SubForge v0.3.0

SubForge 0.3.0 — self-contained, with subtitle sync checking and cleanup.

Install

• Windows: run subforge_0.3.0_x64-setup.exe or the .msi. FFmpeg, MKVToolNix and the Vulkan loader are included.
• Linux: subforge_0.3.0_amd64.AppImage (any distro, tools included), .deb or .rpm (install FFmpeg/MKVToolNix as dependencies).
• Arch / CachyOS / Manjaro: AUR packages are prepared but not yet published; use the AppImage for now.

Highlights

• No Python required. Speech-to-text is the built-in whisper.cpp engine; subtitle conversion, retiming and translation helpers are native.
• Bundled tools. The Windows installer and AppImage include FFmpeg 9.0.2 and MKVToolNix 103.0, so there is nothing else to install.
• GPU speech-to-text through Vulkan, with automatic CPU fallback on machines without a usable GPU.
• Sync tab. Checks a subtitle against its video: duration, bad timings, and PAL/NTSC frame-rate mismatches (25↔23.976 and friends) with a one-click fix. An optional deep check transcribes three short speech windows and reports a constant delay, growing drift or the wrong episode, with the exact shift/frame-rate correction.
• Clean tab. Removes translator credits, social handles, links and channel ads from SRT/ASS/SSA/VTT with a mandatory preview. ASS styles, fonts and drawings are untouched; overwriting keeps a .bak.
• Drag and drop. Drop videos on Subtitle/Audio/Transcribe, a folder on Batch, or a subtitle on Convert/Clean.
• Online subtitle search now reports a timing verdict, and never overwrites an existing subtitle with the same name.
• Settings. API keys are stored in the OS keyring (migrated on first launch), fields show what is saved, keys are masked with a reveal button, and Subsource reports Not configured instead of Unreachable.
• End-to-end CI builds and tests the real Windows installer and AppImage on every pull request.

Upgrading

• Python and its private environment (~/.local/share/subforge/venv, %LOCALAPPDATA%\subforge\venv) are no longer used and can be deleted.
• Whisper models must be downloaded again (new GGML format). Old faster-whisper models can be deleted from the Speech to Text page.
• Saved cuda device settings are treated as Auto.

Bundled-software licences

Every bundled component's licence text ships in licenses/. The corresponding GPL/LGPL source archives are attached to this release; see THIRD_PARTY_NOTICES.md for the written offer covering the rest.

Full details: CHANGELOG.md.

SubForge v0.1.0

See the assets below to download this release.