Skip to content
Manga Translator Wiki

English

简体中文
简体中文

English

简体中文
简体中文

Appearance

Inpaint Only ​

Use the Inpaint Only workflow when you only need to erase the text areas from the artwork and keep a clean background image, without recognizing text content, translating, or typesetting and rendering. It still runs conditional colorization, conditional upscaling, detection, text-line merge, mask refinement, and inpainting, but skips OCR, translation, and rendering. The branch clears the text regions before finishing, so the main output is a clean image without translated text.

Inpaint Only, Colorize Only, and Upscale Only are image-only bypass modes; the difference from the full pipeline is covered in Normal Translation. The overall boundaries of the nine workflows are in Output Directory and Workflow, with a summary table in Workflow Matrix. The mask and inpainting parameters themselves are documented in Mask and Inpainting.

When to use it ​

  • Inputs: the main input images (the same file-discovery rules as normal translation). No workflow sidecar such as a project JSON, TXT, or paired image is required.
  • Stages executed: conditional colorize (when colorizer.colorizer != none) → conditional upscale (when upscale.upscale_ratio is set) → detection → skip OCR and fill detected lines with the literal TEXT → text-line merge → mask refinement → inpainting.
  • Stages skipped: OCR, translation, rendering, and text typesetting. text_regions is cleared before the branch ends, so the result is never rendered with translated text.
  • Output files: the main output image (the inpainted clean image; the un-inpainted work image when there are no text lines, no merged regions, or an AI renderer is selected); an editor base image when conditional colorization or upscaling is active; and a project JSON with empty regions when save_text is enabled (the exact JSON fields may differ by release).
  • Workflow field: combo index 7 writes cli.inpaint_only=true at runtime; GUI switching keeps the eight workflow booleans mutually exclusive.

Run this workflow ​

Select the Inpaint Only workflow ​

  1. Open the translation page and choose “Inpaint Only” in the “Translation Workflow Mode:” combo box.
  2. The page title becomes “Inpaint Only” and the subtitle shows the hint “Tip: Detect text regions and inpaint to output clean images, no translation or rendering”.
  3. The start button becomes “Start Inpainting”; clicking it starts the backend task in this mode.

Selecting a mode only writes configuration and updates the UI texts; it does not start a task. Before starting, add the main input images (“Add Files”, “Add Folder”, or drag-and-drop). No sidecar file is required in this mode.

“Output Directory:” determines where the main output image goes, with the same naming rules as normal translation: the input folder name and relative hierarchy are preserved under the normal output directory, save_to_source_dir=true switches to manga_translator_work/result/ next to the source image, and an empty or none cli.format keeps the original extension.

Processing order ​

Stages and outputs ​

Inpaint Only reuses the first half of the normal-translation preprocessing and finishes inside the “Inpaint Only” branch of _translate_until_translation. The Mermaid diagram below shows the stage order, skip branches, and outputs. It shares conditional colorization, conditional upscaling, and detection with normal translation, but OCR and translation are skipped.

flowchart LR
    Input["Main input images"] --> Pre["Conditional: colorize -> upscale"]
    Pre --> Detect["Detection"]
    Detect --> HasLine{"Text lines and a valid raw mask?"}
    HasLine -->|no| Pass["Use the un-inpainted work image as the result"]
    HasLine -->|yes| Fill["Skip OCR: fill detected lines with literal TEXT"]
    Fill --> Merge["Text-line merge"]
    Merge --> HasRegion{"Merged text regions exist?"}
    HasRegion -->|no| Pass
    HasRegion -->|yes| Refine["Mask refinement"]
    Refine --> AiRenderer{"AI renderer selected?"}
    AiRenderer -->|yes| WorkBase["Skip inpainting; use the work image as the inpaint base"]
    AiRenderer -->|no| HasMask{"Mask empty?"}
    HasMask -->|yes| WorkBase
    HasMask -->|no| Inpaint["Inpainting"]
    WorkBase --> Result["Result image"]
    Inpaint --> Result
    Result --> Clear["Clear text_regions; never render translated text"]
    Result --> Save["Save the main output image"]
    Save -. "skipped" .-> SkipStages["OCR / translation / rendering"]
    Clear -. "skipped" .-> SkipStages
  • Each detected line is filled with the literal TEXT placeholder while OCR is skipped; text-line merge combines the detected lines into text_regions.
  • When there are no text lines, no raw mask, or no merged regions, the un-inpainted work image is returned directly as the result without mask refinement or inpainting.
  • When mask refinement fails, it degrades to a simple dilation driven by kernel_size and mask_dilation_offset.
  • Real inpainting is skipped when an AI renderer (OpenAI/Gemini) is selected or the mask is empty; the work image passes through as the “inpaint base”.
  • Before the branch ends, text_regions is cleared and the inpaint_only_complete flag is set; the save stage skips _complete_translation_pipeline, so the regular post-mask rendering does not run.

Mask and inpainting details ​

  • Mask refinement and inpainting parameters come from the “Inpainting” tab in Settings (inpainting model, inpainting size, precision, per-block inpainting, solid-fill pure bubbles, mask dilation offset, and so on); see Mask and Inpainting.
  • The inpainting input is the work image after conditional colorization/upscaling (load_image(ctx.upscaled)), not the raw input image.
  • The inpainting result is stored in ctx.img_inpainted, converted to a PIL image, and saved as ctx.result as the main output. This mode does not write an extra inpainted/ sidecar.

Concurrency and mutual exclusion ​

  • batch_concurrent is incompatible: both the desktop controller and translate_batch() treat it as an incompatible mode and force non-concurrent handling. Keeping the concurrent configuration in the UI does not turn this mode into a concurrent pipeline.
  • Manually stacking multiple workflow fields is not a supported combination; GUI switching keeps the eight fields mutually exclusive, and the backend translate_batch() only builds the concurrent pipeline when no incompatible mode is present.
  • As with normal translation, preprocessing still runs conditional colorization and upscaling according to colorizer.colorizer and upscale.upscale_ratio; those values are not forced by this workflow.

Inputs, outputs, and limitations ​

  • Input dependency: the main inputs must be images supported by the file service; no project JSON, TXT, or paired image is required.
  • cli.overwrite=false: the GUI checks whether the main output image already exists before starting (it shares the normal-translation check branch with other main-image-only modes).
  • cli.save_text: the default is true. When enabled, the save stage still writes a project JSON with empty regions (including the mask and colorize/upscale info). This is the current behavior; the actual GUI file behavior may vary by release.
  • AI renderer: with the OpenAI/Gemini renderer selected, this mode does not perform real inpainting and outputs the un-inpainted work image. The name “Inpaint Only” therefore differs from the actual behavior in that case; this is current.
  • Detection, mask refinement, and inpainting consume model and VRAM costs according to their parameters; OCR and translation are skipped, so they incur no such costs.
  • The main output directory, save_to_source_dir, and cli.format affect only the main output image. This mode writes no original/translated TXT, so the export template file has no effect here.

See the reference index: Workflow Matrix.