v3.5.1

fix(cli): use bash shell on Windows for tar packaging step

PowerShell doesn't accept 'VAR=value cmd' prefix syntax. Set shell: bash
so EXE_EXT=.exe is set correctly on Windows runners.

SubLens v3.5.0

SubLens v3.5.0 - 项目正式更名

🎨 Branding — Project Renamed to SubLens

• 项目正式更名：HardSubX → SubLens
• 新 Logo 设计：从 H 形重新设计为 aperture/镜头造型
• GitHub 仓库重命名：Agions/HardSubX → Agions/SubLens
• GitHub Description：更新为 SubLens 品牌描述
• UI 全面更新：ToolBar / SettingsView / AboutDialog Logo → SubLens
• 导出格式元数据：ASS/SSA/LRC/SBV/JSON 等 tool 字段 → SubLens
• 所有文档链接：统一指向 Agions/SubLens

📝 Documentation

• README / docs / CONTRIBUTING / SECURITY 全部链接和标题更新

Full Changelog: v3.4.0...v3.5.0

v3.4.0 — Async I/O Performance

v3.4.0 (2026-04-14)

Performance — Async I/O

All std::process::Command and std::fs calls replaced with tokio:: async versions (non-blocking):

File	Changes
scene.rs	get_video_fps, detect_scenes_ffmpeg, get_video_info → tokio async
video.rs	get_video_metadata_ffprobe/ffmpeg, extract_frame_at_time_impl, detect_scenes_ffmpeg → tokio async
ocr.rs	save_frame_to_temp_png, process_with_tesseract → tokio async
ocr_engine.rs	ocr_image_tesseract, process_paddle_ocr, check_paddle_ocr_available, find_python_binary → tokio async
system.rs	check_single_dependency, get_tesseract_languages → tokio::process::Command
file.rs	write/read_text_file, get_file_info → tokio::fs
export.rs	export_subtitles → tokio::fs::write

Effect: Scene detection, OCR processing, and video analysis no longer block the Tauri main thread. UI stays responsive.

Commits: 78047fa + 99c07ae

HardSubX main

HardSubX main

Professional video subtitle extraction tool powered by OCR.

Downloads

Platform	Architecture	File
macOS	Apple Silicon (M1/M2/M3)	HardSubX_macos-apple-silicon.tar.gz
macOS	Intel	HardSubX_macos-intel.tar.gz
Linux	x86_64	HardSubX_ubuntu-x64.tar.gz
Windows	x64	HardSubX_windows-x64.zip

Features

• Video subtitle extraction via OCR (Tesseract.js / PaddleOCR / EasyOCR)
• Multiple export formats: SRT, VTT, ASS, SSA, JSON, TXT, LRC, SBV, CSV
• Multi-language support: Chinese, English, Japanese, Korean, etc.
• Keyboard shortcuts & theme switching
• Desktop GUI + CLI dual mode

Installation

1. Download the archive for your platform
2. Extract and run HardSubX (Linux/macOS) or HardSubX.exe (Windows)
3. macOS: run xattr -cr ./HardSubX if blocked by Gatekeeper

v3.3.1 — Architecture Refactor

🏗️ Architecture Refactor

• 共享 types + utils 模块：统一 BoundingBox / ROI / OCRConfig，消除 3 处重复代码
• TempFileGuard RAII：自动清理临时文件，video.rs / ocr.rs 共用
• ocr.rs 完全重写：使用共享 types + utils，简化 temp 文件管理
• 删除 8 个历史失败 action runs

✅ CI Status

All jobs passing: Quality / Rust Test / Build

HardSubX v3.3.0

What's New in v3.3.0

Design System v2.0

• UI Design System Overhaul: OKLCH color space, professional fonts (DM Sans/Geist), standardized animation curves
• Component Micro-interactions: Button hover, Modal backdrop blur, StatusBar pulse animation, SubtitleList skeleton screen

Performance

• mergeOCRResults optimization: ROI selector performance improvements
• Temp file cleanup: RAII guard for automatic temporary frame file cleanup

Code Quality

• CI Three-way Parallel: quality / build / rust-test fully parallel, eliminates needs dependency
• pnpm cache fix: setup-node cache conflict resolved
• Rust compilation fixes: PathBuf trait inference, dialog trait import, duplicate detect_scenes macro conflict

Dependencies

• base64: downgraded to 0.22 (0.24 does not exist on crates.io)

Full Changelog: v3.2.1...v3.3.0

HardSubX v3.2.0

Whats New in HardSubX v3.2.0

🔧 Renamed: VisionSub → HardSubX

Project renamed throughout: repository, package names, documentation, and all source code references.

🐛 TypeScript Strict Mode

Fixed all any type violations across 7 files (10+ occurrences):

• useOCREngine.ts: Tesseract module/worker/logger types
• useBatchProcessor.ts: OCREngine type
• settings.ts: Error type (unknown)
• All Vue components: strict File/ExportFormat types

🎨 Professional Logo

New SVG vector logo featuring:

• H letterform with ROI targeting crosshairs
• Cyan subtitle track bars
• Film frame accent (top-right corner)
• Deep tech-blue aesthetic matching dark theme

📚 Professional Documentation

New docs/ directory with 4 pages:

• Getting Started — install, ROI, extraction walkthrough
• CLI Reference — full command reference
• Architecture — system design + OCR pipeline

🌐 GitHub Brand

• 18 professional topics added
• English SEO-friendly description
• README version badge fixed (3.1.1 → 3.2.0)

📄 Changelog

See CHANGELOG.md for full history.

v3.1.1

v3.1.1 — 提取链路 + OCR 准确度全面升级

---

🎯 核心改进

提取链路打通

• 开始提取 按钮现在真正触发完整 OCR 流程
• 所有高级选项（多通道/后处理/字幕合并）全部生效

OCR 准确度管道

• 文字后处理：全角→半角标点转换、去除重复字符、中文常见错字修正
• 置信度校准：混合语言/超短文本/重复字符自动降分
• 字幕自动合并：基于 Levenshtein 相似度（≥80%）+ 时间间隙（≤0.5s）去重
• 场景检测：升级为 16-bin 量化直方图 + chi-square 测试，更快更稳

---

🎨 UI 全面升级

组件	改进
VideoPreview	呼吸边框 + drop overlay bounce-in + timeline 气泡
SubtitleList	置信度彩色徽章（绿/黄/红）+ 内联编辑
SidePanel	四标签设计 + SVG 环形进度 + ROI 预设卡片
BatchProcessView	双栏布局（文件列表
ToolBar	主题切换按钮 + 全 SVG 图标化
OCR Tab	准确度仪表盘 + 引擎星级评分 + 语言分组 + 高级面板

---

🔧 代码质量

• 消除全部 any 类型滥用
• switch → dispatch table（exportToFormat）
• O(n) 全量排序 → O(log n) 二分插入（addSubtitle）

v3.1.0

新增

• 导出格式扩展：TXT、Markdown (MD)
• 字幕处理工具函数
• 格式工具 getSupportedFormats()、getFormatDisplayName()

改进

• 导出选项支持 includeTimecode
• 支持 12 种导出格式

VisionSub v3.0.0

VisionSub v3.0.0

Web build - extract and run with any HTTP server.

See CHANGELOG for details.