The rendering pipeline

Rendering turns a Scene Document into an MP4. It happens one frame at a time, and every frame follows the same two-step shape: evaluate, then draw.

Per-frame: evaluate then draw

Evaluate. The timeline computes each node's animated properties at frame n, producing an AnimatedState, and resolves the camera into a CameraState. Simulations contribute their state for frame n here too.
Draw. The camera is projected, and the state is rasterized into an RGBA frame.

Keeping these steps separate is what makes rendering deterministic and seekable — see Determinism.

Two backends

Backend	Stack	Use it for
CPU	cosmic-text + swash	The default. Crisp text and code, no GPU needed.
GPU	wgpu + WGSL	Shaders and the bloom post-process (`--gpu`).

The CPU backend handles high-quality text shaping and rasterization. The GPU backend adds programmable shaders and bloom.

GPU compositing order

When rendering on the GPU, each frame is composited in a fixed order:

background shaders → CPU/content draw → region & overlay shaders → bloom

So a background_shader sits behind everything, region/overlay shaders sample and modify the drawn content, and bloom is applied last:

A GPU frame: background, region, and overlay shaders, then bloom.

Files that made this video

The GPU effects scene: a scene script, one Rust source file, and three WGSL shaders registered in the manifest.

codescene.toml

Project manifest: output size, theme, scene order, transitions, and shader names.

Open full file

scenes = ["scenes/intro.ccs", "scenes/terrain.ccs", "scenes/pathfind.ccs", "scenes/particles.ccs", "scenes/graph.ccs", "scenes/life.ccs", "scenes/boids.ccs", "scenes/tweens.ccs", "scenes/inserts.ccs", "scenes/edit.ccs", "scenes/edit_source.ccs", "scenes/editor_theme.ccs", "scenes/font_set.ccs", "scenes/remove_styles.ccs", "scenes/morph.ccs", "scenes/shapes.ccs", "scenes/effects.ccs"]
theme = "dark_documentary"

[episode]
title = "How Rivers Form"

[output]
fps = 30
width = 1280
height = 720

[shaders]
aurora = "shaders/aurora.wgsl"
glitch = "shaders/glitch.wgsl"
scanlines = "shaders/scanlines.wgsl"

[transition]
kind = "crossfade"
frames = 18

Encoding

Finished RGBA frames are streamed — as raw video — into ffmpeg, which encodes H.264 and muxes any audio into the final container:

RGBA frames → ffmpeg (rawvideo → H.264) → output.mp4

--crf and --preset tune that final encode; --format png skips it entirely and writes frames to disk instead. See the CLI reference.

The whole journey

For where each stage lives in the codebase, see Architecture.