fix: align multimodal client override semantics

tests/test_manual_multimodal_test_client.py

@@ -39,6 +39,15 @@ def test_parse_args_leaves_mode_empty_for_interactive_fallback():
     assert args.mode is None
 
 
+def test_parse_args_rejects_generic_file_argument():
+    try:
+        parse_args(["llm", "--file", "sample.dat"])
+    except SystemExit as exc:
+        assert exc.code == 2
+    else:
+        raise AssertionError("parse_args accepted deprecated --file argument")
+
+
 def test_resolve_mode_config_uses_mode_defaults_when_overrides_absent(tmp_path):
     args = parse_args(["audio-generation", "--output-dir", str(tmp_path)])
 
@@ -91,3 +100,21 @@ def test_resolve_mode_config_prefers_explicit_model_prompt_url_and_response_form
     assert config["prompt"] == "Custom prompt"
     assert config["response_format"] == "b64_json"
     assert config["output_dir"] == tmp_path
+
+
+def test_resolve_mode_config_keeps_explicit_empty_string_overrides(tmp_path):
+    args = parse_args([
+        "video-generation",
+        "--model", "",
+        "--audio-file", "",
+        "--video-file", "",
+        "--response-format", "",
+        "--output-dir", str(tmp_path),
+    ])
+
+    config = resolve_mode_config(args, selected_mode="video-generation")
+
+    assert config["model"] == ""
+    assert config["audio_file"] == ""
+    assert config["video_file"] == ""
+    assert config["response_format"] == ""

tools/manual_multimodal_test_client.py

@@ -56,7 +56,6 @@ def build_parser() -> argparse.ArgumentParser:
     parser.add_argument("--prompt", default=None)
     parser.add_argument("--response-format", default=None)
     parser.add_argument("--output-dir", default="tmp/manual-client-output")
-    parser.add_argument("--file", default=None)
     parser.add_argument("--audio-file", default=None)
     parser.add_argument("--video-file", default=None)
     return parser
@@ -72,12 +71,12 @@ def resolve_mode_config(args: argparse.Namespace, selected_mode: str) -> dict:
         "mode": selected_mode,
         "url": args.url.rstrip("/"),
         "token": args.token,
-        "model": args.model or defaults.get("model"),
+        "model": args.model if args.model is not None else defaults.get("model"),
         "prompt": args.prompt if args.prompt is not None else defaults.get("prompt"),
         "output_dir": Path(args.output_dir),
-        "audio_file": args.audio_file or defaults.get("audio_file"),
-        "video_file": args.video_file or defaults.get("video_file"),
-        "response_format": args.response_format or defaults.get("response_format"),
+        "audio_file": args.audio_file if args.audio_file is not None else defaults.get("audio_file"),
+        "video_file": args.video_file if args.video_file is not None else defaults.get("video_file"),
+        "response_format": args.response_format if args.response_format is not None else defaults.get("response_format"),
     }