✨ Features

• Depth of Field: mipmap-based variable radius blur driven by depth map
• Petzval Swirly Bokeh: tangential stretch on out-of-focus areas
• Rotational Blur: camera rotation around optical axis simulation
• Grain rewrite: moved to output linear space with physical Selwyn law (σ ∝ √(1-lum) × lum)
• Preview downscaled to 1024px for fast interactive processing

🐛 Fixes

• Enable depth feature in release builds (was missing from cargo-dist)
• Prevent worker thread death on panic (catch_unwind + fallback result)
• Fix depth map coordinate mapping (was using pixel coords instead of scaling)
• Remove musl target from dist (ort doesn't provide musl prebuilts)
• Configure Trunk to exclude depth for WASM builds

✨ Features

• Auto Levels: scanner-style black/white point stretch (opt-in checkbox in UI)

🐛 Fixes

• Fix BW films rendering as flat gray in Accurate mode (R/B channels were zero from single-emulsion spectral engine)
• Fix Polaroid B&W 667 all-white output (exposure_offset was 0.00)
• Detect BW films by monochrome flag instead of FilmType::BwNegative (fixes Agfa Scala 200)
• Reorder Accurate pipeline: lens effects (ChromaticAberration, MTF, etc.) now run before develop
• BW grayscale merge uses film's spectral response weights via bw_weights()

✨ Features

• Depth-aware object motion: monocular depth estimation (Depth Anything V2 Small) drives per-object motion blur.
  • Connected region segmentation by depth similarity — each object moves coherently.
  • Near objects get more motion blur, far objects stay sharp.
  • UI: Object Motion slider + depth map preview.
• process_image_with_depth() API for passing depth maps through the pipeline.

🐛 Fixes

• Fix exposure calibration: simulate_gray now uses same output model as create_output_image (density linear + tone_gamma=2.47).
• Fix ObjectMotionStage missing from Accurate pipeline.

🐛 Fixes

• Fix highlight clipping: remove redundant shoulder_softening (sigmoid H-D curve provides natural shoulder).
• Unified output model: density linear mapping + tone gamma=2.47 for both Fast and Accurate modes.
• Norm target raised 0.18→0.32 to compensate scatter/WB losses in Accurate pipeline.

🔧 Improvements

• Physically correct d_min per channel for C-41 color negative films (orange mask residual).
  • Kodak: R=0.14, G=0.16, B=0.19. Fuji: R=0.13, G=0.15, B=0.17.
  • Shadow color cast now emerges naturally from physics.

✨ Features

• Vignetting: cos⁴ lens falloff model in output space (~25% corner darkening).
• Micro motion blur: multi-frequency hand tremor simulation with Brownian trajectory.
  • UI: Motion Blur slider, seed control, trajectory + dwell weight visualization.
• Radial MTF: center-sharp, edge-soft blur (replaces uniform MTF).
• Chromatic aberration: R/B channel radial scaling for edge RGB fringing.
• Film stock analyzer: PSD shape, isotropy, signal dependency, film detection scoring.
  • Parallelized with rayon, xtask runs in release mode.

🐛 Fixes

• Self-calibrating exposure norm via binary search (consistent brightness across presets).

✨ Features

• Full-spectrum dye transmittance output (Yellow/Magenta/Cyan dye spectra × D65 → CIE XYZ → sRGB).
• Self-calibrating exposure norm via binary search (consistent brightness across all presets).
• Film stock analyzer tool (cargo xtask analyze <image>).
• Lucky Color 200 and Ricoh GR Street Night presets.

🐛 Fixes

• Fix negative film inversion in spectral dye output path.
• Fix grain model: Selwyn √D theory, color_correlation 0.93.
• Rewrite grain as physically-based texture field (multiplicative, per-layer, spatially correlated).

📝 Documentation

• Rendering Pipeline chapter with LaTeX formulas in mdBook.

✨ Features

• Default SimulationMode to Accurate (full-spectrum for all paths).

🐛 Fixes

• Fix inhibition darkening neutral gray (use density deviation, not absolute).
• Fix per-channel norm to match Fast mode channel balance exactly.

✅ Tests

• Add strict inhibition behavior tests (neutral unaffected, color separation, mean preserved).
• Replace external photo tests with synthetic scene histogram comparison.

⚡ Performance

• Optimize Accurate mode ~10x via precomputed layer coefficients (eliminates per-pixel exp() calls).

✨ Features

• Add xtask for bump, ci, publish, release automation.
• Add criterion benchmark for Fast vs Accurate mode comparison.

📝 Documentation

• Add Full-Spectrum Engine chapter to mdBook.

✨ Features

• Spectral: Add CIE 1931 XYZ colour matching functions and D65 illuminant lookup tables (380–780nm, 5nm, 81 bins).
• Spectral: Replace blackbody approximation for D65 with CIE standard data.
• Spectral: Replace Gaussian camera sensitivities with CIE XYZ CMF × XYZ→sRGB matrix.
• Film Layer: Add multi-layer film structure model (FilmLayerStack / FilmLayer) with refractive index, thickness, spectral absorption, and scattering coefficients.
• Spectral Engine: Add full-spectrum propagation engine — per-wavelength per-layer light transport with Beer-Lambert absorption, Fresnel interface reflection, and forward+backward (halation) passes.
• Core: Add SimulationMode::Fast / Accurate — Fast uses existing 3×3 matrix path, Accurate uses full-spectrum engine.
• Film: Add layer_stack: Option<FilmLayerStack> to FilmStock for custom per-preset layer structures.
• UI: Develop button triggers Accurate mode (full-spectrum); preview stays Fast (matrix).
• Spectral: Add scattering spatial diffusion — Gaussian blur derived from physical layer scatter coefficients.
• Spectral: Add interlayer interimage effect (DIR coupler developer inhibition matrix).
• Presets: Calibrated layer stacks for Kodak Portra 400, Kodak Tri-X 400, Fujifilm Velvia 50.

✨ Features

• Film: Add FilmStyle system for systematic rendering style modifiers (Accurate, Artistic, Vintage, HighContrast, Pastel).
• Presets: Add artistic variants for popular films (Portra 400, Tri-X 400, Velvia 50, HP5 Plus 400).
• UI: Add FilmStyle selector to both Simple and Professional modes with horizontal button layout.
• UI: Trigger automatic reprocessing when switching between Simple and Professional modes.

🐛 Fixes

• Presets: Fix grain alpha values using correct formula (RMS/1000)² for all 51 films (was incorrectly RMS/1000).
• Presets: Restore original curve parameters for Portra 400, Tri-X 400, HP5 Plus, and Velvia 50.
• Grain: Add visual grain boost factor (25x) to compensate for display/perception differences - grain was physically correct but visually too weak.
• UI: Load preset values on image load to fix initial grain intensity issue (first load showed extreme grain).

💄 Style

• UI: Move Rendering Style to separate section with consistent grouping.
• UI: Change FilmStyle selector from dropdown to horizontal buttons for better UX.
• UI: Force white balance Off in Simple mode during processing.

♻️ Refactor

• Film: Add helper methods to FilmStyle enum (all(), name(), description(), short_description()) for cleaner UI code.

🙈 Chore

• Git: Add .DS_Store to .gitignore.

📄 Documentation

• README: Update snapshot image.

♻️ Refactor

• IO: Separate EXIF metadata building from file I/O operations.
• IO: Unify image encoding and EXIF writing to memory (Vec<u8>) before file output.
• IO: Use write_to_vec() instead of write_to_file() for EXIF metadata to enable cross-platform support.
• IO: Remove platform-specific #[cfg] guards from build_exif_metadata() - now works on both desktop and WASM.
• IO: Consolidate file I/O to single point: std::fs::write() for desktop, handle.write() for WASM.
• UI: Split controls.rs into modular components (film_list, technical, simple).
• UI: Split app.rs into modular components (io, state, update).
• GPU: Split gpu_pipelines.rs into modular pipeline stages.
• Presets: Split into manufacturer-specific modules (kodak, fujifilm, ilford, vintage).
• App: Extract EXIF orientation utils to shared module.

💄 Style

• UI: Update window size to 1200x800 and rename title.

🔧 Fixes

• Presets: Fix imports after module refactoring.
• Presets: Fix get_stocks() function calls in all preset modules.
• Examples: Update preset function names after module split.

✨ Features

• Controls: Improve film stock list row interaction with full row hover/click highlighting, better padding, and rounded thumbnail corners in Simple mode.

✨ Features

• App: Preserve original EXIF metadata when saving images and add Filmr processing info (Software, ImageDescription, Copyright tags).

✨ Features

• App: Improve save dialog with source-based default filename ({source}_FILMR.jpg) and JPEG as default format.

🐛 Fixes

• App: Fix image::ImageOutputFormat API change for image crate 0.25 compatibility.

🐛 Fixes

• Controls: Fix gamma boost (contrast) not applied to preset thumbnails. Thumbnail worker now receives the modified FilmStock with gamma boost instead of using the original preset.

✨ Features

• Settings: Add optional histogram smoothing toggle (3-tap [1,2,1]/4 weighted average) to reduce visual jitter in shadow regions.

♻️ Refactoring

• Core: Extracted spectral matrix computation into FilmStock::compute_spectral_matrix(), eliminating duplicated spectral integration logic across pipeline and processor modules.

⚡️ Performance

• Spectral: Added SIMD (f32x4) optimization to Spectrum arithmetic operators (Add, Mul), reducing per-operation cost for spectral calculations.

✨ Features

• GPU: Enhanced develop shader with spectral/color dual matrix pipeline, shoulder softening compression, and logistic sigmoid H-D curve (replacing erf approximation) for better CPU/GPU consistency.
• Grain: Switched grain shader shadow noise model from inverse-distance to exponential decay (exp(-2D)), providing smoother shadow-to-midtone grain transitions.

🐛 Fixes

• Metrics: Fixed PSD slope calculation to use correct row-then-column 2D FFT instead of flattened 1D FFT.

♻️ Refactoring

• Core: Refactored Spectrum struct to remove Copy trait, forcing explicit data flow and improving performance by avoiding expensive implicit copies.
• Core: Improved numerical stability in new_blackbody and integrate_product functions within spectral.rs, ensuring physical consistency and preventing potential overflows.
• Core: Replaced implicit amplitude scaling with new_gaussian_normalized to guarantee energy conservation in spectral modeling regardless of bandwidth.

✨ Features

• Grain: Implemented resolution-dependent grain scaling. Grain blur radius and noise amplitude now scale automatically with image resolution (reference 2K), ensuring consistent visual graininess across different image sizes.

🚀 Features

• WASM: Implemented multi-threaded image processing using rayon and wasm-bindgen-rayon for significantly improved performance on the web.
• WASM: Added dedicated ComputeBridge and Web Worker infrastructure to handle heavy computations off the main UI thread.
• WASM: Integrated console_log for unified logging in the browser console.

🐛 Fixes

• WASM: Fixed hist_rgb serialization issue by implementing serde-big-array wrapper, ensuring correct histogram data transfer between worker and UI.
• WASM: Resolved "Parking not supported" panic by enabling parking_lot/nightly and wasm-bindgen-rayon/no-bundler features.
• WASM: Fixed worker.js module loading errors by patching import paths and removing problematic modulepreload links.
• CI: Fixed GitHub Actions workflow for WASM builds by switching to nightly toolchain and adding rust-src component (required for build-std and atomics).
• Scripts: Enhanced patch_dist.py robustness with better regex matching and environment variable support.

🚀 Features

• UI: Added "Save" and "Back" buttons in Studio Mode for improved workflow.
• UI: Optimized main layout by removing the top panel and relocating the settings button for a cleaner interface.
• UI: Improved positioning of the UX mode toggle.

🐛 Fixes

• Metrics: Fixed logic for retrieving and displaying film metrics.

♻️ Refactoring

• Core: Refactored FilmStock struct to embed manufacturer and name fields directly, simplifying the presets module and removing redundant tuple wrappers.
• Core: Optimized FilmStock usage to prefer references and moves over cloning, improving performance and reducing unnecessary allocations.

🚀 Features

• UI: Introduced Simple and Professional UX modes. Simple mode focuses on quick adjustments (Brightness, Contrast, Warmth, Intensity), while Professional mode offers full physics-based control.
• UI: Added Split-Screen Comparison view in the central panel for side-by-side before/after comparison.
• UI: Implemented async Preset Thumbnails in the controls panel for visual preview of film stocks.
• Processor: Added warmth and saturation parameters to the simulation engine.
• UI: Added persistency for UX mode preference in config.json.
• UI: Optimized top bar layout with direct mode toggles.

🐛 Fixes

• Metrics: Fixed incorrect metrics display logic and optimized metrics panel visibility for different modes.
• UI: Fixed control panel layout issues and improved visual hierarchy with better spacing and grouping.

⚠ Breaking Changes

• Core: Refactored FilmStock struct to embed manufacturer and name fields directly, removing the need for external name management. Updated get_all_stocks to return Vec<FilmStock> instead of tuples.

⚠ Breaking Changes

• Core: Refactored ReciprocityFailure model to be a struct with beta parameter, removing the description field. This restores Copy trait for FilmStock.
• Core: Removed references to internal documentation IDs ("tec n") from public API comments.

🚀 Features

• Core: Added standard color negative spectral response model (new_color_negative_standard).
• Ops: Implemented structured logging with tracing crate, replacing println! debugging.
• Docs: Established mdBook knowledge base structure in docs/.
• Docs: Added comprehensive Rustdoc documentation to public API (FilmStock, PipelineStage, etc.).

⚡️ Performance

• Bench: Added criterion benchmarks for image processing (1080p).

🐛 Fixes

• Tests: Resolved Clippy warnings and unused variables in test suites.

⚡️ Performance

• UI: Offloaded image decoding, analysis, and resize operations to a background worker thread to prevent UI blocking during file load.