Accept OpenAI multipart image edits

dev/test_router_support.py

@@ -24,6 +24,7 @@ from router_support import (  # noqa: E402
     load_profile_registry,
 )
 from ai_profile_router import (  # noqa: E402
+    Handler,
     STATE,
     _cap_chat_generation,
     _context_matches,
@@ -141,6 +142,45 @@ class ChatImageInputTests(unittest.TestCase):
         self.assertEqual(request, normalized)
 
 
+class ImageEditMultipartTests(unittest.TestCase):
+    def test_openai_image_array_upload_keeps_all_references(self) -> None:
+        boundary = "OpenClawImageBoundary"
+        parts = [
+            (
+                f"--{boundary}\r\n"
+                'Content-Disposition: form-data; name="model"\r\n\r\n'
+                "Qwen-Image-2.1-int8\r\n"
+            ).encode(),
+            (
+                f"--{boundary}\r\n"
+                'Content-Disposition: form-data; name="prompt"\r\n\r\n'
+                "Nur die Farbe ändern\r\n"
+            ).encode(),
+            (
+                f"--{boundary}\r\n"
+                'Content-Disposition: form-data; name="image[]"; filename="a.png"\r\n'
+                "Content-Type: image/png\r\n\r\n"
+            ).encode() + b"PNG-A\r\n",
+            (
+                f"--{boundary}\r\n"
+                'Content-Disposition: form-data; name="image[]"; filename="b.png"\r\n'
+                "Content-Type: image/png\r\n\r\n"
+            ).encode() + b"PNG-B\r\n",
+            f"--{boundary}--\r\n".encode(),
+        ]
+        handler = object.__new__(Handler)
+        files, fields = handler._parse_multipart_parts(
+            b"".join(parts), f"multipart/form-data; boundary={boundary}")
+
+        self.assertEqual(fields["model"], "Qwen-Image-2.1-int8")
+        self.assertEqual(fields["prompt"], "Nur die Farbe ändern")
+        self.assertEqual(
+            files,
+            [("image[]", "a.png", b"PNG-A"),
+             ("image[]", "b.png", b"PNG-B")],
+        )
+
+
 class LlamaCppReasoningTests(unittest.TestCase):
     def test_disabled_values_really_disable_thinking(self) -> None:
         for effort in (None, "none", "off", "disabled", False):

router/ai_profile_router.py

@@ -2223,7 +2223,25 @@ class Handler(BaseHTTPRequestHandler):
             self._image_request(data, [])
 
     def _image_edit(self) -> None:
-        """Edit with local image bytes supplied by the private Hermes plugin."""
+        """Edit with OpenAI multipart uploads or legacy JSON/base64 input."""
+        content_type = self.headers.get("Content-Type", "")
+        if "multipart/form-data" in content_type:
+            try:
+                body = self._read_body()
+                files, data = self._parse_multipart_parts(body, content_type)
+            except ValueError as exc:
+                self._send_error(400, str(exc),
+                                 "invalid_request_error", "invalid_multipart")
+                return
+            images = [payload for name, _filename, payload in files
+                      if name in {"image", "image[]"}]
+            if not images:
+                self._send_error(400, "Referenzbild fehlt",
+                                 "invalid_request_error", "missing_image")
+                return
+            self._image_edit_bytes(data, images)
+            return
+
         data = self._read_image_request()
         if data is None:
             return
@@ -2249,7 +2267,7 @@ class Handler(BaseHTTPRequestHandler):
                              "invalid_request_error", "too_many_images")
             return
 
-        source_files: list[str] = []
+        decoded: list[bytes] = []
         try:
             for item in encoded:
                 if item.startswith("data:"):
@@ -2260,6 +2278,25 @@ class Handler(BaseHTTPRequestHandler):
                     raw = base64.b64decode(item, validate=True)
                 except Exception as exc:
                     raise ValueError("ungültige Base64-Bilddaten") from exc
+                decoded.append(raw)
+            self._image_edit_bytes(data, decoded)
+        except ValueError as exc:
+            self._send_error(400, str(exc),
+                             "invalid_request_error", "invalid_image")
+
+    def _image_edit_bytes(self, data: dict, images: list[bytes]) -> None:
+        if len(images) > 4:
+            self._send_error(400, "höchstens vier Referenzbilder erlaubt",
+                             "invalid_request_error", "too_many_images")
+            return
+        if not images:
+            self._send_error(400, "Referenzbild fehlt",
+                             "invalid_request_error", "missing_image")
+            return
+
+        source_files: list[str] = []
+        try:
+            for raw in images:
                 if not raw or len(raw) > CHAT_IMAGE_MAX_BYTES:
                     raise ValueError(
                         f"Referenzbild muss 1..{CHAT_IMAGE_MAX_BYTES} Bytes groß sein")
@@ -2652,9 +2689,9 @@ class Handler(BaseHTTPRequestHandler):
 
     # ---------- STT (Spracherkennung) ----------
 
-    def _parse_multipart(self, data: bytes, content_type: str
+    def _parse_multipart_parts(self, data: bytes, content_type: str
-                         ) -> tuple[bytes, str, dict]:
+                               ) -> tuple[list[tuple[str, str, bytes]], dict]:
-        """Parst multipart/form-data. Liefert (file_data, filename, fields).
+        """Parst alle Datei- und Textteile eines multipart/form-data-Body.
 
         Nutzt email.parser.BytesParser (Standardbibliothek) für robustes
         MIME-Parsing. Handhabt quoted und unquoted Boundaries, beliebige
@@ -2667,9 +2704,8 @@ class Handler(BaseHTTPRequestHandler):
         if not msg.is_multipart():
             raise ValueError("Kein multipart/form-data")
 
-        file_data = b""
+        files: list[tuple[str, str, bytes]] = []
-        filename = ""
+        fields: dict[str, str] = {}
-        fields = {}
 
         for part in msg.get_payload():
             disposition = part.get("Content-Disposition", "")
@@ -2690,12 +2726,20 @@ class Handler(BaseHTTPRequestHandler):
 
             if part_filename is not None:
                 # Dateifeld (binär, nicht dekodieren)
-                file_data = payload
+                files.append((name, part_filename or "", payload))
-                filename = part_filename or ""
             else:
                 # Textfeld
                 fields[name] = payload.decode("utf-8", errors="replace")
 
+        return files, fields
+
+    def _parse_multipart(self, data: bytes, content_type: str
+                         ) -> tuple[bytes, str, dict]:
+        """Kompatibler Einzeldatei-Wrapper für den STT-Pfad."""
+        files, fields = self._parse_multipart_parts(data, content_type)
+        if not files:
+            return b"", "", fields
+        _name, filename, file_data = files[-1]
         return file_data, filename, fields
 
     def _transcribe(self) -> None: