I was wondering if it's possible to generate fewer than 121 frames as output. When I tried setting num_video_frames=9 in the inference script, I got an error. Then I modified chunk_duration=9 in registry.py, which did produce a video, but it was still around 3 seconds long and the quality seemed lower compared to using chunk_duration=121.
I have a couple of questions:
- Is it possible for the model to generate a different number of output frames, particularly fewer frames at a lower FPS?
- Even when I set both
chunk_duration=9 and --num_video_frames=9, the output video is still ~3 seconds long. What determines the minimum duration or number of frames the model can generate?
I was wondering if it's possible to generate fewer than 121 frames as output. When I tried setting
num_video_frames=9in the inference script, I got an error. Then I modifiedchunk_duration=9in registry.py, which did produce a video, but it was still around 3 seconds long and the quality seemed lower compared to usingchunk_duration=121.I have a couple of questions:
chunk_duration=9and--num_video_frames=9, the output video is still ~3 seconds long. What determines the minimum duration or number of frames the model can generate?