What this guide covers
OCR turns image shapes into text; it does not retrieve an original text layer. OKYA Watch uses a local recognition engine after loading its browser code and required language data. The interface language and OCR language are independent. A French interface can recognize Japanese text if the task selects the Japanese model. Select according to the source, not according to your preferred navigation language.
Set it up
Create a text task, select the visible region and choose its OCR language. Start with a readable, stationary phrase and wait for the first engine or language-data download to finish. Subsequent use may benefit from browser caching, but clearing site data, changing profiles or cache eviction can require another download. Do not describe the first visit as a fully offline experience.
Check the result
Compare the result character by character, including zeros, capital letters, punctuation and accents. Recognition confidence is a model estimate, not proof that a value is correct. Increase source font size, improve contrast and reduce background clutter before adding broader automation. For consequential numbers or names, retain a human review step even when several sample readings agree.
Before you rely on it
The product offers English, simplified Chinese, Spanish, Portuguese, Japanese, Korean, German, French, Italian and Dutch OCR options. Availability in the underlying OCR project does not mean every language supported by that project is exposed in this product. Do not infer handwriting, document layout reconstruction, PDF parsing or audio transcription from a successful screen-text demonstration.
Troubleshooting
A loading failure is different from an empty recognition result. Check the network while models initialize, browser storage constraints and whether the task was canceled. A clear preview with persistent spelling errors suggests language or image-quality problems instead. Retry with one short line to separate model setup from the complexity of the original screen.
Product reference
Models are cached by the browser; task settings and recognized history are separate browser-local data. Downloading a model is not uploading your screenshot. However, selecting cloud translation or external notifications changes the destination of recognized text. Review those actions independently instead of assuming that every later operation shares the OCR engine’s local boundary.
The interface and OCR both support English, simplified Chinese, Spanish, Portuguese, Japanese, Korean, German, French, Italian and Dutch. OCR language models are downloaded on demand the first time they are used and cached in this browser; interface and OCR languages can be configured independently.
Further reading
- Extract subtitles already visible in a video
- Color changes and reference-image matching
- Privacy boundaries and local versus cloud translation