Crop Image: trim an image to its pixels that aren't fully transparent

object.cropImage {trim: true} cuts the selected image to the box of its pixels
with any alpha, copying them as they are (no re-render, no resampling) and
placing them where they were. An image with nothing to cut is left as it is
and isn't an error (trimmed: false), so a script can trim every image it
opens. Print-ready PNGs usually carry transparent padding; with this, Fit to
Artwork Bounds can size to the art.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

ROADMAP.md

@@ -143,7 +143,7 @@ grade by behaviour against `plan/illustrator/`, not by whether a menu item exist
 - **Type:** Text Wrap, Type on a Path effects (Rainbow/Skew/3D Ribbon/Stair Step/Gravity) and Options (Flip, Align to Path, Spacing; OK keeps the type where it is), type on a path's start, centre and end brackets (Selection and Direct Selection drag them to set where the type runs, slide it along its path and, across the path, flip it: #429), Character and Paragraph Styles (override-preserving redefine), paragraph attributes per paragraph (alignment, indents, spacing, hyphenation, hanging punctuation, tab stops, paragraph style, direction, mojikumi and leading model; the Paragraph, Tabs and Paragraph Styles panels act on the paragraphs the Type tool's selection or caret touches, Return continues the paragraph's style, threads and SVG keep them), Area Type Options (rows/columns/inset/first baseline, vertical Align Top/Center/Bottom/Justify per row/column, reflowing non-rectangular frames and text wrap), threaded text across any closed shapes, Fit Headline, the Every-line Composer for justified and ragged text (paragraph-wide breaks for an even rag, Single-line per object from the Paragraph panel menu or `text.setFormat {composer}`). Area type resizes like Illustrator's type areas: a bounding-box handle drag (Selection tool) or Area Type Options Width/Height size the frame and the text reflows at its size (threads too); Direct Selection drags a frame corner or edge (`text.reshapeArea`). Point type still scales. Auto Size: the frame height follows the text after every edit, in the same undo step (rectangular frames of horizontal type in one row; columns get the least height that holds them; a height set by hand or threading turns it off), and the "Auto Size New Area Type" preference gives it to new area type. Shrink Text to Fit (beyond Illustrator): overflowing area type scales its sizes, leading and baseline shifts down by the largest factor (in 0.1 pt steps of the first run, down to a minimum percentage) that fits, at layout time so every export sees it. `text.areaOptions` and `document.inspect` report overflow and the shrink factor. The Selection tool shows the type widget beside selected type's bounding box (hollow on point type, filled on area type); double-clicking it converts between them (Type › Convert To Area Type / Convert To Point Type), keeping the text and its styles in place. Variable fonts list every named instance as a style of their family (Figtree Light to Black and their italics, Bahnschrift, Segoe UI Variable), and the instance chosen is shaped, drawn, outlined and measured at its axis settings; PDF and SVG embed it as that instance. Variation axis sliders are not done. Beyond Illustrator: inline graphics (`text.insertInline`, Type ▸ Insert Inline Symbol) set a document symbol in a run of type as one character that flows, wraps and justifies like a glyph (InDesign-style inline objects, e.g. mana symbols in card rules text): sized to a multiple of the font size, centred on the cap height, opening up tall lines, following Redefine Symbol; a character of its run's size for Character Alignment and Top-to-Top leading, placed by the bidi reordering in right-to-left text (a neutral U+FFFC); drawn on the canvas, kept as live text plus a `<use>` in SVG and as vector art in PDF; Backspace removes it whole. Hebrew and Arabic are laid out right to left with the Unicode bidirectional algorithm (embedded Latin and numbers keep their direction, lines put in visual order after wrapping) in point, area and path type, with caret, selection and arrow keys in visual order; each paragraph's direction comes from its first strong character unless Paragraph Direction sets it (Paragraph panel with Show Indic Options, `text.setFormat {direction}`), and new type aligns to the start of its direction. SVG export outlines right-to-left text. Font Size and Leading are combo boxes in the Character and Properties panels (Font Size in the Control bar too, beside the Character link, font family and style): type, step with ↑/↓ or the spinner, scrub the label, or pick a preset (6–72 pt, shown in the type unit; Leading also Auto), each pick one undo step; Kerning and Tracking list -100…200; a size, leading, kerning or tracking the selected text doesn't share shows blank (#458).
 - **Interface languages, vertical type:** VectorCraft ▸ Language (or Preferences ▸ User Interface) picks Automatic (the system locale) or a registered language, persisted as `interfaceLanguage`. Translations come from per-language catalogs (`i18n/*.tsv`): Traditional and Simplified Chinese, Japanese, Spanish (neutral international Spanish, for every `es-*` locale), French (every `fr-*` locale), Italian (every `it-*` locale) and Russian (every `ru-*` locale) cover the whole interface, Czech and Brazilian Portuguese every menu label, Brazilian Portuguese the shell's `tl!` literals too (tested); Spanish, French, Italian and Russian also translate status-bar and error messages as they are shown (`@msg` templates, every message literal tested; the messages themselves stay English for agents); untranslated text stays English. Japanese and Chinese glyphs come from the craft-fonts build input (release builds) or the installed fonts, Czech, Spanish, French, Italian and Russian ones from the bundled UI fonts. The web app follows the browser's languages (or `?lang=`). The Basic toolbar's Type button holds Vertical Type in its flyout. Vertical Type, Vertical Area Type and Vertical Type on a Path create vertical text (columns right to left, upright CJK glyphs and marks with `vert` alternates, Latin and longer numbers on their side, two- and three-digit numbers set as tate-chu-yoko in one em of the column (kept whole when justified), upright glyphs centred on the column (tracked or justified too), point type's anchor on the first column's centre line); kinsoku (the strict set) applies to horizontal and vertical Japanese, and justified Japanese lines spread their room between characters (not inside Latin words); Paragraph panel › Mojikumi Set (shown with Preferences › Type › Show East Asian Options, `text.setFormat {mojikumi}`): None or Line-end Punctuation Half Width (new type's default, as in Illustrator; older documents and imported text keep None) shares one half-em space between consecutive punctuation, sets a closing mark ending a line half width and an opening bracket starting a wrapped line flush with its start, and puts a quarter em between Japanese and Latin letters or digits; Paragraph panel menu › Top-to-Top / Bottom-to-Bottom Leading measures leading from em box top to top (area type's first line touches the frame) or from baseline to baseline (new type in a Japanese interface starts top to top, with em box centre character alignment, #432); Paragraph panel menu › Burasagari None / Regular / Force (`text.setFormat {burasagari: "none"|"standard"|"forced"}`) hangs an East Asian comma or full stop (、。，．､｡) ending an area type line outside it, Regular (new type's default) when it doesn't fit and Force always (the rest of the line then fills the measure; closing brackets and Latin punctuation don't hang); Character panel menu › Character Alignment (with the East Asian options; `text.setFormat` or `text.setRangeStyle {charAlign}`) lines characters smaller than the largest on their line up on the Roman baseline, the em box top/right, centre or bottom/left, or the ICF (ideographic character face) top/right or bottom/left, from the font's BASE table or, without one, the average ink box of some ideographs and kana; Type ▸ Type Orientation switches existing text; caret, selection, hit testing and arrow keys follow the writing direction. Vertical text exports to PDF as real text and to SVG as outlines.
 - **Transparency:** opacity masks (clip/invert/disable/link), exported as SVG `<mask>` and PDF soft masks.
-- **Advanced art:** live Blends (steps/distance/smooth colour, editable spine, anchor-targeted Blend tool, faithful interpolation, knockout), Envelope Distort (warp/mesh/top object, Reset, full Envelope Options, mesh handles, Edit Contents; type, images, symbols, gradients, patterns and appearance distorted everywhere), Liquify tools with options and pen pressure, Puppet Warp with rotating pins, Perspective Grid with presets, movable planes and projected type (choosing a perspective tool shows the grid; Hide Grid hides it with the tool still chosen), Gradient Mesh, Shape Builder (stroke-only open paths and lines cut the regions they cross and enclose regions of their own; filled open paths count as closed; erase mode while Alt/Option is held: a minus pointer, red highlights of what will go, a click or drag deletes the regions touched, the pieces of open lines and the edges (outline pieces between the points where paths meet, opening their paths there) it passes over, one undo step, `shapeBuilder.merge {erase}` and `shapeBuilder.regions` edges over MCP), Live Paint (every path an edge, open lines included), Image Trace (12 presets, colour layers traced in parallel; a trace past two million anchor points or an image over 64 megapixels is refused with advice rather than running out of memory, and a selection too large to outline shows its bounds (#525); Object › Image Trace › Make and Make and Expand trace at once with the Default preset, #543; with an image selected the Control bar and the Properties panel show Image Trace, whose arrow traces with any preset, Mask (`object.maskImage`: a clipping path on the image's outline, selected so its handles crop), Crop Image and Embed or Unembed in place of Fill and Stroke; with an Image Trace object selected, its Preset (choosing another traces again), View, the Image Trace panel and Expand, plus Release in Properties; the five View modes (Tracing Result, with Outlines, Outlines, Outlines with Source Image, Source Image) redraw a trace on screen without tracing again, are saved with it and `imageTrace.setView` sets them, while exports and Expand keep the result; the panel's Advanced section is a header that stays open or closed as it was left (part of #641); Color mode's Palette: Limited, Full Tone (as many colours as the tones need, Less … More), Automatic (flat art with its own colours, else Full Tone; a heuristic) and Document Library (the most used colours of the document's swatches or a swatch library, filled with the swatch colours as they are), part of #641; Advanced › Create: Fills and/or Strokes, where lines up to the Stroke width trace as stroked centre lines (distance transform, thinning, a skeleton graph with spurs pruned, curve fitting; one width per line; without Fills, wider areas become stroked outlines), part of #641), pattern swatches with pattern editing mode, live Repeat (radial/grid/mirror).
+- **Advanced art:** live Blends (steps/distance/smooth colour, editable spine, anchor-targeted Blend tool, faithful interpolation, knockout), Envelope Distort (warp/mesh/top object, Reset, full Envelope Options, mesh handles, Edit Contents; type, images, symbols, gradients, patterns and appearance distorted everywhere), Liquify tools with options and pen pressure, Puppet Warp with rotating pins, Perspective Grid with presets, movable planes and projected type (choosing a perspective tool shows the grid; Hide Grid hides it with the tool still chosen), Gradient Mesh, Shape Builder (stroke-only open paths and lines cut the regions they cross and enclose regions of their own; filled open paths count as closed; erase mode while Alt/Option is held: a minus pointer, red highlights of what will go, a click or drag deletes the regions touched, the pieces of open lines and the edges (outline pieces between the points where paths meet, opening their paths there) it passes over, one undo step, `shapeBuilder.merge {erase}` and `shapeBuilder.regions` edges over MCP), Live Paint (every path an edge, open lines included), Image Trace (12 presets, colour layers traced in parallel; a trace past two million anchor points or an image over 64 megapixels is refused with advice rather than running out of memory, and a selection too large to outline shows its bounds (#525); Object › Image Trace › Make and Make and Expand trace at once with the Default preset, #543; with an image selected the Control bar and the Properties panel show Image Trace, whose arrow traces with any preset, Mask (`object.maskImage`: a clipping path on the image's outline, selected so its handles crop), Crop Image (`object.cropImage {trim: true}` cuts an image to its pixels that aren't fully transparent, pixel for pixel) and Embed or Unembed in place of Fill and Stroke; with an Image Trace object selected, its Preset (choosing another traces again), View, the Image Trace panel and Expand, plus Release in Properties; the five View modes (Tracing Result, with Outlines, Outlines, Outlines with Source Image, Source Image) redraw a trace on screen without tracing again, are saved with it and `imageTrace.setView` sets them, while exports and Expand keep the result; the panel's Advanced section is a header that stays open or closed as it was left (part of #641); Color mode's Palette: Limited, Full Tone (as many colours as the tones need, Less … More), Automatic (flat art with its own colours, else Full Tone; a heuristic) and Document Library (the most used colours of the document's swatches or a swatch library, filled with the swatch colours as they are), part of #641; Advanced › Create: Fills and/or Strokes, where lines up to the Stroke width trace as stroked centre lines (distance transform, thinning, a skeleton graph with spurs pruned, curve fitting; one width per line; without Fills, wider areas become stroked outlines), part of #641), pattern swatches with pattern editing mode, live Repeat (radial/grid/mirror).
 - **Paint and appearance (M3):**
   - **Swatches:** names follow the colour model and are unique; Swatch Options (process/spot, Global, Gray/RGB/HSB/CMYK/Web, live preview); editing a global or spot swatch recolours every linked fill, stroke and text run in one undo step; multi-select and delete with confirmation; New Swatch and New Color Group dialogs (a group from the selected artwork's colours); swatches in colour groups convert, separate and count like any other. The Swatches panel selects ranges and whole groups, has a find field, and drags swatches to reorder, regroup or paint art; Add Used Colors, Select All Unused, Merge, Ungroup and Sort by Kind; a CMYK document keeps applied colours in CMYK. Swatch libraries: nine generated colour libraries and five gradient libraries in a library panel, saved as `.vcswatches`, `.gpl` or CSS, and loaded from `.vcswatches`, `.gpl` and swatch exchange (`.ase`) files (RGB, CMYK, Lab and Gray; global, spot and process; groups). Tints of global and spot colours ("Name 40%") stay linked and separate as a percentage of their plate. Spot colours can be defined in Lab (Spot Colors options). A built-in [Registration] swatch prints on every plate.
   - **Colour:** a Color Picker (colour field, H/S/B/R/G/B channel slider, HSB/RGB/Lab/CMYK fields, hex, web-only snapping, out-of-gamut correction); the Color panel follows each colour's own model and Alt-click paints the other proxy; proxies look through groups and show "?" for mixed paint; Invert, Complement, Apply Last Color (`,`) and Apply Last Gradient (`.`) keep the colour model; every paint command feeds the recent colours. The Color Guide has 21 harmony rules, Steps and Variation options and Limit to Library. Adjust Color Balance (its Global mode shifts tints) and Saturate preview live; Edit Colors and Recolor also reach meshes, images and patterns. Recolor Artwork has colour reduction, five recolour methods, presets, Limit to Library, a colour-wheel Edit tab and Edit Color Group; a local Color Themes panel saves five-colour themes. Fill/Stroke chips open swatch and mixer popovers (the Properties panel has the Control bar's Fill and Stroke chips, Stroke link and weight spinner, with nothing selected too), panels have Illustrator's Window menu keys (Align Shift+F7, Transform Shift+F8, Pathfinder Cmd+Shift+F9, Layers F7, Character Cmd+T…; every panel can be given one in Edit › Keyboard Shortcuts), and Eyedropper Options pick up and apply chosen attributes (Alt and Shift+Alt, image pixels). Fills and strokes can overprint (Attributes panel, Overprint Black options), and CMYK documents blend in inks on screen, in PDF and when flattening. Document Color Mode converts the art, symbols, pattern tiles and swatches through the colour settings: Gray stays Gray, RGB greys separate into four-colour greys and a rich black as Illustrator does, or go on the black plate only with `grays: "black"` (also for `document.open {colorMode}`).

crates/engine/src/cmd/menucmds.rs

@@ -111,7 +111,7 @@ pub fn specs() -> Vec<CommandSpec> {
             "Crop Image",
             ["Object"],
             None,
-            "{rect?: [x, y, width, height]} crop the selected image (default: to the artboard it sits on) → {id, width, height}",
+            "{rect?: [x, y, width, height] | trim?: true (to the pixels that aren't fully transparent, pixel for pixel; an image with none to cut is left as it is, trimmed: false)} crop the selected image (default: to the artboard it sits on) → {id, width, height, trimmed? (with trim)}",
             has_image,
             crop_image
         ),
@@ -541,6 +541,9 @@ fn crop_image(s: &mut Session, p: &Value) -> Result<Value> {
         .ok_or_else(|| bad(C, "select an image"))?;
     let node = st.doc.node(id).cloned().ok_or(EngineError::NoNode(id))?;
     let NodeKind::Image(im) = &node.kind else { return Err(bad(C, "select an image")) };
+    if p.get("trim").and_then(Value::as_bool) == Some(true) {
+        return trim_image(s, id, im.clone());
+    }
     let ib = node.geometric_bounds().ok_or_else(|| bad(C, "image has no bounds"))?;
     let rect = match p.get("rect").and_then(Value::as_array) {
         Some(a) if a.len() == 4 => {
@@ -594,6 +597,49 @@ fn crop_image(s: &mut Session, p: &Value) -> Result<Value> {
     Ok(json!({ "id": id.0, "width": w, "height": h }))
 }
 
+/// Crop Image's trim: the image cut to the box of its pixels that aren't fully transparent, the
+/// pixels copied as they are (no resampling) and placed where they were.
+fn trim_image(s: &mut Session, id: NodeId, im: ImageObject) -> Result<Value> {
+    const C: &str = "object.cropImage";
+    let st = s.doc()?;
+    let blob = st.doc.images.get(&im.key).ok_or_else(|| bad(C, "the image's pixels are missing"))?;
+    let px = image::load_from_memory(&blob.bytes).map_err(|_| bad(C, "the image's pixels can't be read"))?.to_rgba8();
+    let (w, h) = px.dimensions();
+    let (mut x0, mut y0, mut x1, mut y1) = (w, h, 0, 0);
+    for (x, y, p) in px.enumerate_pixels() {
+        if p[3] > 0 {
+            (x0, y0, x1, y1) = (x0.min(x), y0.min(y), x1.max(x + 1), y1.max(y + 1));
+        }
+    }
+    if x1 <= x0 || y1 <= y0 {
+        return Err(bad(C, "the image is fully transparent"));
+    }
+    if (x0, y0, x1, y1) == (0, 0, w, h) {
+        // Already trimmed: not an error, so a script can trim every image it opens.
+        return Ok(json!({ "id": id.0, "width": w, "height": h, "trimmed": false }));
+    }
+    let cut = image::imageops::crop_imm(&px, x0, y0, x1 - x0, y1 - y0).to_image();
+    let mut png = Vec::new();
+    cut.write_to(&mut std::io::Cursor::new(&mut png), image::ImageFormat::Png).map_err(|e| bad(C, e.to_string()))?;
+    let (cw, ch) = (x1 - x0, y1 - y0);
+    s.edit("Crop Image", |d, _| {
+        let key = unique_key(d, "crop");
+        d.images.insert(key.clone(), vectorcraft_doc::ImageBlob::new("image/png", png));
+        if let Some(n) = d.node_mut(id) {
+            n.kind = NodeKind::Image(ImageObject {
+                key,
+                width: cw,
+                height: ch,
+                xf: im.xf * Affine::translate((f64::from(x0), f64::from(y0))),
+                link: None,
+                placement: Default::default(),
+            });
+        }
+        Ok(())
+    })?;
+    Ok(json!({ "id": id.0, "width": cw, "height": ch, "trimmed": true }))
+}
+
 /// The Control bar's Mask for an image: a clip group of the image and a clipping path on its
 /// outline, with that path selected so dragging its handles crops the image.
 fn mask_image(s: &mut Session, _: &Value) -> Result<Value> {

crates/engine/src/tests_menucmds.rs

@@ -1159,3 +1159,35 @@ fn saved_selection_names_are_trimmed_and_capped() {
     let list = s.execute("select.savedList", &json!({})).unwrap();
     assert_eq!(list[0].as_str().map(|n| n.chars().count()), Some(vectorcraft_doc::SavedSelection::MAX_NAME));
 }
+
+#[test]
+fn crop_image_trim_cuts_to_the_opaque_pixels_as_they_are() {
+    // 100 × 80 px, transparent but for a block at (20, 10)–(50, 40) px; placed at 2 pt a pixel.
+    let mut px = image::RgbaImage::new(100, 80);
+    for y in 10..40 {
+        for x in 20..50 {
+            px.put_pixel(x, y, image::Rgba([x as u8, y as u8, 7, if x == 20 { 1 } else { 255 }]));
+        }
+    }
+    let mut png = Vec::new();
+    px.write_to(&mut std::io::Cursor::new(&mut png), image::ImageFormat::Png).unwrap();
+    let mut s = Session::new();
+    s.execute("document.open", &json!({"name": "art.png", "dataBase64": vectorcraft_format::base64_encode(&png)})).unwrap();
+    s.execute("select.all", &json!({})).unwrap();
+    let img = selected(&s)[0];
+    s.execute("object.scale", &json!({"sx": 200, "origin": [0, 0]})).unwrap();
+    let r = s.execute("object.cropImage", &json!({"trim": true})).unwrap();
+    assert_eq!((r["width"].as_u64().unwrap(), r["height"].as_u64().unwrap()), (30, 30), "{r}");
+    // Placed where those pixels were: (40, 20) pt, 60 × 60 pt at 2 pt a pixel.
+    let b = bounds(&s, img);
+    assert!(close(b.x0, 40.0) && close(b.y0, 20.0) && close(b.width(), 60.0) && close(b.height(), 60.0), "{b:?}");
+    // Pixel for pixel, a barely visible (alpha 1) edge included.
+    let NodeKind::Image(im) = &node(&s, img).kind else { panic!("an image") };
+    let out = image::load_from_memory(&s.doc().unwrap().doc.images[&im.key].bytes).unwrap().to_rgba8();
+    assert_eq!(out, image::imageops::crop_imm(&px, 20, 10, 30, 30).to_image());
+    assert_eq!(r["trimmed"], true);
+    // Nothing transparent left to cut: left as it is, and not an error (scripts trim every image).
+    let r = s.execute("object.cropImage", &json!({"trim": true})).unwrap();
+    assert_eq!((r["trimmed"].as_bool(), r["width"].as_u64()), (Some(false), Some(30)), "{r}");
+    assert_eq!(bounds(&s, img), b);
+}