Video Predictor issue in playground vs real code

I have a cricket video and my task is to detect and mask the white cricket ball that is seen in the video. I have uploaded the video to see how the results are coming

In the playground in efficiently detects and mask the cricket ball and thats why i am trying the video predictor with prompt.

However, for my case, when i use the code:
from sam3.model_builder import build_sam3_video_predictor

video_predictor = build_sam3_video_predictor()
video_path = "" # a JPEG folder or an MP4 video file
# Start a session
response = video_predictor.handle_request(
request=dict(
type="start_session",
resource_path=video_path,
)
)
response = video_predictor.handle_request(
request=dict(
type="add_prompt",
session_id=response["session_id"],
frame_index=0, # Arbitrary frame index
text="",
)
)
output = response["outputs"]

It does not give me any result? Why is that?

Start by reproducing the shown build_sam3_video_predictor, start_session, and add_prompt calls with the referenced video, then inspect the returned response and outputs. Compare this behavior with the playground result; done means identifying the cause of the discrepancy or documenting the missing configuration or error.

Written by the indexing model from the issue text.