Reload Footage (Cmd+Alt+L) updated an item's metadata but left the media
pool's decoded frames and the layer cache in place, so the old image kept
rendering until Edit > Purge. It also replaced the footage with a probe of
the merged document, so a Photoshop layer item lost its layer and its
Retain Layer Sizes bounds.
Reload now re-reads only what the file decides (size, a movie's duration,
streams and frame rate, codec, a data file's text), keeps the layer or
page and the interpretation, finds a Photoshop layer again by name, and
makes the media layer forget the file (FootageSource::forget, which also
drops the EXR aux-channel cache) and clears the layer cache.
Settings > Import > Reload Footage Changed on Disk (on by default): the
desktop app stamps footage files (size + modification time) on a worker
thread every two seconds and when the window comes back to the front, and
reloads what changed. Not an undo step and it doesn't modify the project;
a file still being written is retried on the next check.
footage.reloadChanged runs one check (CLI, MCP, agents).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cause: quantize_px rounded with `(v + 0.5) as u32 as f32`. Converting back
from u32 is several instructions in V8's wasm code on x86 (no unsigned
conversion), so quantising a 1080p layer took about 15 ms in the browser:
every 8/16 bpc layer is quantised after its source and every effect, and
the CPU compositor quantises the comp after every layer.
Fix: round_half_up uses floor in wasm (one instruction there) and keeps the
integer conversion elsewhere (x86-64's baseline has no floor instruction,
where floor is slower). The values are the same: both round non-negative
values half up. About 1.5x faster quantising in the browser (15 -> 10 ms
per 1080p layer in a wasm micro-benchmark), unchanged on the desktop.
Tests: color::tests::rounding_forms_agree: both forms agree on 0..=levels
in sixteenths (exact halves included) and on 100 001 products, at 8 and
16 bpc.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Cause: at Half, Third and Quarter resolution (the viewer's Auto resolution
in most windows) every movie frame was converted from Y'CbCr to f32 RGBA at
full size and then resampled down. In the browser (wasm, one thread) that
was about 8 ms converting a 1080p frame, 6 ms resampling it and 2 ms
zeroing a 33 MB buffer, per layer and frame, for a frame shown at a
quarter or a ninth of its pixels; the footage frame cache held the
full-size frames.
Fix: FootageSource::frame_at_size asks for a frame shown at a given size
(default: the full frame, as before). The renderer asks for it where it
resampled footage (below 3/4 scale; not for proxies, fields or Interpret
Footage's pixel options, which need the full frame). MediaPool makes opaque
4:2:0 movie frames of the footage's own size there: it box-filters them to
half size in Y'CbCr before converting (each 2x2 block of luma, and the
[1 6 1]/8 neighbourhood of its chroma sample, which is the mean of the
upsampled chroma its pixels get), resamples the rest of the way and caches
the result per size. Conversion is linear until colours clip, so this is
the RGB box filter resample starts with, except in blocks where some
colours clip. Other frames convert at full size and resample as before.
In the browser the conversion and resampling take about 4 ms instead of
16. Docs (architecture).
Tests:
- convert::scaled_conversion_matches_resampling_the_converted_frame: half,
third, quarter and odd sizes of an unclipped 4:2:0 frame are within 1e-5
of resampling the converted frame; larger sizes, odd frame sizes and
4:4:4 convert at full size.
- tests_footage_size::reduced_renders_take_footage_resampled_by_the_source:
a source that resamples frames itself renders the same frames as one
that doesn't, at 8 and 32 bpc and every resolution, and is asked only
below 3/4 scale, and not with Invert Alpha.
- footage::h264_frames_at_reduced_sizes (ffmpeg fixture): 1080p H.264
testsrc2 frames, whose saturated colours clip, at 960x540, 640x360,
480x270 and 323x182 are 58-62 dB from the resampled full frames, and
are cached per size.
- convert's ignored 1080p bench also times the conversion at a third.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Cause: CC Particle World projected its particles through a fixed camera
(distance 2 layer widths, looking at the layer centre) and never asked the
host for the comp camera, so moving or orbiting a camera layer left the
particles flat on the layer. It also had no Extras > Effect Camera. Separately,
the layer cache key ignored the comp camera and lights, so the effects that
read them (Card Dance / Shatter / Card Wipe Comp Camera, First Comp Light,
Caustics) kept a stale frame when the camera, a light or the layer moved.
Fix: as the Cycore manual describes, Particle World sees its particles through
the comp's active camera layer when there is one (EffectHost::comp_scene now
says whether its camera is a camera layer's, CompScene::camera_layer), through
card3d's comp-camera projection (factored out of Card Dance as comp_proj, with
Proj::view for depth, size scale and screen velocity). Otherwise it uses the
new Extras > Effect Camera: Distance (2), Rotation X / Y / Z (0) and FOV (50)
orbit the world origin; FOV moves the camera so the origin's plane keeps the
scale Distance gives it, so the defaults reproduce the old fixed camera bit for
bit (golden hashes unchanged) and old projects render as before. The Effect
Camera is left out of the simulation cache key, so animating it doesn't
restart the simulation. The plan is shared, so the GPU compositor follows too.
The layer cache folds the comp scene into the key of layers running an effect
in effects::READS_COMP_SCENE (render::comp_scene is shared with the host).
Tests: sim particle_world_follows_the_comp_camera (panning a camera layer moves
the particles, zooming spreads them, the Effect Camera is ignored then, the
default view keeps the Effect Camera, cameras facing away / zero / NaN draw
nothing) and particle_world_effect_camera (missing params = defaults, rotation,
distance and FOV change the view, extremes stay finite, the sim key ignores the
camera); render particle_world_follows_the_comp_camera (a real camera layer
panned 40 px moves the particles left, and the cached render after the move
equals an uncached one: it failed with the old cache key); gpu
particle_world_through_the_comp_camera (CPU vs GPU at 8 / 32 bpc, full and
half resolution). particle_golden / sim_golden unchanged. effects.md
regenerated; parity.md and gaps.md describe the fix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cause: float footage (OpenEXR, float TIFF) was always sRGB-encoded on
decode, and Interpret Footage could not turn that off, so an OCIO effect
on the layer converted values that were already encoded.
Fix: After Effects' Interpret Footage ▸ Color ▸ Preserve RGB
(Footage::preserve_rgb; file.interpretFootage {"preserveRgb": true}; a
checkbox in the Interpret Footage form; Remember / Apply Interpretation
carry it). The decoder keeps float values linear, and the working-space
conversion from the colour profile and Interpret As Linear Light don't
apply, so the file's values enter the composition unconverted. The media
cache key includes the flag, so both interpretations decode separately.
Tests: pool::preserve_rgb_reads_float_exr_values_unconverted,
tests_color::preserve_rgb_footage_is_not_converted (managed, linearized
and not), tests_proxy::guess_alpha_invert_and_linear_light (command,
render, Apply Interpretation).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Text is vector in After Effects and stays sharp at any Scale. Here a
text layer was drawn at its font size and the bitmap resampled by Scale
unless Continuously Rasterize was on, so text grown with Scale went soft.
This reworks #200 (reverted in 29c4d3a for kinetic_typography).
Why kinetic_typography failed: its time remap key at 1.9 s sits on
frame 23 (1.917 s), so main-comp 0.3 s shows the precomp at 1.6026 s,
but the test compared it with the precomp rendered at 1.6 s. The moving,
motion-blurred title already differed by 0.89 of the 1.0 mean-difference
budget on main (0.987 in all); with continuous rasterisation the "Sub"
line, whose expression scale is 84.00 % at 1.6 s and 84.105 % at
1.6026 s, falls on either side of the 2^-1/4 quarter-octave step and was
drawn at 0.84 at one time and at 1.0 at the other, pushing the total to
1.075.
Fix: raster_scale treats text layers as always continuously rasterised
(shape layers and vector footage still follow the switch), and rounds
the quarter-octave step up past rounding noise only: an unscaled layer
at a fractional position measured its scale as 1 + 1e-15 and was
rasterised at 2^1/4, softening unscaled text. The QA test now compares
the remapped precomp with the precomp at the time the remap really
shows (read back from the property): difference 0.63, all of it the
title's motion blur, the Sub line identical.
Tests: tests_collapse::text_scaled_up_stays_sharp_without_the_switch
(from #200: 10 px text at 800 % matches 80 px text),
tests_collapse::unscaled_text_rasterises_at_the_output_scale (fails
without the noise tolerance), automation qa kinetic_typography.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cause: the interior styles were blended over the layer body with the
content's coverage folded into their opacity. On an edge pixel at 50 %
coverage a Gradient or Color Overlay went over the half-transparent body
as a half-transparent layer, so the result kept a quarter of the layer's
own fill and its alpha rose to 0.75: a light rim (the white fill
showing through) on styled text.
Fix: blend_clipped composites the interior styles (Gradient Overlay,
Color Overlay, Satin, Inner Glow, Inner Shadow and the inside of
Stroke) onto the layer within its shape: the body is divided by the
content coverage, blended with the style at its own opacity, and
multiplied back, so the alpha is kept and an edge pixel mixes the style
with the layer's colour inside the shape only, as a clipped layer does.
Fill Opacity and Blend Interior Styles as Group behave as before.
Tests: tests_styles::overlays_leave_antialiased_edges_clean (a masked
square with half-covered edge columns; Color Overlay, Gradient Overlay
and an inside Stroke now match a plain layer of the overlay colour;
Color Overlay's edge was 0.41 red against 0.23 before).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cause: each glyph's fill was rasterised and composited on its own, so
where two glyphs share an edge (the crossbars of "ff") two half-covered
pixels composited to 0.75 coverage instead of 1: a thin light seam,
where After Effects draws the line as one outline.
Fix: consecutive fill passes of the same opaque colour, without Blur,
under Normal Inter-Character Blending are rasterised as one coverage
mask (the glyphs' outlines in layer space, wound the same way so
overlaps add up under NonZero). Characters of other colours, translucent
or blurred ones, strokes and the other blending modes are drawn one by
one as before.
Tests: text::tests::touching_glyphs_join_without_a_seam (two glyphs
sharing an edge mid-pixel: 0.75 before, 1.0 now; opposite windings and
overlaps stay filled; other colours and Multiply still composite one by
one).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Effects: effect points on shape layers start at the layer centre (#227)
# Conflicts:
# crates/ui-egui/src/panels/puppet_tool.rs
# crates/ui-egui/src/panels/viewer.rs
The QA test's time-remapped precomp no longer matches the precomp rendered
directly (crates/automation/tests/qa/kinetic.rs:78); FreeBSD CI has been red
since. Reverting until the PR is reworked against that test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
While a mask was being drawn with the Pen, its open path was filled as
if closed, so the layer went black except inside the points placed so
far, which made rotoscoping by eye impossible. As in After Effects, an
open mask makes no transparency: the layer shows whole until the path
is closed. Effects that take a mask as a path (Stroke and the others)
still get open masks.
Test: open_masks_make_no_transparency (render).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Shape and text layers have comp-sized effect bounds centred on their
origin, but effects measured positions from the origin: a point at its
default "layer centre" sat at the comp's bottom-right corner, so Twirl
on a centred shape changed nothing and Bulge only its bottom-right
quarter. As in After Effects, effects now work in effect space,
measured from the top-left of the layer's effect bounds.
The renderer moves the layer buffer into effect space for the whole
stack (Buf::rebase, CPU and GPU alike) and hands effects everything
else in it: masks, other layers' pixels for layer parameters (each in
its own effect space, so Set Matte and displacement maps see a whole
shape layer), the layer's own frames (layer_input, also what Puppet,
Roto Brush and the Layer panel read) and the comp camera
(CompScene::rebased). Layer bounds are now right for every effect, so
pin_rect and EffectEnv::bounds_origin go and layer_rect serves edge
pinning too. Adjustment layers still run on the comp below as it is.
effect_bounds / EvalCtx::effect_to_comp replace the repeated "comp size
when there is no source" code; the viewer's effect point handles and
eyedroppers, Puppet pins, the Layer panel (paint, Roto Brush) and the
tracker's Corner Pin output use them.
Project schema 2: opening a schema 1 project moves positions held by
effects on shape and text layers by half the comp size, so set points
keep their place (point parameters with keyframes, Paint paths and
clone positions, Puppet seeds and pins, Liquify and Roto Brush
strokes). Points still at their default become the layer centre;
expression controls, adjustment layers and expressions are left as
they are (docs/architecture.md). The disk cache's render version is
bumped.
Tests: default_effect_points_sit_at_the_centre_of_a_shape_layer and
layer_parameters_see_a_whole_shape_layer (render, both fail before),
effect_points_on_a_shape_layer (GPU vs CPU),
opening_a_schema_1_project_moves_shape_layer_effect_points_into_effect_space
(engine), and the effects migration tests.
Part of #227
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Edge pinning (Turbulent Displace's default Pin All, Wave Warp's
Pinning, Bulge's Pin All Edges) holds the layer bounds still: a
comp-sized rectangle for shape and text layers. That rectangle started
at the layer's origin, but a shape layer's content surrounds its origin
(anchor 0,0), so everything left of or above the origin counted as
outside the layer and was pinned: only the bottom-right quadrant of a
centred shape moved.
The renderer now passes the bounds' top-left in layer coordinates
(EffectEnv::bounds_origin): (-w/2, -h/2) for layers without a source
rectangle, (0, 0) otherwise. util::pin_rect places the bounds there and
the three effects use it on the CPU and the GPU. Effect point
parameters keep their current layer coordinates.
Test: edge_pinning_displaces_all_of_a_centred_shape_layer,
edge_pinning_on_a_shape_layer (GPU vs CPU).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every pin move re-rendered the Puppet's input (source, masks and the
effects before it) at full size on the UI thread and deep-copied it
for the mesh overlay: the cache key of a layer's input hashed all of
its effects, so moving a pin of the Puppet changed the key of its own
input. The input key now covers only the effects that run before the
Puppet, and the overlay shares the cached buffer.
Test: input_keys_ignore_later_effects.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018gnTnH3Jmk5zr4tMq1Mc9f
Both visualisers read their Audio Layer's samples around the frame
time, but they weren't in TIME_DEPENDENT, so the viewer's layer cache
gave the layer one key for every frame and showed the first frame's
spectrum for the whole preview. Export doesn't use that cache, which is
why it rendered correctly.
Test: audio_spectrum_follows_time_through_the_layer_cache renders a
tone that starts at 0.5 s through a shared layer cache after t = 0 and
compares it with an uncached render (it showed the stale baseline).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018gnTnH3Jmk5zr4tMq1Mc9f
A text layer was drawn at its font size and the bitmap resampled by the
layer's Scale unless its Continuously Rasterize switch was on, so text
grown with Scale (the usual way to animate it) went soft. Text layers now
always rasterise at their on-screen scale, as they would with the switch;
shape layers and vector footage still follow the switch.
Text at 10 px scaled 800% now renders identically to text set at 80 px
(795 partially covered pixels each, maximum difference 0; it was 5854
against 795 before).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The per-line tracking compensation in text_geom used only the animators'
Line Anchor, which defaults to 0% (left) when no animator has one, so
tracking on centred or right-aligned text pushed the line to the right
instead of spreading it around its alignment point.
Without a Line Anchor, a line now grows from where its own paragraph's
alignment pins it: 50% for centred, 100% for right-aligned (a justified
paragraph by its last line), read per paragraph rather than from the
document. The Line Anchor property still overrides it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Toggling a layer's Audio, Lock or Shy switch (or Hide Shy Layers) threw
away every cached frame: the RAM preview identity hashes each comp's
address, which any edit replaces, and the layer and disk cache keys
hashed every switch. None of these change pixels.
- Switches::pixels, Layer::same_pixels and Comp::same_pixels (exhaustive
field lists, so a new field must be considered) and Comp::pixel_form.
- Layer cache keys hash Switches::pixels; the disk cache's comp content
key hashes Comp::pixel_form.
- Frames::identity: when a new revision's comps draw the same pixels as
those of the comp's last new identity, that identity (and the project
holding its comps' addresses) is kept, so the frames stay, also when the
switch is toggled back.
With audio on, playback skipped frames that weren't rendered yet. Now the
sound waits: until the frames ahead (to the end of the preview, or what
fits in the RAM preview) are cached, every frame shows as it renders,
silently, like a preview without audio; then the sound starts in sync.
Reaching an uncached frame stops the sound instead of skipping frames.
Tests: audio_lock_and_shy_switches_keep_the_frames (frames.rs, nested
comps and toggling back), audio_lock_and_shy_keep_cached_layers (engine),
the disk key in tests_disk_cache, and
preview_with_audio_shows_every_frame_and_sounds_once_cached.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- FrameKey's identity is the comp's content (frames::comp_content), not the project revision:
an edit keeps the cached frames of comps it doesn't touch and undo finds the frames of the
state it returns to. The identity hashes the addresses of the comps a comp draws (~0.1 ms on
the 1500-layer test project; the full content hash took ~117 ms), by value the footage, solids,
proxies and project settings; with expressions, every item. Frames::content_of keeps a
project snapshot per identity with frames (or in flight), so Arc::make_mut always copies an
edited comp and its address can't be reused meanwhile; past 64 identities the least recently
used go with their frames (long drags).
- drop_stale is gone (frames of other contents are evicted least-recently-shown first).
- disk_cache::comp_content_key covers items' proxies and Use Proxy (it served stale frames).
- Tests: frames (content keys, edits elsewhere and undo, bounded drags, in-place edit safety),
ui_ram_preview (end to end), the disk cache proxy key. Docs: roadmap (M4 done), parity.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Collapsed precomps: the renderer records the precomp layers a nested comp is drawn through;
motion-blur sub-samples re-evaluate their transforms (when their motion blur is on) and map
times down through each one, so the precomp layer's own motion blurs its nested layers. The
containing (outermost) comp's shutter applies to collapsed layers. The GPU compositor gets
this through Renderer::placement.
- Shape and text layers whose content changes within the shutter (compared by its cache key)
draw their source at Samples Per Frame times and average them; the cache key covers every
sub-sample's content.
- Tests: tests_motion_blur_depth (collapsed vs flat, the containing shutter, shape content vs
the same move by Position); each fails without the change.
- Docs: roadmap, parity.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`Renderer::mb_on` (render options × the comp's switch × the layer's switch as the precomp
layers above limit it) now decides everywhere:
- Mask motion blur checked only the comp's switch and the layer's own switch, so Render
Settings ▸ Motion Blur ▸ Off for All Layers and a precomp layer with motion blur off still
blurred masks; it now takes the gate (`masks::MaskBlur`) and the comp's Samples Per Frame
(four in Draft) instead of a fixed eight.
- Effects that read the shutter (`EffectEnv::shutter`) get it only when the layer is motion
blurred in this render.
- Advanced 3D layers follow the precomp layers' Motion Blur switches too.
- The layer cache key of a layer with masks or effects includes whether it is blurred and the
comp's shutter and samples, so a buffer blurred for the viewer is never reused by a render
with motion blur off.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A precomp layer trimmed out (or time-remapped) past its nested comp's end kept drawing the
nested layers whose bars ran on. Like After Effects, the nested comp now shows nothing outside
its own span (`Comp::covers`), flattened, through a proxy or with Collapse Transformations, on
the CPU and the GPU (both go through `Renderer::source` / `collapse_into`).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Composition Settings ▸ Advanced has both options (they were missing from the model, the
renderer and the dialog):
- Preserve frame rate: nested in another comp, the comp shows only its own frames
(`EvalCtx::nested_time` floors the precomp layer's source time to the nested comp's frame
rate; used for the nested frame, its proxy, aux channels, collapsed transformations and the
viewer's internal wireframes), and the Render Queue renders it at its own rate whatever the
Frame Rate override (`RenderSettings::rate`).
- Preserve resolution: nested in a comp rendered at Half / Third / Quarter, it renders at full
size and the buffer's scale places it.
- `comp.settings {preserveFrameRate, preserveResolution}`, `comp.info` reports them, scripts
read and set `CompItem.preserveNestedFrameRate` / `preserveNestedResolution`, Pre-compose
copies them, and the dialog registers `dialog.comp.preserveFrameRate` /
`dialog.comp.preserveResolution`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
render: the debug-only quantisation check uses debug_assert! instead of panic!. gpu: the
test-only panic on a wgpu validation error is cfg(test) with a narrow allow, and release
builds only log. Test-only unreachable!() calls become panic!(). Denies the panic lints in both
crates.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Every purge emptied the viewers' RAM preview; Purge ▸ All Disk Cache and Snapshot now leave
it (Image Cache Memory, All Memory, All Memory & Disk Cache, All and All 3D Cache still
clear it).
- The memory purges also drop the decoded footage frames (`FootageSource::purge`, the media
pool's `clear_frames`; decoders stay open), so footage reads decode again.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- GPU in browser job workers: each job worker opens its own WebGPU device
(deferred readbacks); the Render Queue export and the analyses' frame loops
are futures (render::passes, offload::run_request_async) whose frames
render in passes, awaiting the device between them. GPU particles and
Advanced 3D read back under keys too. workers.jobs reports passes and
readbacks; the smoke test measures them.
- Layer buffers in the browser's OPFS disk cache: frame workers back their
layer caches with a PrefetchStore; the page plans what to fetch before a
request from recent misses (LayerPrefetch over the shared DiskIndex).
- Templates embed footage (default on, size-capped with a warning) and
creating from a template extracts it next to the new project / into
browser storage.
- ICC LUT profiles (lut8/lut16/lutAtoB/lutBtoA, from the public ICC spec) in
Simulate Output > My Custom RGB, baked into a 3D LUT for the viewer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- GPU in the browser: each frame worker opens its own WebGPU device (Gpu::request_deferred) and
renders the GPU compositor and GPU effects with deferred readbacks (effectcraft-gpu deferred):
readbacks keyed by content start with map_async and miss, the pass returns placeholders, the
layer cache's gate drops inserts, and FrameServer::resume renders again once the device
delivered. Backend Auto times whole frames per comp in the workers. Frames with effects go to
GPU workers instead of the page. CPU fallback without WebGPU in workers (?nogpuworkers).
- Fix: a WGSL constant (3.40282347e38) that Chrome rejects invalidated every GPU kernel in Chrome
(the page's compositor too); a test checks every float literal fits f32.
- Disk cache in the Origin Private File System: workers write sealed frame entries with sync
access handles (temp file + move), the page keeps the DiskIndex (LRU under the settings' limit,
at most half the quota) and reads hits instead of rendering; survives reloads; blue cache bar.
- Storage manager: Settings ▸ Disk ▸ Browser Storage (usage/quota, persistent storage, Clear
buttons, automation ids) and storage.info / storage.persist / storage.clear (StorageHost).
- Docs: web.md (GPU workers, deferred passes, disk cache, storage manager, threading design:
one engine instance per worker), architecture, preferences, parity (WEB-1), roadmap.
- Tests: deferred passes vs CPU on a real device, placeholder gating, frame server passes and
Auto, GPU worker routing + disk cache across a reload, storage commands and widgets, store
usage by kind; smoke test: GPU effect frames in workers with pixel parity, disk hits after a
reload, storage manager.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
render: three_d::adv::SplitRun (Renderer::split_adv_run) hands draw_run's split path to an
accelerator as data (main scene, special layers far to near, 3D mattes drawn solo);
Renderer::sky_draw / three_d::SkyDraw for Environment Light Background layers. gpu: the walk
draws skies (adv_sky kernel), resolves each special layer, hides it behind the nearer main
scene (occlude kernel) and composites it through the 2D path without readbacks.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- gpu: pooled working textures (released once every encoder that could read
them is submitted), shared zero images, reused readback staging buffers,
direct readback into Image; NaN-poisoned scratch images under test and
NaN-aware diffs; validation errors panic under test
- render: AutoPick (per comp / scale / path EWMA of CPU vs GPU frame times,
warm-up and periodic re-probe) behind Backend::Auto; viewer uses it too;
cached adjustment footprints (uploaded once)
- gpu fx_warp: Warp, Bezier Warp, CC Bend It, CC Page Turn, Smear, Reshape,
Color Emboss, Cartoon; sliding-histogram median for large radii; CC Power
Pin perspective < 100
- cli: bench --gpu reports Auto and per-frame upload / readback MB
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Advanced 3D runs render entirely on the GPU: every motion-blur sub-sample
rasterised, resolved and averaged, the depth-based iris depth of field
(Classic 3D bokeh spans, highlight boost, progressive levels), the sRGB
encode and the composite over the GPU canvas (Accelerator::render_3d,
Renderer::prepare_adv_run, adv3d.wgsl). Wireframe outlines draw on the GPU.
Fixes NaN environment lookups for straight up/down normals on Metal.
bench --gpu gains an Advanced 3D comp (--adv3d).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Viewer frames render in long-lived Web Workers holding a project replica
(engine::remote: structural JSON diff/patch, Mirror, FrameServer); the
frame cache hands jobs to a RemoteFrames renderer (UI-thread fallback).
Roto Brush propagation runs in a job worker and streams segmentations
back (FrameSeg binary form), relayed to frame workers. Control-channel
wait:true commands whose job is offloaded reply when it ends. The Media
Browser browses a pluggable Browser: the file system on the desktop;
browser storage and File System Access folders on the web.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Depth of field in Advanced 3D comps uses the camera's iris (Fast Rectangle, Triangle to
Decagon, rotation, roundness, aspect, diffraction fringe) and highlight gain/threshold/
saturation through the Classic 3D bokeh kernel. Collapsed precomps in an Advanced 3D comp
add their nested layers as real geometry (meshes, extrusions, cards) lit by the parent;
2D collapsed precomps draw nested 3D layers with the parent's renderer. Text and shape
strokes extrude as bevelled meshes stacked in paint order (model extrude_stacked).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude
Katt
sanyamsingh001
daijro
bflatastic
cdinhoo
Brandon Thomas
EthanSuquet
Dexter