Corregido proveedor NTIA para conectar al Space correcto

app.py

@@ -853,33 +853,91 @@ def generate_video(prompt, model_name, num_frames=16, num_inference_steps=20):
         # Configuración específica por tipo de modelo
         if "zeroscope" in model_name.lower():
             # Zeroscope models
-            video_frames = pipeline(
+            result = pipeline(
                 prompt,
                 num_inference_steps=num_inference_steps,
                 num_frames=num_frames,
                 height=256,
                 width=256
-            ).frames
+            )
         elif "animatediff" in model_name.lower():
             # AnimateDiff models
-            video_frames = pipeline(
+            result = pipeline(
                 prompt,
                 num_inference_steps=num_inference_steps,
                 num_frames=num_frames
-            ).frames
+            )
         else:
             # Text-to-video models (default)
-            video_frames = pipeline(
+            result = pipeline(
                 prompt,
                 num_inference_steps=num_inference_steps,
                 num_frames=num_frames
-            ).frames
+            )
         
         print("Video generado exitosamente")
-        return video_frames
+        
+        # Manejar diferentes tipos de respuesta
+        if hasattr(result, 'frames'):
+            video_frames = result.frames
+        elif hasattr(result, 'videos'):
+            video_frames = result.videos
+        else:
+            video_frames = result
+        
+        # Convertir a formato compatible con Gradio
+        if isinstance(video_frames, list):
+            if len(video_frames) == 1:
+                return video_frames[0]
+            else:
+                return video_frames
+        else:
+            # Si es un tensor numpy, convertirlo a formato de video
+            if hasattr(video_frames, 'shape'):
+                import numpy as np
+                print(f"Forma del video: {video_frames.shape}")
+                
+                # Convertir a formato de video compatible con Gradio
+                if len(video_frames.shape) == 4:  # (frames, height, width, channels)
+                    # Convertir frames a formato de video
+                    frames_list = []
+                    for i in range(video_frames.shape[0]):
+                        frame = video_frames[i]
+                        # Asegurar que el frame esté en el rango correcto (0-255)
+                        if frame.dtype == np.float32 or frame.dtype == np.float16:
+                            frame = (frame * 255).astype(np.uint8)
+                        frames_list.append(frame)
+                    
+                    # Crear video a partir de frames
+                    import imageio
+                    import tempfile
+                    import os
+                    
+                    # Crear archivo temporal
+                    with tempfile.NamedTemporaryFile(suffix='.mp4', delete=False) as tmp_file:
+                        temp_path = tmp_file.name
+                    
+                    # Guardar frames como video
+                    imageio.mimsave(temp_path, frames_list, fps=8)
+                    
+                    print(f"Video guardado en: {temp_path}")
+                    return temp_path
+                    
+                elif len(video_frames.shape) == 5:  # (batch, frames, height, width, channels)
+                    # Tomar el primer batch
+                    frames = video_frames[0]
+                    return generate_video(prompt, model_name, num_frames, num_inference_steps)
+                else:
+                    print(f"Forma no reconocida: {video_frames.shape}")
+                    return None
+            else:
+                return video_frames
         
     except Exception as e:
         print(f"Error generando video: {str(e)}")
+        print(f"Tipo de error: {type(e).__name__}")
+        import traceback
+        traceback.print_exc()
         return f"Error generando video: {str(e)}"
 
 # @spaces.GPU #[uncomment to use ZeroGPU]
@@ -948,21 +1006,52 @@ def generate_video_with_info(prompt, model_name, optimization_level="balanced",
             else:
                 return video_frames
         else:
-            # Si es un tensor numpy, convertirlo a lista
+            # Si es un tensor numpy, convertirlo a formato de video
             if hasattr(video_frames, 'shape'):
-                # Es un tensor, convertirlo a formato compatible
                 import numpy as np
-                if len(video_frames.shape) == 5:  # (batch, frames, height, width, channels)
+                print(f"Forma del video: {video_frames.shape}")
+                
+                # Convertir a formato de video compatible con Gradio
+                if len(video_frames.shape) == 4:  # (frames, height, width, channels)
+                    # Convertir frames a formato de video
+                    frames_list = []
+                    for i in range(video_frames.shape[0]):
+                        frame = video_frames[i]
+                        # Asegurar que el frame esté en el rango correcto (0-255)
+                        if frame.dtype == np.float32 or frame.dtype == np.float16:
+                            frame = (frame * 255).astype(np.uint8)
+                        frames_list.append(frame)
+                    
+                    # Crear video a partir de frames
+                    import imageio
+                    import tempfile
+                    import os
+                    
+                    # Crear archivo temporal
+                    with tempfile.NamedTemporaryFile(suffix='.mp4', delete=False) as tmp_file:
+                        temp_path = tmp_file.name
+                    
+                    # Guardar frames como video
+                    imageio.mimsave(temp_path, frames_list, fps=8)
+                    
+                    print(f"Video guardado en: {temp_path}")
+                    return temp_path
+                    
+                elif len(video_frames.shape) == 5:  # (batch, frames, height, width, channels)
                     # Tomar el primer batch
                     frames = video_frames[0]
-                    return frames
+                    return generate_video_with_info(prompt, model_name, optimization_level, input_image)
                 else:
-                    return video_frames
+                    print(f"Forma no reconocida: {video_frames.shape}")
+                    return None
             else:
                 return video_frames
         
     except Exception as e:
         print(f"Error generando video: {str(e)}")
+        print(f"Tipo de error: {type(e).__name__}")
+        import traceback
+        traceback.print_exc()
         return f"Error generando video: {str(e)}"
 
 def chat_with_model(message, history, model_name):