Every bounding-box handle in the Composition viewer showed the
north-west / south-east resize cursor, so two corners and all four
edges showed the wrong direction.
Each handle now takes the cursor along the direction it scales, worked
out from the on-screen box: a corner points between its two edges (the
box's diagonal whatever its proportions), an edge handle across its
edge. The cursors turn with the layer's rotation, as in After Effects.
Tests: handle_cursors_follow_each_handle_and_the_layer_rotation
(ui_viewer: a wide box at 0°, 90° and 45°).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cause: pipeline::Cx rendered with Backend::Auto, which picks the CPU or the
GPU compositor for each top-level frame from measured frame times
(render::AutoPick: warm-up on both sides, a re-probe every 48 frames). The
two compositors differ slightly, so frames came out differently depending
on which side the timings favoured in that run, and two GPU exports of the
same frames differed.
Fix: the job renders every frame on the project's renderer, as After
Effects renders a job with one renderer: Backend::Gpu with Mercury GPU
Acceleration (frames the GPU declines still render on the CPU, as they do
in every GPU render), Backend::Cpu with Mercury Software Only. The viewer
and previews keep Auto. Docs (architecture, parity, web) say so.
Tests: render_settings::a_job_renders_on_one_compositor exports ten frames
with a stand-in GPU compositor that keeps Auto history and draws solid
green: with Mercury GPU Acceleration every frame is green and two runs are
identical (with Auto the warm-up sent frames to the CPU: it fails), with
Software Only the stand-in is never asked. The web worker GPU job test's
comment no longer describes Auto.
Diagnosis by @pat-the-wizard.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cause: pipeline::Cx::render_at built the job's Renderer without a layer
cache, so every layer of every frame rendered from scratch (source, masks,
effects, styles), and the GPU compositor, whose uploads are keyed by the
cached buffers, uploaded every layer again on every frame.
Fix: each job (Cx) owns a LayerCache, the cache type the viewer uses, with
its default budget (1 GiB, the default Settings > Memory & CPU layer cache),
shared by the job's parallel frames and dropped with the job. With deferred
GPU readbacks it holds the accelerator's miss gate, as the job workers'
caches do. Cx::renderer builds the job's renderer (options, accelerator,
cache) for render_at and the test.
Tests: pipeline::tests::layers_render_once_per_job exports ten frames of a
solid animated only in Transform: the cache misses of the first frame are
the only ones (it fails without the cache: no layer is cached), and every
frame equals the frame rendered without the cache.
Diagnosis by @pat-the-wizard.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
#402 made Effect Controls' rows compact and aligned, as After Effects
draws them, and also added thin separator lines between parameter rows
in Effect Controls and the Timeline, with a Settings > Appearance > Show
Parameter Row Separators preference. After Effects has neither, so the
lines, the preference and the Tokens field go; the compact rows, the
alignment and the automation ids stay.
Saved settings with the old key load as before (unknown keys are
ignored).
Test: ui_parameter_spacing keeps the row geometry, hit target and
nesting tests; the separator tests go with the feature.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Ease Presets panel (#254) is modelled on a third-party panel, so,
as with third-party tools in After Effects, it isn't core any more: it
is `extensions/scriptui-panels/Ease Presets.jsx`, a ScriptUI panel
bundled as an optional extension (Window ▸ Ease Presets.jsx), written
only against the public scripting API. Clicking a preset eases every
pair of neighbouring selected keyframes in one undo step (per
dimension, along the motion path for spatial properties, the keys'
other sides as they play); the fields edit the working curve, which the
panel draws; From Keys, Save, Rename and Delete as before. The twelve
built-in curves are unchanged.
Scripting API fixes it needed (After Effects parity):
- `app.settings` lasts between runs (`script_settings.json` in the
settings folder; `script.settings.get` / `script.settings.save`).
- Per-key methods (`setTemporalEaseAtKey`, `setInterpolationTypeAtKey`,
`removeKey`, the continuous / auto-Bezier / roving setters) no longer
change the user's key selection (they selected the key they edited).
- `keyInTemporalEase` / `keyOutTemporalEase` report linear and
auto-Bezier sides as they play, `keyIn/OutSpatialTangent` the
tangents the motion path plays.
- `selectedProperties` includes the properties of selected keyframes.
- `isInterpolationTypeValid` is false for Linear / Bezier on values that
only hold.
Removed from core: Window ▸ Ease Presets, the `keys.easePreset.*`
commands, `engine::ease_presets` and the `ease_presets.json` store,
`keyframe::EaseCurve` (nothing else used it) and the egui panel. User
presets saved by the v0.6.0 panel move into the panel's app.settings
section the first time settings load (the old file is left as is).
Docs: plugins.md (bundled extensions), parity.md, gaps.md G8, agents.md.
Test: app_settings_last_between_runs,
per_key_methods_keep_the_key_selection_and_report_eases_as_they_play,
spatial_tangents_are_reported_as_the_path_plays (each fails before),
the_ease_presets_panel_eases_every_selected_pair_in_one_undo_step,
a_typed_curve_scales_to_each_dimension_and_follows_motion_paths,
user_presets_save_rename_delete_and_persist,
presets_saved_by_the_old_core_panel_carry_over (script, through the
panel's controls), ease_presets_panel_eases_the_selected_keys_from_its_list
(UI: the panel docks and a click on a preset eases the keys).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
EffectCraft's extension points compared with After Effects' (public
user guide and Scripting Guide) are now in docs/plugins.md, which
becomes "Plug-ins and extensions": scripts, ScriptUI panels and effect
plug-ins, with a short guide and an example panel for writing your own.
ScriptUI panels already came from `Scripts/ScriptUI Panels/` in the
settings folder, were listed at the bottom of the Window menu and
docked as tabs. What was missing: After Effects reopens the panels
that were open when it quit. Now the open panels are remembered in
`scriptui_panels.json` in the settings folder (a panel is added when
it opens and dropped when its tab or its script closes it; quitting
leaves the list), and `window.restoreScriptPanels` opens them again;
the desktop and web apps run it at start-up. A panel whose script is
gone or fails is reported and forgotten, and the others still open.
Where a panel was docked isn't kept, as workspace layouts aren't saved
between launches.
Gaps recorded in plugins.md and gaps.md G8: the Startup and Shutdown
script folders, keyboard shortcuts for scripts, ExtendScript's
`#include` / `#target` directives and `.jsxbin`.
Test: open_panels_open_again_at_the_next_launch (fails before: no
restore command), a_panel_dropped_in_the_folder_is_listed_and_opens_as_a_panel
(script), panels_open_at_quit_dock_again_at_launch (UI: the restored
panel docks with its controls; closing the tab forgets it).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Tools bar's Snapping options menu (#265) offered per-feature
toggles (Edges, Corners, Centers, Anchor Points, Mask and Shape Path
Points) that After Effects' Snapping menu doesn't have. It now has the
two After Effects has:
- Snap Edges Extended (as before).
- Snap to Features in Collapsed Compositions and Text Layers: the
visible layers inside a precomp layer with Collapse Transformations
on are snap targets too, mapped into the outer comp (points always,
edge lines while the precomp layer is unrotated); a snap highlights
the precomp layer. Nested collapsed precomps are followed eight deep.
Every layer feature snaps again (corners, edge midpoints, centre,
anchor point, mask and shape path points); the snapping engine itself
is unchanged. `view.snappingOptions` takes `edgesExtended` and
`collapsedFeatures` (and `toggle` of either); the old keys are refused.
Test: collapsed_precomp_features_snap (fails before: no such option),
snapping_options_command_sets_and_toggles (the per-feature keys are
refused), snapping_options_pick_the_targets (AE behaviour kept: edges
extended or not, hidden layers, guides only), and the Snapping menu UI
test (only the two options are listed).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
After Effects' Interpret Footage has Start Timecode: Use Source File
Timecode, or Override Start, which changes the timecode the footage
shows and leaves its frames alone. The Start Frame field that trimmed or
padded a sequence's head isn't in After Effects, so it is replaced.
Footage.start_timecode (Override Start, in frames at the footage's
rate; None = Use Source, 0 as file timecode isn't read) relabels source
time: the Footage panel's current time and footage.info's `timecode`
count from it. file.interpretFootage takes `overrideStart` and
`startTimecode` (timecode, read at the footage's rate after any frame
rate change, or a frame number, 0 to 1e9). It applies to every footage
with time, not only sequences, and Reload keeps it. A sequence's frames
always run from its first file's number to its last.
Test: start_timecode_only_relabels_source_time (project),
interpret_footage_start_timecode_relabels_source_time (engine: the
reply, the Footage panel's timecode, frame numbers, Use Source, the
footage's rate, bad values, undo), Reload keeps it, and the Interpret
Footage fields in ui_drag_drop.
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
#365 rejects arguments missing from a tool's schema; both tools read `id`
as an alias of `command`, so declare it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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>
The menu bar has had a catalog since #229 (keyed by `(command, English label)`); everything below
it - panel labels, buttons, field names, tooltips, the sentences a panel draws itself - was written
straight into the code and stayed English in every language (#333).
This adds `crates/ui-egui/src/i18n/ui.rs`:
- Rows are keyed by the English source string, so a panel wraps a literal it already has instead of
inventing an id, and a row that is missing falls back to English rather than to an empty label.
- `tr("Rename")` translates a plain string; `tr_args("Proxy Use: {}", &[&quality])` fills the values
of a row that carries them. `tr_args` writes into its own buffer, so it cannot panic on a mismatch,
and a test keeps every row's placeholder count equal to its English source.
- The language in force is set once per frame from the resolved interface language - the same value
the CJK fonts are installed for - and kept in a thread-local, so the panels do not have to thread
it through every function. The render threads never draw interface text.
- Simplified and Traditional Chinese are complete (384 strings); the Japanese rows are still open,
and the catalog takes them without any other change.
Tests: the two Chinese catalogs must list the same source strings, no empty or duplicated rows, every
row must be asked for by a converted panel, and a converted panel that gains a hard-coded string in a
common call shape fails the build.
After Effects shows a missing frame of an image sequence as a
placeholder (colour bars) and has no Hold Previous Frame or Skip choice
in Interpret Footage, so EffectCraft doesn't either.
Interpret Footage loses its Missing Frames field, file.interpretFootage
its missingFrames parameter, and Footage its missing_frames setting. A
gap in a numbered sequence's numbering always shows colour bars. A
sequence imported with Force Alphabetical Order still plays its files
one after another (its numbering doesn't count), as in After Effects;
that is now an internal Footage.alphabetical flag, set on import and
kept by Reload. The Interpret Footage note lists the missing frames.
Projects saved before this have no such field and open with the After
Effects behaviour: a sequence with gaps shows placeholders where it
used to close them up.
Test: project sequence unit tests (placeholders, alphabetical order,
projects saved without the flag), tests_sequence (Start Frame instead
of Hold / Skip), sequence_gaps_show_placeholders (media) and the
Interpret Footage fields in ui_drag_drop.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Settings > Appearance > UI Scale (appearance.uiScale, 75-200 %, default 100)
scales the whole interface on top of the display's own scaling, live as it is
chosen; Cancel puts it back, OK saves it and a saved scale applies from the
first frame. The Settings dialog's choice rows show a number setting with
choices as its text. Automation id: settings.appearance.uiScale.
egui's own Ctrl+= / Ctrl+- / Ctrl+0 zoom is off now: Ctrl+0 (Window > Project)
also reset the zoom behind the app's back, and the scale is a setting.
Test: engine tests_settings ui_scale_is_a_percentage_from_75_to_200; ui_panels
ui_scale_setting_sizes_the_interface.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
As in After Effects, the right end of the Timeline's time ruler has a
composition marker bin (an original icon drawn in code). Dragging out of it
shows a marker following the pointer along the ruler's marker strip, snapping
like a dragged marker (current time, work area, other markers); releasing over
the time ruler's width adds a composition marker there with comp.addMarker (one
"Add Marker" undo step). Dropped back on the bin, nothing is added. The time
graph ends 18 px before the panel edge to leave the bin a column above the
layers' scroll bar. Automation id: timeline.markerBin.
Numpad * (or * typed with Shift+8), the keymap's Add Marker, now works in the
window too: egui has no key for it, so it comes as text.
The marker glyph and colour are shared helpers now (markers, nested markers,
the dragged marker).
Test: ui_timeline_zoom dragging_out_of_the_marker_bin_adds_a_comp_marker.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The in-window menu bar now works from the keyboard as After Effects' does on
Windows: Alt pressed and released on its own focuses the bar, Left / Right move
between its menus and Down, Enter or Space opens one. In an open menu Up / Down
move the highlight (skipping disabled entries, wrapping around), Right opens the
highlighted submenu and Left closes it (on a plain entry they open the next /
previous menu), Enter or Space chooses the entry and Escape closes the menus.
Hovering an entry moves the highlight there, and a highlighted entry scrolls
into view in long menus.
While a menu, context menu or popup list is open, or the bar has the keyboard,
keys go to the menus only: the app's shortcuts don't run (Space started the
preview behind an open menu) and the panels don't see the keys (Enter on an
entry no longer reaches the Project panel's rename).
Tall menus already scroll inside the window since #277 (ui_scrolling's
a_menu_taller_than_the_window_scrolls). F10 stays View > Switch 3D View > Front,
as in After Effects.
Test: ui_menus (shortcuts_wait_while_a_menu_is_open,
arrow_keys_drive_the_menu_bar, enter_chooses_the_highlighted_entry).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review follow-ups on #312:
- `crates/ui-egui/src/theme.rs`'s `japanese_font_tests` still called `install` with two
arguments, so `cargo test` did not compile. Every call site now passes a language.
- The call sites passed the stored `general.language` preference. With the default `system`
that is `""`, which handed the CJK choice back to the environment locale — one macOS and
Windows usually do not set. They now pass `crate::i18n::language(app)`, the resolved
language, so Match System draws Chinese on a Chinese system.
- Changing Settings ▸ General ▸ Language only reached the fonts on the first frame.
`EffectcraftApp` now remembers the language its fonts were installed for and installs them
again when it changes, so the switch applies without a restart.
- Documented that `fonts::set_cjk_locale` is process-global: the Han preference therefore
applies to composition text as well as the interface chrome (docs/preferences.md and the
function's own comment).
Rebased onto main after #309 landed — `header.rs` gained an `install` call site in the
meantime, and the catalogs now resolve through `i18n::language`.
Image sequences were found only by the desktop media prober: one picked
numbered still brought its whole run, with no way to import just that
frame, picking every frame made one sequence item per frame, the browser
couldn't make sequences at all, and a gap in the numbering closed up
silently.
The engine now groups numbered stills itself (engine/src/sequence.rs):
one picked file brings its run from its folder, several picked files of
one run import as that range, a picked or dropped folder imports its
files with each run as a sequence, and Force Alphabetical Order takes
every image of that type in the folder by name. `file.import` takes
`sequence` (default on), `alphabetical` and `frameRate`; Replace Footage,
Reload, proxies and render outputs probe the same way. The media layer
probes one file; the browser lists its file table for the run and
registers every frame with its media pool.
File > Import > File... asks, like After Effects' "PNG Sequence"
checkbox, whether picked numbered stills import as a sequence, with
Force Alphabetical Order and the frame rate. Sequences are named
`shot_[0001-0250].png`.
Interpret Footage gains Missing Frames (Show Placeholder, colour bars as
in After Effects, the new default; Hold Previous Frame; Skip) and Start
Frame (the frame number at the footage's first frame: trims or pads the
head). Saved projects keep Skip, so their sequences play as before.
Form notes keep a readable width and the form fits its rows, so OK stays
in view. docs/footage.md explains importing footage and sequences.
Test: tests_sequence (engine: run, picked range, every frame, Sequence
off, folder, alphabetical, Interpret Footage Missing Frames / Start
Frame, Replace Footage and Reload), project sequence unit tests,
sequence_gaps_show_placeholders_hold_or_close_up (media),
import_dialog_offers_the_image_sequence_option and
dropped_frames_import_as_one_sequence_with_sequence_interpretation (UI).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
EXtractoR's Red, Green, Blue and Alpha are text parameters naming
channels, and Effect Controls drew nothing for text parameters: there
was no way to choose a layer or channel. And a multi-layer OpenEXR file
whose colour is only in named layers (`diffuse.R`, Blender's
`ViewLayer.Combined.R`) didn't import at all: the image decoder wants an
unnamed RGB layer ("does not contain non-deep rgb channels").
- Effect Controls: EXtractoR gets a Layer popup listing the footage's
OpenEXR layers. Picking one sets red, green, blue and alpha to its
channels in one undo step, as After Effects' EXtractoR dialog does (a
one-channel layer such as depth goes in all three colours). Red /
Green / Blue / Alpha are popups of the layer's own R, G, B, A, every
channel of the file, and (none). The lists are read only while open,
so playing an EXR sequence doesn't parse files on the UI thread.
- `layer.channels {layer}` lists a layer's EXR layers and channels and
what each layer shows in red, green, blue and alpha, for agents.
- Media: an EXR without an unnamed RGB layer imports and shows its
colour layer (one named combined, beauty or rgba first); its channels
stay available to EXtractoR and the 3D Channel effects.
- Raster: AuxChannels::layers / layer_rgba and split_channel.
Test: extractor_shows_the_chosen_layer_and_channels_of_a_layered_exr
(host: a layered EXR written with the exr crate imports, layer.channels,
EXtractoR renders spec and depth; it failed to import before),
extractor_offers_the_exr_layers_and_channels (UI: Layer popup, one undo
step, channel popup), layered_exr_without_an_unnamed_rgb_layer_imports
(media).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Keep serving during long renders, recover failed workers, and remove only files opened by the interrupted export. Cover progress, cancellation, EOF and output ownership with regression tests.
Co-Authored-By: OpenAI Codex <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- New TRADITIONAL_CHINESE catalog: the same (command, English label) keys as JAPANESE and
SIMPLIFIED_CHINESE, one entry per fixed menu row (618 rows). The wording follows the
Traditional Chinese catalogs PhotoCraft, PdfCraft, VectorCraft and LightCraft already ship
(`檔案`, `開啟`, `儲存`, `圖層`, `遮色片`, `影格`, `預設集`, `快速鍵`, …).
- Match System maps a Traditional Chinese locale (`zh-TW`, `zh-HK`, `zh-Hant`, `zh_MO`) to
`zh-hant` instead of English, now that there is a catalog for it.
- `simplified()` and `traditional()` replace the Simplified-only `chinese()`; the macOS Services
item is 服务 in Simplified and 服務 in Traditional.
- `general.language` gains 繁體中文; `docs/preferences.md` documents the locale mapping.
- Tests: all three catalogs must cover the same fixed TREE entries; the shape check covers both
Chinese catalogs; label/undo/panel-title/submenu/native-menu behaviour in zh-hant with file and
layer names kept verbatim; the Traditional locale mapping; the preference round-trips.
- The i18n module now selects a catalog by language code instead of testing for Japanese:
`language(app)` returns the code, `lookup` is shared by every catalog, and the undo/redo
and workspace / 3D-view titles are formatted per language.
- New SIMPLIFIED_CHINESE catalog: the same (command, English label) keys as JAPANESE, one
entry per fixed menu row of the engine tree (618 rows).
- `general.language` gains 简体中文 (`zh-hans`). Match System maps a Simplified Chinese locale
(zh, zh-CN, zh_CN.UTF-8, zh-Hans-CN, zh-SG) to it and a Traditional one (zh-TW, zh-HK,
zh-Hant) to English, which has no catalog of its own yet.
- The macOS Services menu item is 服务 in Chinese.
- Tests: both catalogs must cover the same fixed TREE entries; `...`, brackets and spacing
keep the English shape; label/undo/panel-title/submenu/native-menu behaviour in zh-hans
with file and layer names kept verbatim; the locale mapping; the preference round-trips
through prefs.json.
- release.yml: `freebsd` job (FreeBSD 14.3 VM, the freebsd.yml packages) running the new
packaging/freebsd/package.sh; `flatpak` matrix (x86_64 on ubuntu-24.04, aarch64 on
ubuntu-24.04-arm) repackaging the Linux tarball with packaging/linux/flatpak-bundle.sh and the
new ai.storyteller.effectcraft.bundle.yml; both added to the draft release's needs.
- AppImages embed gh-releases-zsync update information and ship a .zsync beside them; the Linux
job installs zsync and checks both.
- version, Linux and web jobs drop `environment: release` (they read no secrets), so a dispatch on
a branch is a full dry run of every unsigned job; the draft release still needs the signed jobs.
- Flatpak finish-args documented and shared by the bundle and Flathub manifests (lint checks they
match); audio, GPU, portals, xdg-pictures/videos/documents, network for the optional model
downloads. The Flathub manifest now also installs effectcraft-cli.
- packaging-lint runs the FreeBSD layout dry run.
- README: Downloads section (every file a release ships) and nav entry; docs/releasing.md updated.
Co-authored-by: Brandon Thomas <shared2@storyteller.ai>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Imported GLB/glTF models were invisible in the Composition viewer while
text showed. Only the Advanced 3D renderer draws model and primitive
layers, and new compositions (and New Comp from Selection) are Classic
3D by default, but adding a model layer left the comp's renderer alone,
so the model drew nothing on every backend. It wasn't the reporter's
GPU: on this machine the viewer showed nothing for a GLB in a new comp
with the GPU compositor (Vulkan) and with Mercury Software Only, and
the model in both once the comp used Advanced 3D.
As After Effects documents ("Advanced 3D gets automatically selected
when you add a 3D model to the composition"), layer.newModel /
layer.addItem of a model and layer.newPrimitive now switch a Classic
3D comp to Advanced 3D inside the same edit (one undo step, Geometry
Options added to its text and shape layers like Composition Settings
does) and toast that the comp changed renderer. Switching a comp that
has model layers back to Classic 3D toasts why they disappear.
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>
#276 documents ["mcp", "--gpu"] and get_project's renderer.active / why in
the Headless setup notes of the same file, so the pitfall only repeated them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S1RGTt6AE3ysiaMQ7X6Nea
The Environment light existed (LightKind::Environment) but layer.newLight's help
and the kind errors of layer.newLight and layer.lightSettings listed only
Parallel|Spot|Point|Ambient, so agents could not find it. The help now lists it
and says how to give it an equirectangular image: prop.set
lightOptions/source to that layer's id, or leave it empty for the comp's
Environment Layer (layer.environment). layer.lightSettings documents its kind
values too.
docs/parity.md (3D) notes that Advanced 3D surfaces reflect only the
Environment light and that Appears in Reflections is stored but has no visible
effect yet. Object-to-object reflections stay tracked in #261.
Test: tests_3d::environment_light_is_listed_and_takes_a_source_layer (help and
both kind errors name every LightKind; an Environment light takes a source
layer by id, with undo).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S1RGTt6AE3ysiaMQ7X6Nea
get_project / render.backend reported "gpuAcceleration": true with
"active": "cpu" and no reason, which read as a broken GPU. The
headless MCP server only attaches the GPU compositor when started with
--gpu, which docs/agents.md didn't mention, and a failed --gpu said
only "no usable GPU adapter" (the adapter's error went to a log no
one sees). The session now keeps the host's reason there is no GPU
compositor (accel_note): the CLI's "headless: renders on the CPU
unless started with --gpu", or the desktop app's compositor setup or
retirement error. The status reports it as `why` whenever it renders
on the CPU (or that the project's renderer is Software Only). --gpu
failures include the adapter's error (Gpu::try_headless).
docs/agents.md documents `mcp --gpu`.
Test: render_backend_says_why_it_renders_on_the_cpu (engine).
Checked with effectcraft-cli mcp and mcp --gpu.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Adds a short Pitfalls section to docs/agents.md: layer time vs comp time
for keyframe tools (#257), layer.timing to set a start time, app.run()
in scripts and undo groups, saving early in headless mode (#258),
`mcp --gpu` and renderer.active (#262), and the Environment light with
lightOptions/source (#261).
Docs only; every statement checked against main and a 0.4.0 build.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S1RGTt6AE3ysiaMQ7X6Nea
Dragging a layer already snapped its feature nearest the pointer to
other layers' edges, corners, centres, anchor points and path points
and to the comp's edges and centre, and Cmd/Ctrl held during the drag
already turned snapping on or off against the checkbox. What was
missing were After Effects' Snapping options beside the Tools bar's
checkbox. The new menu turns Snap Edges Extended off (a layer's edges
then snap only along the layer, not across the comp) and each feature
family (edges, corners, centres, anchor points, mask and shape path
points) on or off, for the dragged layer's snap features and for the
targets alike. Agents set them with view.snappingOptions.
Layers whose eye is off (or that a solo hides) are no longer snap
targets, mask and shape path points of the dragged layer snap too, and
while snapping is on the feature that snaps is boxed during the drag.
Test: layer_drag_snaps_to_other_layers_edges_ctrl_toggles_and_options_apply
(UI), snapping_options_pick_the_targets and
snapping_options_command_sets_and_toggles (engine).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A user suggested py-aep (MIT, Python), which reads and writes .aep
files. G4 now records it for the clean-room decision, with the two
ways it could be used: as an external converter, or ported, which
means reading another implementation of the format.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On some older Intel integrated GPUs (UHD 600, UHD with 2020 drivers)
the graphics driver crashes EffectCraft as its window's DirectX 12 or
Vulkan device is created (exit code 0xC0000005): the process
disappears with no message, and only WGPU_BACKEND=gl helped. A driver
crash can't be caught, so each launch now leaves a marker in the
settings folder until its window draws. A launch that finds the
previous one's marker switches the new Settings ▸ Startup & Repair ▸
Window Graphics (Automatic / OpenGL) to OpenGL and says so in the
status bar. A window that fails to open says the next launch will try
OpenGL. WGPU_BACKEND still wins; macOS has no OpenGL backend, so it
never switches there.
Tests: a_launch_that_never_drew_switches_the_next_to_opengl (app),
window_graphics_is_validated (engine). Checked end to end with a
leftover marker: the app opened on OpenGL with the notice and cleared
the marker.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On macOS 27.2 (rustc 1.98) one of the 404 pinned simulation hashes,
CC Mr. Mercury at non-default settings, 32 bpc, t = 2.2 s, came out
different. The reporter rendered every case at the commit that pinned
the hashes and at current main on that machine, and the frames were
byte-identical. So the code didn't change: the platform's libm did
(the hashes are exact f32 bits through sin, cos and powf). The test
now also accepts listed per-case libm variants, with that hash as the
first, and every other case stays exact. Its docs say how to tell a
libm difference from a code change.
Test: libm_variants_are_accepted_for_their_case_only (sim_golden; runs
on every platform).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The reports of 8 October (#225-#227, #229-#231), their fixes (#235,
#237-#241), what remains of #227 (multi-monitor support, the Project
panel's empty-area menu, effect points on shape and text layers) and
the regression tests for each.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Tools bar had Fill and Stroke swatches and a Stroke Width, kept in
the interface only, so agents couldn't set them and new shapes could
only be a solid fill with a solid stroke. As in After Effects:
- With a shape layer selected and a shape tool or the Pen active, Tool
Creates Shape / Tool Creates Mask choose whether drawing adds shapes
to the layer or masks on it (Fill and Stroke hide while drawing
masks).
- Clicking the word "Fill" or "Stroke" opens Fill Options / Stroke
Options: None, Solid Color, Linear Gradient or Radial Gradient, the
blend mode and the opacity. The swatches show the kind; Alt-click
steps through the kinds.
- New shapes and Pen paths are painted with them: a Gradient Fill or
Gradient Stroke across the shape for the gradient kinds.
The options are engine state (EditorState.shape_tool) set with the new
shape.toolOptions command; shape.newShape and shape.newPath use them
unless given paint parameters (which now include fillType, fillBlend,
fillOpacity and their stroke twins; layer.newShape takes them too). A
Pen path without paint parameters now takes the Tools bar's stroke
(none by default) instead of a fixed 2 px white one. layer.addMask also
draws rounded rectangle, polygon and star masks, so every shape tool
makes a mask on a selected footage or solid layer (rounded rectangles,
polygons and stars made a new shape layer).
Test: shape_tool_options_paint_new_shapes,
masks_of_every_shape_tool_kind,
tool_creates_mask_draws_a_mask_on_the_selected_shape_layer,
fill_options_paint_the_next_shape.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With a shape layer selected, the Rectangle, Rounded Rectangle, Ellipse,
Polygon and Star tools made a new shape layer for every shape. After
Effects draws into the selected shape layer instead: the shape becomes a
new group on top of its Contents ("Rectangle 2", ...), placed by the
group's Transform. The new shape.newShape command does that (in the
layer's space, so a moved or scaled layer still gets the shape where it
was drawn), and makes a new shape layer only when no unlocked shape
layer is selected. The viewer's shape tools use it; masks on other
selected layers are unchanged. layer.newShape and the Pen's
shape.newPath share its layer and Contents helpers.
Test: shape_tools_draw_into_the_selected_shape_layer,
shape_tool_draws_into_the_selected_shape_layer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The interface language was English unless Japanese was chosen in
Settings ▸ General ▸ Language. Language now has a Match System choice
(`general.language` = `system`), the new default: the operating
system's interface language where EffectCraft has a translation
(Japanese), else English, as After Effects installs in the system's
language. English and Japanese can still be chosen; a language chosen
before stays. The browser build stays in English with Match System,
since it has no Japanese font of its own.
The system language comes from sys-locale (MIT OR Apache-2.0, no
dependencies on the desktop platforms), on desktop builds only.
Tests: match_system_uses_the_system_language_where_there_is_a_catalog
(i18n), interface_language_is_validated_persisted_and_backward_compatible
(engine).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
When the page's WebGPU device was lost, the canvas kept showing its
last frame: edits still went through and rendered frames were right,
but the editor never showed them and nothing said why (the status
message about it is drawn in the canvas that can't present any more).
The page now covers the canvas with a notice that offers Save Project
(a download) and Reload. Reload first writes pending changes to
browser storage, so the session snapshot brings the project and its
unsaved changes back; if that write fails it says so and offers
Reload Anyway. Without a kept session (?empty, memory storage) the
notice says to save first.
Test: apps/effectcraft-web/tests/page.mjs (Node, fake DOM).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add Files... and Upload Folder... read every picked file in one go and
rejected on the first file the browser couldn't read; the Media
Browser dropped that rejection, so nothing was added and nothing said
why. Each file is now read on its own: the readable ones are added and
the status bar names the unreadable ones with the browser's reason and
suggests picking them again. Cancelling the picker is not reported.
Test: apps/effectcraft-web/tests/page.mjs (Node, fake DOM).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* About: contributors compiled in from contributors.json (grab bag + table, sortable, name toggle)
About EffectCraft now has tabs About · Contributors · Models, following the craftrules standard
(standards/contributors.md): contributors/contributors.json is generated by
craftrules/scripts/contributors.py and baked into the binary by crates/ui-egui/build.rs as static
tables. Each contributor line: @username, opt-in display/real name, PRs, commits, lines +/−/Δ,
binary assets +/−, first/last commit. Sortable by every column; alphabetical ignores case and
the leading @. Every control registers an automation id (about.tab.*, about.credits.*).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Contributors: self-submitted names in contributors/people.toml; AGENTS.md credit-yourself section
build.rs reads contributors/people.toml next to the generated contributors.json and shows a name only
for entries the generator verified as written by that person (craftrules standards/contributors.md).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Contributors: regenerate (verified people.toml entries)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
A write to browser storage that failed (quota exceeded) was dropped:
effectcraft.flush() resolved anyway, nothing retried it, and a reload
restored the project from before the change. The failed key now stays
pending and is written again after 2 s (doubling to a minute while it
keeps failing; a newer change of the key goes with it), and flush()
rejects with the error.
Test: failed_writes_retry (store). The flush itself runs only in the
browser.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018gnTnH3Jmk5zr4tMq1Mc9f
worker.js left a rejected start (its async init) unhandled, which a
page's Worker error event doesn't see: the page waited for "ready"
forever and the render stayed pending until cancelled. The failed start
is now re-raised where the page's onerror sees it, like the other
worker failures; the job fails with the error and the worker is
terminated.
Test: apps/effectcraft-web/tests/workers.mjs (worker.js with a stub
module whose init rejects; host.js with a worker whose start fails).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018gnTnH3Jmk5zr4tMq1Mc9f
The desktop app opened the demo project whenever it wasn't given a
project file, so File ▸ New Composition on the Home screen added "Comp
4" to the three demo comps. As in After Effects, a launch now starts
with an empty Untitled Project. --demo opens the demo (Help ▸ Open
Demo Project and the Home screen tile still do); --empty is kept.
The arguments are parsed by a function now, with a test
(no_arguments_start_an_empty_project).
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>
add_effect ran effect.apply and then one prop.set per `values` entry as
separate commands: each became its own undo step, and a bad value (an
unknown parameter id) returned an error but left the effect, and any
values set before it, on the layer, so a corrected retry landed at
effects/#2.
It now runs them as one engine.batch, named like effect.apply's own undo
step ("Apply Gaussian Blur"): one undo removes the effect with its values,
and a failing step rolls the whole call back. The reply is unchanged.
Test: add_effect_is_one_undo_step_and_all_or_nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
prop.setExpression keeps an expression with a syntax error but disables
it, and says why only in its own reply. MCP set_property answers with
prop.get instead, and prop.get reported `evaluated` / `expressionError`
only for enabled expressions, so set_property, get_property and prop.get
all showed the broken expression as if it were installed.
prop.get now reports a disabled expression too: `evaluated` is the
static value that renders, and `expressionError` is its syntax error.
Test: expression_syntax_errors_reach_the_reply (set_property and
get_property carry the SyntaxError; runtime errors and valid expressions
report as before).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Args::parse kept any unrecognised `--option` as a valueless flag nobody
read, so `exec comp.new --bogus` ran with defaults and exit 0, and a
misspelt valued option left its value behind as a positional:
`--saveas z.ecproj` opened z.ecproj as the project and failed with an
unrelated "cannot read" error.
Options are now checked against the declared lists (VALUED and the new
FLAGS) before anything runs; unknown ones are a usage error naming them
(exit 2), and `--flag=value` on a flag says it takes no value.
render-frame's `--transparent` (the MCP render_frame option) was one of
the silently ignored options; it now keeps the frame's alpha.
Tests: unknown_options_are_refused, every_option_read_is_declared (every
option the code reads is declared, so none is refused),
unknown_options_are_usage_errors (exit 2, nothing runs or renders, and
--transparent renders alpha).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- #174: the file names are fixed (effectcraft_web.js, effectcraft_web_bg.wasm, worker.js,
sw.js…), not content-hashed, so the guide's year-long immutable caching kept returning
visitors on the old release after an in-place upgrade. The guide, nginx example, _headers
and .htaccess now revalidate (no-cache) the .wasm, .js and index.html, and suggest a
versioned path for longer caching. The file table lists the real files.
- #180: the .wasm is about 58 MB (17 MB gzip, 10 MB Brotli) for v0.3.1, not 13 MB / 5 MB;
docs/web.md's 38 MB figure is updated too.
- #165: preferences live in the origin's browser storage (OPFS, IndexedDB or memory), not
localStorage.
- #164: relative URLs come from the page itself (and package.sh's check), not a Trunk.toml
that doesn't exist.
The iframe example's title no longer says image editor.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Tools table gains history and script_ui, get_layer's depth,
add_keyframe's comp and the bridge screenshot's max_side and path.
agents_doc_tools_table_is_current (effectcraft-automation) fails when a
tool or an argument of its input schema is missing from the table, as
effects_doc_is_current guards docs/effects.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
How a release is cut (version bump, the gate, pushing main to release,
the draft, publishing), test runs, what gets built where, the Linux
glibc baseline behind the ubuntu-22.04 runners, and the optional signing
secrets. release.yml's two pointers now resolve.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A line that is not a JSON object with a string `method` (an HTTP request
line, other non-JSON text, a JSON array, invalid UTF-8, a line over 4 MiB)
now gets one error reply and the connection is closed, so a browser's
cross-origin fetch can't smuggle a command in its body. Connections are
capped at 16. Same transport as LightCraft (storytold/lightcraft#129) and
FilmCraft (storytold/filmcraft#123).
The ui_angles evidence (#141) moves next to G2's other regression
evidence; the Stability row notes the parent-chain hang, the label colour
panic and the contained GPU failures.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gaps.md records the seven reports of 6 October (#85, #88, #89, #93, #103,
#106, #117), what they found in features marked done, their fixes of 7
October and the regression tests; parity.md's Interface, Animation and
Preview rows, the ROADMAP's reliability note and the README's panels
paragraph (drag and drop between panels) follow.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The desktop app shares egui-wgpu's device, which had no uncaptured-error
handler, so wgpu's default one panicked on an out-of-memory error on
every frame thread; the panicking frame stayed "in progress" forever and
never rendered again. The GPU frame budget was a fixed 1 GiB.
- GpuContext::new logs uncaptured errors once its pipelines are built
(errors while building them still panic there, which the app catches
to fall back to CPU compositing).
- Gpu::within_memory runs GPU work in an out-of-memory error scope; on an
error the result is discarded and uploaded layers and pooled textures
are freed.
- Frame workers render each frame within it; out of memory, the frame
renders on the CPU and the RAM preview keeps half as many GPU frames
(down to 64 MiB), since wgpu can't tell how much video memory there is.
- A frame whose render panics renders again on the CPU, or is released,
so the viewer asks for it again.
Tests: a_frame_whose_render_panics_is_released (failed before),
running_out_of_video_memory_halves_the_gpu_frame_budget and
within_memory_returns_the_work_when_there_is_room.
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>
An angle's revolutions were painted text: only the degrees could be
dragged or typed, so full turns needed the dial or typing past 360.
fx_widgets::angle_field draws `0x+45.0°` as two hot numbers: the
revolutions (widgets::hot_int_at, ten pixels a turn, fractions adding up
between frames so slow drags still step; a typed value is truncated) and
the degrees, each keeping the other. Effect Controls, the Timeline and
the Properties panel use it, so they read and edit angles alike (the
Timeline showed `-1x-40.0°` and Properties had its own split).
Automation ids: timeline.prop.<uid>.revolutions,
effectControls.prop.<uid>.revolutions, properties.prop.<uid>.revolutions.
Regression test: angle_revolutions_scrub_and_take_typing drags and types
the revolutions in all three panels.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gaps.md records the Windows report and its fix under real-user experience and G2; parity.md's
Text row, the ROADMAP's reliability note and the README's Text bullet say text can use any
installed font; architecture.md says when the system font scan runs.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019EjTYdt76hTG4Kh2CJFnCr
A Windows user reported that the Properties panel's font menu offered only the bundled Inter,
JetBrains Mono and Noto Serif. The system font folders were scanned only when a project asked for
a family the bundled fonts lack, so until then every font menu (Properties, Character, Essential
Graphics) and `file.findMissing {what: fonts}` saw the bundled families alone, on every
platform. Listing the families now scans the system fonts first, and the desktop app starts that
scan on a background thread at launch so the menus are complete when first opened.
The style menus in the Properties and Character panels list the chosen family's own styles
instead of a fixed Regular / Medium / SemiBold / Bold / Italic. Both menus build their lists only
while open.
New query for agents: `text.fonts {query?, rescan?}` (MCP tool `list_fonts`) lists every family
with its styles, origin (bundled / system / user) and own-language name; `rescan` picks up fonts
installed since launch.
Tests: a fresh-process test that the family list includes every family in the platform's font
folders without a prior scan (failed before), and `text.fonts` filtering, rescan and per-family
styles, and the `list_fonts` MCP tool.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019EjTYdt76hTG4Kh2CJFnCr
- effectcraft-segment: a FaceModel interface (find a face in a region, follow it; a Topology
maps the model's points to the tracker's landmarks and outline), a TensorFlow Lite reader and
interpreter (float ops, float16 weights folded; matches ai-edge-litert to ~1e-6), and
MediaPipe Face Landmarker (Google, Apache-2.0: BlazeFace + Face Mesh V2, 3.8 MB, pinned
SHA-256). The registry gains a task and authors per model; installed weights get a NOTICE.
- effectcraft-track: FaceTracker::new_with uses the model, falls back to the classical fitter
when it finds no face; chin and jaw come from the outline either way so measurements agree.
- Engine: roto_models becomes models (both tasks); face.models / face.model.* commands,
Settings ▸ Face Tracking, `wait` on select/install/download for scripts; track.mask reports
faceModel.
- Credits: ATTRIBUTION.md lists the optional models and the Apache-2.0 specs the code follows.
- Docs: README, ROADMAP (G7), gaps, parity (TRK-3 score unchanged pending G1), agents,
preferences, architecture, segment README.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- New L1 crate effectcraft-segment: the MaskModel interface (prompts → foreground probability),
a registry of open-licensed models (licence must be MIT/Apache/BSD/…; URL, size, SHA-256),
a PyTorch checkpoint reader (zip + pickle subset), dependency-free SHA-256, and MobileSAM
(TinyViT-5M + SAM prompt encoder / two-way transformer / mask decoder) on rayon and ndarray's
safe GEMM. Loads the official mobile_sam.pt as published; checked against candle
(embedding 0.3% RMS, 99.9% equal mask pixels); ~0.7 s/frame encoder at 32 threads, cached
per frame, ~0.1 s decoder.
- track::roto: segment_with / propagate_with prompt a model with the strokes or the flow-warped
matte (box, core points, mask); its probability is a strong, unblurred prior for the same
graph cut (strokes stay hard constraints); a failing model or one disagreeing with the flow
(IoU < 0.5) is ignored. Prior gets a weight.
- effects::roto: Version 2.0 / 3.0 use the model chosen in Settings (set_model / model_for);
the model id is part of the chain seed.
- engine: roto_models (roto.models, roto.model.select / install / download / remove), weights in
the models folder (host::config_dir()/models, shared by the app, CLI and MCP), background
loading, downloads through the system curl, verified; Settings ▸ Roto Brush page (prefs
roto.model). host::config_dir moved from the desktop app.
- Tests: segment (kernels vs direct sums, checkpoint reader, SHA-256 vectors, registry), track
(fake model: prior shapes the cut, failing / disagreeing models ignored; real weights with
EFFECTCRAFT_MOBILESAM: base IoU 0.989, propagated ≥ 0.980 vs classic 0.973), engine (registry,
verification, selection; real install + load), MCP.
- Docs: README, ROADMAP, parity (MSK-4 done: 90/2/0), preferences, agents, architecture,
crates/segment/README.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
bflatastic
Vikram Bhaskaran
Brandon Thomas
Dmytro Matviichuk
cdinhoo
FeiLiuEM
storybored379
DWAA1660
jonassneu
pat-the-wizard
Claude
RicSchonfelder
EthanSuquet
Luskish
Dexter
midasdf
Jonas Neubauer