Output quality

    Inpainting

    Also called Generative fill, Masked editing.

    Inpainting regenerates a region you have masked while leaving the rest of the picture untouched, so a change stays local.

    You draw the boundary, the model repaints inside it. Everything outside the mask is preserved exactly — not re-rendered, not re-graded — which is what makes this the tool for surgical work: remove a bystander, change a sign, fix a hand, swap a colour on one object.

    The mask is most of the craft. Too tight and the new content has nowhere to blend, leaving a visible seam. Too loose and you have handed the model territory you wanted kept. Feathered edges and a little breathing room around the subject produce joins that survive scrutiny.

    The model also reads the surrounding pixels for context, so it matches light and perspective from what is nearby. When it gets that wrong, the fill is competent and lit from the wrong direction, which reads as fake faster than a blurry patch would.

    In practice

    • Mask generously around the object, not tightly along its edge.
    • The prompt describes what should be inside the mask, not the whole picture.
    • Iterate on the same mask at different strengths rather than re-masking each time.

    The mistake to avoid

    Masking the object you want gone and prompting for the object. Prompt the replacement — an empty wall, a clear pavement — not the thing you are removing.

    Where you will run into it

    Related terms

    The all-in-one AI studio for creators. 60+ models for video, image, voice, music and lipsync in a single app.