Every prompt from every AI battle
Every build in these battles came from one plain-English prompt. Copy any of them, paste it into Claude or ChatGPT and you get the same kind of interactive piece in minutes.
What the reel showed
Each battle gives two AI models the exact same prompt and one try each: Claude (through Claude Code) against GPT (through Codex). The builds are recorded with the same mouse path so you can judge them side by side.
No code was written by hand. Everything you see on screen is what the model produced from the prompt on this page.
What this means for your business
A one-paragraph prompt can now produce a polished interactive visual. That is a landing-page hero, a product demo or a trade-show screen that used to cost a designer and a developer days.
The prompts follow one pattern: say what it is, describe how it should feel and move, then set the rules (one file, no libraries, smooth). Reuse that pattern for your own brand visuals.
What I’d use
Run the same prompt in Claude and ChatGPT and keep the one you like, exactly like the reels.
Change the subject and colours to your brand and you have a custom animated background for your website.
Try it yourself
Pick a battle, copy its prompt, and paste it into Claude or ChatGPT. New prompts are added here every week.
Dragon
Paste into Claude or ChatGPT, save the answer as a .html file and open it in your browser.
Build an interactive dragon that follows the mouse cursor. The dragon is drawn with thin light lines, a long segmented spine with ribs, a head and legs, and it should chase the cursor smoothly and look alive. One self-contained HTML file: a full-window canvas on a near-black background, no external libraries or assets. Make it breathtaking: elegant, flowing, alive, with soft glow and fine detail, smooth 60 fps. Reply with only the HTML file.
Phoenix
Paste into Claude or ChatGPT, save the answer as a .html file and open it in your browser.
Build an interactive phoenix that follows the mouse cursor. Fine glowing line art in fire colours (white-gold core to orange and deep red), like a living golden filigree sculpture. A long smooth segmented body and neck with spring physics, two large wings of many long curved feather lines that sweep and flap, a long flowing tail of ribbon-like feathers, a crest and a glowing eye, a gentle trail of embers. About a third of the screen wide. One self-contained HTML file: a full-window canvas on a near-black background, no external libraries or assets. Make it breathtaking: elegant, flowing, alive, with soft glow and fine detail, smooth 60 fps. Reply with only the HTML file.
Koi pond
Paste into Claude or ChatGPT, save the answer as a .html file and open it in your browser.
Build an interactive koi pond where a school of koi follows the mouse cursor, so realistic it looks like real 3D koi in a real pond, seen from directly above. Glossy koi in white, orange and black patterns; the fish must swim, not slide: bend each body along a flexible spine so it undulates, the tail sweeps and the fins flex, with smooth turns and varied speeds. Six to eight koi of different sizes, each with a soft shadow on the pond floor. Animated caustic light shimmering over stones, gentle refraction wobble, ripple rings where the cursor moves. The school gathers around the cursor. One self-contained HTML file: a full-window canvas on a near-black background, no external libraries or assets. Make it breathtaking: elegant, flowing, alive, with soft glow and fine detail, smooth 60 fps. Reply with only the HTML file. Tip: give it a photo of a pond seen from above and a photo of one koi, and tell it to use them as the background and the fish texture.
Jellyfish
Paste into Claude or ChatGPT, save the answer as a .html file and open it in your browser.
Build an interactive bioluminescent jellyfish that follows the mouse cursor through the deep sea. A translucent glowing bell that pulses to swim, with dozens of long trailing tentacles and frilly oral arms that flow with inertia and spring physics, lit in cyan, violet and pink with additive glow. Drifting plankton particles and faint god rays in the dark water. Large, about a third of the screen tall. One self-contained HTML file: a full-window canvas on a near-black background, no external libraries or assets. Make it breathtaking: elegant, flowing, alive, with soft glow and fine detail, smooth 60 fps. Reply with only the HTML file.
Growing city
Paste into Claude or ChatGPT, save the answer as a .html file and open it in your browser.
Build an interactive 3D city that grows wherever the mouse cursor goes, like a miniature tilt-shift diorama at golden hour turning to dusk, seen from a close isometric angle so every building is big and readable on a phone. Along the cursor's path the ground tiles in, roads lay themselves with moving cars, and buildings grow up with a springy build animation: glass skyscrapers with lit windows, brick apartment blocks, little houses, parks with round trees, a stadium, a bridge over a river. Warm sunlight, soft shadows, depth-of-field blur at the edges. The city keeps living: windows flicker on, cars drive, clouds drift. One self-contained HTML file: a full-window canvas on a near-black background, no external libraries or assets. Make it breathtaking: elegant, flowing, alive, with soft glow and fine detail, smooth 60 fps. Reply with only the HTML file.
Formula 1 race
Paste into Claude or ChatGPT, save the answer as a .html file and open it in your browser.
Write a single self-contained HTML file that simulates a Formula 1 race seen from directly overhead. It runs on load with no clicks: ten cars in real team colours race round a recognisable twisting circuit with kerbs, run-off areas, a pit lane and a start/finish line. They brake into corners and accelerate out, follow racing lines, slipstream and overtake, some pit and rejoin, and they never drive through each other. A live timing tower shows positions, gaps and the current lap. Design it for a 1000 x 1000 viewport, whole circuit visible. Colour, hierarchy and motion are part of the job. One attempt. Do not explain. Just build it.
3D fan art in action (from one image)
Paste into Claude or ChatGPT, save the answer as a .html file and open it in your browser.
Attach one reference image of the character you want, then paste this. Animate the character in the attached image as a real 3D model built with three.js (load it from a CDN via an import map): match its design, proportions, colours and materials as closely as you can, but build it from geometry and materials in code, do not show the image. Give it a 10-second action moment that starts on load and loops (for example: an armoured hero hovers on boot thrusters and fires a glowing beam from his palm; a dark knight ignites a red energy sword and swings it). Cinematic lighting with real shadows, rim light, glow, particles and a slowly moving camera; the character fills most of the frame; no text on screen. One self-contained index.html.
Links
Want a visual like this on your website?
I build interactive heroes, demos and brand visuals with AI, faster and cheaper than an agency.