用 Opus 5.5 把放射学论文做成彩铅动画解说
作者把 2024 年放射学论文丢给 Claude Opus 5.5,用一段提示词生成 25 秒彩铅绘本风动画解说,画面与配乐全是代码,无素材库、无生图模型。
I don't know about you, but my feed is full of animations people are creating with @AnthropicAI's Claude Opus 5.5. So I tried one with my 2024 Radiology editorial, "Seeing the Unseen." (https://pubs.rsna.org/doi/full/10.1148/radiol.240935) Every colored-pencil stroke and every note of the music is code. No stock art. No image generator. It even replays my stress test from the paper. Change "absolute" washout to "relative," and the LLM's confident answer turns wrong. Claude built the initial storyboard and draft, and I made a few tweaks. Claude handled the rest. Few tweaks, but seeing what's wrong still takes a human. Earlier this year, I created an entire product video using code (https://app.presto.run/), but this is the first time I feel like these frontier models can generate an animated explainer video that I'd be pretty happy with. Try it yourself. Attach something you wrote or an article and paste this into Claude using Opus 5.5 (as of writing, I find Opus 5.5 does the best job at this): "Read the attached article and make a 25-second animated explainer (1080×1350, 24 fps) in a colored-pencil storybook style, drawn entirely in JavaScript on an HTML canvas, with a soundtrack synthesized using the Web Audio API. Pick 3 main ideas, give each a short caption taken from the text, and use one simple original character. Show me the storyboard first, then render it to MP4." #Radiology #GenerativeAI #ClaudeOpus