---
license: other
license_name: mixed-open-licences
license_link: LICENSE.md
tags:
  - onnx
  - browser
  - background-removal
  - ocr
  - automatic-speech-recognition
---
# Feemag Tools models

Unmodified open models used by the free in-browser tools at
[tools.feemag.com](https://tools.feemag.com). Every tool runs on the visitor's device;
these files are the only thing it downloads, and no user file is ever uploaded.

This repo only mirrors the files. Each model keeps its original licence, listed below
and in the licence file next to it.

| Folder | Model | Source | Licence | Used by |
| --- | --- | --- | --- | --- |
| `background/` | U-2-Net, FP32 ONNX (~176 MB) | [BritishWerewolf/U-2-Net](https://huggingface.co/BritishWerewolf/U-2-Net) @ `c08042b462485d8597c89d3ce831d00be8582c5c`, from [xuebinqin/U-2-Net](https://github.com/xuebinqin/U-2-Net) | Apache-2.0 (`background/LICENSE-U2Net.txt`) | [Background remover](https://tools.feemag.com/background-remover) |
| `ocr/` | PP-OCRv6 small detection and recognition (~30 MB) | Official PaddleOCR archives (`ocr/README` in the tools repo lists the URLs) | Apache-2.0 (`ocr/LICENSE-PaddleOCR.txt`) | [OCR PDF](https://tools.feemag.com/ocr-pdf), [Image to text](https://tools.feemag.com/image-to-text), [PDF to Word](https://tools.feemag.com/pdf-to-word) |
| `speech/whisper-base_timestamped/` | Whisper base, ONNX q8 (~77 MB) | [onnx-community/whisper-base_timestamped](https://huggingface.co/onnx-community/whisper-base_timestamped) @ `608c49e61301901684bc36cac8f74b95ff6b5a8e`, from [openai/whisper](https://github.com/openai/whisper) | MIT (`speech/LICENSE-Whisper.txt`) | [Audio to text](https://tools.feemag.com/audio-to-text) |

## File layout

Large files are split into 50 MiB parts named `<name>-<index>-<hash>.bin`. The JSON file
next to them (`u2net.json`, `<name>.onnx.json`) lists the parts in order with their SHA-256
and the SHA-256 of the whole file; joining the parts in order gives back the original file
byte for byte. `ocr/sha256.json` lists the hashes of the two PaddleOCR archives.

The tools load these files from one fixed commit of this repo and check every hash in the
browser before using a file.