* feat: resize Save for Web previews and added palette editing for Gif/PNG8
* Save for Web: add Turkish and Hungarian rows for the preview navigation hint
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Content-Aware tools: faster, cancellable at once, same pixels
#2202 matched Photoshop's Content-Aware Fill but made the perf nightly slower (#2575): P21
Content-Aware Fill 10 MP 5x, P22 Spot Healing 2.4x, P49 cancel on 24 MP 4x and over its 200 ms
budget. This keeps every result bit for bit and takes the time back:
- The solver works over the hole's bounds grown by a patch: match field, costs, votes and the
per-target tables are kept there (nothing outside changes or is matched), the vote computes
each target's weight once and writes into a buffer instead of a Vec per pixel.
- Colour adaptation reuses each patch's own moments (target sums kept per EM step, source sums
per level where they pay off), summed in the same order as before.
- Exact early outs: SSD per pixel instead of per row (callers only ask "better?"), and the onion
peel's exhaustive search starts from where the filled neighbours took their sources and stops a
source once it can only lose (first-of-equals kept).
- PatchMatch sweeps a band in wavefronts of blocks when there are fewer bands than threads: each
cell sees the same neighbours and random numbers (SplitMix64 skip) as in raster order.
- Parallel pyramid, valid-source and target setup; no copy of the window when there are no
variants (fill_owned / complete_masked_owned); the job reads the window in parallel bands and
keeps only the selection's bounds for the apply step.
- Cancellation is checked per row everywhere, also in Auto sampling and in its max flow
(maxflow_with / grid_cut_cancellable): a 24 MP job now stops in tens of milliseconds.
- algo::par: rayon on native, sequential on wasm.
Proof: the new `content_aware_bench` example hashes every tool's output (Fill with each sampling
and adaptation option, Edit > Fill, Delete and Fill, Spot Healing x3, Patch, Content-Aware Move,
8/16/32-bit, Gray/CMYK/Lab, and the benchmarks' 10 and 24 MP fills): 117 hashes identical to
main, also with 1, 2 and 4 threads. New tests compare the onion peel with a plain scan, full and
clipped patch distances and moments, the valid-source runs with the summed-area test, thread
counts 1/3/16, in-place and copied fills, and cancellation of Auto sampling and max flow.
perf_scenarios --cpu, Ryzen 7 5800X shared with other work (load 4-7), 3 interleaved rounds:
P21 p50 295-306 -> 167-171 ms (max 342-362 -> 180-184); P22 p50 67-68 -> 38-40 ms (max 69-71 ->
40-41); P49 max 378-389 -> 54-77 ms (p50 94-110 -> 47). retouch_bench: Content-Aware Move
-47..-64 %, Spot Healing Content-Aware -40..-61 %.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* Content-Aware: faster Spot Healing dabs than main with #2805, same pixels
Against main with #2805 the branch was slower for small Spot Healing dabs (size 50:
+25-30 %, single-threaded +60 %): its onion peel walked all 49 offsets of every candidate patch,
checking each against the known pixels, and had no bound at the hole's edge.
- Onion peel: the target's known pixels are gathered once (#2805), the search starts from the
filled neighbours' shifted sources or every 16th source at the hole's edge (#2805), the limit
a candidate must stay under is a precomputed sum (no division per pixel), and candidates are
summed four at a time in step: their sums are independent, so the processor overlaps them.
Each is still given up only against the best from before the four, which can only improve.
- Vote: each target's weight and match centre are taken once; a pixel's patches are visited
over the clipped offset range, with the channel count and the colour adaptation fixed at
compile time. Same sums, same order.
- Valid sources: sliding counts per row and stacked runs per column instead of a 7-row check
per pixel; the k-th source is found through per-row and per-64-column counts, so no level
but the coarsest lists them all.
- `integral` stays out of line: inlined into `synthesize` it made Create Texture slower.
Same results: `content_aware_bench -- hash` (112 cases) and `-- hash-big` (the benchmarks' 10 and
24 MP fills) equal main's, also with 1 and 3 threads; the onion peel tests (#2805's and ours)
compare with the plain scan. `content_aware_bench bench spot` runs retouch_bench's Spot Healing
dabs; with `CAF_SPOT_TYPE` (comma-separated) other types, with page faults per dab.
Ryzen 7 5800X, shared (load 5-20), interleaved with main (3af4c46a), min of rounds,
retouch_bench, all threads / one thread:
Spot Healing Content-Aware 50: 49 -> 31 / 204 -> 117 ms; 100: 62 -> 35 / 164 -> 98 ms;
200: 118 -> 67 / 250 -> 158 ms; 300: 212 -> 119 / 411 -> 285 ms. Content-Aware Move -47..-62 %
(one thread -23..-45 %). Rows that don't run this code stay within noise, except Proximity
Match 200 after the Content-Aware dabs in the same process (+8..23 %): it takes ~5800 page
faults there against ~300 on main, because main's heavier dabs left more of the heap faulted
in; alone, or after Content-Aware dabs in `bench spot`, it is equal.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The Elliptical Marquee tested and wrote every pixel one at a time
(Surface::write_pixel), single-threaded, with no shared full tiles. The
add/subtract/intersect modes read both selections over the whole canvas
into f32 buffers. On 6000x4000 an ellipse took ~0.6-1.1 s and a mode
combination ~0.3-0.5 s.
ellipse_surface builds the shape per row: a run of fully selected pixels
(bounded analytically, checked with the exact f32 tests at both ends)
and edge pixels that get the original per-pixel coverage test. Tiles are
built in parallel, and fully selected tiles share one tile.
combine_surfaces maps two GRAY8 selections tile by tile through a
256x256 byte table, sharing tiles that come through unchanged. Both
produce exactly the old bytes (equivalence test against the old
per-pixel code).
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Add the Remove tool with non-local patch completion
Tool::Remove (J flyout) paints a hole that paint.remove fills from its
surroundings as a background job, in one undo step. A stroke around an
object also removes what it encloses; the selection limits the hole.
The fill (photocraft_algo::nonlocal) follows Newson, Almansa, Gousseau and
Pérez (IPOL 2017): texture features in the patch distance, onion-peel
initialisation, match fields upsampled between levels and a best-patch
final copy, plus gradient-domain hiding of edge and interior seams.
Sparse holes are filled in tiles. inpaint::complete (Content-Aware Fill,
Spot Healing, Content-Aware Move) is unchanged.
On 20 holes in public-domain photos (remove_quality) the fill keeps 0.89
of the texture it replaces, against 0.54 with the Wexler/PatchMatch
completion. 24 MP (remove_bench): 100 px ring 120 ms, 1000 px ring 2.5 s.
* Translate the Remove tool's label and tighten comments
Adds "Remove Tool" to every UI catalog, following each one's pattern for
tool names and its existing word for Remove. Corrects comments that no
longer matched the code: the benchmark's description, the tile and margin
minimums, and the reasons given for two parameters.
* Put the Remove tool first in the J group
The J flyout's slot now shows Remove until another J tool is picked, and
J from another group selects it; repeated J goes on to Spot Healing Brush,
Healing Brush, Patch, Content-Aware Move and Red Eye.
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Opening the preset store decoded every imported tip at full resolution, and
each tip then lived twice: in the session's library and in the store's
snapshot of the last sync. A generated 600-brush library (16-bit tips,
800-2000 px, 2.27 GB decoded) cost about 4.5 GB after import and again after
every restart.
Library presets now hold their bitmaps as StoredTile references with a small
saved preview (at most 96 px): photocraft-paint gains TipShape::Stored and
Pattern::Stored (rendered with the full bitmap when loaded, else the preview),
and the store writes each tip's size and preview with its group. Opening the
store reads previews only; once a group is written its presets drop their full
bitmaps. A tip is decoded when its brush is used (picking a preset, a stroke
naming one, exporting) into a 256 MB least-recently-used cache. A loaded stored
tip serialises as `sampled`, so tool presets, actions and brush.get keep the
same JSON. Stores written before this change have no previews: their tips are
decoded once at the next start, eight at a time, and the groups rewritten. A
missing or corrupt tip file is now reported when the brush is used.
Measured with crates/engine/examples/abr_memory.rs on the 600-brush library
from crates/io/examples/gen_abr_library.rs (Windows, release):
after import 4565 -> 38 MB working set, after restart 4551 -> 20 MB,
after using 5 brushes 4560 -> 37 MB; store open 1.47 s -> 0.05 s.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Photoshop's Vertical, Horizontal, Dual Axis and Diagonal paint symmetry presets beside the
path-based symmetry: axes start at the document centre, Transform Symmetry moves and rotates
them (Shift snaps to 15°, Enter applies, Esc cancels), and all copies merge into one coverage
pass so overlaps don't build opacity. New paint.setSymmetry command and preset icons. Fixes#1648.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Indexed Color accepted fractional colour counts and computed its live preview on the UI thread.
Counts are now whole numbers in 2-256 with a linear slider snapping at 8/16/32/64/128/256, and on
native builds the preview runs on one background worker that coalesces slider drags and rejects
stale results. Nearest-colour lookup uses an exact kd-tree above 64 entries (identical indices,
256 colours at 4.2 MP: 3018 -> 1061 ms). Web and inline sessions stay synchronous.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Opening a menu or the command palette looked up command ids by scanning the whole command
list. The registry now keeps its ordered specs and an id-to-index map in one OnceLock (first
match wins for a duplicate id), and is_live caches the shell command ids in a set. A batch of
917 lookups drops from 0.157 ms to 0.013 ms.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Add Pattern Stamp as a Clone Stamp flyout sibling
Introduce paint.patternStamp to paint the current pattern through a
brush stroke. Aligned strokes keep a document-space tile origin;
unaligned strokes restart at the first point; Impressionist jitters
each dab with a seeded RNG. Empty patterns and strokes larger than
4096² pixels return an error instead of allocating.
* Record Pattern Stamp on the tools scorecard
Allow the aligned stamp helper's argument count under clippy.
* Pattern Stamp: raise the stroke cap to 8192² so a 24 MP diagonal fits
The 4096² bounding-box cap refused a corner-to-corner stroke on a 6016x4000 canvas (~24M px).
Unit test covers that diagonal and a still-rejected 9000² box; the hostile-input test uses a
footprint above the new cap. Scorecard note labels Impressionist as an approximation.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* i18n(pl): translate the Pattern Stamp strings
The Polish catalog landed on main after this PR was written; the
translation-coverage tests need every language to carry new UI strings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Tool::ALL holds 51 tools now that Red Eye and Pattern Stamp both landed
The rebase onto main (with #1084's Red Eye tool) leaves Tool::ALL one entry
longer; bump the array length and the count test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Flyout test: qualify egui::Vec2 (panels.rs no longer imports Vec2)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Content-Aware Move: move or extend the selected pixels, the gaps filled content-aware (#213)
New command paint.contentAwareMove {offset, mode: move|extend, structure 1..7,
color 0..10, sampleAllLayers}. The content is pasted at the offset, its colour
fitted to the new place by `color` (0 keeps it, 10 is a full seamless clone),
and an edge band whose width `structure` sets is re-synthesised from the
surroundings. In move mode the old place is then filled with
content_aware::fill, never sampling the moved content (no ghost). The
selection follows the content. One undo step; runs as a background job with
progress and cancel. Works on layer pixels, masks and channels, at any depth.
Timing (retouch_bench, release, 6016x6016, elliptical selections dragged
300 px): 250x250 move 240 ms / extend 129 ms, 500x500 0.93 s / 0.53 s,
1000x1000 2.6 s / 1.5 s.
Known limit: at low Structure the wide band comes out smeared (the PatchMatch
vote blurs a thin band between high-contrast regions); that is the shared
completion, not this command.
generated with Claude Code
* Content-Aware Move Tool in the J group (#213)
Tool::ContentAwareMove next to the Patch Tool: outside the selection a drag
draws a lasso, inside it drags the selection and runs paint.contentAwareMove
on release (as a background job). Options bar: Mode (Move/Extend),
Structure, Color, Sample All Layers. The drag shows the moving outline; a
live preview of the result is not part of this change. New strings are
translated in every catalog. Scorecard TOOL-213-4: missing -> partial
(Transform On Drop and the live preview are still open).
generated with Claude Code
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Add the Patch tool (#213)
paint.patch heals the selected area with texture from where the
selection is dragged (Source) or copies the selection onto the
dragged-to area (Destination), seamless-cloned like the Healing Brush.
The solve is clipped to the canvas, so patches at the edge stay opaque.
Tool::Patch joins the J group: outside the selection it draws a lasso
(Shift/Alt add/subtract), inside it drags the patch, the outline follows
the pointer and the offset is clamped to the canvas. Options bar:
Source / Destination. Strings translated in all six catalogs.
Tests at 8/16/32 bit, a UI gesture test, panic_hunt green. Scorecard
TOOL-213-3 is partial: the drag previews the outline, not the pixels.
retouch_bench: patch 250x250 22 ms, 1000x1000 362 ms (release).
generated with Claude Code
* Patch tool: live preview of the heal while dragging (#213)
The canvas shows the document healed at the pointer's offset
(patch_preview.rs, modelled on the Move tool's preview) and
recomposites only the healed area; release commits the exact
paint.patch and its damage rect refreshes the same area.
The preview (retouch_cmds::patch_preview) solves the colour fit on a
coarse grid (at most 128x128 cells) and interpolates it back: the
Poisson result is source + a harmonic correction, so the texture stays
exact and only the smooth correction is approximated (worst pixel 0.022
at 8x coarser in the test). Selection coverage is read tile-wise and
the interpolation runs per row in parallel.
retouch_bench (release, 6016x6016): preview 250x250 7 ms, 500x500
17 ms, 1000x1000 52 ms (exact patch: 21 / 90 / 354 ms). Scorecard
TOOL-213-3 done. Also applies rustfmt to the first commit's code.
generated with Claude Code
* Patch tool: cap patches at 16M pixels, drop product names from comments (#213)
The solve needs roughly 150 bytes per pixel, so the old 64M cap allowed
about 10 GB and risked an allocation abort instead of an error.
generated with Claude Code
* Regenerate docs/scorecard.md
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Tool::ALL holds both the Patch tool and the Mixer Brush (rebase onto #433)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
- Layers panel rows lay out their right-hand indicators first (lock, fx badge
with an effects disclosure triangle, link, optional blend-mode label) and
give the name the remaining width, cut with an ellipsis. The Studio kind
sub-label and the Effects sub-rows truncate the same way; a narrow panel
squeezes group indentation before the name. Rows leave room for the
overlay scrollbar. Drawn rects are recorded for tests (layer_row_ui).
- New view-state command layer.setEffectsExpanded (DocState::fx_collapsed,
no history step), driven by the fx triangle (Alt-click: all layers).
- Layers panel menu: Collapse All Groups (layer.setExpanded all=false).
Group open/closed state is document data (PSD lsct, .pcraft).
- ui.set {dockWidth} sets the right dock width (250..520) for agents and
snapshots.
- engine example designer_psd: ~140-layer designer PSD (4-deep groups, long,
CJK/Greek/accented names, fx, clipping, masks, vector masks, locks, links,
type/shape/fill/adjustment layers, hidden layers) for visual QA.
- Tests: name never intersects an indicator at 1x/1.5x/2x, 250/290/520 pt and
in every theme; fx triangle hides/shows effect rows without selecting;
Collapse All Groups from the panel menu, state survives PSD and .pcraft.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* photocraft-plugins: sandboxed WebAssembly filter plug-ins (wasmi host, ABI v1)
A new L4 crate runs plug-ins in wasmi (pure Rust, MIT/Apache) with no host
imports, a fuel budget metered in slices with a wall-clock deadline, a memory
cap, a recursion cap and wasmi's strict module limits. Every band gets a fresh
instance; traps, budgets, bad manifests and bad pointers are all Errors.
Pixels go over as interleaved normalised f32 in tile-aligned bands, in
parallel on native, blended by the selection. Includes the example Invert
plug-in (no_std Rust, examples/plugins/invert-rs) built into a test fixture,
hostile-module tests, and docs/plugins.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Engine: plugin.* commands (install, list, run, remove, reload)
plugin.run filters the active pixel layer, a targeted channel or a smart
object (recorded as a smart filter) as one undo step; Preferences >
Plug-ins gains Use Additional Plug-ins Folder, loaded when prefs apply.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* UI: Filter > Plug-ins lists installed plug-ins with generated dialogs
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Plug-ins: bands never outgrow the memory cap; regenerate parity
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Bench: example plug-in Invert vs built-in Invert on 24 MP
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* CMYK documents display and export through their embedded CMYK profile
A thread-local CmykSpace (photocraft_color::convert::with_cmyk_space) replaces the built-in
coated CMYK while the CPU compositor, the GPU compositor's uploads and the PSD merged
composite convert a document's CMYK pixels; GPU residents track the profile they were
converted with.
Refs #46
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* EXR/HDR are linear: tag imports linear sRGB, linearise exports
Refs #46
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Colour-managed canvas: document profile -> monitor profile on GPU and CPU paths
* engine display_color: CanvasDisplay per (document profile, mode, monitor), identity fast
path when they match (sRGB on sRGB costs nothing); linear composites are stored
sRGB-encoded in the 8-bit texture. canvas_lut now always carries the CM transform, with
Proof Colors / Gamut Warning / 32-bit preview on top.
* GPU canvas applies the display LUT whenever the transform isn't the identity (also for
filter/proxy previews); the CPU canvas and Navigator transform the 8-bit composite.
* Edit > Color Settings > Monitor Profile (auto|built-in|.icc). macOS 'auto' reads the main
display's ICC profile at launch via osascript/AppKit (no unsafe); elsewhere sRGB.
* Tests: P3 (hand-computed), Adobe RGB, ProPhoto, Gray Gamma 2.2, synthetic CMYK, linear,
monitor setting; GPU+CPU canvas test via kittest (skips without Rgba32Float).
Fixes#46
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* io: tag EXR/HDR linear in import_flat (adapt to the shared image_to_document)
#65 moved flat decoding into image_to_document(name, img) so camera raws can reuse it; the
linear-EXR tagging needs the file bytes for format detection, so it now runs in import_flat.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
CI's `cargo fmt --all -- --check` has failed on every run because the
repo had no rustfmt config and was never formatted. Use the same settings
as EffectCraft/FilmCraft (max_width 160, use_small_heuristics Max), which
also keeps the churn small. Formatting only; no code changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
