Changelog

Kaption 2.0 — public release

• -- Velopack auto-updates — delta patches install in-place, no reinstall needed
• -- Sign in with Google or Discord — accounts with per-device registration and cross-device access to your translation packs
• -- Per-game translation packs delivered via CDN — no more bundling dictionaries in the installer; machine-bound AES-256 at rest
• -- First-run EULA + crash-report opt-in — versioned so we can re-prompt on legal updates
• -- In-app Dashboard with device list, translation packs, and purchase history
• -- Differential translation workflow — 6.4 → 6.5 patch updates rebuild only the changed lines
• -- Setup Wizard links to the web walkthrough at kaption.one/how-to/region for the full "pick your dialogue region" guide with screenshots
• -- Public safety page at kaption.one/security — every release scanned by ~70 antivirus engines on VirusTotal, with SHA-256 hashes you can verify yourself plus an honest explainer of SmartScreen warnings and Trapmine’s heuristic flag
• -- Guides hub at kaption.one/guides — setup walkthrough, per-game translator pages for Genshin Impact and Honkai: Star Rail, plus safety and spoiler-aware leaks reference material
• -- Safety reassurance block on the home page — answers "will I get banned?" before you have to dig through the FAQ
• -- Send-feedback dialog — drop the developer a short note right from the app; we read every one
• -- Landing-page sticky header fixed on iOS and Telegram in-app browser — scrolled content no longer bleeds through the nav
• -- Landing page respects iPhone safe-area notch + solid fallbacks wherever backdrop-blur was doing load-bearing work
• -- Free during open beta — every tier available at zero cost while we stabilise

GPU pipeline, Setup Wizard, modular overlay

• -- DirectML GPU OCR — ONNX Runtime GPU acceleration, RTX 50-series supported via bundled DirectML.dll v1.15.4
• -- DXGI Desktop Duplication capture — GPU path replaces GDI CopyFromScreen for lower overhead
• -- Serialized matcher index (.gisub / GSMX) — 2–5 s startup vs 30–60 s rebuild on cold start
• -- AES-256-CBC + HMAC-SHA256 machine-bound file protection for translation packs
• -- Three-step Setup Wizard — region, hotkey, test translation on first launch
• -- AI auto-region detection — OCR-based spatial clustering + game-window detection + ratio fallback
• -- Per-game region profiles — capture and answer regions remembered per game
• -- Answer-region translation — second OCR region dedicated to dialogue-choice prompts
• -- Pluggable overlay card system (IOverlayCard) — quest banner, NPC info card, dialogue transcript
• -- Parallel SymSpell index build — halves index construction time on multi-core machines
• -- Global exception handlers + crash reporting via GlitchTip

Dialogue Prediction Engine

• -- Context-aware dialogue prediction — predicts next lines from game dialogue graph, serves from hot cache in <10µs
• -- In-app dialogue graph downloader — auto-fetches game data on first launch, no Python needed
• -- Contextual re-translation tool — re-translates dialogue chains with full conversation context for better grammar and consistency
• -- Dashboard prediction stats — hit rate, cache entries, active chain display
• -- Honkai: Star Rail support — full dialogue coverage with dedicated language packs

OCR Matching & Embedded Illusion

• -- SymSpell fast-path matching — ~5µs per query for exact/near matches (100x faster)
• -- OCR-weighted Levenshtein distance — reduced false matches for visually similar characters (l↔1, O↔0, 5↔S)
• -- Embedded Illusion mode — overlay positioned directly on game dialogue text
• -- DXGI Desktop Duplication capture — overlay invisible to screen capture, preventing OCR feedback loops
• -- NPC role text detection — filters out role titles (e.g. "Owner, Xigu Antiques") from dialogue
• -- Auto-hide subtitles after 2s of no new content
• -- Game UI text filter — ignores "Lv.50", HP numbers, and other non-dialogue text

Rendering & Layout

• -- Outlined text rendering — solid 2px stroke outline for readability on any background
• -- Safe display zone — max height/width constraints with auto-shrink text
• -- Layout engine architecture — swappable ISubtitleLayoutEngine interface
• -- Async dictionary loading — streaming JSON reader with progress bar, no UI freeze
• -- Dashboard tab — quick controls for OCR, region selection, and subtitle visibility
• -- Overlay capture exclusion — WDA_EXCLUDEFROMCAPTURE + MaskOverlayAreas fallback

UI Overhaul

• -- Complete settings UI redesign — modern card-based layout with accent colors
• -- Async startup — non-blocking OCR engine load with loading indicator
• -- Eager OCR preview — shows partial translations while typewriter animation plays
• -- Click-through overlay — mouse input passes through to game by default
• -- Speaker name gold coloring — matches Genshin's NPC name styling
• -- Quick-translate popup (Ctrl+Q) — one-shot OCR with auto-dismiss
• -- Dialogue log system — daily log files for matched translations

Initial Release

• -- Real-time OCR-based subtitle translation for Genshin Impact
• -- PaddleOCR neural network for text recognition
• -- Optimized text matching — n-gram indexed fuzzy search
• -- Configurable capture region with visual overlay
• -- Auto-pause when game is minimized
• -- Subtitle fade-in/fade-out animations
• -- Thread-safe configuration with JSON persistence
• -- Hotkey system — start/stop OCR, select region, show/hide, click-through