...
Removing a simple object from a plain background is easy. Removing a person from a crowd, a reflection from glass, or a shadow from a patterned floor — that's where AI inpainting gets interesting.
You paint a mask over an ex in a group photo, click remove, and the AI fills the gap with… a ghostly smear of background that looks like it was painted by someone who has never seen a wall before. Object removal is easy when the background is sky or grass. It is hard when the background is a brick wall with a specific pattern, a wooden floor with grain lines, or a crowd of people with overlapping bodies.
Our AI object remover uses inpainting — the AI fills masked regions by generating pixels that match the surrounding context. Here is how to get clean results on the hard cases.
Inpainting quality depends more on your mask than on the AI model. A tight mask that hugs the object boundary gives the AI more surrounding context to work with. A loose mask that includes extra background confuses the model — it sees the extra background as "content to be replaced" and tries to fill it, creating visible seams.
Masking technique for clean results:
Removing an object from a brick wall, a tiled floor, or any repeating pattern is the hardest inpainting challenge. The AI has to continue the pattern seamlessly through the masked region — matching brick size, mortar color, alignment, and perspective.
What works: rectangular masks aligned with the pattern direction. If the bricks run horizontally, make your mask wider than it is tall. This gives the AI horizontal context to extend the brick lines.
What fails: irregular masks that cut across pattern lines at odd angles. The AI has to invent a pattern transition that does not exist in reality, and the result looks like corrupted digital art.
When you remove a person from a group photo, the AI has to reconstruct the background that was behind them — which often includes parts of other people. If two people are standing shoulder to shoulder, removing one means reconstructing the other person's shoulder, arm, and clothing edge.
Technique: mask only the person you want to remove, but extend the mask slightly into the space between them and the adjacent person. The AI needs a small margin to blend the reconstructed edge with the remaining person. A mask that stops exactly at the boundary between two people creates a hard seam where the AI-generated pixels meet the real pixels.
What to expect: results are inconsistent. Sometimes the AI reconstructs the adjacent person's shoulder perfectly. Sometimes it gives them an extra arm. For critical photos, be prepared to try 2-3 times with slightly different masks.
Removing sunglasses from a face, a wine glass from a table, or a reflection from a window — these are fundamentally hard because the object is not opaque. The background shows through the object, so the AI has to distinguish "background visible through glass" from "glass to be removed."
What works (sometimes): mask the entire reflective object, including the area where the background shows through. The AI will attempt to reconstruct the background as if the glass was never there. For simple backgrounds, this works. For complex backgrounds, the reconstruction looks smeared.
What does not work: trying to remove only the reflection while keeping the glass. The AI cannot distinguish reflection from transparency — it either removes both or keeps both.
A shadow on a wooden floor, carpet, or grass has texture showing through it. Removing the shadow means brightening the area while preserving the underlying texture — the wood grain, carpet fibers, or grass blades that the shadow was cast onto.
Technique: mask the shadow only, not the object that cast it. Process the shadow separately from the object. This gives the AI a simpler task: "brighten this region while preserving texture" rather than "remove this entire area and fill it from scratch."
Result: shadow removal is more reliable than object removal because the AI has actual pixel data to work with — it just needs to adjust brightness while keeping texture. Success rate is roughly 80% for shadows on uniform textures, dropping to 50% for shadows on complex patterns.
For removing watermarks and text overlays (which follow predictable patterns and are easier than general object removal), our watermark remover is specialized for that task. And for removing entire backgrounds, see our background remover.
AI Object Remover
Remove unwanted objects, people, or text from photos with AI inpainting.
Background Remover
Remove image backgrounds instantly with one click.
Watermark Remover
Erase watermarks, logos, text overlays, and timestamp stamps from images using BRIA Eraser AI inpainting. Canvas mask tool for precise removal area selection with adjustable brush size. Works on semi-transparent watermarks, logo stamps, and photo-bombing objects.