What if you could type a single prompt and get back a cinematic 3D landing page where the camera flies through scene after scene with no cuts, driven entirely by the scroll of a mouse? That is exactly what Scroll World does, and it has accumulated 3,587 GitHub stars in the twelve days since its July 6 release. The open-source agent skill works with Claude Code, OpenAI Codex CLI, and any SKILL.md-compatible AI coding agent to generate complete scroll-scrubbed fly-through 3D worlds from just a brand name and a description of the scenes the camera should visit.
The repo, created by developer oso95, generates the kind of immersive interactive experience normally reserved for Apple product pages and high-budget agency work. As the user scrolls, the camera flies continuously from the outside of each scene into its interior, then flows to the next with no cuts. The effect is a single-shot cinematic experience that makes traditional landing pages feel static by comparison. And it is all generated by AI, from the 3D-style isometric diorama scenes to the video camera flights that stitch them together.
How Scroll World Works
Scroll World is built as a structured agent skill, not a one-shot script. When invoked, it follows a multi-stage pipeline that turns a brand brief into production-ready assets. The skill first interviews the user, gathering the subject, industry, pitch, brand kit, art direction, and the ordered list of scenes the camera will visit. It also asks about budget, showing estimated credit costs before any generation begins. This intake process ensures the results match the brand before a single image is rendered.
Once the brief is approved, Scroll World generates the visual assets through Higgsfield, an AI media platform that produces cohesive isometric diorama scenes. For each scene, the skill generates a still image, a dive-in camera clip, and connector clips that join consecutive scenes. The connectors are generated from the actual rendered frames of their neighbors, ensuring every seam is frame-identical. The result is a seamless flight through the entire world, with no jarring transitions.
For mobile users, Scroll World offers an optional parallel render chain in native 9:16 portrait aspect ratio. These are not cropped versions of the landscape film. They are independently composed for phones, with their own renders, camera flights, and frames. The scrub engine automatically serves the correct version based on the device.
The final output is a vanilla JavaScript scrub engine that is completely stack-agnostic. It works with plain HTML, Next.js, Vue, React, or a Python-served page. The engine handles blob-based video seeking, lazy loading, and seam crossfades automatically.
Installation and Setup
Installing Scroll World takes under a minute. For Claude Code users, the recommended method is the native plugin marketplace:
/plugin marketplace add oso95/scroll-world
/plugin install scroll-world@scroll-world
For Codex and other agents, the skills CLI provides a universal install path:
npx skills add oso95/scroll-world
Manual installation requires cloning the repo and copying the skill folder into the agent's skills directory. The skill depends on the Higgsfield CLI for AI art generation, ffmpeg for frame extraction and encoding, and Python with Pillow for mobile portrait canvas processing. All of these are standard open-source tools, and the skill's README provides clear setup instructions for each dependency.
The skill package is MIT-licensed and ships with a complete set of reference files: prompt templates for every scene type, copy-paste batch scripts for the generation pipeline, the portable scrub engine, a minimal standalone HTML template that mounts the engine, and a background knockout script for floating scene elements. The entire package is designed so that a developer can go from zero to a live scroll-through landing page in under an hour, assuming the generation pipeline runs without issues.
Why This Matters for AI Development
Scroll World is part of a larger pattern that is reshaping how web experiences are built. A year ago, generating a branded 3D scroll-driven landing page required a team of specialized Three.js developers, a creative director, and weeks of iteration. The production cost for a single campaign page could easily run between $50,000 and $200,000 when counting agency fees and production time. Scroll World collapses that timeline to minutes and that cost to whatever the Higgsfield credits and agent compute amount to.
The project demonstrates that AI agent skills are evolving from toy proof-of-concepts into genuine production tools. The skill does not just generate placeholder content. It produces frame-locked video sequences, manages aspect ratio variants for different devices, and outputs a scrub engine that works across any web framework. This is not vibe coding. It is structured creative production, automated end to end.
The rapid star growth, from zero to 3,587 in twelve days, signals strong demand for this capability. Developers and founders are eager to adopt AI-generated interactive experiences for their projects, but they need tools that produce real results, not just prototypes. Scroll World fills that gap by bridging the gap between AI generation and production deployment.
Who This Is For
Scroll World is built for founders, marketers, and developers who want premium interactive landing pages without the premium price tag. If you are launching a startup and need a memorable brand experience to differentiate from competitors, this skill can generate something that looks like it cost $100,000 in agency fees. If you are a developer building an open-source project and want a showcase page that does your work justice, Scroll World offers a path to a production-quality 3D world without hiring a designer or learning WebGL.
The skill is also relevant for agencies and creative studios experimenting with AI-generated content pipelines. The reference files and prompt templates are well-documented enough to serve as a foundation for custom branded experiences. And for anyone building in the agent skill ecosystem, Scroll World is a case study in how to structure a complex multi-stage AI generation task as a composable, reusable skill.
The one caveat is cost. Scroll World relies on Higgsfield credits for AI image and video generation, and the pricing is not transparent from the CLI. The skill estimates costs against your live balance before spending, but heavy use could add up. For a basic three-scene landing page, expect something in the range of a few dollars in credits. For larger productions with mobile variants and more scenes, the cost scales accordingly.