---
license: gpl-3.0
tags:
  - manga
  - comic
  - text-detection
  - ocr
library_name: pytorch
---

# lingva-models

Model files used by [lingva-image-server-pub](https://github.com/MuShanYu/lingva-image-server-pub),
the self-hosted image analysis service for the LingVa manga translation extension.

This repository is a **mirror**. The files are unmodified copies of models published elsewhere,
hosted here so that servers in mainland China can download them through hf-mirror.com
(GitHub release downloads are unreliable there).

## Files

| File | Size | SHA-256 | Original source |
|---|---|---|---|
| `comictextdetector.pt` | 79,948,869 bytes | `1f90fa60aeeb1eb82e2ac1167a66bf139a8a61b8780acd351ead55268540cccb` | [zyddnys/manga-image-translator](https://github.com/zyddnys/manga-image-translator) release [`beta-0.3`](https://github.com/zyddnys/manga-image-translator/releases/tag/beta-0.3) |

`comictextdetector.pt` is the text detection model from
[dmMaze/comic-text-detector](https://github.com/dmMaze/comic-text-detector), as packaged by
manga-image-translator. It detects text regions and produces a text mask for comic and manga pages.

## Usage

The server downloads this file automatically on first start (`scripts/download_models.py`),
pinned to a fixed commit of this repository and verified against the SHA-256 above. In mainland
China, set `HF_ENDPOINT=https://hf-mirror.com` (the install script does this when the China mirror
option is chosen).

Manual download:

```bash
curl -L -o comictextdetector.pt \
  https://huggingface.co/MuShanYu/lingva-models/resolve/main/comictextdetector.pt
sha256sum comictextdetector.pt
```

## License

GPL-3.0, following the original projects. All credit for the model goes to the authors of
comic-text-detector and manga-image-translator.