Preserve filter lookahead alongside main image-cache invalidation and input
bounds. Move HeldFields into its own module and finish scoped cleanup.
Charge repeated USD path names before allocating retained strings.
Reuse the blur transpose-back buffer and release scattered texels before
allocating filter temporaries.
Validated compose/engine release tests, Clippy, layering, wasm and formatting.
Preserve main's decoder, atomic-save, material-payload, and USDC limits.
Bound material reads by the remaining payload budget and trim retained
buffers. Replace quadratic image-name searches with map lookups, and
cache validated linked-material maps before synchronization so they
are not decoded twice.
Add import/undo/redo and payload-boundary regressions, extract
ImageIdentity, remove the unchecked scratch example, and complete
scoped cleanup. Correct the integrated style guide's product name.
Validated 546 release tests, Clippy, layering, wasm, formatting, and
focused allocation and lifetime probes.
Fields kept for the coming composites stayed allocated for as long as a
stroke lasted while tiles still followed, because the end-of-pass cleanup only
runs when none do. A stroke already kept them from being used; now it also
drops them. The test keeps fields first, then starts the stroke.
A stack with a neighbourhood filter (blur, high pass, bevel, glow...) made a
recomposite spread over frames many times slower than the same recomposite in
one call: every chunk filtered its own surroundings again, and on a machine
with many threads the filter's per-texel parallel steps cost more to hand out
than to run on a region of a few tiles.
- Neighbourhood filters split their per-texel work into tasks of at least
16384 texels. Each texel is computed on its own, so results do not change.
- Session::update_for tells the compositor which dirty tiles follow the chunk
in its block (up to 64). The compositor filters the chunk and those tiles at
once and keeps the fields for the composites that follow, compared by
everything the stack reads, and drops them at the end of the pass, after an
edit, and under a live stroke.
Results are identical to the one-call composite: tests cover cold, warm and
partly evicted hot-layer caches, an edit mid-pass and a live stroke.
perf::bench_sliced_filters_4k times both paths.
Applying a material filled the model in tile by tile, and each apply made the
next one slower. Now:
- a big recomposite first draws a coarse look of every tile (sampled texels
repeated across blocks), then refines to the exact result;
- layers hidden behind an opaque layer are skipped (not when anchors or height
masks are involved);
- compositing and upload budgets adapt: larger when idle, smaller while the
user is interacting, and uploads reserve time for conversion;
- no coarse pass during a live stroke or with neighbourhood filters;
- blending at full weight returns the blended value exactly.
Tests: coarse-then-exact, hidden-layer culling gives identical pixels, coarse
pass covers every tile, no coarse pass while a stroke is live.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- Rendering mode traces on the GPU (compute shaders, ~25x the CPU tracer) with CPU fallback;
CLI/MCP render --samples uses the GPU unless --cpu
- Depth of field in the viewport and the path tracer; multiple cameras with glTF import;
transmission/IOR, coat normal and screen-space subsurface blur; exact 8-bit normal decode
- Filter library (18 filters incl. warp, color balance, gradient map, posterize, MatFinish)
with a Filters shelf tab; geometry decals; clone from the current layer; anchors inside a
content stack; instances across texture-set resolutions; drag saved smart materials
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- UV tiles: auto-detected UDIM texture sets as one atlas per set, padding kept inside tiles,
one file per tile with the $udim export token, builtin:udim_plane
- USD: usda/usdc/usdz import (Crate reader with LZ4) and export with UsdPreviewSurface
- Shader lobes: clear coat, sheen, anisotropy and subsurface scattering channels in the
viewport and the path tracer; car paint, velvet, brushed steel, skin and wax materials;
glTF clearcoat/sheen export
- Color swatch popups no longer stay open
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Compositor caches the stack below and replays what lies above the layer being painted,
also inside folders: painting under a smart material is ~10-25x faster per frame and the
stroke commit 12-30x faster, bit-identical to a full composite
- Paint shows in the frame it is drawn; parallel stamp rasterizing and stroke commit;
camera visibility reused between strokes
- 67 procedural brush alphas with an Alphas shelf tab; drag materials, smart masks and
alphas from the shelf onto the layer stack, layers and viewport
- Export output template editor: user presets with per-channel packing, converted maps,
import/export as JSON, validated at load
- Shift+right drag turns the environment; S+drag no longer orbits while moving the stencil
- Frame profiler and scripted paint test on the control channel (debug.frames, debug.paintTest)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An open-source, clean-room implementation of a Substance 3D Painter-style texturing
workflow, written in pure Rust with egui and wgpu.
- Layer stack with 32 blend modes, masks, generators, content effects and anchor points
- Painting, projection, smudge, clone, path, lasso fill, particle brush and quick mask
- 43 procedural materials, 17 smart materials and 10 smart masks
- Mesh map baking with high-poly, cage and match-by-name support
- PBR viewport with HDR environments, TAA and a path-traced Rendering mode
- glTF, FBX, COLLADA, OBJ, PLY and STL import; texture export presets; OBJ and GLB export
- Dockable Painter-style UI, shortcut editor, web build, MCP server and CLI
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>