Skip to main content
Select a text layer on the canvas or in the inspector. Source analysis, semantic text, and visual layout are separate, so you can correct OCR without rebuilding the text box.

Correct the text

A text layer holds source text, a translation, typography, writing direction, and either its own frame or an automatic fit to a detected region. Source text and translation are edited in the expanded layer row of the inspector. Edit source text when OCR is wrong. Edit translation for meaning, tone, names, or line breaks.
Koharu renders translated text only. A source-only layer does not automatically show its OCR text on the canvas or in exports.

Manage layers

The layer view follows the page hierarchy: groups, text, raster layers, and image or artwork layers. The list selects one layer at a time; select several layers on the canvas with a modifier click or Edit → Select All Layers. Use the arrow buttons to reorder a layer among its siblings, toggle visibility, change opacity, and delete a layer with its descendants. Analysis regions describe model findings. They are not visible presentation layers and do not compete with artwork or text during selection.

Review a page

1

Check detection and OCR

Verify the text regions, then correct source recognition errors.
2

Review the translation

Check meaning, tone, terminology, and names against the source.
3

Inspect the artwork

Look for remnants of source lettering and damage around cleanup regions.
4

Check the reading experience

Adjust typography and fitting, then inspect the composite at normal reading size.

Undo and redo

Use Edit → Undo or Redo. The defaults are Ctrl+Z and Ctrl+Shift+Z on Windows/Linux, or Command+Z and Command+Shift+Z on macOS. Each page-stage result that the pipeline commits is its own undo step, so undoing a project run takes one step per page and stage. Closing and reopening a project loads the saved state without restoring the previous undo stack.