Image editing
What is inpainting?
Inpainting is filling a missing, damaged or unwanted area of an image with new content that blends in. AI models use it to remove objects, people or text and rebuild the background behind them, to repair damage, or to replace part of a picture with something new. The rest of the image stays as it was.
By the DreamBench team. Updated .
Inpainting at a glance
At a glance
- Input
- A picture and the area to fill: painted as a mask, or described in words.
- Output
- The same picture with that area replaced.
- Keeps
- Everything outside the area.
- Used for
- Removing objects, people and text; repairing damage; replacing one thing with another.
- Struggles with
- Large areas in busy scenes, faces, text, and shadows left behind.
Which edit do I need?
Take something out of the picture: Inpainting as a removal: the area is filled with the background that was probably behind it. The AI object remover does it from a description, no brush; name the shadow and reflection too.
Take a person or a caption out: The same removal, set up for the job: remove people from a photo and remove text from an image take a description or clear every passerby or caption at once.
Put something else in its place: Inpainting as a replacement: the area is filled with what you describe instead of background. The AI image editor makes that kind of edit from an instruction; keep the new thing about the size of the old one.
Make the picture bigger or change its shape: Not inpainting but outpainting: new scene is added outside the edges while the original stays whole. The AI image expander does it for the frame you pick.
How inpainting works
Inpainting works from a chosen area: you paint a mask over it, or describe it in words, such as “the lamp post on the right”, and the model finds it. The model then reads the pixels around the area and generates new ones to match their light, texture and perspective, so the join disappears. Small areas on plain backgrounds blend best.
Mask
Every pixel is in or out. Fine enough for a solid object, blocky on hair, smoke or motion blur.
Matte
Every pixel has a share of transparency, from 0 to 100%, so loose strands keep their softness.
The hard part is the join. A small removal on a plain background is easy to hide. A large one in a busy scene asks the model to invent more, and the fill can look soft or repeat a texture.
A removal that overlaps the subject shows for a different reason: an arm across a person or a pole in front of a face makes the model rebuild part of the subject as well as the background, and a face is the thing it invents worst. Something standing apart from the subject, on a wall, a sky or a lawn, fills cleanly; a shadow or a reflection it cast stays behind unless you name it with the object.
Mask, brush or words
| You do | Good for | |
|---|---|---|
| Mask | Paint over the area, then optionally describe what should go there | Exact control over where the change goes |
| Healing brush | Brush over a small flaw; the editor copies texture from nearby | Dust, spots and small blemishes |
| Instruction | Write what to remove and where it is | Quick removals, no drawing |
A short history of inpainting
| How it worked | Its limit | |
|---|---|---|
| By hand | Restorers retouched damaged paintings and photos, matching the surrounding brushwork or grain | Slow, and only as good as the restorer |
| Copying pixels | Early software filled a gap by extending lines and copying patches from elsewhere in the picture | Fine for skies and walls, poor where something new had to be invented |
| Learned models | Neural networks, then diffusion models such as latent diffusion, learned what images contain and generate a fill | They invent: the fill is plausible, not recovered |
Where people use inpainting
- Removing people who walked into a shot.
- Erasing power lines, signs, stains or clutter.
- Taking a date stamp or a caption off a photo.
- Replacing one object with another in a product shot.
- Fixing hands, faces or text in an AI-generated image.
- Repairing scratches, tears and stains in old photos.
How DreamBench does inpainting
The AI object remover works without a brush: you write what should go and where it is, and Nano Banana 2 removes it and fills the space with the surfaces, light and perspective around it. Two presets do the same for people and for text.
A removal starts at 7 credits. If a large one leaves a soft patch, run it again with a more precise description. To remove something from a clip instead, see video inpainting.
| Figure | What it means for your photo | |
|---|---|---|
| Input | JPEG, PNG, WebP or HEIC, up to 40 MB | A picture over 2048 pixels on its long edge is scaled to 2048 before the run. |
| The area | A few words: what it is and where it sits | The description is the selection. “The red trash can on the left” works; “the clutter” is vague and can remove too little or the wrong thing. |
| Model | Nano Banana 2, an instruction edit | The whole picture is drawn again with the object gone, instructed to keep everything else. |
| Output | A JPEG in your photo’s proportions | At the model’s own size, so a large upload can come back with fewer pixels; the AI image upscaler adds them back without redrawing. |
| Price | From 7 credits a removal | A larger photo is rendered at a larger size and costs more; the button shows the price before you press it, and a run that fails is refunded. |
| Marks | No visible watermark | The model embeds SynthID, Google’s invisible mark that identifies a picture as made or edited by AI. |
| Your upload | Removed from the tool’s storage after 7 days, with the result | Not used to train a model; upload only pictures you have the right to edit. |
Questions
Is inpainting the same as removing an object?
Removing an object is the most common use of inpainting. Inpainting itself is broader: it fills any chosen area, so it can also replace an object or repair a damaged part of a photo.
Do I need to paint a mask?
Not with an instruction-based editor. DreamBench’s object remover takes a short description of what to remove and where it is. Mask-based tools ask you to paint over the area first.
How do I remove an object from a photo without Photoshop?
To remove an object from a photo without Photoshop, use an instruction-based remover: write what should go and where it is, including its shadow or reflection, and the model fills the space with the surfaces and light around it. DreamBench’s object remover works this way, with no brush, in a phone browser as well as on a computer; keep the area small and run it again if the fill looks soft.
Why does the inpainted area look blurry or smudged?
The model had too little to go on: a large area, a busy background or an imprecise description. Remove a large object in smaller steps, describe exactly what is there, and run the fill again rather than enlarging the area.
How big an object can inpainting remove?
The smaller the better: a removal that covers more than a small part of the picture asks the model to invent a scene, and the fill goes soft or repetitive. Remove a large object in pieces, or re-shoot if you can.
What is the difference between inpainting and outpainting?
Inpainting fills a region inside the picture. Outpainting adds new content outside its edges to make it larger. Both match what they add to what is already there.
Can inpainting be used on photographs as well as AI-generated images?
Yes. It works the same way on both: on photos to remove things or repair damage, and on generated images to fix a hand, a face or a word that came out wrong.
How do you get good results from inpainting?
Keep the area small, describe exactly what should go and where it is, and name shadows and reflections too. If a fill looks soft, run it again rather than enlarging the area.
Does inpainting lower the resolution of my photo?
It can, with an instruction-based editor. The picture is drawn again at the image model’s own size, so a large upload can come back with fewer pixels than it had, in the same proportions; a mask-based tool that regenerates only the painted area keeps the rest at full size. For a large print, enlarge the clean photo afterward with the AI image upscaler, which adds pixels without redrawing anything.
Does inpainting change the rest of the photo?
With a mask, no: only the painted area is regenerated and every other pixel is left as it was. With an instruction-based remover the whole picture is drawn again by the image model, which is told to keep the framing, colors and detail everywhere else; it comes back close to the original, but very small text and fine patterns can shift slightly, so compare the before and after before you download.
Can inpainting remove a watermark or a logo from a picture?
Technically it is a removal like any other: name the mark and where it sits, and the picture behind it is rebuilt from what surrounds it, cleanest over a plain area. Do it only on pictures you own or have the right to edit. A watermark or credit on someone else’s work is not what the tool is for, and removing it doesn’t change who owns the image.
Does inpainting cost anything?
Some apps include a basic eraser. On DreamBench you create a free account and choose a plan, and a removal starts at 7 credits.
Related
DreamBench tools
- AI object removerDescribe what to remove from a photo and the gap is filled in naturally.
- People removerClear strangers, photobombers and crowds out of a photo, the scene filled in behind them.
- Text removerErase captions, date stamps, stickers and overlaid words from a photo, the picture behind them filled in.
- AI image editorDescribe a change to a photo in a sentence and get it back edited.
Terms
- OutpaintingExtending a picture past its edges with new content that matches the scene.
- Video inpaintingFilling a region of a video so it fits every frame, which is how AI removes objects and people.
- Background removalSeparating a photo’s subject from what is behind it, with segmentation or matting models.
- Image upscalingMaking a picture larger with more pixels, the extra detail predicted so it stays sharp.
Sources
DreamBench is an independent platform and isn’t affiliated with or endorsed by the companies named on this page.