Update app.py

app.py

@@ -77,9 +77,24 @@ net, feature_utils, seq_cfg = load_all_models()
 
 # --- INFERENCE FUNCTIONS ---
 
+def get_video_duration(video_path):
+    if video_path is None:
+        return 8
+    try:
+        import torchaudio
+        info = torchaudio.info(video_path)
+        duration = info.num_frames / info.sample_rate
+        return round(duration, 2)
+    except Exception:
+        return 8  # 
+
 @spaces.GPU()
 @torch.inference_mode()
-def video_to_audio(video, prompt, negative_prompt, seed, num_steps, cfg_strength, duration):
+def video_to_audio(video, prompt, negative_prompt, seed, num_steps, cfg_strength, duration=None):
+
+    if duration is None:
+        duration = get_video_duration(video)
+    
     rng = torch.Generator(device=device)
     rng.manual_seed(int(seed)) if seed >= 0 else rng.seed()
     fm = FlowMatching(min_sigma=0, inference_mode='euler', num_steps=int(num_steps))
@@ -103,7 +118,7 @@ def video_to_audio(video, prompt, negative_prompt, seed, num_steps, cfg_strength
     make_video(video_info, path, audio, sampling_rate=seq_cfg.sampling_rate)
     gc.collect()
     return path
-
+    
 @spaces.GPU()
 @torch.inference_mode()
 def image_to_audio(image, prompt, negative_prompt, seed, num_steps, cfg_strength, duration):