[Feature] Automatically slice the video.

[timerring]

Is your feature request related to a problem? Please describe.

In the era of short videos, only a small number of people have the time and patience to watch videos in their entirety.

Therefore, a new function could be developed to extract the most engaging video segments, facilitating the dissemination of the content.

Describe the solution you'd like

Slice the videos based on the density of danmakus and summarize the content with large multimodal models like GLM-4V-PLUS or Gemini.

[timerring]

Cannot insert the metadata into mp4 files in ffmpeg, details in stackoverflow issue. Hence I will try another workflow logic to avoid introducing more packages.