Scroll-world is an agent skill that generates immersive scroll-scrubbed 3D world landing pages where a camera flies continuously through generated scenes as the user scrolls.
The tool solves the problem of creating engaging, cinematic landing pages by automating the entire pipeline: it generates cohesive isometric diorama scenes using AI image generation, then creates camera flight videos between those scenes using video synthesis models, and finally synchronizes the playback to scroll position using a vanilla JavaScript scrubbing engine. The approach chains together multiple AI services—image generation via GPT Image 2 or Codex, and video generation via Seedance or Kling—to produce seamless transitions where the camera moves continuously from outside each scene into its interior and onward to the next, with no cuts.
Adoption suits teams building branded landing pages who want distinctive visual presentation without manual 3D modeling or animation work. The tool is framework-agnostic, dropping into plain HTML, Next.js, Vue, or Python-served pages. It requires external dependencies: the Monid CLI for video generation (pay-per-clip billing), the Higgsfield CLI for scene rendering and fallback video generation (credit-based), ffmpeg for frame handling, Python 3 with Pillow, and optionally the Codex CLI to bill image generation to a ChatGPT subscription instead of Higgsfield credits. The default video backend uses Monid with frame-locking to ensure seamless transitions, with Higgsfield as a fallback when Monid is unavailable.
The project maintains active qualification checks for API endpoint schema changes and keeps fallback probes in the pipeline to adapt when model catalogs shift. The skill re-verifies the Monid endpoint schema on each build and preserves Higgsfield credits as a fallback billing option when the primary service changes.
