Claude builds 5-minute landing pages from single prompts

• Across four days of posts, creators showed Claude turning single long prompts into polished landing pages, with markproduct's first demo, another 5-minute example, and a 10-minute build all framing the workflow as fast concept production.
• viktoroddy's Axion prompt and the longer WISA build spec make the pattern unusually repeatable: they spell out dependencies, color systems, layout structure, animation timing, and even z-index rules instead of vague art direction.
• The stack is already getting modular. viktoroddy's WISA post explicitly combines GPT Image 2, Seedance 2, and Claude, while MengTo's landing-page clip pushes the same idea one step further by turning still images into motion assets.
• The best prompts are not asking for "a nice website." According to the Axion spec, they define shader layers, hover easing, breakpoints, and mobile menu behavior, while the WISA spec goes down to scroll-scrub logic and component props.
• Exact visual fidelity is still the miss. In illscience's critique, the gap is no longer getting a page at all, it is forcing the model to replicate a mock without endless poking and prodding.

You can browse the prompt libraries at Motion Sites and Scene AI, then compare those repositories with the Axion prompt dump and the WISA thread to see how detailed the better results have become. The weird part is how much of the output now comes from production-style specs: the WISA build includes GSAP scroll triggers and a video.seeking guard, while MengTo's clip shows the workflow bleeding into launch-video generation instead of stopping at static page mocks.

The headline claim in this cluster is speed. markproduct's first demo, another landing-page video, the "$2200" post, and the "getting scary" clip all present Claude as a one-shot landing-page generator rather than a chatty coding assistant.

What changed is not just taste level. The posts show working pages with motion, responsive structure, and enough finish that the output reads like a usable concept deck in browser form.

The strongest examples read like creative direction merged with a front-end ticket.

The Axion prompt in viktoroddy's post specifies:

• framework and packages: React, TypeScript, Tailwind, lucide-react, shaders/react
• exact shader order: Swirl, ChromaFlow, FlutedGlass, FilmGrain
• concrete palette values, hover states, and shadow tokens
• mobile and desktop nav behavior
• button microinteractions, including duplicated text-scroll hover treatment
• z-index and pointer-event rules so the shader stays visible but clickable UI still works

The WISA prompt in the longer spec goes even further:

• exact package versions for React 19, Motion, GSAP, Tailwind v4, and Vite
• custom keyframes for arrow and nav-text transitions
• a reusable ScrollReveal component with GSAP ScrollTrigger
• a Reveal wrapper with motion timing values
• scroll-scrub video logic tied to a footer ref
• layout, typography, blur, and glassmorphism tokens for every section

That is the real workflow shift. The prompt is turning into a compact build document.

These demos are not winning on layout alone. They are winning on motion.

Across the examples, the repeated motion patterns are easy to spot:

• shader-backed hero backgrounds, especially in the Axion example
• hover loops for buttons and nav labels, spelled out in the Axion prompt and the WISA prompt
• full-page cinematic video backgrounds, central to the WISA demo
• scroll-linked reveals, where the WISA spec breaks the effect into GSAP opacity, blur, and rotation phases

That makes these posts more relevant to designers than the usual "AI made a website" clip. The visible progress is in interaction design and presentation polish.

One of the clearest reveals in viktoroddy's WISA demo is the tool chain itself: GPT Image 2 for source imagery, Seedance 2 for motion, and Claude for the front-end assembly. MengTo's post points at the same direction from another angle, using a single prompt to turn images into looping video for landing pages.

That breaks the workflow into three separable layers:

1. image generation
2. motion generation
3. code and layout assembly

The prompt libraries in Scene AI and Motion Sites matter because they package that stack into reusable recipes instead of one-off Twitter flexes.

The WISA example is the most useful evidence because the full prompt thread leaks implementation habits, not just output.

It includes several browser-facing tactics that are closer to engineering notes than inspiration:

• a video.seeking guard to avoid frame tearing during scroll scrubbing, per the WISA spec
• fixed video, fixed header, and scrollable content as three separate layers, again in the WISA spec
• pointer-events-none on non-interactive wrapper layers, echoed in both the WISA build and the Axion shader setup
• componentized animation wrappers, with ScrollReveal, Reveal, and animated nav items all defined in the WISA thread

That is why these examples feel more repeatable than earlier vibe-coded site clips. They are specifying interaction architecture, not only surface style.

The sharpest counterpoint came from illscience's critique, which argued that models are still maddeningly bad at reproducing a visual mock exactly. The problem in that post is not whether AI can output a landing page. The problem is the gap between a target design and the generated implementation.

That lands differently against the rest of the evidence. The showcase posts prove Claude can generate attractive motion-heavy concepts fast. illscience's critique shows the harder job still starts after the wow demo, when a team wants precise visual consistency instead of a plausible first pass.