Each region's outline was fitted on its own, so the edge between two regions was fitted twice, a little
differently, and left hairline gaps and overlaps.
- An outline is cut at its junctions: the pixel corners where three or more regions meet, where two parts of a
region touch by a corner only, and where an edge reaches the image's border (each side of the image is a
neighbour of its own, so the image's corners stay exact). A junction lies where the edges around it say, the
same for every region around it.
- Each piece between two junctions is the whole edge between two regions. Both get exactly the same points for
it (shifts are rounded to 1/1024 px so that the two sides agree to the last bit; whether a piece is hard is
decided by the piece alone) and it is fitted the same way from either side (from its end further left), so it
is one and the same curve in both paths. A loop without junctions is fitted clockwise from its leftmost point:
a shape's outline and the hole it leaves in the shape around it are one curve too.
- Stacked layers (Overlapping) share no edges and are cut at the image's border only.
On each preset's own inputs, all sizes and variants (docs/image-trace-quality.md), before -> after:
- Default: gaps 22.2 -> 0.8 px, overlaps 22.4 -> 0.5 px
- 3 Colors: gaps 48.6 -> 0.0, overlaps 29.4 -> 0.0; outlines 0.087 -> 0.104 px off on average
- 6 Colors: gaps 39.0 -> 0.2, overlaps 35.7 -> 0.3; 0.137 -> 0.140 px off, largest 1.09 -> 1.34
- 16 Colors: gaps 34.5 -> 0.2, overlaps 26.4 -> 0.2; 0.133 -> 0.141 px off, largest 1.09 -> 1.31
A photo's noise is as large as the step from one palette colour to the next, so noise decided pixel by pixel
which of two neighbouring colours a pixel got: tens of thousands of specks, or (since the palette change)
neighbouring tones merged into one.
- A continuous-tone image (eight colours leave much of it unsaid; flat art, however noisy or unevenly lit, is
not one) is smoothed before it is quantised: three rounds of a small bilateral filter, which averages noise
away within a tone and keeps the edges between tones where and as sharp as they were.
- With Method: Overlapping, layers are stacked by lightness, from the largest colour's end of the scale,
instead of by area. Each layer reaches under those above it, so where two colours meet every layer between
them in the stack ends on that same edge and shows through its anti-aliasing: by area that was an unrelated
colour's hairline around every region; by lightness it is a neighbouring tone.
- Ignore White with Overlapping: no layer reaches under the white that is left out.
On each preset's own inputs, all sizes and variants (docs/image-trace-quality.md), before -> after:
- High Fidelity Photo: colour error 2.37 -> 1.75 (2.30 at the baseline), edge error 2.59 -> 1.98, 639 -> 121
paths (3,653 at the baseline), 25,604 -> 9,818 anchors, specks 152 -> 10; a 16 megapixel photo-like image is
traced in 3.5 s (31,131 paths) where it was refused
- Low Fidelity Photo: colour error 2.67 -> 2.45, 55 -> 37 paths
- Shades of Gray: 73 -> 17 paths, colour error 9.33 -> 9.26
- The other presets' numbers are unchanged
Color and Grayscale chose their colours from every pixel, so the blends along anti-aliased edges took palette
entries and traced as slivers, and with more colours asked for than the art has, k-means split flat colours
along their noise and the halves traced as specks all over each other.
- The palette is made from the picture's own colours: a pixel that lies between the pixels on opposite sides
of it is a blend and counts for nothing (lines one pixel wide and noise are no blends).
- Colours that noise made of one colour merge: close colours whose pixels are mixed through each other. The
bands of a gradient keep to themselves and stay apart.
- A blend pixel whose nearest palette colour is a third colour (halfway from navy to yellow lies an olive that
is nearer to a green elsewhere) goes to the nearer of the two colours it lies between.
- Boxes are cut where their two sides differ most (Otsu's criterion) instead of at the median, which split the
largest colour of flat art along its noise and left two smaller colours in one box.
- Colours are compared in Oklab when pixels are assigned and in k-means.
On each preset's own inputs, all sizes and variants (docs/image-trace-quality.md), before -> after:
- 16 Colors: 69.9 -> 4.2 paths, 1743 -> 66 anchors, outlines 5.08 -> 0.13 px off on average (largest 85.9 ->
1.09), kinks 1376 -> 4.4, paths in a colour the art doesn't have 5.9 -> 0
- 6 Colors: 21.2 -> 4.0 paths, 4.44 -> 0.14 px off; 3 Colors: colour error 4.86 -> 0.43, 0.33 -> 0.09 px off
- High Fidelity Photo: 3636 -> 639 paths, 81,597 -> 25,604 anchors, colour error 2.50 -> 2.37; a 1 megapixel
photo-like image 1.98 -> 0.42 s (51,389 -> 4,298 paths), a 4 megapixel one traced (27,583 paths) where it was
refused
- Low Fidelity Photo: 128 -> 55 paths, colour error 2.59 -> 2.67; Shades of Gray: 91 -> 73 paths, 9.29 -> 9.33
Each crack between an inside and an outside pixel is moved to where the pixels across it say the edge is (the
blend of the two tones in the column of pixels across the crack; a thin line by the centroid and sum of its
tone; hard-edged images are read as staircases), and the outline is fitted as Flat Logo mode fits its own:
corners put back, straight runs as lines, a round outline as a circle, the rest as few curves as Paths allows.
Snap Curves To Lines straightens slightly curved runs and makes arcs. A fit that strays from its outline
gives way to a plainer one, and at last to the outline's own points.
On each preset's own inputs, all sizes and variants (docs/image-trace-quality.md):
- Default: outlines 0.313 -> 0.091 px off on average, largest 1.51 -> 1.02; edge error 17.2 -> 10.1;
gaps between abutting paths 63.5 -> 22.2 px, overlaps 67.0 -> 22.4 px
- Black and White Logo: 0.270 -> 0.064 px off, kinks 15.3 -> 0.3, anchors 3.78 -> 2.73 times the art's
- Technical Drawing: 0.201 -> 0.086 px off; Line Art: 0.158 -> 0.093
- Worse: anchors on blurred and scanned thin lines (Line Art 25 -> 77 on average), High Fidelity Photo
(colour error 2.30 -> 2.50, anchors 46,866 -> 81,597, 1 megapixel photo-like image 0.37 -> 1.98 s)
crates/trace/tests/quality draws its inputs from known vector art (no image files), traces them with every
built-in preset and measures fidelity, distance from the true outlines, economy, artefacts, time and memory.
docs/image-trace-quality.md records the numbers on main before any change, beside the same measures of an
open tracer run as a black box. No change to the tracer.
An image larger than the tracer's 64 megapixels was refused. It is now
decoded and box-filtered down to the largest size of its aspect within
64 megapixels, and the traced paths are mapped back to the image's own
size. Images over 192 megapixels are still refused from their header,
before decoding.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XbTizgRsDpFPjtBgZAvfrZ
The colour histogram used the top 5 bits of each channel, so tones less
than 8 values apart always shared a bin and could never get their own
palette entry, however many colours were asked for. A dark, low-contrast
photo (a book-cover texture with RGB 33..66) traced to 5 colours when 12
were requested, and eight grey stripes 32..39 traced to one:
vectorcraft-cli run --in stripes-32-39.png --cmd select.all \
--cmd imageTrace.makeAndExpand \
--params '{"params":{"mode":"color","colors":8,"noise":0,"method":"overlapping"}}'
-> {"anchors":4,"colors":1,"id":4,"paths":1}
The histogram keeps its 32x32x32 bins, now fitted to each channel's range
among the opaque pixels: a channel spanning at most 32 values keeps every
value apart, a full-range channel keeps the old 5-bit bins.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Remove Noise merged each small component into its most common neighbour
in one pass. When a speck was merged into a small region that was itself
merged away later, the region's flood fill skipped the speck's pixels
(already seen), so they stayed behind as a new island of the region's old
label. Higher Noise values merge more regions, so they left more such
islands: on noisy images the path count rose with Noise (70 -> 126 -> 206
paths from Noise 50 to 200 on one test image). denoise now repeats its pass
until one changes nothing; each merge joins two components, so it ends
(with a cap of 64 passes as a backstop). Labels shorter than w x h are left
alone instead of indexed past their end.
Tests: the minimal one-row case, a property test on random blocky noise
(no mergeable region under the minimum is left, and a larger minimum never
leaves more regions), and two noisy images whose path count must not rise
with Noise (both fail without the fix).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A fourth Image Trace mode for flat-colour logos. The existing tracer follows hard pixel
edges, so an outline can only be as good as the pixel grid; here each pixel is read as a
blend of the logo's flat colours, the blend is interpolated smoothly (cubic B-spline), and
every outline is placed where two colours are equally present, to a fraction of a pixel.
Outlines are then cleaned the way a person would:
- corners the blur rounded are put back (line fits either side of a bend, intersected);
- near-straight runs and near-circular arcs become exact lines and arcs (Snap Curves To Lines);
- the rest is fitted with as few cubic Beziers as the tolerance allows (Schneider);
- touching colours overlap by 0.06 px so no hairline gap opens between independent fits;
- shapes are painted biggest first; transparent areas fully enclosed by artwork (letters
knocked out of a badge, the counters of an "o") are painted white so the logo reads the
same on any background, or become real holes with Ignore White (each hole belongs to the
smallest outline containing it);
- an outline whose fit strays more than 1.4 px from the raw contour is fitted plainly.
Colours are found automatically (and completed from pixels no blend of the known colours
explains, so a small accent is not lost) or named with `logoColors: ["#rrggbb", ...]` (white is
always its own class; colours closer than 12 count as one; at most ten). Gradients and photos
are refused with advice to use Color (a flat-colour score of the solid areas, 0.93). A
hard-edged source with no anti-aliasing is softened by under a pixel and fitted with looser
snap tolerances. Images over 1024 px on the long side are averaged down first (further, if
they are busy enough to make the unmixing expensive) and the result is mapped back exactly.
More real outlines than we will fit are refused, not silently truncated.
Also: a "Flat Logo" preset, the mode in the Image Trace panel (Method hidden: it always
stacks), the command docs and MCP prompt, es/fr/it/ru translations for the two new messages
and the mode name, and the ROADMAP.
Never-crash: no unwrap/expect/panic, every size and float is bounded and checked, with
tests for degenerate and extreme input (NaN/infinite parameters, 0x0 and 1x1 images, sparse
outlines, an anchor budget, a 200,000-entry palette) and a 250-run randomised no-panic
property test. The B-spline interpolation follows Unser et al. and Thevenaz et al., and the
curve fitter follows Schneider (Graphics Gems); both are written from the papers.
Measured (local experiment, not in CI): a flat-colour logo's vector original rendered down
to a soft 199x100 px PNG and traced. Mean edge error vs the original at 1600 px wide:
built-in colour presets 2.5 to 3.4 px, Flat Logo 0.78 px (0.48 px from a crisp source);
region overlap 0.88-0.91 vs 0.966. About 0.1 s for a 200 px logo.
Tests: 60 new in the trace crate (unmix, field, contours, geometry, Beziers, corners,
outline fitting, palette, end-to-end on synthetic shapes made in code) and 3 in the engine.
cargo xtask ci passes (fmt, clippy, test, assets, brands, layers, wasm): 4,290 tests.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Advanced › Create: Fills and/or Strokes, with Stroke (the widest line, px).
With Strokes, each colour's 8-connected parts no wider than the stroke width
(chamfer distance transform) trace as open centre lines instead of filled
outlines: the line is thinned to one pixel (Zhang-Suen with each removal
re-checked, so two-pixel lines survive; staircase corners dropped), its
skeleton becomes runs between ends and junctions (junction pixels merged,
spurs pruned, runs through two-way junctions joined, rings closed), and each
run is simplified and fitted with curves. The stroke width is the line's mean
width (pixels over length). Traced lines get round caps and joins in the
document. Without Fills, wider areas become 1 px stroked outlines.
`imageTrace.make` takes fills, strokes and strokeWidth, and refuses neither
fills nor strokes and a non-positive width.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Color mode's colours now come from a Palette (`palette` in imageTrace.make's
params), shown in the Image Trace panel under Mode:
- Limited: the `colors` best colours (what Color did before; the default).
- Full Tone: median cut until no colour box spans more than 96 (Less) … 8
(More) levels per channel, up to 256 colours, set by `colorDetail` 0-100
(the panel's Colors slider for this palette).
- Automatic: flat art (90% of the pixels in a few tight colours) traces with
exactly its colours, anything else as Full Tone.
- Document Library: the most used (at most `colors`) colours of the
document's swatches or of a swatch library (`library`: "document" or a
library id or name, chosen in the panel's Library list), used as they are;
traced shapes fill with the swatch's own colour, its colour model kept.
The library's colours are looked up at every trace, not saved.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Tracing a 2048 x 2048 photo with High Fidelity Photo made 733,000 paths
and 7.8 million anchors. The selected result's outlines and anchor
squares then made an egui mesh of 1.8 GB, over the GPU's largest buffer:
wgpu panicked while painting and the app quit (or ran out of memory
first, at 6.4 GB).
- Canvas: a selection with more than 100,000 anchors shows its bounds
instead of every outline and anchor, and the hover highlight skips
such an object. The Contextual Task Bar and that outline read the
selection's bounds from a per-revision cache (VectorcraftApp::
selection_bounds): a selected trace drew a frame in 20 ms, not 170.
- Image Trace: the colour layers are traced in parallel (26 s to 1.9 s
for that photo; the web traces them in turn). trace_within stops once
the paths pass an anchor budget; imageTrace.make allows two million
and otherwise reports the trace as too complex, with what to change.
Images over 64 megapixels are refused from their header, before they
are decoded.
- The trace crate's messages are translated (es, it) and scanned.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Shipped code no longer unwraps, expects or panics: every one of the 124
unwrap()/expect()/panic!/unreachable! sites outside tests now handles the
failure (Result/Option, ?, let-else, defaults) and reports errors to the user.
- Lints: workspace clippy denies unwrap_used, expect_used, panic, todo,
unimplemented and unreachable; clippy.toml allows them in tests, and
integration tests, examples and testkit opt out explicitly.
- Safety net (vectorcraft_engine::guard): a panic in a top-level command rolls
the document back and returns EngineError::Internal; a panicking tool is
reset; the MCP server answers with an internal error and keeps serving; the
UI frame and each control request are guarded.
- APIs that hid failures now return them: Rendered::to_png/to_jpeg/to_webp
return Result (JPEG/WebP errors used to write empty files),
FontDb::face returns Option, Region::top returns Option.
- Fuzzing: engine/tests/import_fuzz.rs feeds garbage, hostile and mutated SVG
and PDF through import → render → export (30k cases locally, no panics).
- Docs: AGENTS.md/CLAUDE.md rule, docs/development.md "Robustness" section,
README pointer and a PR template checklist.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
bflatastic
Fabian Grajales
japbcoelho
mateususa
Jax Spenser
Brandon Thomas