Fix Gradio display compatibility while preserving PIL Images for processor

Store images as temp files for Gradio display (path field) while keeping PIL Image objects in _pil_image metadata for the processor. This fixes ValidationError from Gradio's Chatbot component which can't display PIL objects directly.

app.py

@@ -83,23 +83,32 @@ def user(user_message, image_data, history: list):
     stripped_message = user_message.strip()
     image_obj = _ensure_image_object(image_data)
 
+    # Store image as temp file for Gradio display, but keep PIL object in metadata
+    if image_obj is not None:
+        import tempfile
+        fd, tmp_path = tempfile.mkstemp(suffix=".png")
+        os.close(fd)
+        image_obj.save(tmp_path, format="PNG")
+    else:
+        tmp_path = None
+
     # If we have both text and image, combine them in a single message
-    if stripped_message and image_obj is not None:
+    if stripped_message and tmp_path is not None:
         updated_history.append({
             "role": "user",
             "content": [
                 {"type": "text", "text": stripped_message},
-                {"type": "image", "image": image_obj}
+                {"type": "image", "path": tmp_path, "_pil_image": image_obj}
             ]
         })
         has_content = True
     elif stripped_message:
         updated_history.append({"role": "user", "content": stripped_message})
         has_content = True
-    elif image_obj is not None:
+    elif tmp_path is not None:
         updated_history.append({
             "role": "user",
-            "content": [{"type": "image", "image": image_obj}]
+            "content": [{"type": "image", "path": tmp_path, "_pil_image": image_obj}]
         })
         has_content = True
 
@@ -221,17 +230,27 @@ def _prepare_processor_history(history: list[dict[str, Any]]) -> list[dict[str,
                     if item_type == "text":
                         formatted_content.append({"type": "text", "text": item.get("text", "")})
                     elif item_type == "image":
-                        # Include the PIL Image directly
-                        formatted_content.append({"type": "image", "image": item.get("image")})
+                        # Extract PIL Image from _pil_image field or load from path
+                        pil_image = item.get("_pil_image")
+                        if pil_image is None and "path" in item:
+                            from PIL import Image
+                            pil_image = Image.open(item["path"])
+                        if pil_image is not None:
+                            formatted_content.append({"type": "image", "image": pil_image})
             if formatted_content:
                 processor_history.append({"role": role, "content": formatted_content})
         elif isinstance(content, dict):
             # Legacy format or single image
-            if "image" in content:
-                processor_history.append({
-                    "role": role,
-                    "content": [{"type": "image", "image": content["image"]}]
-                })
+            if "image" in content or "_pil_image" in content:
+                pil_image = content.get("_pil_image") or content.get("image")
+                if pil_image is None and "path" in content:
+                    from PIL import Image
+                    pil_image = Image.open(content["path"])
+                if pil_image is not None:
+                    processor_history.append({
+                        "role": role,
+                        "content": [{"type": "image", "image": pil_image}]
+                    })
             else:
                 # Try to extract text
                 text = _content_to_text(content)