GitHub Copilot CLI: More Than Just Code - Part 2

In Part 1, I used GitHub Copilot CLI to resize a banner image through four iterative attempts. That was fun, but it was just the beginning. This time, I needed to work with video.

#The Setup

I was wrapping up my WordPress-to-Astro migration post. I wanted a demo video showing GitHub Copilot CLI in action. I had a ZoomIt screen recording of the Copilot startup animation. That beautiful ASCII art banner when you launch it. The recording was raw. 18MB. Full screen. A few seconds of nothing at the start.

I also needed a cover photo for the blog post. Something that showed the transformation from old to new. WordPress on one side, the shiny new Astro site on the other.

Both tasks screamed “open a video editor” and “open Photoshop.” I will not lie, part of me almost did. Instead, I opened my terminal.

#Task 1: Taming an 18MB Recording

My first question to Copilot was simple:

Copilot did not even hesitate. It installed ffmpeg via winget install Gyan.FFmpeg. Then generated a compression command:

ffmpeg -i copilot-start.mp4 \
  -c:v libx264 -crf 18 -preset slow \
  -vf "fps=24" -an \
  -movflags +faststart \
  copilot-start-compressed.mp4

18MB down to 369KB. Near-lossless quality. Copilot knew exactly why these settings made sense and chose them without being told.

#Speeding Things Up

The recording was good, but at normal speed it felt slow. I asked Copilot to speed it up 3x. It added a video filter:

-vf "setpts=PTS/3,fps=24"

The setpts=PTS/3 filter divides each frame’s timestamp by 3, making playback three times faster. Combined with the fps filter, Copilot kept the output smooth and the file size tiny. 7.7 seconds, 158KB. That’s smaller than most images on my site.

#Precision Trimming

There was still a problem: the first second or so was just a blank terminal. I told Copilot to trim 1 second from the start.

-ss 1

Then I watched it and asked for another half second off. Copilot adjusted to -ss 1.5 without missing a beat. Each trim was instant. No scrubbing through a timeline, no dragging handles. Just a conversation.

#Extracting the Banner Clip

Later, I wanted just the Copilot ASCII art banner from a different, longer recording. I told Copilot: “I only want the banner part. It starts around 3 seconds and is visible until about 5.5 seconds.”

Copilot trimmed the 7-second video to 1.9 seconds. Cropped the window title bar. Optimized it to 37KB:

ffmpeg -i github-copilot-banner.mp4 \
  -ss 3.8 -t 1.9 \
  -vf "crop=1444:862:0:36" \
  -c:v libx264 -crf 23 -preset slow -an \
  copilot-cli-banner.mp4

The crop filter (crop=width:height:x:y) removed exactly the top 36 pixels where the tab bar sat. Copilot figured out the dimensions from the source video.

#Task 2: Creating a Cover Photo

Now for the cover photo. I wanted a side-by-side comparison: the old WordPress site on the left, the new Astro site on the right. The kind of “before and after” that tells the story at a glance.

Copilot reached for Playwright to capture screenshots of both sites:

from playwright.sync_api import sync_playwright

with sync_playwright() as p:
    browser = p.chromium.launch()

    # Old WordPress site
    page = browser.new_page(viewport={"width": 1400, "height": 900})
    page.goto("https://azureviking.com")
    page.screenshot(path="wordpress.png")

    # New Astro site (dark mode)
    page = browser.new_page(
        viewport={"width": 1400, "height": 900},
        color_scheme="dark"
    )
    page.goto("http://localhost:4321")
    page.screenshot(path="astro.png")

Notice how it captured the new site in dark mode using color_scheme="dark". I did not even ask for that. Copilot just knew the new site had a dark theme and made it look its best.

#The Diagonal Stripe

I wanted a clean diagonal stripe separating the two screenshots. Not a boring vertical split. Something with a bit of style. This led to an iterative process that reminded me of the banner resizing from Part 1.

Attempt 1: Copilot added a diagonal with arrow icons. Too busy. “Drop the arrows, just the stripe.”

Attempt 2: Clean diagonal stripe, but the transition was soft and blurry. “Make it sharper.”

Attempt 3: Sharp diagonal with the WordPress side desaturated and dimmed to create contrast:

from PIL import Image, ImageEnhance

# Desaturate the old site
old_img = ImageEnhance.Color(old_img).enhance(0.3)
old_img = ImageEnhance.Brightness(old_img).enhance(0.7)

The old WordPress side looked faded and dated. The new Astro site popped with vibrant dark mode colors. The diagonal stripe cut between them cleanly.

#Centering the Content

But the cover photo looked off-center. The visual content was shifted to the left. I mentioned this, and Copilot did something clever:

It used numpy to analyze pixel brightness across columns. Found the actual visual center of the content.

import numpy as np

arr = np.array(img)
col_brightness = arr.mean(axis=(0, 2))
content_cols = np.where(col_brightness > threshold)[0]
visual_center = (content_cols[0] + content_cols[-1]) // 2

The content spanned from pixel 13 to 794 in a 1444-pixel frame. Mathematical center was at 722, but the visual center was at 403. Copilot calculated the right padding to center it perfectly.

#What Stood Out

ffmpeg is powerful but cryptic. I know ffmpeg exists. I have used it before. But I never remember the filter syntax, the codec flags, or the crop coordinates. Copilot handled all of that. I just said what I wanted, and it translated that into the right command.

The iteration loop is fast. Each adjustment (speed, trim, crop) was a single prompt and instant result. In a GUI video editor, I would have been scrubbing timelines and waiting for renders. Here, each attempt took seconds.

Mixing tools is natural. In one session, Copilot used ffmpeg for video, Playwright for screenshots, Pillow for image composition, and numpy for pixel analysis. It picked the right tool for each sub-task without me needing to think about it.

Creative work is just problem-solving. Writing Terraform modules or composing a cover photo? Same process. Describe. Evaluate. Refine. Copilot fits into that loop perfectly.

Huge thanks to the ffmpeg, Playwright, and Pillow communities. These open-source tools are incredible, and Copilot made them feel like they were built into my terminal.

#Your Turn

Got a creative task that feels like it needs a GUI? I really encourage you to try your terminal first. The fastest path from “I need this” to “done” might just be a Copilot conversation.

Part 3 is brewing. Next time, I’m bringing Copilot into my day job: Azure architecture, infrastructure code I didn’t write, and real-world troubleshooting. I cannot wait to share that one. Stay tuned.

Missed Part 1? Check out how I used Copilot CLI to resize a banner image through four iterative attempts.

Got your own Copilot CLI stories? I would love to hear them. Find me on LinkedIn or GitHub. 🛠️