Multiple illustrated layers stack at different depths. As you scroll, the camera "dollies in" — depth parallax separates each layer naturally. The Pixar-short opener feeling.
Where 2.1 (Parallax) moves layers vertically at different speeds, this technique uses real 3D perspective and pushes the whole scene forward in Z space as you scroll. Layers at different Z-depths separate at different rates because of perspective math — exactly how a real camera dolly works. Used by Pixar-style illustrated openers, ad campaigns, and any "step inside this world" hero moment.
The container has perspective: 1000px. Inside, a stage with transform-style: preserve-3d holds 5–7 full-bleed layers, each at a different translateZ (sky furthest back at -1200px, trees in front at 0px). The stage's own transform translates Z forward as scroll progresses: transform: translateZ(calc(var(--p) * 800px)). Because of perspective, layers further back appear to move slower while close layers fly past — true 3D parallax with one variable.
Scroll inside. The camera dollies into the mountain landscape — far layers stay put, foreground trees rush past.
That's the dolly. Scroll up to pull out.
Build a "camera dolly" scroll parallax scene. Outer scroll container with a sticky pin (top:0 height:100vh) inside a 2400px-tall section. Pin has perspective:1000px and perspective-origin:50% 50%. Inside the pin, a STAGE div with transform-style:preserve-3d that contains 7 full-bleed layers, each on its own translateZ depth: sky (-1200px, vertical gradient sunset), stars (-1000px, scattered radial-gradient dots), moon (-300px, a glowing radial-gradient circle), far mountains (-700px, conic-gradient triangles in dark purple), mid mountains (-450px, darker triangles), fg silhouette layer (-200px, dark vignette ellipses), foreground trees (0px, dark tree silhouettes). The STAGE itself has a transform of translateZ(calc(var(--p, 0) * 800px)) — driven by a scroll-listener that updates --p from 0 to 1 as the user scrolls. As --p grows, the entire stack pushes FORWARD in Z; because of perspective, near layers (trees) move past quickly while distant layers (sky) barely shift — true 3D parallax. Add a hero label at the top that fades out as --p grows, and a "you're inside" caption at the bottom that fades in after --p > 0.6. Optional: small mono "Z: NNNpx" readout in the corner showing the current dolly depth.