A drag tick requested the loupe's draft and the frame ended; the draft,
done in 1.5-3 ms on the GPU, was picked up at the start of the next frame,
a frame after the slider moved. Now, in a drag, the loupe waits up to 40 %
of a frame for its draft when the last loupe render took well under that
(not with a render offload, nor zoomed in, where what changes on screen is
the window's draft). A CPU render, which would only make the frame late, is
not waited for.
JobPool::wait waits for the next finished job, which try_recv then hands
out; Renderer::poll's collection of finished jobs is split out so the wait
can turn the draft into the texture.
Co-Authored-By: Claude <noreply@anthropic.com>
Each drag tick changed the edited photo's settings, so the filmstrip asked
for a new thumbnail render every frame. For an unedited raw shown with its
embedded preview, the thumbnail-level source had never been decoded, so each
of those renders decoded the whole raw (a 64 MP ARW: over a second), several
side by side, and the loupe's drafts queued behind them.
- The edited photo's thumbnail is not re-rendered during a slider drag while
it has one; it follows once the drag ends.
- A thumbnail of a photo whose preview-level source is decoded (the one in
the loupe) is cut from that source (SourceRef::Fit) instead of decoding the
file again; not when the photo has a denoised picture.
Co-Authored-By: Claude <noreply@anthropic.com>
A drag renders drafts, which have no HDR histogram, so an HDR edit's histogram dropped to the
SDR layout until the drag ended. Now, during a drag of an HDR edit, a small full-quality render
(Slot::HdrHistogram, 192 px, its own stage cache) runs beside the drafts and feeds the panel at
its own pace: the drafts don't wait for it. The panel also keeps an HDR edit's last HDR histogram
when a render without HDR bins arrives (a draft, a cached thumbnail or stand-in), so it never
swaps to the SDR layout mid-drag.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The idle trim (3 s without renders) released the shared device copy of the original a zoom
window is cut from, so pausing between pans made the next pan upload it again: 800 MB for a
64 MP raw, ~1.4 s of "gpu sample" measured (5 ms when it is still there). While a Detail window
texture is held the renderer is not idle; once the windows are gone (zoomed out, another photo,
or another view, which now frees the Detail view's windows) the trim runs as before.
A photo whose render failed while the loupe showed its embedded JPEG or a thumbnail kept
that stand-in up as if it were the photo; "This photo can't be opened" only showed when
nothing at all could be drawn. Now a caution pill at the "Rendering…" spot says so:
"Original offline · no smart preview" when that is why, else "This photo can't be opened",
with the reason on hover and for screen readers (widget `loupe:renderFailed`). Agents see
it as `ui.inspect` → `loupe.failed`, so one waiting for a finished loupe knows when there
won't be one.
A library change retries failed renders; while the retry of the same key runs, the pill
stays and "Rendering…" doesn't take over. The renderer remembers which photo a failure
belongs to (`failure_of`): the centre message no longer shows the previous photo's error
for a moment after a step. Tests share one render offload (`headless::RenderGate`) that
holds renders and can fail chosen slots.
Co-Authored-By: Claude <noreply@anthropic.com>
A panic inside RenderJob::run or QuickJob::run (a raw decoder or pipeline bug) unwound the
job pool's worker without sending a result. The slot stayed pending and its key was never
asked for again, so the loupe or the thumbnail rendered forever; and the pool never
replaces a dead worker, so after as many such panics as it has workers (2–6) no photo
rendered until LightCraft was restarted.
Both now catch the panic (guard::catch) and return it as the job's failure, answering the
request it came from, so the renderer records it like any failed render: no retry until
the catalog changes, and the workers live on. run() is the one place every render goes
through (the pool, export, Build Previews, contact sheets, face and AI-mask workers);
callers that had their own guard keep working, and the watermark preview's, now
redundant, is gone. The web build aborts on panic, so nothing changes there.
Co-Authored-By: Claude <noreply@anthropic.com>
On merge: the quick preview path orients like the whole-file path (preview_orientation with the
raw's header prefix, so a portrait raw whose embedded JPEG carries no orientation is not sideways),
and its doc links point at preview_window.
Compare and Reference drew each pane from a whole-frame render of at most
2560 px, so 1:1 was a magnified preview and focus could not be checked.
Each pane now also renders the window on screen from the photo's own
pixels (Slot::CompareRegion), as the loupe does (#323). The media cache
keeps up to two full-size originals while either view is open, so one
pane's window does not evict the other's source.
Docs: gaps.md entry 4 shrinks to overlays and soft proofing; checklist
LR-VIEW-ZOOM / LR-VIEW-COMPARE and ui-parity zoom row updated.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UyHnpE9573a7nJFQqnGNex
The grid already showed the camera's embedded preview first, then queued a render of the whole raw
to replace it. For a 512 px thumbnail that render decodes 50 of a 56 MB Fuji RAF's bytes for a
picture the 4 416 × 2 944 embedded JPEG already covers — 45 s a photo on the 1.2 MB/s share this was
measured on, 8.3 GB of traffic for a 149-photo folder, and a queue that delayed the render the user
was actually waiting for (measured: the first loupe render of a cold photo finished 174 s after the
click, most of it spent behind thumbnails). Idle for a minute afterwards, the same folder read 12
previews (26.9 MB) and decoded zero raws for thumbnails, against 638 MB of whole-file reads before.
- `Session::embedded_thumb_is_final` answers whether a photo is an unedited raw whose camera JPEG the
viewer is showing — the same `has_import_look` gate the preview path uses.
- `grid::skips_the_real_render` is the decision (unit-tested): a cached render as stand-in, an edited
photo and a raw without an embedded JPEG all still render, because their thumbnail must show the
pipeline's output rather than the camera's.
- `Renderer::remember_embedded_thumb` records the quick render's key and the bucket, and
`thumb_quick_is_final` / `thumb_quick_pending` let the per-frame grid request stop at the embedded
preview instead of rebuilding the render job (and its cache key) every frame.
- The thumbnail key still follows the settings hash, so editing a photo re-renders its thumbnail
through the ordinary background path.
`docs/raw-preview-reads.md` and the `LR-IMP-PREVIEW-READ` parity row record the behaviour and the
numbers.
Co-Authored-By: DeepSeek V3.2 <noreply@deepseek.com>
A render still running on a worker thread while the process exits crashes in the
driver (SIGSEGV in the NVIDIA Vulkan driver on Linux), where neither the panic
hook nor catch_unwind helps: pool workers were detached, the devices are statics
that are never dropped, and nothing on the way out waited for GPU work.
- gpu: every entry into a device (render, device creation, pool trim, denoise
runner and tiles) is counted as work in flight (exit.rs). begin_shutdown()
lets nothing new in; wait_idle(timeout) waits for what is inside.
- preview: JobPool keeps its workers. shutdown(timeout) and Drop discard the
queued jobs and wait, bounded, for the running ones (2 s on drop); never for
the calling thread, and not twice.
- engine: a render job caught by the shutdown gives up instead of rendering on
the CPU, so quitting doesn't wait for it.
- app: on_exit ends with LightcraftApp::shutdown (2 s for everything), after
settings and library are saved; the macOS menu-bar Quit goes through on_exit.
Not changed: the export, merge, denoise and import threads are still detached;
their GPU work goes through the gate, their CPU work is cut off at exit as before.
Tests: pool drop waits / is bounded / survives a panicked job and its own job
dropping it; the gate; GPU renders refused and the device not created after
begin_shutdown; the app's shutdown leaves no render worker. The crash itself
needs Linux + NVIDIA and was not reproduced here (Windows, DX12): 9 of 9 quits
of the app mid-render exit with code 0 in under 0.8 s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Previews were always sRGB, which wide-gamut monitors show oversaturated.
Settings > Display > Choose Profile... (and the UI command
app.displayProfile) now shows photos through the monitor's ICC profile,
matrix/TRC or LUT-based.
- codecs: DisplayProfile (moxcms): the display's primaries (colorant tags,
or measured through the A2B tables of a LUT-only profile), the proxy ->
device correction (per-channel curve tables for matrix/TRC profiles, a
CMS transform for LUT profiles) and sRGB <-> display transforms.
- pipeline: DisplaySpace / RenderRequest::display: the per-pixel stage
renders into the display's primaries and gamut maps into its gamut
(CPU and GPU: only to_out / out_luma change). Soft proofing's display
gamut warning uses the display's gamut.
- engine: RenderJob::with_display; display::present converts sRGB
results (thumbnails, cached previews, embedded JPEGs). Histograms and
the preview caches stay sRGB; exports are unchanged.
- ui: the loupe, Before, Compare, hover and second window render for the
display; everything else is converted on the render workers. Settings >
Display tab, ui.json setting, file picker opening in the system's ICC
folder.
Tests: profile parsing and transforms (sRGB, P3, gamma 2.2, a synthetic
LUT-only profile, corrupted files), display renders equal 8-bit renders
into the matching output space, sampled histograms, and a headless UI test
of the whole flow. docs/display-profiles.md describes the design.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WEuxxPyG8t6KcADbZSEAUj
Keep current upstream UI behavior and the later detector contract checks. Validate candidate models before replacing working weights and records, with a synthetic regression that failed before the fix. Remove the unrelated GPU equivalence-test relaxation. All seven quality gates pass with the pinned hosted fonts.
Generated with Codex.
A review measured 1 GB of GPU stages for a Before/After at 1:1 on a 2800 px
canvas against a 400 MB budget, with nothing bounding it. The open photo's window
and before caches are now cut (the before side first) past four times the budget,
the whole-frame view's never; idle caches still go at the budget. The one shared
device copy of a big original is reported (sharedSourceBytes in the renderer's
memory report) and a misleading comment about it is corrected.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A review measured the drag at 100 % on a 2800 px canvas taking 87 ms a tick
against 24 ms with a bigger budget: the open photo's caches were over the budget
(a 5 MP window alone is 230 MB) and a "hard" budget cleared them every frame, so
every tick rebuilt them. The whole-frame, zoom window and before caches are what a
drag reuses; only idle caches (hover, compare tiles, second window) are trimmed
now, the least useful first, and the open photo's are counted and reported. The
drag test no longer waits on a tick count.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A drag supersedes every request before it finishes, and the source a superseded
draft had decoded was thrown away with it, so a window cut from a full-size
original decoded it again for every tick. A view slot's superseded draft that is
newer than what the view shows now also hands its source to the session (late
results of other slots stay ignored, as before). Render results become textures
with a plain copy for the opaque case instead of egui's per-pixel premultiplying
conversion, on the UI thread.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The Before image was a stand-in capped at 2560 px beside an After at full
resolution, so a 1:1 comparison misled. Both sides now get the same treatment:
the same whole-frame size and, once zoomed, a window each (Slot::RegionBefore
beside Slot::Region) cut from the same frame, in side by side, top/bottom, the
\ original view and both wipes (the Before window is clipped to its side of the
line). Each side's windows and tiles are tracked apart, a side is freed when it
is no longer shown, and ui.inspect reports regionBefore.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The loupe's, the zoom window's and the before / hover renders' stage caches hold
decoded-image sized buffers that the engine's budget did not count. A quarter of
the memory budget is now theirs: each frame the renderer clears what exceeds it,
the least reusable first (hover, before, compare, second window, the zoom window)
and the open photo's own view last (stage_trim_order, pure and tested). The
inspector's memory report shows the budget and how often caches were trimmed.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
WireJob carries the request's window (an old job without one has none), and the
browser build asks for window renders like the desktop; windows there are capped
at 3072 px a side to respect wasm memory.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
An adversarial review found that a slider drag at any zoom started a window
render (the 0.6-scale draft looked stretched), that a tile made for an earlier
look kept being drawn and that the tile table was cleared mid-drag, that the
fixed 256 px margin was far below the wide stages' kernels at depth, that spot
removal reads outside the window, and that the window texture stayed resident
after zooming out. Window renders now wait for the drag to end, a tile is drawn
only while photo, zoom and look match, the margin grows with the zoom (768 px
cap) and the window is capped at 6144 px, photos with spots keep the whole-frame
render, the texture and stages are freed when unneeded, and wasm never asks.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
When the whole-frame render is smaller than the zoomed view (a 1:1 or 8:1 view
of a large photo, or a capped Preview size) the loupe also requests a window
render (Slot::Region) of the pixels on screen plus a margin, and draws it over
the whole-frame render at its place of the frame. Stateless, grid-snapped
windows mean small pans reuse the render. Not used in the browser build, while
comparing, hovering a look, soft-proofing or showing an overlay. ui.inspect's
loupe reports the window.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
#243 dropped every result whose request id wasn't the newest for its slot. The loupe requests a new
draft every frame while a slider moves, so when a draft took longer than a frame nothing landed
until the drag paused. Main, Before and Hover now take a superseded draft that is newer than what
they show (never one from before a library or preview-cache reset, never an error or its source).
Also saturate the xtask RAM rounding.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An unedited raw first shows its embedded preview (ThumbQuick), then the
grid queues the real thumbnail at BACKGROUND_THUMB_PRIORITY. The grid asks
at its own priority (10 on screen), so thumb_current found no match while
the stand-in ran (nothing pending for the real slot) and while the real job
was queued (pending at 3, asked at 10): request_thumb rebuilt the job every
frame without changing anything. thumb_inputs now records whether the
request went through a stand-in; such a request is current while its
stand-in is pending or its real job is pending/rendered at the background
priority. Direct requests keep their priority check, so a higher priority
still promotes a queued job.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Changing the disk cache size (library.preferences cacheMb) re-attached a
new PreviewCache object, and the renderer treats a new cache object as a
library reset: every texture, the loupe included, was dropped and
re-rendered. Thumbnails are keyed by content, so a new budget leaves them
valid. The budget now changes in place on the attached cache (same object,
same generation); the next write trims to it, as before. A preview clear,
a generation bump and a new library still invalidate.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Skip unchanged thumbnail jobs and reuse bounded settings hashes (both maps
capped at 1024 entries). Reject stale render results and cache writes after
edits, library changes and preview clears: render requests carry a request
id and cache generation, PreviewCache writes and reads hold the generation
lock that clear() takes, and library.clearPreviews also drops cached
sources. The web worker index follows the same generation rule.
The decoder guards, Windows path fixes and test fixture changes that were
bundled in a327b98 are separate commits; the benchmark follows this one.
Co-authored-by: Codex <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review follow-ups on the contributor's click-zoom:
- Ratios 1:1 (default), 2:1, 3:1, 4:1, 8:1; set through the new
view.clickZoom {ratio} command (bottom-bar dropdown, CLI, MCP).
- Z / Space zoom to the same ratio as a click, with the same ease.
- The loupe requests its render at the final size; the animation only
changes how it is drawn (no render per animation frame). The test
checks every animation frame asks for one render.
- docs/parity.md: LR-VIEW-ZOOM.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The active photo, fitted, in its own native window (an egui immediate viewport) with its own
render slot sized for that window; closing the window turns it off. Without native windows
(web, headless) egui shows it as a floating panel.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- library.missing lists photos whose file isn't where the library expects it; photo.relink
points one at its new location; library.findMissing {folder} relinks every missing photo
whose file (same name and size) is somewhere in a folder, in one undoable step. New
Op::Relink: like a rename in the catalog, but undo/redo never move files.
- UI: File → Find Missing Photos…, photo menu → Locate Missing File… (when the file is gone).
- Renderer: a remembered failure is retried once the catalog changes (a relinked file).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- library.browse {path, subfolders}: a folder's photos are read in place as "local" photos
(catalog flag, SetLocal op) that only folder views list — not All Photos, albums, date
groups, keyword counts or stats. photo.addToLibrary promotes them; importing the folder for
real promotes instead of reporting duplicates. The browsed folder persists with the view.
- UI: Local section (Pictures, Desktop, Downloads, Home, the browsed folder, Browse Folder…);
folder views get a breadcrumb header (each part opens that folder), Include subfolders and
"Add N to My Photos".
- Renderer: a job that fails (unreadable / missing file) is remembered by slot + key and not
requested again every frame — the grid re-queued it forever (a render loop per broken
file). The loupe says "This photo can't be opened" with the reason; grid cells show "!".
ui.inspect lists pending render slots.
- Headless app.export (MCP, lightcraft-cli run) exports the selection like the desktop app.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
library.importPreview scans/probes (probes reused by the import) and marks duplicates;
library.import gains preset, keywords and albumName and is one undo step; File > Add Photos
opens a review grid (thumbnails, checkboxes, duplicates unchecked), add in place or copy into
the dated Originals folders, batched import with progress; engine + headless tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Settings (Cmd+, / app menu on macOS, Edit elsewhere) with General, Import,
Performance and Interface tabs. App settings (startup view, delete
confirmation, GPU, preview size, filmstrip/grid badges, last library) live in
UiState.settings (ui.json); library settings (raw / non-raw / per-camera
import presets, XMP, thumbnail cache size) go through library.preferences into
prefs.json. Import applies the matching default preset; the photo remembers it
as its import look (unedited; Reset returns to it). Open Library… switches
libraries at runtime.
View modes: F full-screen preview (photo on black, arrows step, I cycles the
info overlay, Esc exits), Cmd+Shift+F window full screen, Cmd+I cycles the
loupe info overlay (off / file + date / exposure), Navigator mini map while
zoomed (click/drag pans).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- memory::budget (default min(25% RAM, 1.5 GiB); LIGHTCRAFT_MEMORY_MB, app.memoryBudget):
half for engine caches (sources + rendered previews, LRU across them via a global
use tick; the photo on screen stays), a quarter for decodes in flight (WorkGate:
thumbnails, prefetch and import probes wait; loupe/export loads never wait), an
eighth for the GPU free pool.
- GPU: best-fit buffer reuse (<= 25% slack), buffers released outside a render are
pooled at once (no stranded retired lists), pool trimmed after 3 s idle.
- Decodes drop file bytes and raw samples early, colour/crop/orient in place.
- macOS: apps install malloc_zone_pressure_relief as the release hook (after each
decode, import and idle trim): freed large blocks no longer stay resident.
14 raws + loupe + 12 steps (M4 Pro): peak RSS 1.9-2.3 GB -> 0.83-0.99 GB, footprint
2.5-2.6 GB -> 0.91-0.96 GB, heap peak 930 -> 656 MB; stepping median ~13 ms.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Once the current photo's loupe render is done, the next and previous photos in filmstrip
order get a background loupe job (Slot::Prefetch, priority below on-screen thumbnails,
once per settings): the decoded 2560 px source is kept in the engine's bounded preview
LRU and the view render is cached, so stepping to a neighbour shows its developed image
in ~50 ms and then re-renders only on the GPU.
- lightcraft_gpu::warm_up (called when the app is created) builds the device and kernels on
a background thread; gpu::ready() answers without blocking and ui.inspect uses it, so the
UI thread never waits for device creation.
- docs/gpu-pipeline.md: opening-a-photo section.
Stepping through 24 MP raws in the loupe (real app): full render 220–580 ms → 45–62 ms.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- engine: QuickJob — what to show the moment a photo opens: the loupe's last render for the
current settings (kept by Session::loupe_job under a size-independent "view" key in the
preview cache, memory + disk with a library; the disk JPEG is written off the render
thread via PreviewCache::put_deferred), else the embedded camera JPEG of a raw that still
has its import look (import::has_import_look), else a thumbnail-level render.
- Grid / filmstrip: an unedited raw without a thumbnail shows its cached thumbnail, else its
embedded preview (Slot::ThumbQuick) at once; the real rendered thumbnail follows at
background priority (same cache keys, so it still ends up matching the develop render;
edited photos render directly).
- Loupe (detail, compare, filmstrip selection): Slot::Preview holds the stand-in until
Slot::Main has the photo; the full render then replaces it in place (no blank frame).
- files::embedded_preview_srgb + PreviewLoader hook (desktop/CLI/MCP and web).
- ui.inspect → loupe {photo, source: render|cached|embedded|small|thumb|none, pending}.
Real app (24 MP raws, loaded machine): first image 15–90 ms (thumbnail, then embedded
preview), full render 0.2–0.5 s (was 1.0–2.4 s with nothing shown before it).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Renders run in dedicated module Web Workers, each a second instance of the app's wasm module
(index.html compiles it once and posts the WebAssembly.Module to worker.js). A job crosses as
JSON (WireJob: source, settings, request); the worker reads originals and cached thumbnails from
browser storage itself, keeps decoded sources and per-view stage caches, and transfers RGBA back.
Jobs follow the worker that decoded the photo; the loupe waits for it. No wasm threads or
SharedArrayBuffer needed; falls back to inline rendering if no worker starts (?workers=0).
Thumbnail disk cache on the web: thumbs/<key>.jpg in OPFS/IndexedDB with an index on the main
thread, pruned to the desktop budget (2 GB → 80 %, LRU).
ui-egui: RenderOffload hook in the Renderer (queue/priorities/dedup stay in the UI).
engine: RenderJob.origin + MediaCache::origin_ref; preview: encode_jpeg/decode_jpeg public.
xtask web: copies worker.js; --serve sends COOP/COEP/CORP (documented in docs/web.md).
Headless Chrome, cold load + grid scroll, 24 demo photos: main-thread busy 12.4 s → 0.7–1.1 s,
long tasks 23 (max 1.1 s) → 0; grid filled in 6.3–8.4 s vs 13–16 s; slider job 4 ms.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Integrated with M4 optics/Upright: render_impl resolves Upright and uses frame_for (lens + auto CA)
on the cached source; defringe runs in the cached linear stage and optics settings are part of its key.
Stage-cache equivalence test extended with optics/geometry steps.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- ui-egui/softpaint: pure-Rust rasterizer for tessellated egui output (gamma-space premultiplied
blending like egui-wgpu, bilinear texel-centre sampling, scissor clips, tie-break rule so shared
edges blend once, parallel row bands, deterministic) + CPU texture mirror from TexturesDelta.
- ui-egui/headless: HeadlessView (offscreen context with our fonts/theme) and Headless (windowless
app session: logic+ui per frame, answers viewport Screenshot/InnerSize/Close, settle loop,
control requests through the shared handler).
- ui.screenshot {headless: true} in the running app renders from the logic tick (works while the
window is occluded / display asleep); windowed captures fall back to headless after 2 s.
Screenshots now wait for in-flight renders, not just queued ones.
- lightcraft-cli snapshot [--demo|--library DIR] [--script f.jsonl] -o out.png --size WxH --scale S.
- Fixes found with it: dialog backdrop was painted over the dialog itself (now an area below the
window); dialog content padding; crop overlay options use our buttons (registered widget ids).
- Docs: control-protocol.md (headless section), AGENTS.md (prefer headless when unattended).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- pipeline: StageCache + render_cached: resampled source, white-balanced/retouched/denoised image
and each spatial plane (log L, guided base, clarity, texture, dark channel + airlight) are keyed
by exactly their inputs; keeps a draft-size and a full-size entry warm.
- Exposure moved after the spatial filters (applied per pixel; planes shifted by EV): identical up
to rounding (test: <= 2 LSB vs. exposing the source), so exposure drags hit the cache too.
- engine: RenderJob::draft / with_stages; ui: loupe and before views own a StageCache, loupe
renders at Draft quality while interacting.
- engine example render_bench (min-of-N wall + process CPU time).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
