[Bugfix] ComfyUI image-to-image DALL-E endpoint cases #2886 (#2980)

Signed-off-by: david6666666 <530634352@qq.com>

vllm_omni/entrypoints/openai/api_server.py

@@ -16,6 +16,7 @@ from argparse import Namespace
 from collections.abc import AsyncIterator
 from contextlib import asynccontextmanager
 from http import HTTPStatus
+from numbers import Integral
 from typing import Annotated, Any, Literal, cast
 
 import httpx
@@ -1826,10 +1827,24 @@ def _get_max_edit_input_images(raw_request: Request, engine_client: Any) -> int
         # config is not exposed on the serving surface.
         return None
 
-    if not bool(getattr(od_config, "supports_multimodal_inputs", False)):
+    supports_multimodal_inputs = getattr(od_config, "supports_multimodal_inputs", None)
+    if not isinstance(supports_multimodal_inputs, bool):
+        # Older serving surfaces and mocked engines may expose a placeholder
+        # object instead of a real diffusion config. Treat that as "unknown"
+        # so existing single-image flows keep working.
+        return None
+
+    if not supports_multimodal_inputs:
         return 1
 
-    return getattr(od_config, "max_multimodal_image_inputs", None)
+    max_input_images = getattr(od_config, "max_multimodal_image_inputs", None)
+    if max_input_images is None:
+        return None
+    if isinstance(max_input_images, bool) or not isinstance(max_input_images, Integral):
+        return None
+    if max_input_images < 1:
+        return None
+    return int(max_input_images)
 
 
 def _get_lora_from_json_str(lora_body):