𝗪𝗵𝗮𝘁 𝘄𝗲 𝘂𝘀𝗲𝗱 𝘁𝗼 𝗽𝗮𝘆 𝗮 𝗪𝗲𝗯𝗚𝗟 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿 𝗳𝗼𝗿, 𝗜 𝗽𝘂𝘁 𝘁𝗼𝗴𝗲𝘁𝗵𝗲𝗿 𝗺𝘆𝘀𝗲𝗹𝗳 𝗶𝗻 𝟮 𝗱𝗮𝘆𝘀. 𝗦𝘁𝗿𝗲𝘀𝘀-𝘁𝗲𝘀𝘁𝗶𝗻𝗴 𝗚𝗲𝗺𝗶𝗻𝗶 𝟯.𝟭 𝗣𝗿𝗼 𝘁𝗼 𝘁𝗵𝗲 𝗺𝗮𝘅 🤯 The previous post, where I recreated the first part of our Three.js animation from izum.study with the help of a neural network, got a lot of reactions. So I decided to go further and take the experiment all the way to the end. 𝗚𝗼𝗮𝗹: build the FULL site animation using prompts only. Without a single line of code written by hand. So you understand the scale — back in the day, a live developer used to make a similar animation for us for a whole month, and it was accompanied by constant nerve-wracking and edits. 𝗡𝗼𝘄 𝘁𝗵𝗲 𝗿𝗲𝘀𝘂𝗹𝘁 𝘄𝗮𝘀 𝗿𝗲𝗮𝗱𝘆 𝗶𝗻 𝟮 𝗱𝗮𝘆𝘀. 𝗔𝗻𝗱 𝗶𝘁 𝗴𝗲𝗻𝘂𝗶𝗻𝗲𝗹𝘆 𝗶𝗺𝗽𝗿𝗲𝘀𝘀𝗲𝗱 𝗺𝗲. Gemini 3.1 Pro produced a fully working project: a wave of thousands of particles collapses into a 3D sphere on scroll, then smoothly breaks apart into three separate spheres (like in the original) and merges back. All of this is driven by GLSL shader math and tied to scroll via GSAP. 𝗪𝗵𝗮𝘁 𝗜 𝗵𝗮𝗱 𝘁𝗼 𝘁𝗶𝗻𝗸𝗲𝗿 𝘄𝗶𝘁𝗵: ⚡ Morphing logic. Splitting one form into three with particle flow is complex geometry. The neural net wrote the shaders for a smooth split and reverse merge by itself. ⚡ Synchronization with GSAP. Camera movement, scaling, and the split triggers are perfectly fitted to the page scroll. After that, all that was left was to bring everything to the final look and polish the animation so that it feels as smooth live as in the original. What used to require a month of work from a dedicated specialist and a solid budget is now solved over a weekend. You just need to act as the art director and steer the model’s logic correctly. Neural nets really are changing the rules. 𝗥𝗮𝘁𝗲 𝘁𝗵𝗲 𝗳𝗶𝗻𝗮𝗹 𝗿𝗲𝘀𝘂𝗹𝘁 𝗳𝗿𝗼𝗺 𝟭 𝘁𝗼 𝟭𝟬 🔥 And the main question: what do you think — is it already time for WebGL developers to start quietly updating their resumes, or can the leather bags still sleep peacefully for now? Write in the comments 👇