Summary
Bart Slodyczka walks through a complete, repeatable six-step process for building visually sophisticated, scroll-animated websites using GPT-6 Astra in Codex mode, integrated with Higsfield’s MCP plugin and Cance 2.5 for AI video generation. The tutorial covers brand discovery, visual storytelling concept development, AI-driven site generation, desktop refinement, and mobile optimization — all from a blank project folder in the ChatGPT desktop app.
The core technique is constructing a “visual story” around a product: a smoothie brand shows berries dropping and blending as the user scrolls, a luxury watch store zooms through internal components before revealing the watch on a wrist, and a pizza shop animates dough being tossed, cooked, and boxed. GPT-6 Astra dynamically generates each scene’s starting image (using integrated image generation models), then calls the Higsfield MCP to create short video clips from those frames, which are embedded as scroll-scrubbed sequences directly in the site’s HTML and CSS. Slodyczka provides all prompts in a public GitHub repository.
Practical guidance throughout includes choosing “light” reasoning mode once prompts are refined (sufficient quality at lower token cost), using Nana Banana image generation as a cheaper alternative to GPT Image 2 for scene starters, and treating mobile optimization as a separate final-pass prompt rather than part of the initial build. The video demonstrates how a toolchain combining GPT-6 Astra, Higsfield MCP, and Cance 2.5 can compress what would historically be a multi-week freelance frontend project into a single agent session.
📺 Source: Bart Slodyczka · Published September 07, 2026
🏷️ Format: Tutorial Demo
