Repository navigation
Camera: a world larger than the frame, per-segment easing, handheld shake, camera-aware validation #428
Description
Activity
Root cause of the first item, found while implementing #427 and verified at 1920×1080. I had filed it separately as #451 before noticing it is this item; closing that one as a duplicate and keeping the diagnosis here.
The viewport clip is issued after the camera transform is already on the canvas, in
crates/rustmotion/src/engine/render/scene.rs(two sites, around lines 490 and 1518):let camera_guard = match effective_camera(scene) { Some(camera) if plane_cam.is_none() => { let g = super::CanvasGuard::new(canvas); apply_camera_transform(canvas, scene, camera, time as f32, w, h); Some(g) } _ => None, }; let clip_guard = super::CanvasGuard::new(canvas); canvas.clip_rect(Rect::from_wh(config.width as f32, config.height as f32), ClipOp::Intersect, true);
Rect::from_wh(width, height)is in local coordinates, and the canvas matrix already carries the camera translation, so the clip rectangle travels with the camera instead of staying pinned to the surface.Reproduction: 1920×1080, camera
x: 0 → -3000over 2 s, a red rectangle at localx: 200and a green one at localx: -2800, so the green one lands at screenx: 200when the pan completes. Counting sampled pixels of each colour inrustmotion still:timecamera xred green 0.0 0 1900 0 1.0 -1500 1026 0 2.0 -3000 0 0 At
time: 1.0the clip sits at device[1500, 3420], so red — at screen1700..2100— survives only over1700..1920, 55% of it; 1026/1900 is 54%. Attime: 2.0the clip is at[3000, 4920], entirely off a 1920-wide surface, and the frame is blank although the green rectangle is at screen200..600.So "panning to an element placed at
x: 3000shows an empty frame" is not a missing feature — the content is drawn and then clipped away. The clip is the surface, so it belongs before the camera transform, or expressed in device coordinates. That is separable from the other half of the item (validaterejecting the element as a viewport overflow, which is the third checkbox's camera-aware validation).camera.motion_blurfrom #427 inherits the same ceiling, since each sub-frame exposure goes through the same path.- added a commit that references this issue
on Sep 29, 2026
Context
User feedback from a real production (a 15 s one-take product film, 1920×1080, hand-written scenario). These are the blockers that made them keep an HTML/browser-based renderer for that kind of film.
To do
x: 3000shows an empty frame, andvalidaterejects the element as a viewport overflow. Theworldview steps from beat to beat with a crossfade, so an object cannot travel from one beat to the next. Wanted: a scene whose world extends beyond the frame, a camera that moves through it, and elements that travel continuously across it.{"version":"1.0","video":{"width":1920,"height":1080,"fps":30,"background":"#101018"},"scenes":[{"duration":2,"camera":{"keyframes":[{"property":"x","easing":"linear","values":[{"time":0,"value":0},{"time":2,"value":2100}]}]},"children":[{"type":"div","position":"absolute","x":3000,"y":440,"style":{"width":200,"height":200,"background":"#FF3366"}}]}]}validatereports a viewport overflow.wiggle/shakeapplicable to the camera.Definition of done
mainis written first. For anything visual it is a render test (scenario → frame → pixel assertions); tests that buildAnimatedPropertiesby hand are not enough — that is exactly how the regressions listed here slipped through.rustmotion validate, render, image inspected.cargo fmt --all --check,cargo clippy --workspace --all-targets -- -D warnings,cargo test --workspacepass.crates/rustmotion/skills/SKILL.md, the relevantrules/file and the exported schema. A field the validator accepts must have an effect, otherwise the validator rejects it or warns.<type>: <verb> <message>. No AI attribution in commits, PRs or issue comments.