Remove Object from Photo
Erase any unwanted object — signs, cables, trash, gear, clutter — with SDXL generative fill. Better than simple inpainters on large or complex shapes. 3 free removes a day, 5 after sign-in.
Before / After
Erase unwanted objects and clutter
Why use a generative fill remover
Object removal tools fall into two broad camps. The first is reconstructive inpainting (LaMa, Content-Aware Fill, and their cousins), which removes by extrapolating from surrounding pixels. Fast, surgical, and excellent when the background is simple or the object is small. The second camp is generative inpainting (SDXL, Stable Diffusion, Dall-E's fill), which redraws the masked area from a prompt + context. Slower and slightly less pixel-faithful, but dramatically better on large, complex, or overlapping objects.
This page runs SDXL inpainting with a prompt tuned specifically for generic object removal: "seamless background, natural continuation of the surrounding scene, clean empty space, photorealistic". The model is told nothing about what kind of object you are removing — it just tries to make the painted region blend into its surroundings. That generality is what lets it handle an abandoned bicycle on a brick street as comfortably as a trash can on a grass lawn.
Pair this with the homepage Object pill (LaMa-based, 10× faster and pixel-surgical) and you have two complementary tools: start simple on the homepage, escalate to SDXL here when the homepage result is not clean enough. Both run on the same daily free pool and shared credit balance.
How to remove an object in 3 steps
- 1
Upload the photo
Drag any JPG or PNG into the upload card, or tap to browse. The tool accepts any aspect ratio — SDXL inpainting handles portrait, landscape, square, and panoramic inputs natively. Images larger than 1536 px on the longest edge get client-side downscaled before upload; you do not need to resize beforehand.
- 2
Brush over the object plus margin
Paint every pixel of the thing you want gone — the whole outline, including any shadow on the ground or reflection on water. Add 6–10 pixels of margin around the edge so SDXL has clean surrounding context to rebuild from. For big or tall objects, a larger brush (80–120 px) covers in fewer strokes. Undo and Clear are on the right panel if you overshoot.
- 3
Download the generative fill
Tap Remove objects. SDXL inpainting takes ~20–30 seconds to redraw the masked region with a plausible continuation of the scene — grass, sky, pavement, fabric, whatever fits. The output returns at the same aspect ratio and dimensions you uploaded. Hit Download to save as PNG.
Where SDXL generative fill outperforms
Large objects across the frame
Parked cars, picnic tables, benches spanning a big area — simpler inpainters give up. SDXL has enough capacity to invent coherent grass / pavement / walls underneath.
Objects in front of detailed backgrounds
A sign in front of a brick wall, a bag on a patterned rug, a bottle on a forest floor. SDXL matches texture and lighting much closer than LaMa-family tools.
Overlapping objects
Cables crossing a fence, signs layered over tree branches — the generative model handles these where reconstructive inpainters leave ghost lines.
Objects with hard shadows
Paint the object + its shadow together. SDXL redraws the ground with matching lighting, so the scene looks naturally unlit rather than half-lit.
Tips for clean edits
Paint generously, include shadows and reflections
The single biggest factor in SDXL quality is mask coverage. A tight mask leaves the object's shadow or reflection behind — looks like a ghost. Always paint 6–10 px past the object silhouette, and include any shadow on the ground or reflection on water / glass.
Use the default settings first
Strength 0.95, guidance_scale 7, steps 30, K_EULER scheduler are tuned for most object-removal scenarios. Reach for the playground (/playground/sdxl-inpaint) only if you need to fine-tune for a specific challenging image.
Try the Object pill on the homepage if results look overdone
The homepage Object pill (LaMa-based) is faster and more surgical on simple removals. If SDXL invents too much new texture for your taste, the homepage tool might produce a subtler edit.
Output is near-original resolution
Unlike SD 1.5 inpainting, SDXL does not cap at 1024 px. Your upload dimensions are preserved. Very large uploads (over ~1 megapixel) get internally downscaled for inference and resampled back up — expect some softening in the masked region on 4K+ photos.
The honest limits
Pixel-perfect preservation
SDXL slightly redraws the WHOLE image (even unmasked areas) as a side effect of being generative. For watermark or timestamp removal where 99% of the photo must stay byte-identical, use the Watermark Remover (LaMa-based, byte-preserving outside mask).
Very small fine objects on busy backgrounds
Tiny text on dense foliage, fine wires in a detailed scene — SDXL's large-scale generation can soften or miss these. A LaMa-based tool sometimes does cleaner work on small, subtle removals.
Strict colour matching
Studio product photography where precise colour reproduction matters. SDXL can shift colours slightly in the rebuilt area. For product shoots, a pure LaMa inpainter is safer.
Questions about removing objects
Other removers on MagicRemover
Free AI Object Remover
The homepage Object pill — LaMa-based, faster, preserves surroundings byte-for-byte. Best starting point for simple removals.
Remove People from Photo
Same SDXL model, prompt-tuned for crowds and photobombers specifically.
Remove Watermark from Photo
LaMa-based watermark remover. Byte-preserving — best when the rest of the image must stay exactly as shot.