Claude as a video editor

Recently I wanted to speed up a demo video 10 times. Having minimum experience with video editing but convinced that such a simple action should be possible with a built-in Windows editor, I mindlessly searched "windows video speed up" on DuckDuckGo and clicked the first credible-looking result. According to the tutorial, the speed-up could be done in the native Photos app on Windows. Great!

Following the tutorial, I

1. launched the Photos app
2. imported and selected the video
3. created a new video project – wait, how do I create a new video project?

Scrolling below in the tutorial, there was a screenshot with a red box (I just love those) directing me to simply click the “New” button after selecting a video. Easy. Unfortunately, for some mysterious reason I could not find the button in the app, and after a minute of clicking around I hopelessly thought: What do I do now? Google¹ “windows photos new video project”? Download some shady app the tutorial definitely recommends as an alternative option?

Finally, I thought: wait, I'm a programmer. I have zero interest in learning video editing, I just want to speed up a stupid demo video for my half-baked project. There's definitely a way to achieve this with python. And what better way to write a simple python script than instructing an LLM using a natural language?

I greeted Claude with a generic “Write a python script to speed up an mp4 video 10x.”, blindly copied the code, rewrote the input and output path, pip-installed a required library, runned the script, and voila, it worked liked charm! No need to navigate confusing Windows UI or install any apps I probably wouldn’t use. For reference, here’s the script:

        from moviepy.editor import VideoFileClip  

def speed_up_video(input_path, output_path, speed_factor=10):
    # Load the video clip
    video = VideoFileClip(input_path)
    
    # Speed up the video
    fast_video = video.speedx(speed_factor)
    
    # Write the result to a file
    fast_video.write_videofile(output_path)
    
    # Close the video clips
    video.close()
    fast_video.close()

# Example usage
input_video = "input_video.mp4"
output_video = "output_video_10x.mp4"
speed_up_video(input_video, output_video)

print(f"Video has been sped up 10x and saved as {output_video}")

    

This is just a simple example, but more and more in the recent two years have I found myself turning to LLMs as the first solution. True, usually for simple tasks such as this one, but with recent focus on agentic workflows, I think it’s safe to say the capabilities will be only more powerful. Specialized software tools will always be necessary for more fine-grained control of the output, but more and more often we’ll be able to get the result only by explaining it to an AI. Great time to be alive!