Yomi Ninja v0.2.1 (Patreon)

Hello everyone! We have a new release!
Version 0.2.1 introduces new features, including one that might assist those facing compatibility issues with specific hardware configurations.
Additionally, Version 0.3.0 is available for Early Access tier members, featuring built-in popup dictionaries and the ability to install dictionary extensions originally designed for web browsers.

TL;DR:

• Option to invert captured image colors: Great for text on a dark background.
• WebSocket support for texthookers.
• Option to alter the OCR inference runtime.

Download Yomi Ninja v0.2.1

Invert Captured Image Colors

This feature can significantly enhance recognition accuracy in specific scenarios, especially when text is placed on a dark background.

Let's look at an example:

Without color inversion, the recognized text is completely nonsense.

However, with color inversion enabled, the results are notably improved:

This new option can be found in the settings menu, and it's recommended to enable it when working with light text on a dark background.

WebSocket for Texthookers

This feature was requested by a user named MobApache on GitHub. It allows YomiNinja to send OCR-extracted text via WebSockets, effectively reducing latency when using texthookers like Texthooker UI.

To use this feature, simply set the texthooker to connect to "ws://localhost:6677".

Option to Change the OCR Inference Runtime

Some users have reported difficulties with OCR functionality.

If you're experiencing this issue, please consider changing the inference runtime and letting us know if it resolves the problem. Additionally, provide information about your hardware, which will help us work on a more effective solution.

I've conducted numerous tests to reproduce the issue on my devices, but with no success. Thanks to the patience of our supporter, Risho, who assisted in extensive debugging, we discovered that the problem is likely caused by incompatibility with hardware I don't have access to.

Another user recently reported a similar issue, using hardware closely resembling Risho's.
These issues seem to primarily affect Ryzen 7000 series CPUs. As all my CPUs are Ryzen 5000 series, I couldn't replicate the problem. Risho managed to resolve the OCR issue by switching the inference runtime from OpenVino to ONNX. As a result, I've added a runtime selector to the settings menu.

Subscribe to the Early Access tier and get the latest features and bug fixes before the public release. Your support allows us to keep improving Yomi Ninja.

Early releases  |  Public releases