requirements

app.py

@@ -219,10 +219,7 @@ def compute_video_embedding(video_reader, min_len):
     video_frames = torch.cat(video_frames, dim=0)
     video_frames = rearrange(video_frames, "f h w c -> f c h w")
     torch.cuda.empty_cache()
-    # Assuming video_embedding is a PyTorch tensor
-    video_embedding_safe = encoded.detach().cpu().numpy().tolist()
-    video_frames_safe = video_frames.detach().cpu().numpy().tolist()
-    return video_embedding_safe, video_frames_safe
+    return encoded, video_frames
 
 
 @spaces.GPU(duration=120)
@@ -598,6 +595,8 @@ def sample(
     return complete_video
 
 
+@spaces.GPU(duration=600)
+@torch.no_grad()
 def process_video(video_input, audio_input, max_num_seconds):
     """Main processing function to generate synchronized video"""