An image-editing workflow should alter one specified region while retaining the surrounding image. Which supported model feature should be evaluated?

1. Text summarization only
2. Regenerating an unrelated image with no original input
3. A higher text token limit alone
4. Mask-based editing or inpainting with the original image