🧊 Scene kinds: SDF vs. rasterized mesh, one G-buffer

Note to self: mockup to test the {% shader mode="postfx", scene="…" %} scene selector (SDF presets + mesh-* primitives). Every embed below runs the same post-process — only the scene= value changes. Safe to delete before launch.

The postfx pipeline fills a G-buffer (albedo / normal / depth / mask) and then runs your mainImage over it. How that G-buffer gets filled is now selectable, so the identical effect can be shown on two very different techniques. Drag to orbit, + / − to zoom, and use the buffers switch under each canvas to inspect the raw channels.

Baseline — the default SDF scene

scene="sdf" (the default): a raymarched signed-distance field. This is the original box + torus. The post-process is a plain albedo passthrough, so buffers → normal / depth / mask shows exactly what the scene wrote.

void mainImage(out vec4 O, in vec2 uv) {
    O = vec4(sampleAlbedo(uv), sampleMask(uv));
}

The same post-process on a rasterized mesh

scene="mesh-sphere" swaps the raymarcher for real triangle geometry through the vertex/rasterizer pipeline (with a depth buffer). The mainImage body is byte-for-byte the same as above — if the normal and depth buffers look right here, the mesh scene is writing the G-buffer correctly.

void mainImage(out vec4 O, in vec2 uv) {
    O = vec4(sampleAlbedo(uv), sampleMask(uv));
}

mesh-cube (flat-shaded faces — a good normal-buffer check) and mesh-torus:

void mainImage(out vec4 O, in vec2 uv) {
    O = vec4(sampleAlbedo(uv), sampleMask(uv));
}

void mainImage(out vec4 O, in vec2 uv) {
    O = vec4(sampleAlbedo(uv), sampleMask(uv));
}

An effect that reads normal + depth (SDF preset)

scene="sdf-spheres" is the depth-testbed preset. Here the post-process is a live-editable depth+normal edge detector, so you can confirm derivative effects behave (and don't band) regardless of which scene feeds them. Edit the kernel and watch it recompile.

void mainImage(out vec4 O, in vec2 uv) {
    // Laplacian of depth + normals -> outline, over the beauty pass.
    float c  = sampleDepth(uv);
    float l  = sampleDepth(uv - vec2(iTexel.x, 0.0));
    float r  = sampleDepth(uv + vec2(iTexel.x, 0.0));
    float u  = sampleDepth(uv + vec2(0.0, iTexel.y));
    float dn = sampleDepth(uv - vec2(0.0, iTexel.y));
    float depthLap = abs(l + r + u + dn - 4.0 * c);

    vec3 nc = sampleNormal(uv);
    vec3 nl = sampleNormal(uv - vec2(iTexel.x, 0.0));
    vec3 nr = sampleNormal(uv + vec2(iTexel.x, 0.0));
    vec3 nu = sampleNormal(uv + vec2(0.0, iTexel.y));
    vec3 nd = sampleNormal(uv - vec2(0.0, iTexel.y));
    float normalLap = length(nl + nr + nu + nd - 4.0 * nc);

    float edge = smoothstep(0.02, 0.20, depthLap * 12.0 + normalLap * 1.5);
    vec3 col = mix(sampleAlbedo(uv), vec3(0.04, 0.05, 0.05), edge);
    O = vec4(col, edge);
}

What to check

• All four channels populate on every scene — flip beauty / normal / depth / mask under each canvas. Background = far depth (white), mask 0; object = mask 1.
• The wire view (mesh scenes only) shows a hidden-line wireframe — the sphere/torus grid and the cube's flat faces, with back edges correctly occluded. SDF scenes have no wire button (there's no mesh to draw).
• Framing matches between the SDF and mesh embeds (same orbit target, FOV, and zoom).
• No banding in the edge pass — the G-buffer is RGBA16F where supported.
• Editing still works on the last embed, with last-good-kept-on-error behaviour.