Update pipeline.py

pipeline.py

@@ -773,19 +773,20 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
 
         # save frames
         if output_path is not None:
-            output_batch_size = 10 #this prevents out of memory errors with large videos
-            num_frames = latents.size(2)  #latents' shape is [batch, channels, frames, height, width]
+            output_batch_size = 10  # prevents out of memory errors with large videos
+            num_frames = latents.size(2)  # latents' shape is [batch, channels, frames, height, width]
+            num_digits = output_path.count('#')  # count the number of '#' characters
+            frame_format = output_path.replace('#' * num_digits, '{:0' + str(num_digits) + 'd}')
+            
             for start_idx in range(0, num_frames, output_batch_size):
                 end_idx = min(start_idx + output_batch_size, num_frames)
                 video_tensor = self.decode_latents(latents[:, :, start_idx:end_idx, :, :])
                 video = tensor2vid(video_tensor, self.image_processor, output_type=output_type)
 
                 for batch_idx, frame_batch in enumerate(video):
-                    for frame in frame_batch[0][0]:
-                        digit_substring = ''.join(filter(str.isdigit, output_path))
-                        frame_number = int(digit_substring) + start_idx + batch_idx
-                        new_output_path = output_path.replace(digit_substring, str(frame_number).zfill(5), 1)
-                        frame.save(new_output_path)
+                    for frame_idx, frame in enumerate(frame_batch[0][0]):
+                        frame_number = start_idx + batch_idx * len(frame_batch[0][0]) + frame_idx
+                        frame.save(frame_format.format(frame_number))
             return output_path
             
         # Post-processing