image-inpainting

To remove objects from a photo, mask-driven image inpainting regenerates specified regions using a grayscale mask where white indicates areas to replace. You provide the source image, mask image, and a prompt describing what should be preserved outside the masked region.

Watermark removal without a mask is possible using a description-based fallback that routes edits via spatial instructions. This allows targeted region cleanup by describing the watermark location in the prompt rather than providing an explicit mask image.

Mask-driven image inpainting is a localized photo retouching technique where a grayscale mask dictates region replacement. White pixels in the mask identify areas to regenerate, while black pixels preserve the original image content.

Yes, image inpainting works for blemish retouching and localized edits by targeting small regions with a mask. It performs controlled local edits to clean up blemishes while preserving the identity and style of the surrounding photo.

To run image inpainting for region replacement, you need to invoke the RunComfy CLI with your source image URL, a mask image URL, and a prompt. The system routes the request to the tongyi-mai/z-image/turbo/inpainting model using a required JSON schema.