Wink Pings

84-second animation with zero video-generation cost: Python frame-by-frame drawing, with Claude Code lending a hand

Yanchuan used Python to draw an 84-second music animation frame by frame, with 36 storyboard shots cut to the beat, and zero cost for video-generation models. Here, AI writes the storyboards, writes code, and checks frames, while humans still make the final decisions.

An 84-second music animation short, set to Zhao Benshan's "What Are You Panicking About?", delivers a message to office workers in high-rises: What's the panic? AI is here, so repetitive work can be handed off first.

The short's creator, Yanchuan, listed three approaches in the tweet.

First, no video-generation model was used. The starry sky, sea of clouds, staircase of light, and walking protagonist were all drawn frame by frame in Python, at a video-generation cost of 0 yuan.

Second, the visuals follow the music. First calculate the song's beat, then time the cut points of the 36 storyboard shots to the beat, place emotional turns on the downbeats, and add white flashes and shake effects.

Third, the division of labor is clear. Yanchuan defines requirements, reviews the footage, and makes final decisions; Claude Code writes storyboards, writes code, and checks frames. The workflow is to first produce an overview image to spot problems, then render the full piece after revisions.

In the comments, some joked that the money saved on Sora all went to render farms, and that the real winners are Python illustrators. Others said drawing everything with code is really pushing it. Yanchuan replied bluntly: it tests Claude; video-generation models are too expensive.

What's interesting about this short is its cost structure. The default AI video path is to use models to generate visuals, paying per generation or by duration. Here, generation is split apart: AI writes code and checks it, and the visuals are rendered by code. A video-generation cost of 0 does not mean the total cost is 0. Writing storyboards, debugging code, reviewing frames, and rendering all take time. The money is just spent in different places.

For AI practitioners with a development background, what's worth watching in this case is the workflow: humans set goals and acceptance criteria, models fill in code and checks, and rendering is left to repeatable scripts. For nontechnical readers, at least one thing becomes clear: when AI enters the office, it doesn't necessarily start by making decisions for people; it first takes over repetitive actions.

Original tweet link: https://x.com/yanchuanlab/status/2122671985725452560

发布时间: 2026-11-17 21:26