: Some(ImageDetail::High),
                     },
                 ]),
diff --git a/codex-rs/core/src/session/tests.rs b/codex-rs/core/src/session/tests.rs
index f9adf42a67..56ca71705d 100644
--- a/codex-rs/core/src/session/tests.rs
+++ b/codex-rs/core/src/session/tests.rs
@@ -74,6 +74,7 @@ use codex_protocol::models::FunctionCallOutputBody;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::PermissionProfile;
 use codex_protocol::models::SandboxEnforcement;
 use codex_protocol::openai_models::ModelServiceTier;
@@ -2575,11 +2576,15 @@ async fn prepares_image_failures_before_history_insertion() {
                     text: "before".to_string(),
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "data:image/png;base64,%%%".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,%%%".to_string(),
+                    },
                     detail: Some(ImageDetail::High),
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "https://example.com/image.png".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "https://example.com/image.png".to_string(),
+                    },
                     detail: Some(ImageDetail::High),
                 },
             ]),
@@ -2644,11 +2649,15 @@ async fn prepares_resumed_history_before_installing_it() {
         role: "user".to_string(),
         content: vec![
             ContentItem::InputImage {
-                image_url: "data:image/png;base64,%%%".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,%%%".to_string(),
+                },
                 detail: Some(ImageDetail::High),
             },
             ContentItem::InputImage {
-                image_url: "https://example.com/image.png".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "https://example.com/image.png".to_string(),
+                },
                 detail: Some(ImageDetail::High),
             },
             ContentItem::InputText {
@@ -11270,7 +11279,9 @@ async fn task_finish_emits_turn_item_lifecycle_for_leftover_pending_user_input()
                 text: "late pending input".to_string(),
             },
             ContentItem::InputImage {
-                image_url: image_url.clone(),
+                image: ImageReference::Inline {
+                    image_url: image_url.clone(),
+                },
                 detail: Some(ImageDetail::Original),
             },
         ],
diff --git a/codex-rs/core/src/tools/code_mode/response_adapter.rs b/codex-rs/core/src/tools/code_mode/response_adapter.rs
index 635eb94900..f28d5da934 100644
--- a/codex-rs/core/src/tools/code_mode/response_adapter.rs
+++ b/codex-rs/core/src/tools/code_mode/response_adapter.rs
@@ -2,6 +2,7 @@ use codex_code_mode::ImageDetail as CodeModeImageDetail;
 use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 
 trait IntoProtocol<T> {
     fn into_protocol(self) -> T;
@@ -36,7 +37,7 @@ impl IntoProtocol<FunctionCallOutputContentItem>
             }
             codex_code_mode::FunctionCallOutputContentItem::InputImage { image_url, detail } => {
                 FunctionCallOutputContentItem::InputImage {
-                    image_url,
+                    image: ImageReference::Inline { image_url },
                     detail: detail
                         .map(IntoProtocol::into_protocol)
                         .or(Some(DEFAULT_IMAGE_DETAIL)),
diff --git a/codex-rs/core/src/tools/context_tests.rs b/codex-rs/core/src/tools/context_tests.rs
index 7a393d35d7..bd0cb0c5b5 100644
--- a/codex-rs/core/src/tools/context_tests.rs
+++ b/codex-rs/core/src/tools/context_tests.rs
@@ -1,5 +1,6 @@
 use super::*;
 use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::SearchToolCallParams;
 use core_test_support::assert_regex_match;
 use pretty_assertions::assert_eq;
@@ -217,7 +218,9 @@ fn mcp_tool_output_response_item_preserves_content_items() {
                             text: "Wall time: 0.5000 seconds\nOutput:".to_string(),
                         },
                         FunctionCallOutputContentItem::InputImage {
-                            image_url: image_url.to_string(),
+                            image: ImageReference::Inline {
+                                image_url: image_url.to_string()
+                            },
                             detail: Some(DEFAULT_IMAGE_DETAIL),
                         },
                     ]
@@ -295,7 +298,9 @@ fn custom_tool_calls_can_derive_text_from_content_items() {
                 text: "line 1".to_string(),
             },
             FunctionCallOutputContentItem::InputImage {
-                image_url: "data:image/png;base64,AAA".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,AAA".to_string(),
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
             FunctionCallOutputContentItem::InputText {
@@ -315,7 +320,9 @@ fn custom_tool_calls_can_derive_text_from_content_items() {
                     text: "line 1".to_string(),
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "data:image/png;base64,AAA".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,AAA".to_string(),
+                    },
                     detail: Some(DEFAULT_IMAGE_DETAIL),
                 },
                 FunctionCallOutputContentItem::InputText {
diff --git a/codex-rs/core/src/tools/handlers/view_image.rs b/codex-rs/core/src/tools/handlers/view_image.rs
index e447208fd4..fd4e3910db 100644
--- a/codex-rs/core/src/tools/handlers/view_image.rs
+++ b/codex-rs/core/src/tools/handlers/view_image.rs
@@ -7,6 +7,7 @@ use codex_protocol::models::FunctionCallOutputBody;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseInputItem;
 use codex_protocol::openai_models::InputModality;
 use codex_utils_image::data_url_from_bytes;
@@ -238,7 +239,9 @@ impl ToolOutput for ViewImageOutput {
     fn to_response_item(&self, call_id: &str, _payload: &ToolPayload) -> ResponseInputItem {
         let body =
             FunctionCallOutputBody::ContentItems(vec![FunctionCallOutputContentItem::InputImage {
-                image_url: self.image_url.clone(),
+                image: ImageReference::Inline {
+                    image_url: self.image_url.clone(),
+                },
                 detail: Some(self.image_detail),
             }]);
         let output = FunctionCallOutputPayload {
diff --git a/codex-rs/core/tests/suite/client.rs b/codex-rs/core/tests/suite/client.rs
index 033126a16e..79d0eb4f6d 100644
--- a/codex-rs/core/tests/suite/client.rs
+++ b/codex-rs/core/tests/suite/client.rs
@@ -40,6 +40,7 @@ use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::LocalShellAction;
 use codex_protocol::models::LocalShellExecAction;
 use codex_protocol::models::LocalShellStatus;
@@ -1078,7 +1079,9 @@ async fn resume_replays_legacy_js_repl_image_rollout_shapes() {
                 id: None,
                 role: "user".to_string(),
                 content: vec![ContentItem::InputImage {
-                    image_url: legacy_image_url.to_string(),
+                    image: ImageReference::Inline {
+                        image_url: legacy_image_url.to_string(),
+                    },
                     detail: Some(DEFAULT_IMAGE_DETAIL),
                 }],
                 phase: None,
@@ -1218,7 +1221,9 @@ async fn resume_replays_image_tool_outputs_with_detail() {
                 namespace: None,
                 output: FunctionCallOutputPayload::from_content_items(vec![
                     FunctionCallOutputContentItem::InputImage {
-                        image_url: image_url.to_string(),
+                        image: ImageReference::Inline {
+                            image_url: image_url.to_string(),
+                        },
                         detail: Some(ImageDetail::Original),
                     },
                 ]),
@@ -1247,7 +1252,9 @@ async fn resume_replays_image_tool_outputs_with_detail() {
                 name: None,
                 output: FunctionCallOutputPayload::from_content_items(vec![
                     FunctionCallOutputContentItem::InputImage {
-                        image_url: image_url.to_string(),
+                        image: ImageReference::Inline {
+                            image_url: image_url.to_string(),
+                        },
                         detail: Some(ImageDetail::Original),
                     },
                 ]),
diff --git a/codex-rs/core/tests/suite/code_mode.rs b/codex-rs/core/tests/suite/code_mode.rs
index 1663fa0a1a..acb1f4b26e 100644
--- a/codex-rs/core/tests/suite/code_mode.rs
+++ b/codex-rs/core/tests/suite/code_mode.rs
@@ -42,6 +42,7 @@ use codex_protocol::dynamic_tools::DynamicToolResponse;
 use codex_protocol::dynamic_tools::DynamicToolSpec;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::PermissionProfile;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::openai_models::InputModality;
@@ -5662,7 +5663,9 @@ async fn code_mode_node_repl_screenshots_can_be_captured_without_guardian_transc
     assert_eq!(
         evidence.images(),
         vec![ContentItem::InputImage {
-            image_url: SCREENSHOT.to_owned(),
+            image: ImageReference::Inline {
+                image_url: SCREENSHOT.to_owned()
+            },
             detail: Some(ImageDetail::Low),
         }]
     );
diff --git a/codex-rs/core/tests/suite/image_rollout.rs b/codex-rs/core/tests/suite/image_rollout.rs
index 3316742feb..e12d6a7f57 100644
--- a/codex-rs/core/tests/suite/image_rollout.rs
+++ b/codex-rs/core/tests/suite/image_rollout.rs
@@ -10,6 +10,7 @@ use codex_protocol::config_types::Settings;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::PermissionProfile;
 use codex_protocol::models::ResponseInputItem;
 use codex_protocol::models::ResponseItem;
@@ -69,7 +70,10 @@ fn find_user_message_with_image(text: &str) -> Option<ResponseItem> {
 fn extract_image_url(item: &ResponseItem) -> Option<String> {
     match item {
         ResponseItem::Message { content, .. } => content.iter().find_map(|span| match span {
-            ContentItem::InputImage { image_url, .. } => Some(image_url.clone()),
+            ContentItem::InputImage {
+                image: ImageReference::Inline { image_url },
+                ..
+            } => Some(image_url.clone()),
             _ => None,
         }),
         _ => None,
@@ -178,7 +182,7 @@ async fn copy_paste_local_image_persists_rollout_request_shape() -> anyhow::Resu
                 ),
             },
             ContentItem::InputImage {
-                image_url,
+                image: ImageReference::Inline { image_url },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
             ContentItem::InputText {
@@ -269,7 +273,7 @@ async fn drag_drop_image_persists_rollout_request_shape() -> anyhow::Result<()>
         role: "user".to_string(),
         content: vec![
             ContentItem::InputImage {
-                image_url,
+                image: ImageReference::Inline { image_url },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
             ContentItem::InputText {
@@ -341,7 +345,9 @@ async fn resumed_history_only_emits_resize_notices_for_new_images() -> anyhow::R
     historical_content.insert(
         /*index*/ 0,
         ContentItem::InputImage {
-            image_url: original_image_url.clone(),
+            image: ImageReference::Inline {
+                image_url: original_image_url.clone(),
+            },
             detail: Some(ImageDetail::High),
         },
     );
@@ -480,7 +486,9 @@ async fn resumed_history_only_emits_resize_notices_for_new_images() -> anyhow::R
             ResponseInputItem::Message {
                 role: "user".to_string(),
                 content: vec![ContentItem::InputImage {
-                    image_url: original_image_url,
+                    image: ImageReference::Inline {
+                        image_url: original_image_url,
+                    },
                     detail: Some(ImageDetail::High),
                 }],
                 phase: None,
diff --git a/codex-rs/core/tests/suite/rmcp_client.rs b/codex-rs/core/tests/suite/rmcp_client.rs
index 78999fca06..23405bf9fc 100644
--- a/codex-rs/core/tests/suite/rmcp_client.rs
+++ b/codex-rs/core/tests/suite/rmcp_client.rs
@@ -52,6 +52,7 @@ use codex_protocol::mcp_policy::McpServerRequirement;
 use codex_protocol::mcp_policy::PluginMcpRequirements;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::PermissionProfile;
 use codex_protocol::models::PermissionProfileSnapshot;
 use codex_protocol::models::ResponseItem;
@@ -1549,7 +1550,9 @@ async fn interrupt_during_mcp_startup_preserves_user_input_in_history(
     };
     assert!(
         content.contains(&ContentItem::InputImage {
-            image_url: OPENAI_PNG.to_string(),
+            image: ImageReference::Inline {
+                image_url: OPENAI_PNG.to_string()
+            },
             detail: Some(ImageDetail::Original),
         }),
         "interrupted input must use the current model's unified image budget"
diff --git a/codex-rs/ext/guardian-v2/src/async_scorer/budget_tests.rs b/codex-rs/ext/guardian-v2/src/async_scorer/budget_tests.rs
index 171dc00518..9454713ef8 100644
--- a/codex-rs/ext/guardian-v2/src/async_scorer/budget_tests.rs
+++ b/codex-rs/ext/guardian-v2/src/async_scorer/budget_tests.rs
@@ -2,6 +2,7 @@
 
 use super::*;
 use crate::async_scorer::authorization::ScoreAuthorization;
+use codex_protocol::models::ImageReference;
 use pretty_assertions::assert_eq;
 
 #[derive(Clone, Copy)]
@@ -139,7 +140,7 @@ async fn assert_catalog_budget(evidence: BudgetEvidence) -> Result<()> {
                     unreachable!("user instruction is a message");
                 };
                 content.push(ContentItem::InputImage {
-                    image_url: "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAIAAACQd1PeAAAADElEQVR4nGPgEpEDAABoAD1UCKP3AAAAAElFTkSuQmCC".to_owned(),
+                    image: ImageReference::Inline { image_url: "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAIAAACQd1PeAAAADElEQVR4nGPgEpEDAABoAD1UCKP3AAAAAElFTkSuQmCC".to_owned() },
                     detail: None,
                 });
                 (None, vec!["optional old commentary ".repeat(/*n*/ 500)])
diff --git a/codex-rs/ext/guardian-v2/src/async_scorer/extension_tests.rs b/codex-rs/ext/guardian-v2/src/async_scorer/extension_tests.rs
index 31aac3bed6..38440c4f6c 100644
--- a/codex-rs/ext/guardian-v2/src/async_scorer/extension_tests.rs
+++ b/codex-rs/ext/guardian-v2/src/async_scorer/extension_tests.rs
@@ -38,6 +38,7 @@ use codex_protocol::models::ContentItem;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::InternalChatMessageMetadataPassthrough;
 use codex_protocol::models::LocalShellAction;
 use codex_protocol::models::LocalShellExecAction;
@@ -1641,7 +1642,9 @@ async fn contributor_includes_transcript_images_by_default() -> Result<()> {
                     text: "Review what is shown on screen.".to_owned(),
                 },
                 ContentItem::InputImage {
-                    image_url: user_image.to_owned(),
+                    image: ImageReference::Inline {
+                        image_url: user_image.to_owned(),
+                    },
                     detail: Some(ImageDetail::High),
                 },
             ],
@@ -1667,7 +1670,9 @@ async fn contributor_includes_transcript_images_by_default() -> Result<()> {
                     text: "Screenshot captured.".to_owned(),
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: tool_image.to_owned(),
+                    image: ImageReference::Inline {
+                        image_url: tool_image.to_owned(),
+                    },
                     detail: Some(ImageDetail::High),
                 },
             ]),
diff --git a/codex-rs/ext/guardian-v2/src/async_scorer/transcript_tests.rs b/codex-rs/ext/guardian-v2/src/async_scorer/transcript_tests.rs
index ed1bc9ab6a..a3add7fc0a 100644
--- a/codex-rs/ext/guardian-v2/src/async_scorer/transcript_tests.rs
+++ b/codex-rs/ext/guardian-v2/src/async_scorer/transcript_tests.rs
@@ -7,6 +7,7 @@ use codex_protocol::models::AgentMessageInputContent;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::MessagePhase;
 use codex_protocol::models::ReasoningItemContent;
 use codex_protocol::models::ReasoningItemReasoningSummary;
@@ -1077,7 +1078,9 @@ fn transcript_preserves_outputs_with_call_ids_or_explicit_names() {
     if let ResponseItem::FunctionCallOutput { output, .. } = &mut items[0] {
         *output = FunctionCallOutputPayload::from_content_items(vec![
             FunctionCallOutputContentItem::InputImage {
-                image_url: "data:image/png;base64,image".to_owned(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,image".to_owned(),
+                },
                 detail: None,
             },
         ]);
@@ -1222,7 +1225,9 @@ fn transcript_omits_media_payloads_and_keeps_readable_content() {
                     text: "Review this screenshot.".to_string(),
                 },
                 ContentItem::InputImage {
-                    image_url: format!("data:image/png;base64,{oversized_image}"),
+                    image: ImageReference::Inline {
+                        image_url: format!("data:image/png;base64,{oversized_image}"),
+                    },
                     detail: None,
                 },
                 ContentItem::InputAudio {
@@ -1242,7 +1247,9 @@ fn transcript_omits_media_payloads_and_keeps_readable_content() {
                     text: "Screenshot captured.".to_string(),
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "data:image/png;base64,tool-image".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,tool-image".to_string(),
+                    },
                     detail: None,
                 },
                 FunctionCallOutputContentItem::InputAudio {
diff --git a/codex-rs/ext/history-notes/src/tools.rs b/codex-rs/ext/history-notes/src/tools.rs
index ad7422c961..114ee2ece8 100644
--- a/codex-rs/ext/history-notes/src/tools.rs
+++ b/codex-rs/ext/history-notes/src/tools.rs
@@ -11,6 +11,7 @@ use codex_extension_api::parse_tool_input_schema;
 use codex_protocol::models::FunctionCallOutputBody;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseInputItem;
 use codex_tools::JsonToolOutput;
 use codex_tools::ResponsesApiNamespace;
@@ -367,7 +368,9 @@ impl HistoryNotesToolOutput {
                     serde_json::from_value(image.get("detail").cloned().unwrap_or(Value::Null))
                         .map_err(|_| invalid_image())?;
                 content.push(FunctionCallOutputContentItem::InputImage {
-                    image_url: format!("data:{mime_type};base64,{data}"),
+                    image: ImageReference::Inline {
+                        image_url: format!("data:{mime_type};base64,{data}"),
+                    },
                     detail,
                 });
             }
diff --git a/codex-rs/ext/history-notes/src/tools_tests.rs b/codex-rs/ext/history-notes/src/tools_tests.rs
index 55b99cdc35..9ec63f6621 100644
--- a/codex-rs/ext/history-notes/src/tools_tests.rs
+++ b/codex-rs/ext/history-notes/src/tools_tests.rs
@@ -1,6 +1,7 @@
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseInputItem;
 use codex_tools::ToolOutput;
 use codex_tools::ToolPayload;
@@ -73,19 +74,27 @@ fn preserves_images_as_separate_output_items_without_logging_bytes() {
                     encrypted_content: "enc_payload".to_string()
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "data:image/png;base64,cG5n".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,cG5n".to_string()
+                    },
                     detail: Some(ImageDetail::Original)
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "data:image/jpeg;base64,anBlZw==".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/jpeg;base64,anBlZw==".to_string()
+                    },
                     detail: Some(ImageDetail::Low)
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "data:image/gif;base64,Z2lm".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/gif;base64,Z2lm".to_string()
+                    },
                     detail: None
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "data:image/webp;base64,d2VicA==".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/webp;base64,d2VicA==".to_string()
+                    },
                     detail: None
                 },
             ])
diff --git a/codex-rs/ext/image-generation/src/tests.rs b/codex-rs/ext/image-generation/src/tests.rs
index c7bca61e8a..32b0493775 100644
--- a/codex-rs/ext/image-generation/src/tests.rs
+++ b/codex-rs/ext/image-generation/src/tests.rs
@@ -12,6 +12,7 @@ use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
 use codex_protocol::models::FunctionCallOutputBody;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseInputItem;
 use codex_protocol::models::ResponseItem;
 use codex_tools::ResponsesApiNamespaceTool;
@@ -266,7 +267,9 @@ fn generated_output_returns_image_input_and_output_hint() {
         content_items,
         vec![
             FunctionCallOutputContentItem::InputImage {
-                image_url: format!("data:image/png;base64,{RESULT}"),
+                image: ImageReference::Inline {
+                    image_url: format!("data:image/png;base64,{RESULT}")
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
             FunctionCallOutputContentItem::InputText { text: output_hint },
@@ -311,7 +314,9 @@ fn generated_output_omits_oversized_output_hint() {
     assert_eq!(
         content_items,
         vec![FunctionCallOutputContentItem::InputImage {
-            image_url: format!("data:image/png;base64,{RESULT}"),
+            image: ImageReference::Inline {
+                image_url: format!("data:image/png;base64,{RESULT}")
+            },
             detail: Some(DEFAULT_IMAGE_DETAIL),
         }]
     );
@@ -319,14 +324,18 @@ fn generated_output_omits_oversized_output_hint() {
 
 fn input_image(image: &str) -> ContentItem {
     ContentItem::InputImage {
-        image_url: format!("data:image/png;base64,{image}"),
+        image: ImageReference::Inline {
+            image_url: format!("data:image/png;base64,{image}"),
+        },
         detail: None,
     }
 }
 
 fn image_output(image: &str) -> FunctionCallOutputPayload {
     FunctionCallOutputPayload::from_content_items(vec![FunctionCallOutputContentItem::InputImage {
-        image_url: format!("data:image/png;base64,{image}"),
+        image: ImageReference::Inline {
+            image_url: format!("data:image/png;base64,{image}"),
+        },
         detail: None,
     }])
 }
diff --git a/codex-rs/ext/image-generation/src/tool.rs b/codex-rs/ext/image-generation/src/tool.rs
index ac36240cc4..8fc408b442 100644
--- a/codex-rs/ext/image-generation/src/tool.rs
+++ b/codex-rs/ext/image-generation/src/tool.rs
@@ -29,6 +29,7 @@ use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
 use codex_protocol::models::FunctionCallOutputBody;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseInputItem;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::protocol::EventMsg;
@@ -494,7 +495,10 @@ fn recent_images(history: &[ResponseItem], count: usize) -> Vec<ImageUrl> {
         match item {
             ResponseItem::Message { content, .. } => {
                 image_urls.extend(content.iter().rev().filter_map(|item| match item {
-                    ContentItem::InputImage { image_url, .. } => Some(image_url.clone()),
+                    ContentItem::InputImage {
+                        image: ImageReference::Inline { image_url },
+                        ..
+                    } => Some(image_url.clone()),
                     ContentItem::InputText { .. }
                     | ContentItem::InputAudio { .. }
                     | ContentItem::OutputText { .. } => None,
@@ -542,7 +546,10 @@ fn output_image_urls(output: &FunctionCallOutputPayload) -> impl Iterator<Item =
         .flatten()
         .rev()
         .filter_map(|item| match item {
-            FunctionCallOutputContentItem::InputImage { image_url, .. } => Some(image_url.clone()),
+            FunctionCallOutputContentItem::InputImage {
+                image: ImageReference::Inline { image_url },
+                ..
+            } => Some(image_url.clone()),
             FunctionCallOutputContentItem::InputText { .. }
             | FunctionCallOutputContentItem::InputAudio { .. }
             | FunctionCallOutputContentItem::EncryptedContent { .. } => None,
@@ -651,7 +658,9 @@ impl ToolOutput for GeneratedImageOutput {
     /// Returns generated bytes and persisted-artifact context for model follow-up.
     fn to_response_item(&self, call_id: &str, _payload: &ToolPayload) -> ResponseInputItem {
         let mut content = vec![FunctionCallOutputContentItem::InputImage {
-            image_url: format!("data:image/png;base64,{}", self.result),
+            image: ImageReference::Inline {
+                image_url: format!("data:image/png;base64,{}", self.result),
+            },
             detail: Some(DEFAULT_IMAGE_DETAIL),
         }];
         if let Some(output_hint) = &self.output_hint {
diff --git a/codex-rs/ext/web-search/src/history.rs b/codex-rs/ext/web-search/src/history.rs
index fb1630c8dc..6d64e22124 100644
--- a/codex-rs/ext/web-search/src/history.rs
+++ b/codex-rs/ext/web-search/src/history.rs
@@ -84,6 +84,7 @@ mod tests {
     use codex_api::SearchInput;
     use codex_protocol::ResponseItemId;
     use codex_protocol::models::ContentItem;
+    use codex_protocol::models::ImageReference;
     use codex_protocol::models::ResponseItem;
     use pretty_assertions::assert_eq;
 
@@ -158,7 +159,9 @@ mod tests {
                     text: "previous user".to_string(),
                 },
                 ContentItem::InputImage {
-                    image_url: "data:image/png;base64,image".to_string(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,image".to_string(),
+                    },
                     detail: None,
                 },
             ],
diff --git a/codex-rs/guardian-context/src/budget.rs b/codex-rs/guardian-context/src/budget.rs
index 7f172a9c01..b29ff2c66f 100644
--- a/codex-rs/guardian-context/src/budget.rs
+++ b/codex-rs/guardian-context/src/budget.rs
@@ -5,6 +5,7 @@
 use std::io::Write;
 
 use codex_protocol::models::ContentItem;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::protocol::TruncationPolicy;
 
@@ -97,7 +98,10 @@ impl SectionCost {
             ContentItem::InputText { text } | ContentItem::OutputText { text } => {
                 self.text_bytes = self.text_bytes.saturating_add(text.len());
             }
-            ContentItem::InputImage { image_url, .. } => {
+            ContentItem::InputImage {
+                image: ImageReference::Inline { image_url },
+                ..
+            } => {
                 self.image_bytes = self.image_bytes.saturating_add(image_url.len());
                 self.image_count = self.image_count.saturating_add(1);
             }
@@ -206,7 +210,11 @@ pub(super) fn content_framing_tokens(item_count: usize) -> usize {
 
 fn adjusted_tokens(mut bytes: usize, content: &[ContentItem]) -> usize {
     for item in content {
-        if let ContentItem::InputImage { image_url, .. } = item {
+        if let ContentItem::InputImage {
+            image: ImageReference::Inline { image_url },
+            ..
+        } = item
+        {
             let payload = ByteCount::measure(|counter| serde_json::to_writer(counter, image_url));
             if payload == usize::MAX {
                 return usize::MAX;
diff --git a/codex-rs/guardian-context/src/budget_tests.rs b/codex-rs/guardian-context/src/budget_tests.rs
index 5af8e7460a..33fc174ad7 100644
--- a/codex-rs/guardian-context/src/budget_tests.rs
+++ b/codex-rs/guardian-context/src/budget_tests.rs
@@ -4,6 +4,7 @@ use super::*;
 use crate::Budgeted;
 use crate::composition::SectionOutput;
 use crate::composition::user_message as message;
+use codex_protocol::models::ImageReference;
 use pretty_assertions::assert_eq;
 
 #[test]
@@ -17,7 +18,9 @@ fn section_costs_keep_multimodal_payloads_separate() {
                         text: "évidence".to_owned(),
                     }),
                     Budgeted::required(ContentItem::InputImage {
-                        image_url: "data:image/png;base64,AAAA".to_owned(),
+                        image: ImageReference::Inline {
+                            image_url: "data:image/png;base64,AAAA".to_owned(),
+                        },
                         detail: None,
                     }),
                 ]),
@@ -59,7 +62,9 @@ fn section_costs_keep_multimodal_payloads_separate() {
 fn request_estimate_reserves_images_independently_of_encoded_size() {
     let image = |payload: &str| {
         message(vec![ContentItem::InputImage {
-            image_url: format!("data:image/png;base64,{payload}"),
+            image: ImageReference::Inline {
+                image_url: format!("data:image/png;base64,{payload}"),
+            },
             detail: None,
         }])
     };
diff --git a/codex-rs/guardian-context/src/composition.rs b/codex-rs/guardian-context/src/composition.rs
index 29c7427870..555fbb2163 100644
--- a/codex-rs/guardian-context/src/composition.rs
+++ b/codex-rs/guardian-context/src/composition.rs
@@ -6,6 +6,7 @@
 
 use codex_context_fragments::ContextualUserFragment;
 use codex_protocol::models::ContentItem;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::protocol::TruncationPolicy;
 use codex_protocol::user_input::UserInput;
@@ -214,7 +215,10 @@ impl CollectedContext {
                                 Ok(Budgeted::required(ContentItem::InputText { text }))
                             }
                             UserInput::Image { image_url, detail } => Ok(Budgeted::optional(
-                                ContentItem::InputImage { image_url, detail },
+                                ContentItem::InputImage {
+                                    image: ImageReference::Inline { image_url },
+                                    detail,
+                                },
                                 BudgetPriority::Image,
                             )),
                             _ => Err(SectionError::UnsupportedDelivery {
@@ -281,9 +285,10 @@ impl ComposedContext {
                         }));
                         continue;
                     }
-                    ContentItem::InputImage { image_url, detail } => {
-                        UserInput::Image { image_url, detail }
-                    }
+                    ContentItem::InputImage {
+                        image: ImageReference::Inline { image_url },
+                        detail,
+                    } => UserInput::Image { image_url, detail },
                     ContentItem::InputAudio { .. } | ContentItem::OutputText { .. } => {
                         return Err(SectionError::UnsupportedDelivery {
                             section: section.id,
diff --git a/codex-rs/guardian-context/src/enforcement.rs b/codex-rs/guardian-context/src/enforcement.rs
index 99c88dff12..54cf00d076 100644
--- a/codex-rs/guardian-context/src/enforcement.rs
+++ b/codex-rs/guardian-context/src/enforcement.rs
@@ -7,6 +7,7 @@ use std::collections::HashSet;
 
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::protocol::TruncationPolicy;
 
 use crate::ComposedContext;
@@ -90,7 +91,10 @@ impl ComposedContext {
     pub fn retain_images(&mut self, mut admit: impl FnMut(&str, &mut Option<ImageDetail>) -> bool) {
         for section in &mut self.sections {
             retain_content(section, &mut self.truncations, |_, item| match item {
-                ContentItem::InputImage { image_url, detail } => admit(image_url, detail),
+                ContentItem::InputImage {
+                    image: ImageReference::Inline { image_url },
+                    detail,
+                } => admit(image_url, detail),
                 _ => true,
             });
         }
@@ -273,7 +277,10 @@ fn retain_content(
         if !keep {
             let original_bytes = match &item.content {
                 ContentItem::InputText { text } | ContentItem::OutputText { text } => text.len(),
-                ContentItem::InputImage { image_url, .. } => image_url.len(),
+                ContentItem::InputImage {
+                    image: ImageReference::Inline { image_url },
+                    ..
+                } => image_url.len(),
                 ContentItem::InputAudio { audio_url } => audio_url.len(),
             };
             truncations.push(TruncationObservation {
diff --git a/codex-rs/guardian-context/src/enforcement_tests.rs b/codex-rs/guardian-context/src/enforcement_tests.rs
index 893e3fea64..88599eb177 100644
--- a/codex-rs/guardian-context/src/enforcement_tests.rs
+++ b/codex-rs/guardian-context/src/enforcement_tests.rs
@@ -3,6 +3,7 @@
 use super::*;
 use crate::budget::section_tokens;
 use crate::composition::user_message;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseItem;
 use pretty_assertions::assert_eq;
 
@@ -146,7 +147,9 @@ fn budget_reserves_existing_context_and_preserves_required_messages() {
         .unwrap()
         .into_message();
     let image = ContentItem::InputImage {
-        image_url: "data:image/png;base64,AAAA".to_owned(),
+        image: ImageReference::Inline {
+            image_url: "data:image/png;base64,AAAA".to_owned(),
+        },
         detail: None,
     };
     let make_context = || ComposedContext {
@@ -241,7 +244,9 @@ fn image_omission_preserves_text_and_later_eviction_policy() {
             delivery: SectionDelivery::UserContent(vec![
                 Budgeted::optional(
                     ContentItem::InputImage {
-                        image_url: "rejected-image".to_owned(),
+                        image: ImageReference::Inline {
+                            image_url: "rejected-image".to_owned(),
+                        },
                         detail: None,
                     },
                     BudgetPriority::Image,
@@ -307,11 +312,15 @@ fn image_omission_preserves_text_and_later_eviction_policy() {
     );
 
     let older = ContentItem::InputImage {
-        image_url: "older-image".to_owned(),
+        image: ImageReference::Inline {
+            image_url: "older-image".to_owned(),
+        },
         detail: None,
     };
     let newer = ContentItem::InputImage {
-        image_url: "newer-image".to_owned(),
+        image: ImageReference::Inline {
+            image_url: "newer-image".to_owned(),
+        },
         detail: None,
     };
     let image_section = |images: Vec<ContentItem>| SectionOutput {
diff --git a/codex-rs/guardian-context/src/history_tests.rs b/codex-rs/guardian-context/src/history_tests.rs
index 4c03904ba2..f505003037 100644
--- a/codex-rs/guardian-context/src/history_tests.rs
+++ b/codex-rs/guardian-context/src/history_tests.rs
@@ -1,4 +1,5 @@
 use codex_protocol::models::ExecutedToolCall;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ReasoningItemContent;
 use pretty_assertions::assert_eq;
 use serde_json::json;
@@ -147,7 +148,9 @@ fn oversized_user_images_preserve_text_and_metadata_in_order() {
         content.insert(
             /*index*/ 1,
             ContentItem::InputImage {
-                image_url: format!("data:image/png;base64,{}", "A".repeat(image_bytes)),
+                image: ImageReference::Inline {
+                    image_url: format!("data:image/png;base64,{}", "A".repeat(image_bytes)),
+                },
                 detail: Some(codex_protocol::models::ImageDetail::Original),
             },
         );
diff --git a/codex-rs/guardian-context/src/images.rs b/codex-rs/guardian-context/src/images.rs
index 1d50b90930..ccae031095 100644
--- a/codex-rs/guardian-context/src/images.rs
+++ b/codex-rs/guardian-context/src/images.rs
@@ -10,6 +10,7 @@ use crate::SectionScope;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseItem;
 use std::collections::VecDeque;
 
@@ -65,7 +66,11 @@ impl TranscriptImages {
             while images.len() >= MAX_TRANSCRIPT_IMAGES
                 || image_bytes + image_url.len() > MAX_TRANSCRIPT_IMAGE_BYTES
             {
-                let Some(ContentItem::InputImage { image_url, .. }) = images.pop_front() else {
+                let Some(ContentItem::InputImage {
+                    image: ImageReference::Inline { image_url },
+                    ..
+                }) = images.pop_front()
+                else {
                     break;
                 };
                 image_bytes -= image_url.len();
@@ -73,7 +78,9 @@ impl TranscriptImages {
             }
             image_bytes += image_url.len();
             images.push_back(ContentItem::InputImage {
-                image_url: image_url.to_owned(),
+                image: ImageReference::Inline {
+                    image_url: image_url.to_owned(),
+                },
                 detail,
             });
         };
@@ -84,7 +91,11 @@ impl TranscriptImages {
                     if matches!(role.as_str(), "user" | "assistant") =>
                 {
                     for item in content {
-                        if let ContentItem::InputImage { image_url, detail } = item {
+                        if let ContentItem::InputImage {
+                            image: ImageReference::Inline { image_url },
+                            detail,
+                        } = item
+                        {
                             include_image(image_url, *detail);
                         }
                     }
@@ -95,8 +106,10 @@ impl TranscriptImages {
                 {
                     if let Some(content) = output.content_items() {
                         for item in content {
-                            if let FunctionCallOutputContentItem::InputImage { image_url, detail } =
-                                item
+                            if let FunctionCallOutputContentItem::InputImage {
+                                image: ImageReference::Inline { image_url },
+                                detail,
+                            } = item
                             {
                                 include_image(image_url, *detail);
                             }
@@ -108,7 +121,11 @@ impl TranscriptImages {
         }
         if input.include_tool_outputs {
             for image in input.node_repl_images {
-                if let ContentItem::InputImage { image_url, detail } = image {
+                if let ContentItem::InputImage {
+                    image: ImageReference::Inline { image_url },
+                    detail,
+                } = image
+                {
                     include_image(image_url, *detail);
                 }
             }
diff --git a/codex-rs/guardian-context/src/images_tests.rs b/codex-rs/guardian-context/src/images_tests.rs
index d39b7e8910..090b5c0531 100644
--- a/codex-rs/guardian-context/src/images_tests.rs
+++ b/codex-rs/guardian-context/src/images_tests.rs
@@ -5,12 +5,15 @@ use codex_protocol::models::ContentItem;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseItem;
 use pretty_assertions::assert_eq;
 
 fn image(url: &str) -> ContentItem {
     ContentItem::InputImage {
-        image_url: url.into(),
+        image: ImageReference::Inline {
+            image_url: url.into(),
+        },
         detail: Some(ImageDetail::High),
     }
 }
@@ -32,7 +35,9 @@ fn image_selection_preserves_source_policy_order_and_both_limits() {
             namespace: None,
             output: FunctionCallOutputPayload::from_content_items(vec![
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "tool".into(),
+                    image: ImageReference::Inline {
+                        image_url: "tool".into(),
+                    },
                     detail: Some(ImageDetail::High),
                 },
             ]),
diff --git a/codex-rs/protocol/src/models.rs b/codex-rs/protocol/src/models.rs
index 7286f9865f..e2f588eaa3 100644
--- a/codex-rs/protocol/src/models.rs
+++ b/codex-rs/protocol/src/models.rs
@@ -879,7 +879,8 @@ pub enum ContentItem {
         text: String,
     },
     InputImage {
-        image_url: String,
+        #[serde(flatten)]
+        image: ImageReference,
         #[serde(default, skip_serializing_if = "Option::is_none")]
         #[ts(optional)]
         detail: Option<ImageDetail>,
@@ -892,6 +893,13 @@ pub enum ContentItem {
     },
 }
 
+#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, JsonSchema, TS)]
+#[serde(untagged)]
+#[ts(untagged)]
+pub enum ImageReference {
+    Inline { image_url: String },
+}
+
 #[derive(Debug, Clone, Serialize, Deserialize, PartialEq, JsonSchema, TS)]
 #[serde(tag = "type", rename_all = "snake_case")]
 pub enum AgentMessageInputContent {
@@ -1843,7 +1851,7 @@ fn local_image_content_items(
         });
     }
     items.push(ContentItem::InputImage {
-        image_url,
+        image: ImageReference::Inline { image_url },
         detail: Some(detail),
     });
     if label_number.is_some() {
@@ -2006,7 +2014,7 @@ impl ResponseInputItem {
                         image_index += 1;
                         let detail = detail.unwrap_or(DEFAULT_IMAGE_DETAIL);
                         vec![ContentItem::InputImage {
-                            image_url,
+                            image: ImageReference::Inline { image_url },
                             detail: Some(detail),
                         }]
                     }
@@ -2080,7 +2088,8 @@ pub enum FunctionCallOutputContentItem {
     },
     // Do not rename, these are serialized and used directly in the responses API.
     InputImage {
-        image_url: String,
+        #[serde(flatten)]
+        image: ImageReference,
         #[serde(default, skip_serializing_if = "Option::is_none")]
         #[ts(optional)]
         detail: Option<ImageDetail>,
@@ -2138,7 +2147,7 @@ impl From<crate::dynamic_tools::DynamicToolCallOutputContentItem>
             }
             crate::dynamic_tools::DynamicToolCallOutputContentItem::InputImage { image_url } => {
                 Self::InputImage {
-                    image_url,
+                    image: ImageReference::Inline { image_url },
                     detail: Some(DEFAULT_IMAGE_DETAIL),
                 }
             }
@@ -2383,7 +2392,7 @@ fn convert_mcp_content_to_items(
                     format!("data:{mime_type};base64,{data}")
                 };
                 FunctionCallOutputContentItem::InputImage {
-                    image_url,
+                    image: ImageReference::Inline { image_url },
                     detail: meta
                         .as_ref()
                         .and_then(serde_json::Value::as_object)
@@ -2656,7 +2665,9 @@ mod tests {
         assert_eq!(
             content_item,
             ContentItem::InputImage {
-                image_url: "data:image/png;base64,abc".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,abc".to_string(),
+                },
                 detail: Some(ImageDetail::Auto),
             }
         );
@@ -2711,7 +2722,9 @@ mod tests {
         assert_eq!(
             items,
             vec![FunctionCallOutputContentItem::InputImage {
-                image_url: "data:image/png;base64,Zm9v".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,Zm9v".to_string(),
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             }]
         );
@@ -3043,7 +3056,9 @@ mod tests {
         assert_eq!(
             items,
             vec![FunctionCallOutputContentItem::InputImage {
-                image_url: "data:image/png;base64,Zm9v".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,Zm9v".to_string(),
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             }]
         );
@@ -3100,7 +3115,9 @@ mod tests {
                 text: "line 1".to_string(),
             },
             FunctionCallOutputContentItem::InputImage {
-                image_url: "data:image/png;base64,AAA".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,AAA".to_string(),
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
             FunctionCallOutputContentItem::InputText {
@@ -3119,7 +3136,9 @@ mod tests {
                 text: "   ".to_string(),
             },
             FunctionCallOutputContentItem::InputImage {
-                image_url: "data:image/png;base64,AAA".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,AAA".to_string(),
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
             FunctionCallOutputContentItem::InputAudio {
@@ -3148,7 +3167,9 @@ mod tests {
                 text: "line 1".to_string(),
             },
             FunctionCallOutputContentItem::InputImage {
-                image_url: "data:image/png;base64,AAA".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,AAA".to_string(),
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
         ]);
@@ -3423,7 +3444,9 @@ mod tests {
                     text: "caption".into(),
                 },
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "data:image/png;base64,BASE64".into(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,BASE64".into(),
+                    },
                     detail: Some(DEFAULT_IMAGE_DETAIL),
                 },
             ]
@@ -3498,7 +3521,9 @@ mod tests {
             name: None,
             output: FunctionCallOutputPayload::from_content_items(vec![
                 FunctionCallOutputContentItem::InputImage {
-                    image_url: "data:image/png;base64,BASE64".into(),
+                    image: ImageReference::Inline {
+                        image_url: "data:image/png;base64,BASE64".into(),
+                    },
                     detail: Some(DEFAULT_IMAGE_DETAIL),
                 },
             ]),
@@ -3563,7 +3588,9 @@ mod tests {
         assert_eq!(
             items,
             vec![FunctionCallOutputContentItem::InputImage {
-                image_url: "data:image/png;base64,BASE64".into(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,BASE64".into(),
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             }]
         );
@@ -3595,7 +3622,9 @@ mod tests {
         assert_eq!(
             items,
             vec![FunctionCallOutputContentItem::InputImage {
-                image_url: "data:image/png;base64,BASE64".into(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,BASE64".into(),
+                },
                 detail: Some(ImageDetail::Original),
             }]
         );
@@ -3627,7 +3656,9 @@ mod tests {
         assert_eq!(
             items,
             vec![FunctionCallOutputContentItem::InputImage {
-                image_url: "data:image/png;base64,BASE64".into(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,BASE64".into(),
+                },
                 detail: Some(ImageDetail::High),
             }]
         );
@@ -3650,7 +3681,9 @@ mod tests {
                 text: "note".into(),
             },
             FunctionCallOutputContentItem::InputImage {
-                image_url: "data:image/png;base64,XYZ".into(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,XYZ".into(),
+                },
                 detail: None,
             },
         ];
@@ -3859,7 +3892,7 @@ mod tests {
         match item {
             ResponseInputItem::Message { content, .. } => {
                 let expected = vec![ContentItem::InputImage {
-                    image_url,
+                    image: ImageReference::Inline { image_url },
                     detail: Some(DEFAULT_IMAGE_DETAIL),
                 }];
                 assert_eq!(content, expected);
@@ -4005,7 +4038,7 @@ mod tests {
                 assert_eq!(
                     content.first(),
                     Some(&ContentItem::InputImage {
-                        image_url,
+                        image: ImageReference::Inline { image_url },
                         detail: Some(ImageDetail::Original),
                     })
                 );
@@ -4208,7 +4241,7 @@ mod tests {
                 assert_eq!(
                     content.first(),
                     Some(&ContentItem::InputImage {
-                        image_url,
+                        image: ImageReference::Inline { image_url },
                         detail: Some(DEFAULT_IMAGE_DETAIL),
                     })
                 );
diff --git a/codex-rs/thread-store/src/local/rollout_migration/rollback_plan.rs b/codex-rs/thread-store/src/local/rollout_migration/rollback_plan.rs
index 588ac37787..07da7b7042 100644
--- a/codex-rs/thread-store/src/local/rollout_migration/rollback_plan.rs
+++ b/codex-rs/thread-store/src/local/rollout_migration/rollback_plan.rs
@@ -12,6 +12,7 @@ use std::collections::HashSet;
 
 use codex_protocol::ResponseItemId;
 use codex_protocol::models::ContentItem;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::protocol::EventMsg;
 use codex_protocol::protocol::UserMessageEvent;
@@ -492,7 +493,10 @@ fn user_response_matches_event(content: &[ContentItem], event: &UserMessageEvent
     for item in content {
         match item {
             ContentItem::InputText { text: item_text } => text.push_str(item_text),
-            ContentItem::InputImage { image_url, .. } => images.push(image_url.as_str()),
+            ContentItem::InputImage {
+                image: ImageReference::Inline { image_url },
+                ..
+            } => images.push(image_url.as_str()),
             ContentItem::InputAudio { audio_url } => audio.push(audio_url.as_str()),
             ContentItem::OutputText { .. } => return false,
         }
diff --git a/codex-rs/tools/src/image_detail_tests.rs b/codex-rs/tools/src/image_detail_tests.rs
index 0fb26c10c4..c0949576fe 100644
--- a/codex-rs/tools/src/image_detail_tests.rs
+++ b/codex-rs/tools/src/image_detail_tests.rs
@@ -2,6 +2,7 @@ use super::*;
 use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::ImageDetail;
+use codex_protocol::models::ImageReference;
 use codex_protocol::openai_models::ModelInfo;
 use pretty_assertions::assert_eq;
 use serde_json::json;
@@ -88,11 +89,15 @@ fn sanitize_original_falls_back_to_high_without_support() {
             text: "header".to_string(),
         },
         FunctionCallOutputContentItem::InputImage {
-            image_url: "data:image/png;base64,AAA".to_string(),
+            image: ImageReference::Inline {
+                image_url: "data:image/png;base64,AAA".to_string(),
+            },
             detail: Some(ImageDetail::Original),
         },
         FunctionCallOutputContentItem::InputImage {
-            image_url: "data:image/png;base64,BBB".to_string(),
+            image: ImageReference::Inline {
+                image_url: "data:image/png;base64,BBB".to_string(),
+            },
             detail: Some(ImageDetail::Low),
         },
     ];
@@ -106,11 +111,15 @@ fn sanitize_original_falls_back_to_high_without_support() {
                 text: "header".to_string(),
             },
             FunctionCallOutputContentItem::InputImage {
-                image_url: "data:image/png;base64,AAA".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,AAA".to_string()
+                },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
             FunctionCallOutputContentItem::InputImage {
-                image_url: "data:image/png;base64,BBB".to_string(),
+                image: ImageReference::Inline {
+                    image_url: "data:image/png;base64,BBB".to_string()
+                },
                 detail: Some(ImageDetail::Low),
             },
         ]
diff --git a/codex-rs/tools/src/tool_output.rs b/codex-rs/tools/src/tool_output.rs
index e95e662941..a299d172ee 100644
--- a/codex-rs/tools/src/tool_output.rs
+++ b/codex-rs/tools/src/tool_output.rs
@@ -2,6 +2,7 @@ use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
 use codex_protocol::models::FunctionCallOutputBody;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
+use codex_protocol::models::ImageReference;
 use codex_protocol::models::ResponseInputItem;
 use serde_json::Value as JsonValue;
 
@@ -219,12 +220,13 @@ fn response_input_to_code_mode_result(response: ResponseInputItem) -> JsonValue
                     | codex_protocol::models::ContentItem::OutputText { text } => {
                         FunctionCallOutputContentItem::InputText { text }
                     }
-                    codex_protocol::models::ContentItem::InputImage { image_url, detail } => {
-                        FunctionCallOutputContentItem::InputImage {
-                            image_url,
-                            detail: detail.or(Some(DEFAULT_IMAGE_DETAIL)),
-                        }
-                    }
+                    codex_protocol::models::ContentItem::InputImage {
+                        image: ImageReference::Inline { image_url },
+                        detail,
+                    } => FunctionCallOutputContentItem::InputImage {
+                        image: ImageReference::Inline { image_url },
+                        detail: detail.or(Some(DEFAULT_IMAGE_DETAIL)),
+                    },
                     codex_protocol::models::ContentItem::InputAudio { audio_url } => {
                         FunctionCallOutputContentItem::InputAudio { audio_url }
                     }
@@ -254,11 +256,10 @@ fn content_items_to_code_mode_result(items: &[FunctionCallOutputContentItem]) ->
                 FunctionCallOutputContentItem::InputText { text } if !text.trim().is_empty() => {
                     Some(text.clone())
                 }
-                FunctionCallOutputContentItem::InputImage { image_url, .. }
-                    if !image_url.trim().is_empty() =>
-                {
-                    Some(image_url.clone())
-                }
+                FunctionCallOutputContentItem::InputImage {
+                    image: ImageReference::Inline { image_url },
+                    ..
+                } if !image_url.trim().is_empty() => Some(image_url.clone()),
                 FunctionCallOutputContentItem::InputAudio { audio_url }
                     if !audio_url.trim().is_empty() =>
                 {
diff --git a/codex-rs/utils/output-truncation/src/lib.rs b/codex-rs/utils/output-truncation/src/lib.rs
index f4f502ab28..da59e7b86d 100644
--- a/codex-rs/utils/output-truncation/src/lib.rs
+++ b/codex-rs/utils/output-truncation/src/lib.rs
@@ -3,6 +3,7 @@
 use codex_protocol::models::FunctionCallOutputBody;
 use codex_protocol::models::FunctionCallOutputContentItem;
 use codex_protocol::models::FunctionCallOutputPayload;
+use codex_protocol::models::ImageReference;
 pub use codex_utils_string::approx_bytes_for_tokens;
 pub use codex_utils_string::approx_token_count;
 pub use codex_utils_string::approx_tokens_from_byte_count;
@@ -90,12 +91,15 @@ pub fn formatted_truncate_text_content_items_with_policy(
         text: formatted_truncate_text(&combined, policy),
     }];
     out.extend(items.iter().filter_map(|item| match item {
-        FunctionCallOutputContentItem::InputImage { image_url, detail } => {
-            Some(FunctionCallOutputContentItem::InputImage {
+        FunctionCallOutputContentItem::InputImage {
+            image: ImageReference::Inline { image_url },
+            detail,
+        } => Some(FunctionCallOutputContentItem::InputImage {
+            image: ImageReference::Inline {
                 image_url: image_url.clone(),
-                detail: *detail,
-            })
-        }
+            },
+            detail: *detail,
+        }),
         FunctionCallOutputContentItem::InputAudio { audio_url } => {
             Some(FunctionCallOutputContentItem::InputAudio {
                 audio_url: audio_url.clone(),
@@ -159,9 +163,14 @@ pub fn truncate_function_output_items_with_policy(
                     remaining_budget = 0;
                 }
             }
-            FunctionCallOutputContentItem::InputImage { image_url, detail } => {
+            FunctionCallOutputContentItem::InputImage {
+                image: ImageReference::Inline { image_url },
+                detail,
+            } => {
                 out.push(FunctionCallOutputContentItem::InputImage {
-                    image_url: image_url.clone(),
+                    image: ImageReference::Inline {
+                        image_url: image_url.clone(),
+                    },
                     detail: *detail,
                 });
             }
diff --git a/codex-rs/utils/output-truncation/src/truncate_tests.rs b/codex-rs/utils/output-truncation/src/truncate_tests.rs
index d58dace05d..bbcf617714 100644
--- a/codex-rs/utils/output-truncation/src/truncate_tests.rs
+++ b/codex-rs/utils/output-truncation/src/truncate_tests.rs
@@ -7,6 +7,7 @@ use crate::truncate_function_output_items_with_policy;
 use crate::truncate_text;
 use codex_protocol::models::DEFAULT_IMAGE_DETAIL;
 use codex_protocol::models::FunctionCallOutputContentItem;
+use codex_protocol::models::ImageReference;
 use pretty_assertions::assert_eq;
 
 #[test]
@@ -114,7 +115,9 @@ fn truncates_across_multiple_under_limit_texts_and_reports_omitted() {
         FunctionCallOutputContentItem::InputText { text: t1.clone() },
         FunctionCallOutputContentItem::InputText { text: t2.clone() },
         FunctionCallOutputContentItem::InputImage {
-            image_url: "img:mid".to_string(),
+            image: ImageReference::Inline {
+                image_url: "img:mid".to_string(),
+            },
             detail: Some(DEFAULT_IMAGE_DETAIL),
         },
         FunctionCallOutputContentItem::InputText { text: t3 },
@@ -142,7 +145,9 @@ fn truncates_across_multiple_under_limit_texts_and_reports_omitted() {
     assert_eq!(
         output[2],
         FunctionCallOutputContentItem::InputImage {
-            image_url: "img:mid".to_string(),
+            image: ImageReference::Inline {
+                image_url: "img:mid".to_string()
+            },
             detail: Some(DEFAULT_IMAGE_DETAIL),
         }
     );
@@ -183,7 +188,9 @@ fn truncate_function_output_items_with_policy_discards_empty_text() {
             text: "caption".to_string(),
         },
         FunctionCallOutputContentItem::InputImage {
-            image_url: "img:one".to_string(),
+            image: ImageReference::Inline {
+                image_url: "img:one".to_string(),
+            },
             detail: Some(DEFAULT_IMAGE_DETAIL),
         },
         FunctionCallOutputContentItem::InputAudio {
@@ -253,7 +260,9 @@ fn formatted_truncate_text_content_items_with_policy_merges_text_and_appends_med
             text: "abcd".to_string(),
         },
         FunctionCallOutputContentItem::InputImage {
-            image_url: "img:one".to_string(),
+            image: ImageReference::Inline {
+                image_url: "img:one".to_string(),
+            },
             detail: Some(DEFAULT_IMAGE_DETAIL),
         },
         FunctionCallOutputContentItem::InputText {
@@ -266,7 +275,9 @@ fn formatted_truncate_text_content_items_with_policy_merges_text_and_appends_med
             text: "ijkl".to_string(),
         },
         FunctionCallOutputContentItem::InputImage {
-            image_url: "img:two".to_string(),
+            image: ImageReference::Inline {
+                image_url: "img:two".to_string(),
+            },
             detail: Some(DEFAULT_IMAGE_DETAIL),
         },
     ];
@@ -281,14 +292,14 @@ fn formatted_truncate_text_content_items_with_policy_merges_text_and_appends_med
                 text: "Warning: truncated output (original token count: 4)\nTotal output lines: 3\n\nabcd…6 chars truncated…ijkl".to_string(),
             },
             FunctionCallOutputContentItem::InputImage {
-                image_url: "img:one".to_string(),
+                image: ImageReference::Inline { image_url: "img:one".to_string() },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
             FunctionCallOutputContentItem::InputAudio {
                 audio_url: "audio:one".to_string(),
             },
             FunctionCallOutputContentItem::InputImage {
-                image_url: "img:two".to_string(),
+                image: ImageReference::Inline { image_url: "img:two".to_string() },
                 detail: Some(DEFAULT_IMAGE_DETAIL),
             },
         ]
diff --git a/sdk/python/scripts/update_sdk_artifacts.py b/sdk/python/scripts/update_sdk_artifacts.py
index 68234bfc88..61d98e29e4 100755
--- a/sdk/python/scripts/update_sdk_artifacts.py
+++ b/sdk/python/scripts/update_sdk_artifacts.py
@@ -575,6 +575,7 @@ def generate_v2_all(schema_dir: Path) -> None:
             ],
             cwd=sdk_root(),
         )
+    _preserve_inline_image_class_names(out_path)
     _require_nullable_chatgpt_account_email(out_path)
     _preserve_reasoning_effort_enum(out_path)
     _preserve_thread_source_enum(out_path)
@@ -582,6 +583,23 @@ def generate_v2_all(schema_dir: Path) -> None:
     _normalize_generated_timestamps(out_path)
 
 
+def _preserve_inline_image_class_names(out_path: Path) -> None:
+    """Keep the public class names used before ImageReference was introduced."""
+    source = out_path.read_text()
+    stable_names = {
+        "ImageUrlContentItem": "InputImageContentItem",
+        "ImageUrlFunctionCallOutputContentItem": "InputImageFunctionCallOutputContentItem",
+    }
+    for generated_name, stable_name in stable_names.items():
+        if source.count(f"class {generated_name}(") != 1:
+            raise RuntimeError(f"Generated SDK is missing a unique {generated_name} class")
+        if re.search(rf"\b{re.escape(stable_name)}\b", source):
+            raise RuntimeError(f"Generated SDK already defines {stable_name}")
+        source = re.sub(rf"\b{re.escape(generated_name)}\b", stable_name, source)
+
+    out_path.write_text(source)
+
+
 def _require_nullable_chatgpt_account_email(out_path: Path) -> None:
     """Preserve required-but-nullable email semantics in the generated SDK model."""
     source = out_path.read_text()
diff --git a/sdk/python/src/openai_codex/generated/v2_all.py b/sdk/python/src/openai_codex/generated/v2_all.py
index 621298ce6c..80839540a3 100644
--- a/sdk/python/src/openai_codex/generated/v2_all.py
+++ b/sdk/python/src/openai_codex/generated/v2_all.py
@@ -7784,8 +7784,8 @@ class InputImageContentItem(BaseModel):
         populate_by_name=True,
     )
     detail: ImageDetail | None = None
-    image_url: str
     type: Annotated[Literal["input_image"], Field(title="InputImageContentItemType")]
+    image_url: str
 
 
 class ContentItem(
@@ -7796,9 +7796,7 @@ class ContentItem(
     model_config = ConfigDict(
         populate_by_name=True,
     )
-    root: (
-        InputTextContentItem | InputImageContentItem | InputAudioContentItem | OutputTextContentItem
-    )
+    root: InputTextContentItem | InputImageContentItem | InputAudioContentItem | OutputTextContentItem
 
 
 class ExperimentalFeature(BaseModel):
@@ -7980,10 +7978,10 @@ class InputImageFunctionCallOutputContentItem(BaseModel):
         populate_by_name=True,
     )
     detail: ImageDetail | None = None
-    image_url: str
     type: Annotated[
         Literal["input_image"], Field(title="InputImageFunctionCallOutputContentItemType")
     ]
+    image_url: str
 
 
 class FunctionCallOutputContentItem(
diff --git a/sdk/python/tests/test_artifact_workflow_and_binaries.py b/sdk/python/tests/test_artifact_workflow_and_binaries.py
index d9c6617ed7..7fae0b7a57 100644
--- a/sdk/python/tests/test_artifact_workflow_and_binaries.py
+++ b/sdk/python/tests/test_artifact_workflow_and_binaries.py
@@ -536,6 +536,20 @@ def test_generated_chatgpt_account_email_is_required_nullable() -> None:
         ChatgptAccount.model_validate({"planType": "pro", "type": "chatgpt"})
 
 
+def test_generated_inline_image_class_names_remain_stable() -> None:
+    """Keep the existing Python class names when image references expand."""
+    from openai_codex.generated.v2_all import (
+        InputImageContentItem,
+        InputImageFunctionCallOutputContentItem,
+    )
+
+    assert InputImageContentItem.__name__ == "InputImageContentItem"
+    assert (
+        InputImageFunctionCallOutputContentItem.__name__
+        == "InputImageFunctionCallOutputContentItem"
+    )
+
+
 def test_runtime_package_template_has_no_checked_in_binaries() -> None:
     runtime_root = ROOT.parent / "python-runtime" / "src" / "codex_cli_bin"
     assert sorted(