Claude Opus 5.5 Is Writing Videos — 703 Prompts to Steal

Creators on X discovered that Claude Opus 5.5 will write an animation as a self-contained HTML file — Canvas, SVG, Three.js, whatever the scene needs — and you screen-record the browser tab. No generation credits. No watermark. One line changed in the prompt and the video updates. A GitHub repository called awesome-opus5-5-videos launched September 27, 2026, collected 703 of those prompts with links to the original creator posts, and on October 12 dropped 285 more. It now has 8.6k stars. This is not a product launch — it is a workflow that spread faster than most product launches.

What Happened

The awesome-opus5-5-videos repository (MIT licensed — the code and text in the repo, not the videos themselves) is a structured collection: a prompts/ folder and a data/videos.json file with 703 entries across four categories — motion graphics, explainers, 3D scenes, and interactive games. Each entry links back to the creator’s original X post. 418 of the entries have live side-by-side remakes on Skillry (skillry.dev), a platform that rebuilt each video from the shared prompt so you can compare the original to the remake. Skillry provides the viewing layer and benefits from the list’s traffic — the comparison coverage is comprehensive partly because Skillry built it to be. That does not make the prompts less useful, but it explains the economics.

The October 12 update added 285 prompts, mostly explainers and 3D scenes. The repo gained 1.9k stars in its first seven days and appeared on GitHub’s trending lists for nine consecutive days after launch.

The workflow is three steps. Paste a prompt from the list into Claude Opus 5.5 (the claude.ai app or Claude Code both work). Ask it to output a single self-contained HTML file with no external dependencies. Open the file in a browser and record it with QuickTime, OBS, or a frame-exact renderer like Remotion if you need clean 60fps output. A 30-second recipe explainer — one of the common entry types in the list — runs start to finish in under ten minutes.

Why This Matters for Creators

Text-to-video generators — Runway, Kling, Sora — solve a specific problem: generating footage-like imagery from a description. They are the right tool when you need something that looks photographed. They are the wrong tool when your video needs exact text, precise numbers, or brand colors you control, because diffusion models hallucinate all three.

Coded animation has none of those failure modes. The text is literal HTML. The numbers are literal numbers. The colors are hex codes you set. Revisions are prompts, not re-rolls. And the cost structure is flat: Claude Pro costs $20/month as of October 2026, with no per-second generation fees and no watermarks on export.

The 703-prompt library is the fastest path into this workflow. You do not need to invent prompting patterns from scratch. Copy a prompt that produces the output category you need, swap in your content, and iterate. Community reaction confirms this is the actual use: “Such a good resource for learning how to prompt better motion” was a top reaction on Skillry’s site within days of launch.

One sharp caveat: some entries in the list hold only partial prompts — the creator shared enough to show the technique but not enough to reproduce the exact video. Treat those as patterns to study, not recipes to run verbatim.

When to Use a Generator Instead

Coded animation is not a Runway replacement. It is an alternative for a specific slice of the creator workflow — the explanatory slice.

Use Runway or Kling when you need:

• Photoreal B-roll — landscapes, people moving, product shots with natural lighting
• Talking avatars — HeyGen or Synthesia own this category; Claude writes no faces
• Stylized cinematic motion — handheld camera feel, depth of field, film grain

The cleaner mental model: if your video shows something that exists in the physical world and needs to look real, use a generator. If your video explains something — how a process works, what the numbers mean, how a product flows — write it in code.

The Take

The awesome-opus5-5-videos list is the signal, not the story. The story is that a large category of “AI video” was never really video — it was animation, and animation has always been code. Creators who figured this out first are producing explainers faster, cheaper, and with more editorial control than anyone using a diffusion pipeline for the same job.

The library is a shortcut into that approach. Use the prompts as patterns — study the structure, copy the technique, write your own content. Do not repost the videos in the list; every one of them belongs to its original creator, not to the MIT license on the repo’s scaffolding.

For a 30-second stats explainer or a product walkthrough with branded colors, the coded approach beats a generator on every axis that matters to a creator: cost, editability, text accuracy, and turnaround time. For anything that needs to look like footage, Runway and Kling remain the right tools.

Start with the library. Pick a prompt in the explainer category. Run it. You will have a working motion graphic in the time it takes a generator to queue.