Update

app.py

@@ -34,7 +34,7 @@ pose_image_processor = AutoProcessor.from_pretrained(pose_model_name)
 pose_model = VitPoseForPoseEstimation.from_pretrained(pose_model_name, device_map=device)
 
 
-@spaces.GPU
+@spaces.GPU(duration=5)
 @torch.inference_mode()
 def process_image(image: PIL.Image.Image) -> tuple[PIL.Image.Image, list[dict]]:
     inputs = person_image_processor(images=image, return_tensors="pt").to(device)
@@ -107,6 +107,7 @@ def process_image(image: PIL.Image.Image) -> tuple[PIL.Image.Image, list[dict]]:
     return vertex_annotator.annotate(scene=annotated_frame, key_points=keypoints), human_readable_results
 
 
+@spaces.GPU(duration=60)
 def process_video(
     video_path: str,
     progress: gr.Progress = gr.Progress(track_tqdm=True),  # noqa: ARG001, B008
@@ -133,9 +134,6 @@ def process_video(
     return out_file.name
 
 
-process_video.zerogpu = True  # type: ignore
-
-
 with gr.Blocks(css_paths="style.css") as demo:
     gr.Markdown(DESCRIPTION)