Yomika

A local-first workspace for translating, cleaning, typesetting, and exporting manga.

Documentation · First-page tutorial · Issues · Discussions

English | 日本語 | 简体中文 | Português (Brasil) | Türkçe kurulum

Yomika combines text and speech-bubble detection, OCR, inpainting, translation, review, typesetting, and export in one page-aware desktop app. The vision pipeline and downloaded translation models can run on your device. Hosted providers and Codex are optional and send data only when you choose those workflows.

What Yomika Does

• Processes individual pages or multi-page projects through a staged pipeline
• Runs local vision models with candle and local GGUF language models with llama.cpp
• Supports local, hosted, machine-translation, and OpenAI-compatible providers
• Renders vertical CJK, right-to-left text, font fallback, strokes, and effects
• Uses Google Fonts plus installed OpenType, TrueType, and variable fonts
• Exports rendered images, editable layered PSD files, and Yomika project archives
• Exposes the same runtime through the desktop UI, headless Web UI, HTTP API, and MCP
• Separates model downloads from loading, with cancellation and storage controls

The desktop shell uses Tauri and Rust; its embedded interface is built with Next.js.

Installation

Windows portable release

Download either the portable .exe or .zip from GitHub Releases. The ZIP contains the same executable; extract it anywhere you can write files, then run Yomika-<version>-windows-x64.exe. Yomika does not use an installer.

Source Build Requirements

• Rust 1.95 or later
• Bun 1.0 or later
• LLVM/Clang with a usable libclang shared library
• Platform-specific Tauri prerequisites
• CUDA Toolkit 13.0 for the default Windows and Linux CUDA build, or Apple Silicon for the macOS Metal build

Build From Source

git clone https://github.com/proxlavee/yomika.git
cd yomika
bun install --frozen-lockfile
bun run build

The binary is written to target/release/yomika or target/release/yomika.exe. First launch initializes runtime libraries and downloads the default vision/OCR models. Optional local translation models use an explicit Download action and can be cancelled before you choose Load.

See Build From Source for Windows, Linux, macOS, and WSL notes.

Yomika checks GitHub for a newer release at startup and from Settings → About. An available-update notice opens the Releases page; the app never downloads or installs application updates automatically.

First Page

1. Import one or more PNG, JPEG, or WebP pages.
2. Run Detect → OCR → Inpaint → Translate → Render.
3. Review text blocks, repair masks, and adjust the lettering.
4. Export a finished image or layered PSD.

Read Translate Your First Page for the complete walkthrough.

Usage

Hotkeys

Canvas:

• Mouse Wheel: Zoom in/out around the pointer
• Hand tool + Drag: Pan the canvas
• Ctrl + Drag: Pan the canvas

Tools:

• V: Select tool
• M: Block tool
• B: Brush tool
• E: Eraser tool
• R: Repair Brush tool
• [ / ]: Decrease / increase brush size

History and selection:

• Ctrl + Z / Cmd + Z: Undo
• Ctrl + Shift + Z / Cmd + Shift + Z: Redo
• Ctrl + A / Cmd + A: Select all text blocks on the current page

For the full list and customization details, see Keyboard Shortcuts.

Desktop, Headless, and MCP

Run without the desktop window:

# macOS / Linux
yomika --headless --port 4000

# Windows
yomika.exe --headless --port 4000

Open http://127.0.0.1:4000/ for the Web UI, use http://127.0.0.1:4000/api/v1 for the HTTP API, or connect an MCP client to http://127.0.0.1:4000/mcp. See Run GUI, Headless, and MCP Modes.

Models, Providers, and Acceleration

Yomika downloads required detection, OCR, inpainting, and font-analysis models on demand. Active downloads can be cancelled, and completed model files can be deleted or downloaded again from Settings → Runtime. The model library can use its default app-data location or a folder you choose. Translation can use a local GGUF model or a configured OpenAI, Gemini, Claude, DeepSeek, DeepL, Google Cloud Translation, Caiyun, LM Studio, OpenRouter, or other compatible endpoint. API keys are stored through the platform credential store.

Codex image generation is a separate opt-in workflow that sends the source page and prompt to the ChatGPT Codex backend. Use the staged local pipeline when pages must remain on-device.

See Models and Providers and Acceleration and Runtime for supported models, provider setup, and fallback behavior.

Export and Project Files

• Rendered export produces a finished, flattened page.
• PSD export preserves editable text and helper layers for manual cleanup.
• .ymk archives move complete Yomika projects between installations.

See Export Pages and Manage Projects.

Development

bun install --frozen-lockfile
bun run dev

Before opening a pull request, run the complete verification suite:

bun run verify

While iterating, use bun run verify:rust or bun run verify:ui to check one side of the workspace.

Read CONTRIBUTING.md before submitting changes.

Troubleshooting

Use yomika --debug (or yomika.exe --debug) for detailed startup, download, GPU, and model logs. The troubleshooting guide covers first-run downloads, CUDA fallback, source-build failures, headless access, and exports.

Contributors

Thanks to everyone helping improve Yomika.

License

Yomika is licensed under the GNU General Public License v3.0.