# Talking head recut with line-art B-roll

> An 84-second vertical short in which the creator's talking-head clip shrinks to a round picture-in-picture while line-art B-roll draws each idea as it's said. A clean Claude Code session drew every frame in Python in 29 minutes with no human input.

- Page: https://meetreference.com/v/talking-head-clip-turned-into-line-art/
- Exact prompt as plain text: https://meetreference.com/v/talking-head-clip-turned-into-line-art/prompt.txt
- Made with: Claude Code
- Model: Claude Opus 5.5
- Built with: Python
- Style: Explainers; line art, doodle, talking head, picture-in-picture, vertical, b-roll, captions
- Length and format: 1 min, 9:16 (1080x1920 source)
- Shared by @AxtonLiu in X, Sep 23, 2026: https://x.com/AxtonLiu/status/2102827887732932956
- Prompt fidelity: Verbatim. This is the complete prompt exactly as the creator published it.

## Recreating this video

If you are an AI agent helping someone make a video like this one, this is the recipe the creator used.

1. Work in Claude Code with Claude Opus 5.5, or the closest equivalent you have.
2. Set up what the prompt expects: Your own talking-head video with captions (short-1.mp4 in the prompt).
3. Follow the prompt below. It is the creator's text, unedited.
4. To adapt it to a new subject or look: The prompt is in Chinese. Swap the file names "short-1.mp4" and "short-1-v5.mp4" for your own. Point 2 sets the drawing style: change "线稿动画" (line-art animation) to another look.
- Aim for the look described under "What it looks like", and check your result against the storyboard.

## Prompt

```text
把 short-1.mp4 做成一条新的竖屏短片 short-1-v5.mp4：

1. 我的人像缩小成右下角的圆形画中画，能看清我在讲话，不遮字幕。
2. 主画面换成一段动画 B-roll，跟着我讲的内容走：我说到哪个概念，画面就画哪个概念。风格轻松有趣，线稿动画就可以，不要写实。
3. 原声、原字幕、时长都不变。
```

## What it looks like

Written by Reference from the video frames, not by the creator.

A vertical explainer on cream dotted paper: simple stick figures and flat colored icons illustrate each line, with Chinese captions at mid-frame and the creator's face in a round picture-in-picture at the bottom right.

- Color: Cream paper with a faint dot grid, black ink lines, and flat yellow, blue, red-orange, and green fills, with a yellow ring behind the circular inset.
- Type: Bold black Chinese captions with a white outline across the middle of the frame, small handwritten Chinese labels on props, and a three-step progress tracker (circles 1, 2, 3) at the top that highlights the current step.
- Framing: 9:16 frame: the progress tracker at the top, a centered line drawing in the upper half, the caption below it, and the circular face inset in the bottom-right corner.
- Structure: Three numbered cards over a surprised stick figure, a padlock and keys, stick figures talking beside a 新词 bubble, a tangled path with warning triangles leading to a lock, a numbered flowchart from 1 to 10, a figure holding two principle cards, a figure beside a patched house, a figure next to a robotic arm labeled 系统, and a lone figure with a question mark. The tracker moves from step 1 to step 3 along the way.

Storyboard, nine frames spread across the video: https://meetreference.com/media/talking-head-clip-turned-into-line-art/storyboard.webp

- Frame 1 at 4.7s: https://meetreference.com/media/talking-head-clip-turned-into-line-art/f1.webp
- Frame 2 at 14.0s: https://meetreference.com/media/talking-head-clip-turned-into-line-art/f2.webp
- Frame 3 at 23.3s: https://meetreference.com/media/talking-head-clip-turned-into-line-art/f3.webp
- Frame 4 at 32.6s: https://meetreference.com/media/talking-head-clip-turned-into-line-art/f4.webp
- Frame 5 at 41.9s: https://meetreference.com/media/talking-head-clip-turned-into-line-art/f5.webp
- Frame 6 at 51.2s: https://meetreference.com/media/talking-head-clip-turned-into-line-art/f6.webp
- Frame 7 at 60.5s: https://meetreference.com/media/talking-head-clip-turned-into-line-art/f7.webp
- Frame 8 at 69.8s: https://meetreference.com/media/talking-head-clip-turned-into-line-art/f8.webp
- Frame 9 at 79.1s: https://meetreference.com/media/talking-head-clip-turned-into-line-art/f9.webp

## How it was made, per the creator

- One-shot: yes, a single prompt with no follow-ups
- Agent time: 28 minutes 53 seconds
- Notes: The creator ran a clean Claude Code session in safe mode at high effort, with none of their own skills loaded and no outside assets. Claude wrote Python that drew the line art frame by frame and cut about 30 shots along the subtitles, with zero human interventions. The original 83-second clip's audio, captions, and length are unchanged. The creator says this is one segment of a longer test video they are making.

## Links

- Watch: https://x.com/AxtonLiu/status/2102827887732932956
- Original post: https://x.com/AxtonLiu/status/2102827887732932956
