Export Original Text

Use this mode when you want to export the text on the original images as text, translate it manually outside the app, and render it back later. This is the export-original workflow among the nine workflows; its UI name is “Export Original Text”. It runs only conditional colorization/upscaling, detection, OCR, and text-line merging on each main input image, then skips translation, inpainting, and rendering and writes the per-image project JSON and <stem>_original.<template-format> original-text template. After translating those templates manually, Import Translation and Render reads them back and renders.

See Output Directory and Workflow for the nine-mode overview and output-directory settings, and Mode-Specific Workflows and Template Alignment and CLI Batch and Output for the cli.template and cli.save_text parameters.

When to use it

• This guide focuses on the “Export Original Text” workflow (combo index 2) among the nine workflows. Selecting it clears all eight mutually exclusive workflow fields and sets only cli.template=true; at runtime the export branch is entered only when cli.save_text=true as well.
• Inputs are the main input images and a readable translation template; outputs are the project JSON and the original-text sidecar, with no main output image. Each image's work directory is keyed by its <stem> without the extension.
• This guide does not repeat the parameter algorithms of detection, OCR, colorization, upscaling, masking, inpainting, or rendering; selecting a workflow is not translator selection or API-candidate rotation (see Translator selection).

Run this workflow

Select the Export Original Text workflow

1. Open the “Translation” page and click the “Translation Workflow Mode:” combo box in the “Translation Task” card.
2. Select “Export Original Text”. The UI sets only cli.template=true, clears the other seven workflow fields, and saves the configuration; the title changes to “Export Original Text” and the subtitle shows the matching tip.
3. Click “Generate Original Text Template” to start the task. Switching modes does not start a task automatically; while a task is running, the button shows states such as “Stop Translation”; see Progress, Stop, and Task State.

imagename in the tip is the program's example name for the input <stem>, not a private user file name; manga_translator_work/originals/ is a fixed subdirectory under each image's work directory.

Processing order

The “Export Original Text” mode enters the export branch only when template=true and save_text=true (is_template_save_mode in source). Core translate_batch() forces batch_size=1 for per-image write-out and lists this mode as incompatible with batch_concurrent; the desktop controller also sets the concurrency local variable to false for this run. The high-quality translator flow is skipped for import/export modes as well.

flowchart LR
    Input["Main input image"] --> Pre["Conditional: colorize -> upscale"]
    Pre --> Detect["Detection"] --> OCR["OCR"] --> Merge["Text-line merge"]
    Merge --> Export["Export handler: mask refinement as needed -> save project JSON -> generate original-text template"]
    Export --> Tpl["_original.<format> original-text template"]
    Export --> Json["_translations.json project JSON"]
    Tpl -. "after manual translation" .-> Import["Import Translation and Render"]

The diagram expresses only the stage order: _translate_until_translation() runs conditional colorization/upscaling, detection, OCR, and text-line merging; _handle_template_and_save_text() then refines the mask as needed, saves the JSON, and generates the original-text template. When there are no text regions, an empty JSON and an empty template file are still written.

Input and discovery rules

• Main inputs must be images supported by the file service. Adding a folder searches recursively in natural order and skips directories named manga_translator_work. Archive and document extensions are recognized by the same service, but archive-sidecar pairing for this workflow may vary by release.
• Each image's work directory is keyed by the input's original path and its <stem> without the extension: the original-text sidecar is written to manga_translator_work/originals/<stem>_original.<template-format>.
• A readable template is required; when config/translation_template.json is missing or unreadable, output_format falls back to json.
• With detector.import_yolo_labels=true and imported labels present, detection uses the imported boxes directly and marks the run as “template mode”.

Skipped and kept stages

• Skipped: translation, inpainting, rendering, and main-output-image saving; no translation service is called, so no API translation requests are produced.
• Kept: conditional colorization → conditional upscaling → detection → OCR → text-line merging; mask refinement runs when non-empty regions and an original mask exist.
• Exception: imported-YOLO-label export modes skip mask refinement and do not save a mask in the JSON.
• Boundary: the GUI sets only template; the configuration defaults set save_text=true. If an external configuration sets save_text=false, the export branch is not entered and the actual fallback path may vary by release.

Mask and JSON details

• The project JSON is written by _save_text_to_file() to manga_translator_work/json/<stem>_translations.json (the new location is preferred; the legacy image-directory location is the fallback).
• Export Original Text writes skip_font_scaling=false into the JSON so a later import-and-render run re-runs smart typesetting instead of inheriting old font sizes; because translation did not run, region translation values are still the originals.
• The refined ctx.mask is saved with mask_is_refined=true; without a refinement result, mask_raw is saved with mask_is_refined=false. Imported-YOLO-label export modes do not save a mask.
• generate_original_text() renders each region as an <original> line per the template, using the original text as a placeholder when translation is empty; with no text regions it logs and creates an empty file.

Output files

Output	Path	Notes
Project JSON	manga_translator_work/json/<stem>_translations.json	Regions, dimensions, mask, and export markers; read by import-and-render
Original-text template	manga_translator_work/originals/<stem>_original.<template-format>	Extension comes from the template output_format; defaults to json; an empty file is created when there are no text regions
Main output image	not written	Rendering is skipped, so no main image is produced
Editor base image	manga_translator_work/editor_base/<original-filename>	Written conditionally only when colorization or upscaling is enabled

Inputs, outputs, and limitations

• Requires cli.save_text=true and a readable template; incompatible with batch_concurrent (both the frontend and core process it non-concurrently), and Export Original Text additionally forces batch_size=1.
• With cli.overwrite=false, the existing <stem>_original.<template-format> is checked before starting; existing files are skipped and recorded as skipped.
• Shares the template and JSON write path with Export Translation, and pairs with Import Translation and Render as “export original → manual translation → import and render”; see Import Translation and Render.
• The display name describes the goal and does not automatically enable or disable colorization/upscaling models; the colorizer and ratio are still governed by colorizer.colorizer and upscale.upscale_ratio.

Read next

• Other workflows: Normal Translation · 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.