I turned a Pokémon card from my binder into a real holographic shader with Kimi K3.1 Pro and Three.js. Diffraction grating, 8 foil patterns, drag to tilt. $2.10 and 38 minutes. The rule that made it work: never ask for "A holo card shader". Ask for one layer at a time. Step 1 -> One card texture from my own binder, a plane in Three.js, orbit controls. Nothing shiny yet. Step 2 -> The diffraction grating itself in a single prompt: a rainbow that moves with the view angle, not a static gradient. Step 3 -> Foil patterns as toggles: hairline, pillar beams, crosshatch, cracked ice, glitter, water ripple, cat eye. One prompt each, test each one before the next. Step 4 -> Sliders for foil exposure, star intensity, pattern scale. Auto-rotate on a Lissajous path so it never sits still. Step 5 -> Export PNG and WebM straight from the canvas, so the card can leave the browser. Step 6 -> K3.1 Pro reviews its own shader, finds the banding on the cat-eye pattern, fixes it before I even notice. The split: > Three.js: the plane, the camera, the controls. > GLSL: every pattern, written by K3.1 Pro. > K3.1 Pro: the shader math, the UI, the fixes. > Me: tilt the card and decide which foil looks real. The part that matters: the whole thing cost about half of what Opus 5.5 charges for the same session and roughly 5x less than GPT-6 Astra, and on the high-speed version the shader came back in seconds, not minutes. Commit after every pattern that works. A bad prompt costs seconds, not an evening.