Inpainting - Walkthrough

Inpainting: Enhancing Specific Areas

Inpainting is a great way to improve specific areas within an image. Let's say you’ve generated a fantastic image. Everything is perfect and exactly as you imagined, except for one small detail. Re-prompting the entire image would change it completely and would be another guessing game to get back to where you were.

Instead, you can mask specific areas and inpaint them. Here’s how to do it.

Example: Correcting Eyes

Take this example: Many aspects of the image are already correct, but the eyes are not quite right.

In the top left of the generated image, you’ll find options for further processing. Click on the brush icon and then select Inpainting in the next step. This will automatically take you to the Inpainting area.

Alternatively, you can drag and drop an image into the upload field next to the “Create” button on the bottom right.

Inpainting Settings

In this area, you will find new settings. The most important setting is the size of your brush, as it determines how precisely or broadly the area you want to change can be marked.

Below that, you’ll find the usual image generation settings. In this case, there is also “Image Weight.” This slider determines how much the image should be altered.

Applying the Mask

Once you are satisfied with your settings, draw a mask over the area you want to change.

Editing the Masked Area

In the next step, determine what exactly should happen in the masked field. In our case, the eyes should look better. It’s best to use the entire original prompt and only change the object. This ensures that all other image mood information remains the same. So, the prompt part “female actress” becomes “beautiful brown eyes, makeup, smokey eyes.”

Finally, click the large “Inpaint” button and wait for your image to be regenerated. Here is our example.

Further Tutorials

If you enjoyed this tutorial, feel free to check out our other tutorials in the blog-section where we explain the major features of Cogniwerk.ai. Such as finetuning models or ControlNet.