8000 GitHub - watchsound/clip-master: create clip from video
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

watchsound/clip-master

Repository files navigation

intro

The project involves taking a full-length movie video as input and generating a concise 2-5 minute highlight reel or clip from it. The goal is to automatically identify and extract key scenes that capture the essence of the movie, providing a brief yet compelling summary of the content.

The process may involve analyzing the movie's audio, visual elements, and narrative structure to determine the most impactful moments. These selected scenes are then compiled into a short, cohesive video clip that represents the movie's storyline, major plot points, or most visually striking sequences.

This tool could be useful for creating promotional material, summarizing movies for viewers, or generating quick previews for content recommendations.

implementation

step1:

download video

step2:

if video does not contains srt file, download srt file

step3:

get abstraction

step4:

use chatgpt to get scenes from abstraction

step5:

decompose video into v-scenes

step6:

add object to v-scene images

step7:

from v-scenes get cluster dialogs

step7.5

use srt to adjust v-scenes (merge v-scenes if belong to one sentence in srt)

step8:

ask chatgpt to find best match cluster for each scene

step9:

add sentiment score

step9.5:

add distance between objects and scene.

step10:

alignment.

step11:

create video

About

create clip from video

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

0