← all repositories
joeseesun/opus-video-prompts

A field guide to making videos with code instead of cameras

A curated archive of 18 copy-ready prompts and 54 real cases where Claude Opus 5.5 rendered entire videos by writing code, not pixels.

Collecting fresh signals — velocity needs a few days of history.
collecting data…
star history

What it does

This repo collects prompts people actually used to get Claude Opus 5.5 to produce videos — 18 full prompts and 54 public cases from the model’s first week, each with a link back to the original post. The trick is that Opus never outputs a video file: it writes code (HTML Canvas, p5.js, Three.js, Remotion, Manim, even Blender scripts) to draw every frame, then a headless browser screenshots frame-by-frame and FFmpeg stitches it into an MP4. Music comes from Web Audio or Python synthesis; narration hooks into TTS.

The interesting bit

The collection is honest about the ceiling: this approach shines at motion graphics, line-art and ink-wash animation, explainers, launch videos, and lyric MVs — and fails at photorealistic live-action. For realism, the documented pattern is to let Opus orchestrate while delegating actual footage to video models like Seedance or Runway. Also notable: some of the best results came from absurdly short prompts, including a startup launch video reportedly done in one minute for about two dollars.

Key highlights

  • 18 complete prompts, from one-liners to an 847-character cinematic brief for a Battle of Austerlitz film
  • Covers 9 technical routes: Canvas, p5.js, Three.js, Remotion, Manim, Blender, TTS narration, real-asset pipelines, and hybrid video-model workflows
  • Includes practical field wisdom — e.g. one-author-at-a-time revision (“slow it down,” “more energy”), and a .claude/settings.json deny rule so Claude can’t read your TTS API key
  • Two marathon prompts documented by outline only: a ~3000-character Japanese pixel-art runner and a ~2000-word MV pipeline mixing fal, Seedance 2.5, and frame-by-frame JS repainting
  • Bilingual throughout — Chinese and English prompts both represented

Caveats

  • It’s a prompt-and-casebook, not a tool: no code of its own to run, just curated text with source links
  • Two of the longest prompts are only summarized with links, not reproduced in full
  • Case quality is self-reported by authors on X, Bilibili, and forums; nothing here is independently verified

Verdict

Worth a browse if you want to try code-rendered video and want proven starting prompts instead of guessing. Skip it if you expected a library, a renderer, or anything you install — the repo’s entire value is other people’s homework, helpfully organized.

Frequently asked

What is joeseesun/opus-video-prompts?
A curated archive of 18 copy-ready prompts and 54 real cases where Claude Opus 5.5 rendered entire videos by writing code, not pixels.
Is opus-video-prompts open source?
Yes — joeseesun/opus-video-prompts is an open-source project tracked on heatdrop.
How popular is opus-video-prompts?
joeseesun/opus-video-prompts has 634 stars on GitHub.
Where can I find opus-video-prompts?
joeseesun/opus-video-prompts is on GitHub at https://github.com/joeseesun/opus-video-prompts.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.