Skip to content

Releases

FrameMind Studio 0.1.8 - UNSIGNED development preview

Choose a tag to compare

@Lukedaca Lukedaca released this 21 Nov 12:31

FrameMind Studio 0.1.8

21 November 2026 · Windows x64 · unsigned development preview · culling local-2.0

Adaptive strictness and optical defect detection

Release 0.1.8 extends the local-2.0 local culling engine introduced in 0.1.7:

  • Automatic batch-adaptive strictness (Auto) + manual override (Conservative / Balanced / Strict):
    recommendCullingStrictness inspects the set's optical blur baseline, sharp share, noise level and
    series density to pick the appropriate strictness automatically — without any fixed reject quota
    (an all-sharp set still rejects 0 photos). Switching strictness or delivery output rescores immediately
    without remeasurement.
  • Directional motion blur (subjectDirection): opticalMotionBlur reads the four orientation bins
    measured by cullingOptics.ts so linear motion blur is detected instead of masked by sharp edges
    parallel to movement. Intentional motion, creative and stacking intents remain protected.
  • Back-focus / out-of-focus subject & higher-ISO tolerance: distinguishes mild softness (Review)
    from strongly out-of-focus subjects (Reject) even beside a sharp background, removes false texture
    vetoes on blurry high-contrast frames, and raises the blur-evidence noise ceiling for realistic higher-ISO
    shooting while keeping extreme noise protected.
  • Optical series comparison: verified burst series prefer optically sharper subjects and can reject
    substantially softer alternatives directly against a sharp representative.

The bundled U²-NetP model and local optical measurements do not recognize eyes, expressions, decisive
moments or artistic intent. Limited local calibration is not general photographic-accuracy evidence.
RAW import continues to use the embedded JPEG preview, not developed sensor data.

Workflow and darkroom interface

  • Photo-by-photo focus mode: queue filter (All / Review / Reject / Keep / Undecided) with live
    counts, inline series indicator and one-click/shortcut series comparison (C), direct editor handoff
    button, shortcuts (Z or click for 1:1 on the detected subject/focus point, M for blur map,
    Backspace for step back, Ctrl+Z for Undo, ? for shortcut help), and photographic EXIF strip
    (1/1600 s · f/2.8 · ISO 800 · 200 mm).
  • Whole-set Contact Sheet (Light Table): unobstructed photos with a 2 px top status bar and matte
    metadata footer below the image; expanded series open at the top of the viewport; live ETA during
    analysis; unified amber color for Review.
  • Darkroom UI & cross-app flow: quiet obsidian active sidebar state with spectral edge hairline,
    non-clipping segmented controls, single-row editor header, non-overlapping top-right notifications,
    explicit Add to current set / Start new set choice on import, smart culling restore on launch, and
    Apply to whole set in AdjustPanel with edited indicators on the filmstrip.

Verification

  • TypeScript (npm run typecheck): passed with 0 errors.
  • Unit tests (vitest run): 32 files, 277 tests passed (including 9 new tests in
    tests/cullingStrictness.test.ts covering motion blur, back-focus, high-ISO blur, optical series
    ranking, strictness levels and recommendCullingStrictness).
  • Web/PWA production build (npm run build), desktop frontend build (npm run build:desktop), Rust
    application build and Windows NSIS packaging (npm run desktop:build) passed locally.
  • Ten current-interface screenshots (studio-v018-20261121-*.webp) were captured using
    scripts/capture-release-ui.mjs in an isolated Chromium profile with public demonstration images,
    passing heading and zero-page-error checks. They are interface evidence, not clean-machine installation
    or native runtime checks.
  • Earlier native WebView2 executable checks remain documented in release 0.1.7 and
    release 0.1.6. Clean Windows installation without an existing WebView2 runtime and general
    photographic accuracy across genres and cameras remain unverified.

Download and limits

Download v0.1.8
and verify the installer against SHA256SUMS.txt attached to that release.
The 317.75 MB (303.03 MiB) installer includes the offline WebView2 installer and the bundled culling model.

The Windows installer is unsigned. A signing certificate is not configured;
signed client distribution, clean Windows installation without an existing
WebView2 runtime and every GPU configuration remain unverified. Do not disable
Windows protection. Updates are installed manually using the installer.

The entire Studio is not fully offline: retouch/lasso models and fonts may be
downloaded, and two optional editor operations use Gemini with the user's key.
No cloud sync or public gallery hosting is added. Local data and disk originals
retain their documented boundaries.

Czech guide · English guide ·
Windows guide · Documentation index

FrameMind Studio 0.1.7 - UNSIGNED development preview

Choose a tag to compare

@Lukedaca Lukedaca released this 14 Nov 05:21
8ec4459

FrameMind Studio 0.1.7

14 November 2026 · Windows x64 · unsigned development preview · culling local-2.0

Photo-by-photo review

After local analysis, Photo by photo / Fotka po fotce opens with the current
photo, its explained suggestion and a queue of all photos. K/R/X or the decision
buttons save a manual choice and advance. Repeating a choice keeps it set.
Arrows navigate the full set; the back button and queue allow revisiting photos.
The final decision opens a summary. Reviewing again preserves manual choices.

Whole set / Celá sada opens the existing grid, profile, intent and comparison
controls. Deliver as / Odevzdám jako in the header rescores for 100%, A4 print
or web without measuring again. The sharpness map is available when optical
measurements are reliable. Detail 1:1 opens the original viewer at 100%.

Continuing from the summary confirms removal of rejected photos from the working
set before opening the editor. Cancelling preserves the set; editor Undo restores
removed photos and their culling metadata. Original disk files are kept.

Local automatic analysis

This release includes the previously unpublished local U²-NetP salient-region
model, measured original-size detail, up to four optional manual regions,
per-photo intent and verified identical-copy checks. Edge blur is measured in
pixels across the frame and object, then judged at the delivery size.
Manual decisions remain authoritative. Existing automatic results need a fresh
analysis. See the culling guide and the
dated implementation history.

The model does not recognise eyes, expressions, decisive moments or artistic
intent. Limited local calibration is not general photographic-accuracy evidence.
RAW uses the embedded JPEG preview, not developed sensor data.

Interface

  • Larger left sidebar with numbered steps and completion marks
  • Working-set overview with actual previews and keep/review/reject counts
  • New-set entry and the five workflow steps; imports still append to the working
    set, so use Clear photo set when starting an unrelated set
  • Larger import area, blurred photo backdrop, floating editor thumbnails and
    clearer panel headings
  • Consistent project/client empty states and Czech/English labels
  • Notifications no longer intercept clicks on the header controls

Verification

TypeScript, 267 tests across 31 files, web/PWA build, desktop frontend, Rust
application and Windows NSIS packaging passed locally.

Seven isolated browser checks passed: focus review, local culling, selected
regions, original-image inspection, automatic selection with real JPG fixtures,
editor history and clearing the workspace. They cover manual choices across
reruns, original bytes and pixel scale, cancellation/confirmation, Undo,
persistence and unchanged original fixture hashes. Culling checks reported zero
external requests and page errors.

The actual packaged WebView2 application passed the focus review in a separate
verified profile, including keyboard navigation, repeated choices, 100% detail,
summary, cancelled/confirmed native dialogs and editor handoff. Page errors and
external culling requests were zero.

The packaged Windows application also passed automatic selection with the real
sharp/blurred JPG fixtures: automatic object detail, blur and identical-copy
suggestions, cancelled/confirmed native dialogs, editor handoff, Undo restoring
metadata and unchanged fixture hashes. External culling requests and page errors
were zero. It used another verified isolated profile and an existing WebView2
runtime; it does not establish clean-machine installation or general accuracy.

Ten current-interface screenshots use public demonstration images in an isolated
browser profile. They are interface evidence, not clean-machine installation tests.

Download and limits

Download v0.1.7
and compare the installer with SHA256SUMS.txt attached to that release.
The 237.68 MB installer includes the offline WebView2 installer and the bundled culling model.

The Windows installer is unsigned. A signing certificate is not configured;
signed client distribution, clean Windows installation without an existing
WebView2 runtime and every GPU configuration remain unverified. Do not disable
Windows protection. This release does not install itself into an existing local
workspace; updates use the installer manually.

The entire Studio is not fully offline: retouch/lasso models and fonts may be
downloaded, and two optional editor operations use Gemini with the user's key.
No cloud sync or public gallery hosting is added. Local data and disk originals
retain their documented boundaries.

Czech guide · English guide ·
Windows guide · Documentation index

FrameMind Studio 0.1.6 - UNSIGNED development preview

Choose a tag to compare

@Lukedaca Lukedaca released this 25 Oct 15:14

FrameMind Studio 0.1.6

25 October 2026 · Windows x64 · unsigned development preview

Previous versions restored the last working photo set at startup and appended
new imports to it, but offered no general action to clear the set. This patch
adds Clear photo set / Vyprázdnit sadu below the workflow in the left sidebar.

Clearing the working set

  • Available from import, culling and the editor, with confirmation. Cancelling
    preserves the photos and their history.
  • Confirmation clears the working set and its complete Undo/Redo history,
    detaches the open project and returns to an empty import screen. Export work
    outside a saved project before clearing it.
  • Saved projects, clients and original disk files are kept. Reopening a saved
    project deliberately opens that project's photos again.
  • One IndexedDB transaction saves the empty set and deletes image records no
    saved project or any of its past/present/future history entries needs.
    Startup, ordinary saves and installation do not trigger this cleanup.
  • An import already in progress cannot add its photos back after clearing or
    switching projects. Clear remains available when only Undo/Redo history remains.
  • Save errors and Retry remain visible outside the editor. Wait for Saved locally
    before closing. If a clear fails, the previous persisted state and image records
    remain intact and can return after restart until a retry succeeds.

Verification

Checks completed locally on 25 October 2026:

  • TypeScript and 208 tests across 27 files passed. The final test run used
    node node_modules/vitest/vitest.mjs run --maxWorkers=2.
  • The desktop frontend, Rust application and Windows NSIS installer built successfully.
  • The existing browser history/persistence check passed, including save failure,
    retry, project isolation and complete browser shutdown/restart.
  • The new browser clear check passed cancellation, delayed imports, atomic
    rollback after cleanup was aborted, Retry, preservation of saved projects and
    their history image assets, a fresh import, a Redo-only set and a browser restart.
  • The same clear check passed in the installed 0.1.6 WebView2 application with
    synthetic photos and a separate profile. A separate complete process restart
    preserved the empty set, saved project and its three referenced image assets.
    Original test-file hashes were unchanged and page errors were zero.
  • All 346 files in the normal IndexedDB/localStorage workspace matched their
    pre-check SHA-256 hashes during the update and tests, before reopening the normal app.

The installed executable was tested directly rather than assuming it had the
same hash as the unbundled executable. Tauri changes its bundle-type marker when packaging NSIS.
The installed executable that passed native checks had SHA-256
16af65de3558ee00abb40d189d58a1463a2da21aa427979a3fa5bd5a49b306cd.

Native export, model inference and general photographic accuracy were not retested
for this patch. Culling remains local-1.6; retouch response and model reuse remain
as described in 0.1.5. Earlier features and evidence limits are in
0.1.4. These synthetic checks do not validate every existing project
or every Windows/GPU configuration.

Download and limits

Download the installer and SHA256SUMS.txt from
GitHub v0.1.6.
The approximately 230 MB Windows x64 installer remains unsigned and is a
development preview. A signed client release and installation on clean Windows
without an existing WebView2 runtime remain unverified. Do not disable Windows protection.

The entire Studio is not fully offline: model/font downloads and optional Gemini
editor functions retain their documented behavior. GitHub hosts source code and
downloads; no Studio hosting deployment is part of this release. Installation
keeps local data; clearing the working set requires the explicit action above.

Czech guide · English guide ·
Windows guide · Documentation index

FrameMind Studio 0.1.5 - UNSIGNED development preview

Choose a tag to compare

@Lukedaca Lukedaca released this 24 Oct 20:39

FrameMind Studio 0.1.5

24 October 2026 · Windows x64 · unsigned development preview

This patch addresses retouch startup, repeated model preparation and brush latency.
It preserves the existing models, repair resolution, mask pixels, local workflow
and culling engine local-1.6. Studio itself is not deployed to a hosting service.

  • Ready models are reused across editor openings and photos. Concurrent callers
    share one preparation; a worker crash clears readiness before retry.
  • Reading cached model bytes is shown as preparation, rather than a new download.
    Cached files and completed downloads are checked against the pinned byte size.
  • Pinned MI-GAN/LaMa weights are read in the window scope and transferred into
    the model worker. This avoids the desktop worker's cache/network path; model
    parsing and inference remain in the worker. Recovery rereads cached bytes
    instead of retaining a second full copy of the large model in memory.
  • Desktop supplies the packaged WASM binary and a Blob URL of the packaged
    standalone module. This enables WASM threads without fetching their module
    through the desktop origin. A forced one-thread recovery remains available.
  • MI-GAN starts on threaded WASM without repeated benchmark repairs. LaMa
    retains GPU support, requests the high-performance adapter and keeps CPU fallback.
    An initialized WASM runtime is reused instead of constructing an extra large
    CPU model before GPU initialization.
  • The photograph has a separate canvas surface. Brush feedback and processing
    animation no longer rescale the full original on every frame. Cursor movement
    avoids React state updates, and drawing is limited to one animation frame.
  • Soft-stroke blur runs on the stroke bounds instead of the complete mask, at
    most once per frame; releasing the pointer always commits its final pixels.
  • Context variants are planned and deduplicated before inference. The mask is
    read and labelled once, and progress shows the actual number of distinct crops.

Verification: TypeScript and 23 targeted runtime, region and pixel-preservation
tests passed. The bounded synthetic comparison is
node scripts/retouch-performance-smoke.mjs; it compares the 0.1.4 brush with
this version on a 24 MP image, checks identical final masks, model reuse and real
MI-GAN inference without external model requests. Native verification is separate
from browser evidence. These checks are not a general photographic benchmark.

LaMa computation, particularly 1024-pixel input on CPU, still needs more time than
MI-GAN. This patch removes redundant work; it does not claim instant completion
on every device. Initial model downloads still require internet.

Measured in isolated Chromium on this machine: dispatching a burst of 48 brush
movements on the same 6000 × 4000 synthetic image took 2613 ms before and 2.14 ms
after. Final mask SHA-256 matched exactly. Blur covered 134,184,960 pixels before
and 85,618 after; processing animation drew the full photo 30 times in 500 ms
before and zero after. This is a bounded event-burst measurement, not a promise
of the same speedup for normal interaction or other hardware.

Preparation from cached MI-GAN bytes took 2004 ms. Five subsequent preparation
calls together took 0.11 ms and shared the same model session. Actual first
MI-GAN inference took 2451 ms on WASM; external requests and new model downloads
were zero. Native installation and GPU behavior are verified separately.

The final native WebView2 check passed without request overrides: cached-model
preparation took 3680 ms, returning to retouch took 322 ms, and the first complete
repair/save took 7743 ms. A subsequent repair/save took 6439 ms. Inference itself
took approximately 6244 and 5054 ms. Native GPU MI-GAN was measured separately
and was slower, so the release keeps MI-GAN on threaded WASM. No universal
instant-inference claim is made.

Native cloning, PNG saving, Undo, actual MI-GAN inference and model reuse passed.
The complete imported JPEG SHA-256 remained unchanged. There were no external
requests or repeat model downloads. The check used a separate profile and kept
all 345 normal workspace files unchanged. Native export and native LaMa/1024
inference were not retested for this patch.

The local 0.1.5 update also passed: the installed executable matches the exact
native-tested SHA-256, and all 345 existing workspace files remained unchanged.

Download: GitHub v0.1.5.
The installer remains unsigned, following previous preview releases. Do not
disable Windows protection. Earlier features and limits: 0.1.4.

FrameMind Studio 0.1.4 - UNSIGNED development preview

Choose a tag to compare

@Lukedaca Lukedaca released this 24 Oct 16:35

FrameMind Studio 0.1.4

24 October 2026 · Windows x64 · unsigned development preview · culling local-1.6

This release runs locally on Windows. GitHub hosts source code and downloads;
no Studio application is deployed to Netlify or Vercel. Optional editor cloud
features and initial model downloads are documented in the READMEs.

Download the installer and SHA-256 checksum from
GitHub release v0.1.4.
The unsigned preview follows the existing development distribution policy.
Windows may show SmartScreen; do not disable Windows protection. A signed client
release and installation on clean Windows without WebView2 remain unverified.

Current verification and limits

  • The final suite passed: 206 tests across 27 files, and TypeScript passed.
  • Windows NSIS and both frontend builds passed for this version.
  • Actual 512/1024 LaMa inference passed in isolated Chromium with unchanged
    source and outside-mask pixels. Higher resolution is slower and is not a
    guarantee of better photographic reconstruction.
  • Native WebView2 verified context comparison, Cancel/Escape, Apply and exact
    Undo/Redo before the final brush additions. The later native Quick check
    verified a real clone source, a PNG working copy and an unchanged original.
    Native 1024 inference and a complete final adaptive UI run were not verified.
  • The local installed update preserved all 341 existing IndexedDB/localStorage
    files by SHA-256. This was the preceding local 0.1.3-metadata build containing
    these features, not a clean installation of the versioned 0.1.4 package.
  • Native export evidence belongs to 0.1.2; it was not repeated for this release.
    Earlier culling calibrations below are dated historical checks, not a fresh
    local-1.6 accuracy benchmark. No general photographic quality claim is made.

Updating existing culling results

Run culling again after upgrading to engine local-1.6. Older automatic results
need new measurements; manual K/R/X decisions remain authoritative. Region
selection is manual and never implies automatic face or subject recognition.

Adaptive retouch crops and optional higher resolution — 24 October 2026

  • Distant disconnected mask regions receive independent crops, preserving their
    own input scale. Nearby regions with overlapping surroundings are merged, with
    at most 16 processed regions. Connected objects are not split into tiles.
  • Balanced, Fine detail and Wider context change the surrounding scene supplied
    to the model. Optional comparison prepares up to three distinct crops from the
    same unchanged source. Only Apply changes the photograph, as one history step;
    Cancel and Escape discard the previews. Identical crops appear once.
  • Standard input remains 512 × 512. Explicit Detailed/LaMa mode offers Higher
    detail at up to 1024 × 1024; small crops stay at 512 × 512. This reuses the same
    pinned model weights, changing only input dimension metadata in memory. The
    downloaded model and cached bytes are unchanged. Automatic and Fast use 512.
  • Higher resolution uses more memory and time and is not a guarantee of better
    photographic detail. Source dimensions, lossless PNG working copies, immutable
    originals and the existing brush/lasso workflow are retained.
  • CPU runtime initialization completes before a GPU trial. A GPU timeout restarts
    retouch in a fresh CPU worker, avoiding concurrent initialization of the shared
    WASM module. Pending histogram results are ignored after leaving adjustments.

During the adaptive-retouch stage, 201 unit tests and TypeScript passed, including recovery from GPU and
initialization timeouts, concurrent waiting callers, and a separate compute limit.

The synthetic adaptive check is node scripts/retouch-adaptive-smoke.mjs --hd;
see contributor instructions. The earlier culling
verification below does not validate this retouch change.

Sample brushes and conservative culling — 24 October 2026

  • Clone copies real source pixels selected with Alt + click. Heal transfers that
    sample's fine texture while adapting its low-frequency color to the destination.
    Both retain the existing mask strength, lossless working copy and shared history.
  • Inpainting can optionally match a consistent smooth-boundary color difference.
    Strong edges and clipped colors are excluded; correction is capped and remains
    restricted to the repair mask. The option can be disabled.
  • In full-photo comparison, Mark the main subject selects a 128/256/384-pixel
    region. Its original-scale sharpness replaces background sharpness in the score.
    Weak subject detail is reviewed; it is not automatically rejected against a sharp
    background. Regions persist, can be cleared, and never override manual K/R/X.
  • A miniature signature alone no longer proves a removable duplicate. JPEG/PNG
    candidates require identical decoded source pixels, compared in bounded tiles;
    RAW candidates require identical original bytes. Differences remain for review.
  • Preview and selected-subject measurements compare noise-corrected edge energy
    at two scales. Conflicting noisy detail adds a review reason, never a sole reject.
    Culling stays local and deterministic. Older automatic results need reanalysis;
    manual decisions remain in force.

These changes have not been evaluated as a general photographic accuracy benchmark.

Full original photographs

  • Click a culling photograph to open the entire original image file, replacing
    the fixed 384 × 384 px detail viewer. Display uses the imported bytes without
    another image conversion or compression.
  • The initial view fits the complete frame without cropping. Use the wheel or
    100%, 200% and 400% controls to zoom, then drag or use arrows to pan. Physical
    pixel scaling accounts for Windows display density.
  • Select up to four frames with + to compare full photographs. K/R/X apply to
    the active frame; 0 fits the image, 1 selects 100%, and Esc closes the viewer.
  • RAW still uses its embedded JPEG, not developed sensor data. The viewer uses
    the import source rather than later editor adjustments or retouching.

Detail-guided culling in local-1.5

The nine fixed original-resolution samples now have a budget of up to four
additional samples guided by preview detail. The extra locations are measured
at original resolution and do not replace fixed coverage. They can veto a blur
rejection; fixed-grid measurements still determine scores and series references.
If only the extra locations show strong detail, the frame requires review with
an explanation of the difference between regions. Sharp background detail does
not prove that the photographer's subject is in focus.

No photographer's selection is turned into universal taste rules. Manual choices
retain priority, and an unexported photograph is not treated as a bad-photo label.
The read-only culling-selection-audit.mjs compares source basenames with exports,
uses the normal RAW preview/import path and keeps private reports under ignored output.

Verification on 21 October 2026:

  • 155 tests and TypeScript passed. The browser UI check covers a sharp patch
    between fixed samples, pervasive softness, manual overrides, profile changes,
    confirmed removal and cancellation. Culling requests and page errors were zero.
  • Web and desktop frontend builds and the Windows NSIS build passed. The same
    UI check passed in the installed WebView2 application using a separate profile.
    The local update preserved all 341 existing IndexedDB/localStorage files by
    SHA-256, and the installed executable matched the verified installer payload.
    Package metadata remains 0.1.3; this does not update the public 0.1.3 download.
  • Two RAW sets contained 146 and 81 originals with 39 matching exported frames.
    Reject suggestions among those exported frames changed from three to zero.
    Keep counts were unchanged; ambiguous detail moved to review. This is one
    photographer's selection, not a measure of general photographic accuracy.
  • The earlier 114-JPEG set changed from 94 keep / 13 review / 7 reject to
    94 / 19 / 1. All 341 original-file SHA-256 hashes across the three sets remained
    unchanged. Unexported photographs were never labelled as bad ground truth.

Earlier culling changes in local-1.4

Engine local-1.4 corrects cases where small residual edge energy prevented a
soft original from receiving a reject suggestion. A verified series can also
identify weak original-resolution detail hidden by downscaling. That path
requires matching source dimensions, enough informative samples, low noise,
independently weak candidate detail and a sharp reference. A sampled sharp patch
protects the candidate from that original-resolution rule.

Manual choices still take priority. Culling remains local and deterministic,
without AI, models, API keys or photo uploads. Suggestions do not delete original
files; removing rejects from the working set still requires confirmation.

Verification of the earlier local-1.4 update

  • 147 tests across 22 files, TypeScript, web and desktop frontend builds, and the
    Windows NSIS installer build passed.
  • Browser and native WebView2 checks verified identical original bytes, complete
    frames without clipping, physical pixel scaling, drag/wheel zoom, four-frame
    comparison and manual decisions across reruns. Browser DPR was 1.5; native DPR
    was 1.25. Culling external requests and page errors were zero.
  • A read-only run on the same 114 sports JPEGs changed suggestions from
    94 keep / 17 review / 3 reject to 94 / 13 / 7. All original SHA-256 hashes
    remained unchanged. This is one calibration set, not a general accuracy claim.
  • Native checks used a separate WebView2 profile. The local installed update
    preserved all 341 existing IndexedDB/localStorage files and matched the exact
    NSIS executable payload. This records that local update, not a public release.

Bounded automatic samples...

Read more

FrameMind Studio 0.1.3 - UNSIGNED development preview

Choose a tag to compare

@Lukedaca Lukedaca released this 20 Oct 17:29

Unsigned development preview for testing. A signed client release is not available.

Download FrameMind-Studio_0.1.3_x64-setup.exe and SHA256SUMS.txt below (about 230 MB). This is the same installer verified during the local 0.1.3 update.

FrameMind Studio 0.1.3 — 20 October 2026

Windows x64 development preview. The installer is unsigned; a signed client
release is not available. Download and SHA-256 checksums.

Changes

  • Culling cards show the full uncropped frame. Double-click for original-resolution
    detail, or select up to four frames with + and compare them side by side.
  • Choose a region in the full frame, link crop positions or move them independently,
    and inspect at physical 1:1 or 2:1 scale. Arrow keys move the crop; K/R/X apply
    to the active frame. Region metrics assist manual review without changing
    automatic suggestions. The source is the imported original or embedded RAW JPEG.
  • Repeated analysis in the same application session reuses completed measurements
    of the exact same inputs. Scores and groups are recalculated. Remeasure all
    bypasses the cache; restarting the application requires new measurement.
  • Engine local-1.3 requires preview and native-sample detail to agree before
    relative sharpness alone can reject a frame in a verified series. Native
    comparisons require sufficient evidence and matching source dimensions.
  • Grouping reuses pair evidence already collected during admission, avoiding
    a second all-pairs pass. Scene checks, group limits and manual choices remain.

Czech culling guide · English feature guide

Verification

All 144 tests across 22 files, TypeScript, web and desktop frontend builds passed.
Browser and native WebView2 checks covered four-frame comparison and its limit,
linked/independent regions, 1:1 and 2:1 display, and manual decisions across reruns.
The checks used display pixel ratios 1.5 and 1.25 respectively. Worker checks
verified pixel equality with the decoded original, source hashes, cache reuse
and forced remeasurement. Culling external requests and page errors were zero.

A read-only run measured 114 authorized sports JPEGs and preserved all source
SHA-256 hashes. The final scoring/grouping replay returned 94 keep / 17 review /
3 reject. This is one calibration set, not a general accuracy or speed benchmark;
the final replay did not decode the photos again.

The local 0.1.3 update was installed and opened. Its executable matched the actual
installer payload, and all 12 existing IndexedDB/localStorage files were unchanged
before first launch. Native comparison used a separate QA profile and identifier.
See Windows installation and verification for the evidence boundaries.

Limits and data

Fixed automatic samples can miss a small sharp subject. Culling does not recognise
eyes, expressions, the decisive moment or focus on a particular subject. Scores
are technical indicators, not probabilities of photographic quality. More labeled
shoots, cameras and real RAW sets still need calibration.

RAW uses an embedded JPEG, not sensor development. Selected region positions and
measurement cache are not persisted across application sessions. Original files
are not overwritten or deleted by culling. Keep originals and independent backups.

The whole Studio is not fully offline: retouch/lasso models and fonts download,
and optional Gemini editor operations send a photo/preview with the user's own
API key. This release does not add face/eye detection or cloud culling.

The installer includes an offline WebView2 prerequisite, making it about 230 MB.
Installation on clean Windows without an existing runtime and compatibility with
every GPU remain unverified. A checksum verifies matching bytes, not a publisher
signature. See the release assets for the exact download hash.

FrameMind Studio 0.1.2 - UNSIGNED development preview

Choose a tag to compare

@Lukedaca Lukedaca released this 15 Oct 19:56

Unsigned development preview for testing. A signed client release is not available.

Download FrameMind-Studio_0.1.2_x64-setup.exe and SHA256SUMS.txt below (~230 MB). Windows SmartScreen may warn or block installation. A SHA-256 checksum detects changed bytes; it does not replace a publisher signature. Do not disable Windows protection.

Changes

  • Adjustment previews use a cached decoded source in a worker, a maximum long
    edge of 1600 px, and the same rendering code as export. Obsolete preview
    requests are replaced by the latest one. Export uses the selected full size.
  • Adjustments, auto adjustment, cropping, retouching and original restoration
    share Undo/Redo history. One slider gesture is one step; up to 60 steps are
    retained in each direction.
  • IndexedDB stores projects, the workspace, image copies and edit history in
    the local application profile. Image assets are separate from edit metadata.
    Completed saves are confirmed; errors can be retried. Legacy JSON metadata
    remains intact, but missing legacy image bytes require reimport.
  • Retouch working copies use lossless PNG. Non-square model crops retain
    their proportions with reflected padding. ONNX runtime assets use explicit
    matching local URLs; temporary tensors are disposed after each run.
  • Culling engine local-1.2 adds sampled 384 × 384 crops at original resolution
    for larger images and compensates small image shifts when comparing series.
    Explained reject suggestions remain editable; removing files from a set
    requires confirmation and never deletes the original disk files.

Verification

TypeScript, 136 tests across 21 files, and web/desktop frontend builds passed.
Synthetic browser checks exercise preview/export parity, history, persistence,
save failure/retry, and both real inpainting models through CPU/WASM.
Retouch tests found zero changed pixels outside the patch, zero changes after
four PNG saves, and unchanged original hashes through brush, Undo/Redo and reload.

Native installation and runtime evidence is recorded in the Windows guide.
The native installer contains an offline WebView2 prerequisite; clean Windows
installation without a pre-existing runtime and a signed client release remain unverified.

Limits and data

No full RAW development, cloud synchronization, or shared public galleries.
Retouch model input remains 512 × 512; large fills are limited by that resolution.
Synthetic checks do not establish photographic quality on real shoots or
compatibility with every GPU. The verified model path was CPU/WASM.

Ordinary work is local. Retouch/lasso models download from Hugging Face and fonts
from Google Fonts. Text-based photo edits and object search still use optional
Gemini calls with the user's own API key. See the current README for data flows.

Updating preserves the application identifier and origin. Do not clear the local
profile when updating. Keep originals and independent backups: local working
copies are not cloud storage, and GitHub does not back up private photos.

FrameMind Studio 0.1.1 - UNSIGNED development preview

Choose a tag to compare

@Lukedaca Lukedaca released this 04 Oct 22:04

Unsigned development preview for testing. A signed client release is not available.

Local culling now suggests Keep, Review and Reject. Previously the automatic decision path could never return Reject. Version 0.1.1 adds explained reject suggestions for almost empty clipped previews, redundant near identical frames and substantially weaker detail than a stronger frame in a verified series. Weak measurement evidence no longer drops sharpness weight and inflates the score. Equal scores choose a stable filename rather than a random import ID.

Manual K/R/X decisions survive reruns and profile changes. Removing rejects requires confirmation and removes images from the application set; original files on disk remain unchanged. Culling still uses local pixel calculations without AI, API keys, model downloads or network requests. No percentage of images is forced to be rejected.

Validation: 113 tests across 17 files, typecheck, web and desktop frontend builds, a Windows installer build, and native WebView2 regression checks using a separate application identifier. The native test produced two automatic reject suggestions from three synthetic JPEGs, preserved manual decisions, checked cancelled/confirmed removal and stopped a 60-file run. Culling recorded zero external requests and zero page errors. This verifies the regression cases; calibration on the user's football gallery has not been performed.

Download FrameMind-Studio_0.1.1_x64-setup.exe and SHA256SUMS.txt below. The per-user Windows installer includes an offline WebView2 installer, Czech/English setup and the MIT license. The download is approximately 223 MB.

The application and installer are unsigned. Windows SmartScreen may warn or block installation. This preview is not approved for client distribution. A SHA-256 checksum helps detect changed bytes; it does not replace a publisher signature. Do not disable Windows protection or install custom trust certificates. No developer registration, account or trust-store change was made for this release.

Close Studio before updating. Imported photo objects are not automatically restored after restarting, so retain originals and export results you need before closing. The entire editor is not fully offline; some features use Gemini, while retouch models and fonts are downloaded. Installation on a clean computer without WebView2 remains unverified.

Source code and Czech/English documentation are in this repository under MIT. No website deployment is part of this release.

FrameMind Studio 0.1.0 - UNSIGNED development preview

Choose a tag to compare

@Lukedaca Lukedaca released this 04 Oct 18:48

Unsigned development preview. This is not a client release.

The application and installer in this release have no trusted publisher signature. Windows SmartScreen may warn or block installation. Do not distribute this installer to clients. Disabling Windows protection, unblocking downloaded files, or installing a custom trust certificate is not part of the installation procedure. The SHA-256 checksum detects changed bytes when compared with the published checksum; it does not replace an Authenticode publisher signature.

No signed client installer is available yet. The source now has a separate signed-release preparation command that requires an existing trusted code-signing certificate. It verifies publisher identity, Authenticode signatures and timestamps before preparing client artifacts. It does not create accounts or change Windows certificate trust. Signing alone does not guarantee immediate SmartScreen reputation. See Microsoft's explanation.

This preview uses Tauri 2, Rust and Microsoft WebView2. The per-user installer supports Czech and English and creates a desktop shortcut. It includes an offline WebView2 installer, making the download approximately 223 MB.

JPEG/PNG export and batch export use Windows dialogs. Culling uses local pixel calculations without AI, API keys, model downloads or network requests. Manual choices, reruns, cancellation and confirmed removal were tested in the installed application; original files remain unchanged.

Application validation: typecheck, 106 unit tests, web and desktop frontend builds, and smoke checks in the installed application. Culling recorded 0 external requests and 0 page errors. Export covered cancelled saving, JPEG, PNG dimensions, a two-file batch and unchanged original SHA-256 hashes. Release verification was separately tested with unsigned, correctly signed, wrong-publisher and tampered executables; actual FrameMind signing has not been performed.

Installation on a clean computer without WebView2 has not been tested. The entire editor is not fully offline: some editor features still use Gemini, while retouch models and fonts are downloaded. Imported photo objects are not automatically restored after restarting the application; retain originals on disk and export your results.

Source code: MIT. See the Czech/English README and docs/desktop.md for current behavior and limitations. The application is not deployed to a website.