AI inpainting for the part of a photo that needs repair

A mark across plain plaster is a different task from a mark across a face or printed label. The plaster may only need compatible grain and lighting. The face needs recognizable identity, and the label needs exact characters. Before using AI inpainting, name the details that make the region correct rather than merely plausible.

This workflow focuses on local repair and redraw. Use the dedicated removal tools when the main job is identifying an unwanted person or object. Use object addition when you want something new in an empty space. AI inpainting shares the idea of regional generation with those tasks, but the review here is continuity: does the repaired material belong to the same surface?

An extremely tight region can leave part of a mark behind. A region that is much too broad can invite unrelated changes. Look at the complete damage, including faint edges, and at the nearby texture that should guide the repair. A small amount of surrounding context helps explain what should continue through the affected area.

The editor can identify the requested target from your description, and you can refine a selection when necessary. You do not need to label every pixel before starting. Still, inspect the proposed task in relation to neighboring details. With AI inpainting, a seam, a rim or a printed line near the target should be named if crossing it incorrectly would spoil the result.

For a mark that crosses two materials, identify both sides of the boundary. A scratch across a window frame should not turn the adjacent glass into painted wood. Keep that division visible in your description, then inspect it after the patch is filled.

Choose a small, clearly defined issue first. Upload JPG, PNG, WebP or HEIC up to 20MB, using the best available original.

1. 1

   Identify the damaged region

   Describe the position and appearance of the patch, such as the pale scratch on the lower-left wooden panel. If several marks look similar, add a selection or another positional cue.

2. 2

   Describe the replacement surface

   Ask for the visible material to continue: woven blue fabric, painted plaster or horizontal wood grain. State any structural detail, such as a seam, that must cross the area.

3. 3

   Inspect inside and outside the repair

   Compare the original and new version at the same zoom. Check the patch itself, then the transition around it and finally the whole object for changes beyond the intended region.

Illustrative prompt, not a tested output: "Repair the small tear in the front blue cushion with matching woven fabric. Continue the seam along the lower edge. Preserve the cushion's folds, piping and position on the chair." This tells AI inpainting which material should replace the damage and which features should constrain it.

The seam is the strongest test. A repair can reproduce blue texture while moving the seam or changing its thickness. Compare the seam on both sides of the patch and look for a consistent curve. If the source is too blurred to show its construction, the result can only infer it. Describe the desired appearance without calling that inference a restoration of the original fabric.

Different surfaces reveal different errors. Choose the inspection that fits your image instead of relying on a quick full-frame glance.

Once a scratch or object has covered part of the source, those pixels do not contain an unobstructed record of the scene. AI inpainting uses the surroundings to generate a possible replacement. That distinction matters for family archives, product details and documentary images: a believable result is not proof of what was originally there.

Be especially cautious with identity, signatures, diagrams and exact lettering. A reconstructed character can change meaning while looking tidy. For those cases, a clean reference or the original editable document gives you something concrete to verify against. Keep generated repairs in a separate version, and label them appropriately when the image's history matters.

AI inpainting can also introduce a subtle texture mismatch. Inspect grain at full size, then step back to the size at which the image will be used. The first view catches artifacts; the second tells you whether the repair draws attention away from the photograph.

If the repaired area is wrong, adding more edits to it may make comparison harder. Select the original or an earlier accepted version before submitting the next brief. Describe the visible problem directly: the wood grain should run horizontally, or the seam should remain at the lower edge. Keep subjective style preferences separate from those observable requirements.

An AI inpainting task returns one image. Generation costs 10 credits at 1K, 20 at 2K and 30 at 4K; Auto normally uses 1K. Internal retries do not add a charge. A new variation you request is a new task, while reviewing or selecting an existing version is free. Check the displayed price before submitting a higher-resolution attempt.

An AI inpainting repair should support the image's intended use. For a small online photo, a visually consistent patch may be sufficient; an archival reproduction can require a much more exact reference and specialist workflow. Do not keep generating simply because a different texture might be possible. Decide what would make the version acceptable before paying for another attempt.

Export the accepted result as PNG, JPG or WebP, and keep the source separately. Basic crop and resize tools can handle the final delivery dimensions without another AI inpainting request. If the missing area lies beyond the original frame, move to the image extender; if it contains a specific unwanted person, use the person-removal workflow for more targeted guidance.