What is image inpainting?
Image inpainting is an operation where the missing part of an image gets updated with imagery that matches the image scene or the instructions the user provides. Initially, this process was used in image restoration as a repair function. In AI image generation workflows AI inpainting is combined with a prompt to alter specific parts of an existing image.
Inpainting workflow
Unlike prompt-to-text, inpainting starts with an image. This can be uploaded from your device or an image you've generated from a prompt.
Once you've selected your starting image, you'll need to create a mask layer that indicates the areas you want the AI to update. To help speed up this process, Freeway uses computer vision AI to detect the image layers automatically. Clicking on an item in the list will automatically mask a region of the image. If the AI missed some spots you wanted masked, Edit Layer provides a brush to touch up the mask layer.
Once the layer is complete, the next step is to provide instructions for the inpainting operation. Options for enhancement, variations, and style change are available, but the most impactful (and fun) operation is to enter a prompt for the item you want to appear in the masked area.
Here are some fun ideas to try:
The possibilities are endless!
Inpainting for Image Clean up
Inpainting also enables you to clean up an automatically generated image. Below, you'll see our "monkey with sunglasses" generation was great, but the AI created an extra pair of ears. With image inpainting we could remove the ears and blend in the background.
Search your recent images
The number of images generated can accumulate quickly, and finding the image from yesterday's session is challenging. Now you can search your recent or saved image generations by the prompt or the style.
Faster load times
If you had a few hundred or more images generated, the recent images tab took a while to load, especially if you were on mobile LTE. We've reduced the load times with image thumbnails.
Lastly, we've fixed a handful bugs and improved user flows. If you have additional feedback we want to hear from you! Email us at support@ or join our Discord to start a conversation.
