Clean Up a Photo — Pro, on our server
Clutter is the easy half of object removal, and almost nobody treats it that way. Difficulty in an inpainting job scales with the area you cover, not with the number of things you cover, because every masked region is filled from its own immediate surroundings. A traffic cone on tarmac, a bin against a hedge, a downpipe on brick, a crisp packet on grass: each is a small hole ringed entirely by material that explains it, and each closes without ceremony. Fifteen of those in one mask is fifteen independent easy problems solved in a single run. Compare that with one van parked across a shopfront, which is a single hard problem, and the arithmetic of what to spend an HD run on becomes obvious.
Find the clutter systematically, because your eye is already ignoring it. Whoever took the picture was looking at the subject, and the brain edits out the periphery in the moment and then again on the screen. So work a fixed route instead of scanning. Start at the four edges, where the objects a viewer notices first almost always live. Then run along the ground plane — feet, kerbs, skirting, lawn — which collects everything dropped or left. Then the sky, for wires and aerials, which have their own page under cable removal. Then vertical surfaces for signage, switches and stains. Then anything reflective, because a window or a wet pavement can hold a copy of something you already removed. Viewing the photograph as a thumbnail is a good final check: what still bothers you at that size is what genuinely needs to go.
Know when to stop. There is a threshold past which a cleaned photograph stops reading as a place and starts reading as a render — pavements with no marks, lawns with no leaves, walls with no fixings. Nobody consciously notices the missing detail; they notice that the picture feels synthetic and trust it a little less. The useful rule is to remove what pulls attention away from the subject and to leave what makes the scene credible. A hose across a lawn goes; the slight wear on the path stays. And if a whole corner of the frame is beyond saving, a crop costs nothing, invents nothing, and is often the better answer than a heroic mask.
Where clean-up sits among the other tools, and the terms. Furniture in an empty-room shot is its own case, on furniture removal; a parked vehicle is on car removal; a person is on person removal; the shared mechanism behind every one of them is described on the object remover. Once the picture is tidy, sizing it for a listing or a message thread is free, on-device work in the image compressor, which uploads nothing at all. The clean-up run itself is Cloud HD — Pro-only, opt-in, on our GPU in the United States, with the photo and mask encrypted in transit, deleted when the run finishes and in all cases within 24 hours, and never used to train anything. Pro includes 150 runs a month on Monthly and 200 on Annual, of which 10 may be HD.