Current implementation gaps, experimental workflows, and platform or format limits. Status labels distinguish incomplete foundations from workflows that are gated or still experimental; linked guides describe what works today.
Reviewed 21 October 2026 against the v0.5.1 source tag, retained implementation, release notes, and linked guides. The color entry distinguishes native document values and working precision from browser display surfaces; print entries reflect the current export pipeline: bleed is configurable per page, visualized on the canvas, and honored by PDF/X. “Partial” means foundations exist but the complete user-facing path is not yet guaranteed. Some limits below are deliberate or qualification-dependent, not promised fixes for the next release.
The Linux AppImage for 0.5.1 can fail to start on systems that do not have WebKitGTK 4.1 (libwebkit2gtk-4.1) installed. The AppImage also needs FUSE 2 (libfuse2) to run. A fix that bundles WebKitGTK has already landed and will ship in a future release.
Workaround: Install WebKitGTK 4.1 and FUSE 2 from your distribution, or use the .deb or .rpm package instead.
The current local Fill and Remove paths perform promptless reconstruction; entering a prompt does not make those models follow a natural-language instruction. Prompt-conditioned modes remain gated. Browser AI-quality Expand requires the user to install the local LaMa model and still awaits independent real-photo qualification; the browser Fast/PatchMatch route stays disabled after visible edge striping.
Workaround: Use the documented promptless workflow where available, inspect the mask edge and result at 100%, and keep the source recoverable. Do not rely on a prompt to direct these local reconstruction paths.
The browser build exposes a limited workspace set and browser-local storage. Desktop-only Print production, on-device inference, and Motion and Email authoring are outside the demo; installed-app and offline behavior also depend on browser support and storage permission.
Workaround: Review the demo guide before moving important work into the browser, and use a supported desktop installer for workflows the demo does not include.
Presentation output is a raster screen PDF or numbered PNG archive; PDF text is not searchable and the PDF has no tagged accessibility structure. Local plugins are bounded Wasm packages for selected-layer analysis and batch renaming, not a marketplace or a general-purpose extension host.
Workaround: Check the task-specific guide before depending on either workflow for production delivery. Keep an editable source document and verify exported presentation output separately.
Varve can import and re-import supported DTCG token files, review a three-way merge, and export canonical JSON. It does not watch a file in the background, use a Git-backed source, write changes into the original file, or claim vendor-dialect interoperability.
Workaround: Keep the source under your normal version control, re-import it when it changes, review conflicts, and check exported tokens in the downstream tool that will consume them.
Mockups support flat, four-corner perspective, bounded front-facing cylindrical, and folded-mesh envelope surfaces. They do not provide cameras, hidden backsides, lighting, calibrated displacement maps, or Photoshop smart-object re-rendering. PDF/X stops until a mockup is flattened; code exports warn that mockups are outside vector-structure output.
Workaround: Use a supported surface and export format, then inspect the composed result. Flatten the mockup before PDF/X delivery.
RAW development covers a documented DNG camera subset, not every camera format or encoding. Some image workflows need an optional local model download; a capability available in the desktop app is not automatically available in the bounded browser demo.
Workaround: Check the supported inputs and model requirements in the RAW/HDR and image-tool guides before planning a production workflow.
Varve now has a standards-based shaping contract, Unicode source maps, and a HarfBuzz-compatible WASM adapter in the engine. The live Canvas2D renderer still uses browser measurement and fillText in this migration stage, so Arabic, Devanagari, Thai, and other contextual scripts are not yet guaranteed to match the canonical glyph layout across fonts and exports.
Workaround: For production-critical output, verify the selected font and export result. Basic Latin and CJK workflows remain the most predictable while the shaped-glyph renderer is integrated.
The engine now resolves UAX #9 embedding levels, script itemization, per-line visual run ordering, word-level line breaking (never inside a grapheme), cluster-safe caret stops, hit testing, and selection geometry through one canonical paragraph layout pipeline. The inline editor and pointer tools still use the legacy textarea/canvas path, and export consumers are being migrated incrementally.
Workaround: Keep mixed-direction text short while validating it, and check the final export. Explicit paragraph direction can make intent clearer than Auto for authored text.
Logical paragraph/run indexing, grapheme-safe formatting transactions, property removal, text replacement, and mixed bold/italic inspector state are now in place. Insertion-style state, full paragraph controls, composition transactions, and canonical shaped rendering for every span are still incomplete.
Workaround: Use the rich-span inspector for supported character properties and verify mixed-style text after editing. Separate text nodes remain the safest fallback for production-critical compositions.
The engine exposes a derived layout snapshot containing glyph clusters, lines, caret stops, selection fragments, bounds, diagnostics, and cache identity. The existing canvas, text-on-path, mask, and export consumers are being migrated incrementally and may still use independent browser measurement paths.
Workaround: Treat complex-script, mixed-direction, and font-loading results as beta behavior and inspect exported output before delivery.
Collaboration features (PresenceIndicator, CollabCursorOverlay) are UI scaffolding with no transport or wire protocol. Real-time collaboration does not work.
Workaround: Use version control (git) for collaboration.
State machine inspector for complex prototype flows is not implemented.
Workaround: Use timeline-based animation for now.
Rigging and inverse kinematics (IK) are not implemented. Types are defined but rendering is deferred.
Workaround: Use manual keyframing for animation.
Native CMYK document values and precision-aware working buffers are supported. Browser raster decode and Canvas2D ImageData surfaces remain on the sRGB/RGBA8 display path, while the current WebGPU preview target is also RGBA8. Profile-aware raster decode, CMYK raster-plane import, and high-precision GPU targets are still being integrated.
Workaround: Use native desktop PDF/X or print export as the print authority, and treat the browser preview as directional. Verify the exported proof when color accuracy is production-critical.
Bleed is part of the print workflow: it is configured on the page, visualized on the canvas, and honored by PDF/X exports (page boxes and trim geometry). Raster formats (PNG/JPEG) still export the artwork’s visual bounds only — the canonical page-plus-bleed bounds exist in the export plan but the standard export dialog does not use them.
Workaround: For raster deliverables with bleed, export at a larger scale and trim in post, or use the PDF/X path when the destination accepts PDF.
The native PDF/X path now emits distinct MediaBox, BleedBox, TrimBox, CropBox, and ArtBox values for the single-page production export path. Multi-page PDF execution and complete printer-mark preview are still being integrated.
Workaround: Export and inspect a PDF/X proof per page when the document contains multiple print pages.
The plain PDF export writes a MediaBox only. TrimBox and BleedBox are emitted only by the PDF/X path, so screen PDFs carry no print geometry metadata.
Workaround: Use PDF/X for any file destined for print.
Bleed and trim guides are visualized on the canvas, and trim corner marks are shown with the print guides. The full printer mark set (crop/registration marks, color bars) is rendered only at PDF/X export time — there is no canvas print-preview of the exported mark layout.
Workaround: Export a PDF/X proof to review the mark layout before sending to print.
Preflight flags documents with no bleed, bleed below the recommended minimum, or non-uniform bleed, but cannot detect artwork that stops exactly at the trim edge. A production-preflight coverage check that inspects painted content in the bleed region is a future enhancement.
Workaround: Turn on the bleed guides while designing and check that backgrounds and images reach the bleed boundary by eye.
Get started
See the contribution guidelines to help implement these features.