Start with what the person is hiding
You can try removing a passerby from a photograph of your venue using an uploaded image and a custom prompt in RealShoot. Treat the result as an edit to inspect, not a recovered view of what was behind the person. People removal can invent hidden background. Before uploading, identify what the person covers: a small patch of plain wall is a different review task from a doorway, staircase, sign or furniture arrangement. If you cannot verify the hidden area, take an unobstructed photograph instead. This tutorial offers a suggested workflow, not a tested result or a guarantee that the space will remain unchanged.
Prepare the photograph and a reference for checking
Keep the original file and work from a copy. Choose a clear photograph in which the space is already framed as you want it. Confirm that you have permission to upload and edit the image, including appropriate consent for identifiable people; removing someone later does not resolve permission for the original upload. Examine the person, their shadow and any reflection before writing the prompt. Note nearby landmarks, such as the edge of a doorway or a floor-tile junction, so you can describe the target precisely. If available, keep a current unobstructed photograph of the same area beside you for manual comparison. A reference from before a renovation or furniture change cannot verify today's scene.
Decide whether to edit, crop or reshoot
Try a limited edit only when you can check the affected area against the actual venue or a current photograph. Consider an ordinary crop if the person is near the edge and cropping still shows the space accurately. Reshoot when several people obscure most of the room, a person blocks an important fixture, or you cannot establish what belongs behind them. For example, someone standing against an uninterrupted wall gives you a simpler inspection task than someone covering a handrail and steps. These are decision examples, not measured success rates. Do not use an empty-looking edit to imply that a busy shared space is private or normally unoccupied.
Upload the image and try this prompt
Use the photograph as the input for image editing. Text-only generation cannot document your actual venue. Replace the bracketed description in this suggested prompt before using it: Remove only [the passerby in the blue coat beside the left doorway]. Preserve all other people and the existing framing, perspective, room proportions, walls, doors, windows, fixtures, furniture, signs, flooring and visible wear. Do not add amenities, conceal damage, or remove accessibility or safety features. Limit background changes to the area occupied by that person. Remove only shadows or reflections clearly belonging to that person, preserving other shadows and reflections. Keep surrounding surfaces and architectural lines unchanged. This prompt requests constraints; it does not ensure that the output follows them.
Inspect the removal area and the rest of the frame
Compare the original and edited images side by side, first at the intended display size and then enlarged. Check for leftover hands, shoes, clothing edges, silhouettes or reflections. Trace floor joints, wall edges, doorframes and railings through the edited area: reject breaks, bends or newly invented geometry. Look for repeated textures, smudged signs and patches that do not match nearby surfaces. Compare the reconstructed background with your current reference, including object positions and visible condition. Finally, inspect areas outside the requested edit. A clean removal is still unacceptable if the output changes a window, widens a doorway, alters furniture or erases wear elsewhere.
Choose the next action from the failure
If a small visible remnant remains and the underlying scene is verifiable, you can try a more precise instruction from the original image, naming the remnant and its location. Review the whole new output again. If the model invents a fixture, changes room proportions or reconstructs an area you cannot verify, reject the result and reshoot rather than repeatedly polishing a plausible invention. Keep a short record of the requested change and any differences you found. Passing an artifact check only means you did not find those artifacts; it does not establish that hidden details are accurate.
Keep the original and make the edit clear
Save an accepted edit separately from the original and label it clearly, for example: AI-edited photograph; passerby removed. Use it only where that disclosure and the image still represent the venue honestly. Do not present it as an untouched record of attendance, an incident, property condition or another documentary event. Do not use people removal to conceal defects or accessibility and safety features, even if those features were partly obscured in the original. Your next step is to select one photograph, identify the exact area you would need to verify, and decide whether a prompt-based edit is worth trying. If verification depends on guessing, take a new photograph.