Automatic subtitle generator with word-by-word alignment, inline editing, and styled video rendering.
A standalone desktop app built with Electron + React + TypeScript on top of a Python FastAPI backend powered by WhisperX and NVIDIA CUDA.
| Platform | Installer |
|---|---|
| Windows | CapForge-Setup-2.4.0.exe |
| macOS | CapForge-2.4.0.dmg |
See the changelog for what's new in each version, or all releases for older builds.
The installer is ~155 MB. On first launch a setup wizard downloads the embedded Python runtime and the Whisper model (large-v3-turbo, ~1.6 GB) into %APPDATA%\CapForge\.
- Word-level timestamps — every word gets a precise start and end time
- Speaker diarization (optional) — identifies who said what (via pyannote-audio). Requires a free Hugging Face access token and one-time gating acceptance for pyannote/speaker-diarization-3.1 and pyannote/segmentation-3.0.
- 99 languages with auto-detection
- GPU auto-detection — recommends the optimal model size and precision based on your VRAM
- Inline subtitle editing — Text view (click-to-edit, search, split/merge, keyboard nav) and Groups view (drag, merge, split, reorder words, inline speaker editing)
- Per-word style overrides — set color, size, font, animation, or position per word (right-click a word in Groups view)
- Timeline retiming — drag caption blocks on the canvas timeline with edge snapping, an adaptive ruler, and hover tooltips; waveform and timeline zoom/scroll stay in sync
- Autosave & crash recovery — the live session is autosaved to app data; if the app closes or crashes unexpectedly, it offers to restore your work on next launch
- Video / audio preview — synced waveform, video playback, canvas timeline, live caption overlay
- Multiple animations — Fade, Slide, Pop entrance + Highlight, Underline, Bounce, Scale, Karaoke, Reveal word styles
- Custom Render — full control over resolution, fps, format, mode, bitrate
- Quick Render — one-click MP4 (baked) or MOV (transparent overlay) at source resolution + 40 Mbps
- Multi-format export — SRT (word-level & standard), VTT, JSON, CapForge project file
- Custom fonts & presets — import
.ttf/.otfand save complete style presets - Project save/load — save transcription + edits + style settings to a
.capforgefile - Light & dark themes — full theme support across UI and canvas timeline
┌─────────────────────────────────────────────────────┐
│ Electron Shell │
│ electron/main.js · preload.js · python-manager.js │
│ runtime-setup.js · update-check.js │
├─────────────────────────────────────────────────────┤
│ Renderer (Chromium) │
│ React 19 + TypeScript + Tailwind v4 │
│ WaveSurfer.js · Canvas timeline · CSS theme tokens │
├──────────────┬──────────────────────────────────────┤
│ REST / WS │ http://127.0.0.1:53421 │
├──────────────┴──────────────────────────────────────┤
│ Python Backend (FastAPI) │
│ WhisperX · pyannote · Pillow · FFmpeg │
└─────────────────────────────────────────────────────┘
Electron spawns the Python backend on startup. The renderer talks to the backend via REST (commands, edits, exports) and WebSocket (live progress for transcription and rendering). Audio and video files are streamed back through GET /api/serve-audio.
| Layer | Technology |
|---|---|
| Desktop Shell | Electron 33 |
| Renderer | React 19, TypeScript 6, Tailwind CSS v4 |
| Build | electron-vite 5 + Vite 7 |
| Audio / Waveform | WaveSurfer.js 7.12 |
| Backend | Python 3.11 (embedded), FastAPI, uvicorn |
| ASR Engine | WhisperX, faster-whisper, CTranslate2 |
| Diarization | pyannote-audio, speechbrain |
| ML Framework | PyTorch 2.6.0 + CUDA 12.4 |
| Video Render | Pillow 11, FFmpeg 8.1 |
| Packaging | electron-builder 26 (NSIS / DMG) |
- Python 3.11 (or 3.10) for the dev backend
- Node.js 22.12+
- NVIDIA GPU with CUDA 12.4-compatible driver (≥ 550), or CPU fallback
- FFmpeg on PATH (for video rendering in dev mode)
# Clone the repo
git clone https://github.com/FRSname/CapForge.git
cd CapForge
# Python venv for backend dev
python -m venv .venv
.venv\Scripts\activate # Windows
# source .venv/bin/activate # Linux / macOS
# Install backend deps — order matters, see DOCS.md "torch install order trap"
pip install whisperx fastapi[standard] uvicorn[standard] websockets pillow
pip uninstall -y torch torchaudio torchvision
pip install torch==2.6.0 torchaudio==2.6.0 torchvision==0.21.0 --index-url https://download.pytorch.org/whl/cu124
# Node deps
npm installnpm run dev:react # electron-vite dev — Electron + React HMR + backend
npm run typecheck # tsc --noEmit
npm run backend # standalone backend (no Electron)npm run build:react # build renderer + main + preload
npm run dist:win # NSIS installer
npm run dist:mac # DMG
npm run dist:dir # unpacked build (debug)- Drop a file — drag any audio/video into the app, or click to browse
- Configure — pick language (or auto-detect), toggle speaker diarization, adjust model
- Transcribe — WhisperX runs the full pipeline: load → transcribe → align → diarize
- Review — synced video, waveform, and canvas timeline with caption overlay
- Edit — Text view for line-by-line editing, Groups view for merge/split/drag/per-word style overrides
- Style — Custom Settings sidebar: typography, colors, layout, animations
- Render or export — Quick MP4 / MOV, Custom Render with full control, or SRT / VTT / JSON
- Save project —
Ctrl+Swrites a.capforgefile with everything to resume later. Your work is also autosaved continuously and offered for recovery after an unexpected close
| Shortcut | Action |
|---|---|
| Space / K | Play / Pause |
| J / L | Scrub back / forward |
| ← / → | Step one frame |
| , / . | Previous / next caption |
| Tab / Shift+Tab | Next / previous segment |
| M (Groups view) | Merge focused group with the next |
| Enter (Groups view) | Split focused group in half |
| Ctrl+Z / Ctrl+Shift+Z | Undo / redo edits |
| Ctrl+S | Save project |
| Ctrl+O | Open project |
| Enter | Start transcription (file screen) |
| Escape | Cancel job |
| Ctrl+Wheel (timeline) | Zoom |
| Wheel (timeline) | Pan |
| Format | Extension | Description |
|---|---|---|
| Word SRT | .srt |
One word per entry with precise timestamps |
| Standard SRT | .srt |
One sentence per entry |
| WebVTT | .vtt |
WebVTT format |
| JSON | .json |
Full transcription data |
| CapForge | .capforge |
Project file: transcription + edits + style settings |
| Mode | Format | Use case |
|---|---|---|
| Baked | MP4 (H.264) | Subtitles burned into the source video |
| Overlay | MOV (ProRes 4444) or WebM (VP9 alpha) | Transparent subtitle layer for compositing |
Quick Render uses the source resolution + fps + 40 Mbps. Custom Render exposes resolution presets (1080p / 4K / portrait / square), 24–60 fps, and a bitrate selector.
The app auto-detects your GPU and recommends settings:
| VRAM | Model | Compute |
|---|---|---|
| ≥ 10 GB | large-v3 | float16 |
| ≥ 6 GB | large | float16 |
| ≥ 4 GB | medium | int8 |
| ≥ 2 GB | small | int8 |
| CPU | base | float32 |
CapForge/
├── backend/ # FastAPI Python backend
│ ├── main.py # REST + WebSocket
│ ├── engine/ # WhisperX pipeline + GPU detection
│ ├── exporters/ # SRT, VTT, JSON, CapForge, video render
│ └── models/schemas.py
├── electron/
│ ├── main.js # Electron main process
│ ├── preload.js # contextBridge (window.subforge)
│ ├── python-manager.js # backend lifecycle
│ ├── runtime-setup.js # first-launch installer
│ └── update-check.js # GitHub release check
├── src/renderer/src/ # React renderer (TypeScript)
│ ├── App.tsx
│ ├── main.tsx
│ ├── components/
│ │ ├── screens/ # DropZone, Progress, Results
│ │ ├── player/ # AudioPlayer + canvas timeline
│ │ ├── editor/ # SubtitleEditor, GroupEditor, WordStylePopup
│ │ ├── studio/ # StudioPanel, ExportPanel, CustomRenderPanel, …
│ │ └── ui/ # ColorSwatch, FontPicker, Toggle
│ ├── hooks/ # useWaveSurfer, useTimeline, useRender, …
│ ├── lib/ # api, render, project, presets, fonts, groups
│ ├── styles/globals.css # Tailwind v4 theme tokens (light + dark)
│ └── types/ # app types + global.d.ts (window.subforge)
├── resources/
│ ├── bin-win/ · bin-mac/ # bundled FFmpeg binaries
│ └── python/ # embedded Python archives
├── DOCS.md # detailed technical documentation
├── DEVELOPMENT_PLAN.md # roadmap
└── package.json
See DOCS.md for the deep architectural and packaging notes.
MIT