Practical overview
When it fits
Use it when you want a local, manually guided inpainting workflow for anime images and can prepare the Python environment and model weights.
Who it is for
- Anime-image editors who want to fill marked damaged or missing regions.
- Users who want to guide repairs by drawing edges in an optional editing window.
- Python users interested in training an inpainting model on their own dataset.
Documented uses
- Mark a defective region and generate a fill using the input image.
- Add edge guidance to a marked region before patching it.
- Save a completed repair from the desktop tool.
- Train a model using the project's training workflow.
Requirements & limits
The repository README lists Windows and Linux as platforms. It documents a locally installed Python workflow with model weights.
- The interface is based on OpenCV and the backend on Edge-Connect.
- The documented pretrained model is anime-face-trained and is cautioned against for large, full-character images.
- English prerequisites list NVIDIA GPU, CUDA and cuDNN; the Chinese instructions say CPU can be selected through DEVICE in config.yml.
- The README points to PI-REC as a later drawing method.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install and prepare Anime Inpainting
Set up the repository and model files needed before starting the image tool.
Before you start
- Python 3
- PyTorch 1.0 and dependencies
- NVIDIA GPU, CUDA and cuDNN are listed in the English prerequisites; the Chinese instructions describe CPU selection through config.yml
- A trained model weights file
Steps
Clone the Anime-InPainting repository.
Install PyTorch and its dependencies, then install the repository's Python requirements with pip install -r requirements.txt.
Download trained weights, extract the .7z file, and place the .pth file under ./model/getchu/ as shown in the README.
Optionally review ./model/getchu/config.yml.
Expected outcome
The software, dependencies and model weights are arranged for the documented tool-launch commands.
Guide limitations
- The README's supplied weights were trained on anime faces and are described as performing poorly on large, full-character images.
- Review the Edge-Connect work and its license, as the project README advises.
Mark and save an image repair
Use the documented controls to mark an area, optionally guide it with edges, and save the output.
Before you start
- The repository, dependencies and model weights prepared as described in the README
- The tool launched in its standard or edge-window mode
Steps
In the input window, use the left mouse button to mark the defective area.
If the edge window is open, draw guidance edges with the left mouse button and erase edges with the right mouse button.
Press n to patch the marked black area using the input image, or press e to use the input image and edited edge when the edge window is open.
Press s to save the output; press r to reset the setup if needed.
Expected outcome
A patched image can be saved from the tool window.
Guide limitations
- The e operation works only when the edge window is open.
- The README warns that its supplied face-trained weights perform poorly on large, full-character images.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 18 October 2026
- Original website title
- youyuge34/Anime-InPainting · Official homepage title · retained in its source language