Skip to content

Releases

KomaShift v1.0.0 - Local Manga Reader & AI Translation

KomaShift v1.0.0 - Local Manga Reader & AI Translation Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 11 Nov 11:34
v1.0.0
4ee0b62

Welcome to the official v1.0.0 release of KomaShift! KomaShift is a fast, modern desktop manga reader, library organizer, and AI translation powerhouse that leaves your original manga files completely untouched.

What Makes KomaShift Special

1. Distraction-Free Manga Reading

  • Continuous Cross-Folder Rollover: Flip seamlessly past the final page of a chapter directly into the next folder without touching the library tree.
  • Immersive Fullscreen (F12): Hide all toolbars, panels, and window chrome with a single keystroke for an uninterrupted reading experience.
  • Smooth Asynchronous Rendering: Instant page loads with persistent zoom presets (Fit, 50%, 75%, 100%, 120%) and anti-aliased image scaling.

2. Non-Destructive On-Screen Translation

  • Instant Toggle (T key): Switch back and forth between translated dialogue and the original Japanese artwork in real time.
  • Sharp Vector Overlays: Speech bubble contours are whitened and overlaid with clean typography, dynamic line wrapping, and automatic hyphenation without altering source image files.

3. Permanent Rasterization & Export for Sharing (Ctrl+B)

  • Share Anywhere: When you want to share translated chapters with friends or upload them to web readers, click the Rasterize button or press Ctrl+B (Bake Translated Pages to Sibling Folder).
  • Standalone Image Output: KomaShift flattens the cleaned speech bubbles and rendered typography into high-quality image files inside a <folder_name>_translated sibling folder, fully readable on any phone, tablet, or website.

4. Dual Translation Pipelines

  • Token-Efficient Mode (Local Detection + Text LLM): Speech bubble segmentation (ONNX Runtime) and Japanese text recognition (Manga OCR) run 100% offline on your device. Only extracted dialogue text is sent to your configured OpenAI-compatible API (~2,502 tokens per page), running smoothly even on basic laptops.
  • Highest Quality Mode (Local ONNX Geometry + Vision LLM): Bubble contours are extracted locally to lock in precise coordinates, while the page image is analyzed by vision models (Gemini, GPT-4o) for full contextual storytelling nuance and typo-free reading (~3,702 tokens per page).

5. Background Queue Manager

  • Batch Processing: Enqueue entire chapters or specific pages for translation and rasterization.
  • Full Job Control: Real-time progress bars, pause, resume, cancel, and automatic retry handling for network timeouts.

6. Hardware Acceleration & Custom Models

  • GPU Speedup: Automatically detects and leverages DirectML (Windows), CUDA (NVIDIA), CoreML (macOS), or OpenVINO (Intel).
  • Custom Manga OCR: Point to custom HuggingFace repositories or local directories containing ONNX models directly in Settings.

7. Library Organization & Batch Tools

  • Deep Sub-folder Scanning: Recursively discovers and catalogs every image folder across massive nested archives.
  • Batch Rename: Clean up folder names in place with natural numerical sorting.
  • Recycle Bin Protection: All folder removals route safely through the Windows Recycle Bin via Send2Trash.
Assets 3
Loading