475 Viral AI Video Prompts, Free to Steal (Made with Claude — Works with Any AI)

What this collection is

In late September 2026, developers started posting a new kind of viral video: not AI-generated video in the text-to-video sense, but animations that Claude Opus 5.5 wrote as code — a single HTML file with Canvas, SVG, or Three.js doing all the motion. A curator gathered the best of them into awesome-opus5-5-videos: 475 videos and counting, each entry linking the creator's original post, a watchable remake, and the prompt that produced it.

The collection is MIT-licensed and updated regularly, and every entry is organized by category: motion graphics, explainers, 3D scenes, and games & interactive pieces. Think of it less as a gallery and more as a prompt library — 475 field-tested starting points for making your own animated videos with an AI agent.

Why these prompts work with any AI

Here's the useful part: almost none of these prompts are Claude-specific. They don't call model APIs or use proprietary features. They describe an animation — the look, the pacing, the elements on screen — and ask the model to write it as a self-contained web page. Any frontier model that writes good front-end code can take the same prompt and produce something in the same spirit.

That means the whole library doubles as a prompt collection for Gemini, Grok, Meta's Muse, or ChatGPT. Paste one in, get a working animation file back, then iterate: tighten the timing, swap the palette, change the aspect ratio to 9:16 for reels. The original prompt gets you 80% of the way; your revisions supply the taste.

Expect variation, not replication. A different model will interpret the same prompt differently — different easing, different composition. Treat each prompt as a creative seed rather than a deterministic recipe, and you'll get more out of the collection than anyone chasing pixel-perfect remakes.

What's inside: the four categories

Motion graphics (288 entries) is the bulk of the collection — logo animations, kinetic typography, abstract loops, the kind of 10-second pieces that dominate reels and shorts. If you make content, start here; the prompts are short and the results are immediate.

Explainers (62 entries) are the most practical category: product walkthroughs, recipe animations, concept visualizations. These prompts tend to be longer and more structured — they brief the model like a motion designer, with beats, timing, and visual language spelled out.

3D scenes (55 entries) lean on Three.js — rotating product shots, landscapes, particle systems. They demand more from the model, so they're the best stress test when you're comparing which AI writes better 3D code.

Games & interactive (70 entries) are playable: tiny arcade games, interactive toys, physics demos. Beyond fun, they're proof that a single well-written prompt can produce a complete interactive experience — useful inspiration if you're prototyping.

Five prompts worth stealing

You don't need to browse all 475. These five show the range of what's possible — each quote is a fragment; the full prompt lives at the link.

An infinite-zoom collage: "Build a looping 'infinite zoom' animation, After Effects style: the camera travels from landscape to landscape by flying through vintage objects." (koldo2k's piece) — a good template for any seamless-loop background.

A recipe explainer: "render a recipe motion graphic animation using javascript or html to show the full recipe from start to finish (empty glass to finished drink)" (Ror_Fly's piece) — the structure works for any step-by-step process, not just cocktails.

A startup explainer: "make a modern slick and punchy video for a modern startup that works on inference" (deedydas's piece) — short brief, big result; a reminder that confident art direction beats long specs.

A logo reel: "Make this logo into a reel. Pretend you are a world-class motion designer. Make a 10-second dynamic video." (0xValure's piece) — the persona framing ("pretend you are a world-class motion designer") is a technique worth copying into your own prompts.

A product tour: a prompt that asks the model first for the product URL, 8–12 UI states, brand colors, and required formats (1:1, 16:9, 9:16) before generating anything (verbove's piece) — the rare prompt that interviews you before it works, and the output shows it.

How to actually use one

Pick a video whose style matches what you want — don't start from the prompt, start from the result. Open its prompt file, copy it, and paste it into your agent of choice with one addition: tell it what format you need. "Single HTML file, 9:16, 15 seconds" turns a cool demo into a shippable reel.

Then iterate in small loops. The first output is a draft. Ask for specific changes — "slow the zoom by 30%", "switch the palette to warm neutrals", "hold the final frame for two seconds" — rather than rewriting the whole brief. Animation prompts respond well to directional tweaks because the model can see the structure it already built.

If the model you use can't run or preview the result, ask it to keep everything dependency-light: vanilla Canvas or SVG with no build step, so you can open the file in a browser and screen-record it. Most of the viral pieces were built exactly this way.

Give credit where it's due

Every prompt in the collection came from a real creator who posted their work publicly — the repo links each entry to the original post. If a prompt becomes the backbone of something you publish, link back to the creator and the collection. It's MIT-licensed, which is permissive, but attribution is what keeps collections like this growing.

And if you make something good with one of these prompts, consider sharing it back. The collection is updated regularly, and the curator takes additions — your variation could be someone else's starting point.