Scroll-world turns a plain landing page into a continuous fly-through video that you control by scrolling. Under the hood it stitches together AI-generated isometric diorama scenes and seamless transition clips. Each scene still is made with GPT Image 2 (via Higgsfield or Codex), and the camera moves come from Seedance on Monid by default, falling back to Higgsfield models that can frame-lock. You get a config-driven scrub engine in vanilla JS, ready for plain HTML, Next.js, Vue or Python pages. Installation works through Vercel’s skills CLI, plugin marketplace commands or manual cloning into your agent’s skills folder. You need Monid CLI (Seedance 2.0 backend), Higgsfield CLI (for stills and fallbacks), Pillow for mobile canvases, ffmpeg/ffprobe, plus Python 3. Optionally the Codex CLI can handle image generation under your ChatGPT subscription. Once set up, invoking `/scroll-world` or `$scroll-world` launches an interactive prompt: you specify your industry, brand kit, art direction, scene order, mobile portrait needs, and budget tiers. The skill shows estimated credit costs—Monid bills per clip (~$27 for a six-scene 1080p chain) and Higgsfield credits cover the rest. After you approve the quote, the system generates one still and one “dive-in” clip per scene, plus connector clips that lock frames at each seam for a perfectly continuous flight. If you ask for mobile, it renders a separate 9:16 chain with its own frame-lock process. Assets run through a background pipeline—generate, extract frames, build connectors, encode—and the JS scrub engine lazy-loads and crossfades at seams. Everything is config-driven and stack-agnostic. Once ready, you drop in the index-template.html, link the clips, and your scroll-scrubbed world springs to life.