Normal Translation

"Normal Translation" is the default choice of the "Translation Workflow Mode:" combo box on the translation page and the only one of the nine workflows that runs the full translation chain. Use this mode when you want to detect text on images, recognize and translate it, inpaint the original text areas, and render the translation; the other eight modes skip most of these stages, see Output Directory and Workflow for the overview.

This guide focuses on the inputs, full processing stages, skip conditions, and output files of the normal mode. Adding files, list states, and drag-and-drop are covered by File List and Input; start, stop, cancel, and progress states by Progress, Stop, and Task State; and the parameter algorithms of each stage by the corresponding settings pages (Detection, OCR Filter and Merge, Translation, Mask and Inpainting, Typesetting and Rendering, Upscale and Colorization, CLI Batch and Output).

When to use it

Input: main input images added with "Add Files", "Add Folder", or drag-and-drop. When a folder is added, the supported image extensions are discovered recursively, collected in natural sort order, and directories named manga_translator_work are skipped.
Processing stages: conditional colorization → conditional upscaling → detection → OCR → textline merge → translation → mask refinement → inpainting → rendering → saving the main output image.
Skip conditions: no textlines after detection, no text after OCR, empty regions after translation, cancellation, and AI-renderer inpainting skip, as listed under "Skip conditions" below.
Outputs: the main output image; the project JSON and inpainted image when cli.save_text (the "Editable Image" UI setting) is enabled; and an editor base image when colorization or upscaling is enabled.
Normal mode is the only one of the nine workflows allowed to enter the batch_concurrent concurrent pipeline; the other eight are treated as incompatible in both the desktop controller and the core translate_batch() and run non-concurrently.
This guide does not explain the parameter algorithms of individual detectors, OCR engines, translators, inpainting models, or renderers; those belong to the settings pages. The workflow combo box, output-directory controls, and the mutually exclusive writes of the nine modes are covered by Output Directory and Workflow.

Run this workflow

Add inputs and select the workflow

Open the "Translation" tab. The header title defaults to "Normal Translation", with the subtitle "Tip: Standard translation pipeline with detection, OCR, translation and rendering".
Click "Add Files" or "Add Folder" to add images, or drop files/folders onto the page; click "Clear List" to clear the current list.
In the "Translation Workflow Mode:" combo box, confirm that "Normal Translation" is selected. Changing the combo box first clears all eight mutually exclusive workflow fields, then sets and saves only the field for the selected mode; normal mode leaves all eight fields false.
Enter a path in the "Output Directory:" field or drag an output folder into it; the placeholder is "Select or drag output folder...". Click "Browse..." to choose a directory, or "Open" to open the selected directory with the operating system.
Click "Start Translation" to start the task. While running, the button text changes to "Stop Translation"; clicking it again requests a stop, and the button then enters the "Stopping..." state.

Processing order

Main pipeline

Normal mode processes each image in the order below; whether colorization, upscaling, or inpainting actually happens is decided by the corresponding parameters, not forced by the mode. When detection finds no textlines or OCR recognizes no text, the image returns early and the remaining stages are skipped.

flowchart LR
    Input["Main input image"] --> Colorize{"colorizer.colorizer\n!= none?"}
    Colorize -- "no" --> UpscaleQ{"upscale.upscale_ratio\nset?"}
    Colorize -- "yes" --> Colorized["Colorize"]
    Colorized --> UpscaleQ
    UpscaleQ -- "no" --> Detect["Detection"]
    UpscaleQ -- "yes" --> Upscaled["Upscale"]
    Upscaled --> Detect
    Detect --> HasLines{"Textlines found?"}
    HasLines -- "no" --> SkipRegion["Skip: output input/upscaled image\n(skip-no-regions)"]
    HasLines -- "yes" --> OCR["OCR"]
    OCR --> HasText{"Text recognized?"}
    HasText -- "no" --> SkipText["Skip: output input/upscaled image\n(skip-no-text)"]
    HasText -- "yes" --> Merge["Textline merge"]
    Merge --> Translate["Translate"]
    Translate --> Mask["Mask refinement"]
    Mask --> AIQ{"Renderer is\nAI renderer?"}
    AIQ -- "yes" --> RenderBase["Skip inpainting: work image as render base"]
    AIQ -- "no" --> Inpaint["Inpaint"]
    RenderBase --> Render["Render"]
    Inpaint --> Render
    Render --> Output["Main output image"]
    Render -. "save_text or text_output_file" .-> Json["Project JSON"]
    Inpaint -. "save_text" .-> Inpainted["Inpainted image"]
    Colorized -. "colorize or upscale enabled" .-> EditorBase["Editor base image"]
    Upscaled -. "colorize or upscale enabled" .-> EditorBase

The diagram expresses the stage order, skip branches, and output branches; it does not claim every run passes every stage. colorizer.colorizer=none, an empty upscale_ratio, no text, an AI renderer, and cancellation each take their documented bypass. The inpainted and editor base images are written only under the listed conditions.

Skip conditions

Condition	Trigger point	Result
No textlines after detection (`textlines` empty)	After detection in `_translate_until_translation()`	Progress status `skip-no-regions`; result set to the input or upscaled image; OCR, translation, mask refinement, inpainting, and rendering are skipped
No text after OCR (`textlines` empty)	After OCR in the same function	Progress status `skip-no-text`; result set to the input or upscaled image; translation and rendering are skipped
`text_regions` empty after translation	`_complete_translation_pipeline()`	Progress status `error-translating`; result set to the input or upscaled image
Translation returns `cancel`	Same function	Progress status `cancelled`; result set to the input or upscaled image
`renderer` is an AI renderer (`openai_renderer` / `gemini_renderer`)	Inpainting step	Inpainting is skipped; the work image is used as the render base
Mask empty or all zero	Inpainting step	Inpainting is skipped; `img_inpainted = img_rgb`
`revert_upscaling=true`	Before saving	Progress status `downscaling`; the result is resized back to the input size

"Input or upscaled image" means that on an early no-text return ctx.result = ctx.upscaled, and with revert_upscaling enabled it is then resized back to the original size. cli.skip_no_text (the "Skip Images Without Text" UI setting) is a stored CLI field; the the current code shows no consumer of it on the main translation path. The built-in early exit for textless images is triggered by the skip-no-regions / skip-no-text states; whether the two interact remains a runtime item.

Outputs and file writes

Main output image: determined by _calculate_output_path(), which preserves the input folder name and relative hierarchy under the output directory; with save_to_source_dir=true (the "Save to Source Directory" UI setting) the output moves to manga_translator_work/result/ next to the source image; an empty or none cli.format keeps the original extension, otherwise the configured extension is used. Saving goes through _save_and_cleanup_context().
Project JSON: when save_text (the "Editable Image" UI setting) or text_output_file is enabled, manga_translator_work/json/<stem>_translations.json is written with regions, original_width, original_height, the mask_raw mask (when save_mask is enabled), upscale/colorization information, and rendered-state flags; this is the basis for later "Import Translation and Render" and editor write-back.
Inpainted image: when save_text is enabled and the image has an img_inpainted, manga_translator_work/inpainted/<stem>_inpainted.<original-extension> is written; when an AI renderer skips inpainting, the saved file is the work image.
Editor base image: when colorization or upscaling is enabled, manga_translator_work/editor_base/<original-filename> is written as the editable base for the editor.
cli.overwrite (the "Overwrite Existing Files" UI setting): when false, the GUI filters files before starting by checking whether the main output image already exists; if all files are skipped, the task ends before translation and asks the user to delete same-named files or enable overwrite.

Inputs, outputs, and limitations

batch_size and batch_concurrent: normal mode is the only workflow allowed to use the concurrent pipeline; the other eight modes are treated as incompatible in both the desktop controller and the core translate_batch() and are forced to run non-concurrently. Concurrency does not mean all images request the API at once; stage-level parallelism, backpressure, and failure isolation are covered by CLI Batch and Output.
cli.save_text: controls project JSON and inpainted-image writes in normal mode as well; the default is true.
Whether colorization, upscaling, detection, OCR, inpainting, and rendering actually run is decided by their parameters (for example colorizer.colorizer, upscale.upscale_ratio, render.renderer); see the settings pages.
Context pages, glossary, replacement rules, API-candidate rotation, and retries affect translation and rendering quality but do not change the stage order of normal mode.
The exact dialog text for a missing, unwritable, or unopenable output directory has not been runtime-checked; static conclusions are not presented as runtime results.

Other workflows: Export Original Text · Export Translation · Translate JSON Only · Import Translation and Render · Colorize Only · Upscale Only · Inpaint Only · Replace Translation
Selecting a workflow, output directory, and mutually exclusive writes: Output Directory and Workflow
Inputs, skipped stages, and outputs of all nine workflows: Workflow Matrix
Mutually exclusive workflow fields, parameter overrides, and template alignment: Mode-Specific Workflows and Template Alignment

See the reference index: Workflow Matrix.

Normal Translation ​

When to use it ​

Run this workflow ​

Add inputs and select the workflow ​

Processing order ​

Main pipeline ​

Skip conditions ​

Outputs and file writes ​

Inputs, outputs, and limitations ​

Read next ​