Fix demo dropdown state and ZeroGPU duration args

app.py

@@ -113,6 +113,7 @@ DEFAULT_VOICE_VALUE = (
 DEMO_ENTRIES = load_demo_entries()
 DEMO_ENTRY_MAP = {entry.demo_id: entry for entry in DEMO_ENTRIES}
 DEMO_AUDIO_PATH_MAP = {str(entry.prompt_audio_path): entry for entry in DEMO_ENTRIES}
+DEMO_ENTRY_NAME_MAP = {entry.name: entry for entry in DEMO_ENTRIES}
 VOICE_FILE_TO_NAME = {
     preset.prompt_audio_path.name: preset.name
     for preset in _VOICE_PRESETS.values()
@@ -146,6 +147,18 @@ def maybe_delete_file(path: str | Path | None) -> None:
         logging.warning("failed to delete temporary file: %s", path, exc_info=True)
 
 
+def normalize_demo_case_id(demo_case_id: str | None) -> str:
+    normalized = str(demo_case_id or "").strip()
+    if not normalized:
+        return ""
+    if normalized in DEMO_ENTRY_MAP:
+        return normalized
+    matched_entry = DEMO_ENTRY_NAME_MAP.get(normalized)
+    if matched_entry is not None:
+        return matched_entry.demo_id
+    return ""
+
+
 @functools.lru_cache(maxsize=2)
 def get_tts_service(runtime_has_cuda: bool) -> NanoTTSService:
     return NanoTTSService(
@@ -249,6 +262,7 @@ def apply_demo_case_selection(
     voice: str,
     prompt_audio_path: str | None,
 ):
+    demo_case_id = normalize_demo_case_id(demo_case_id)
     if not demo_case_id:
         preview_path, prompt_source = refresh_prompt_preview(voice, prompt_audio_path, "")
         return (
@@ -323,23 +337,14 @@ def build_status_text(
 
 
 def estimate_gpu_duration(
-    text: str,
-    voice: str,
-    prompt_audio_path: str | None,
-    max_new_frames: int,
-    voice_clone_max_text_tokens: int,
-    do_sample: bool,
-    text_temperature: float,
-    text_top_p: float,
-    text_top_k: int,
-    audio_temperature: float,
-    audio_top_p: float,
-    audio_top_k: int,
-    audio_repetition_penalty: float,
-    seed: float | int,
+    *args,
+    **kwargs,
 ) -> int:
-    del voice, prompt_audio_path, do_sample, text_temperature, text_top_p, text_top_k
-    del audio_temperature, audio_top_p, audio_top_k, audio_repetition_penalty, seed
+    text = kwargs.get("text", args[0] if len(args) > 0 else "")
+    max_new_frames = kwargs.get("max_new_frames", args[6] if len(args) > 6 else 375)
+    voice_clone_max_text_tokens = (
+        kwargs.get("voice_clone_max_text_tokens", args[7] if len(args) > 7 else 75)
+    )
     text_len = len(str(text or "").strip())
     estimated = 75 + (text_len // 12) + int(max_new_frames) // 8 + int(voice_clone_max_text_tokens) // 10
     return max(90, min(240, estimated))
@@ -454,6 +459,7 @@ def build_demo():
                     value="",
                     label="Default Case",
                     info="Select a built-in case to auto-fill the text and prompt preview.",
+                    allow_custom_value=True,
                 )
                 text = gr.Textbox(
                     label="Target Text",