Key takeaways
- Identify one subject by position and appearance.
- Keep the first edit narrow; do not combine unrelated removals.
- Inspect the removed subject’s boundary, shadow and reflection.
- Confirm you have a lawful basis to edit and publish the image.
Prepare the source and the permission context
Start with the original image rather than a screenshot or compressed social-media copy. More visible background around the subject gives the model more context for reconstruction, but no input guarantees a clean result.
Before editing a recognizable person, consider who owns the image, whether the person has privacy or publicity rights, and whether the final use changes the meaning of the scene. ImgObjectRemover’s Terms require the uploader to hold the rights or other lawful basis needed for the image and instruction.
- Use JPG, PNG or WebP up to 16MB.
- Keep the person and surrounding background visible.
- Avoid presenting an edited documentary image as an unaltered record.
- For sensitive, advertising or high-impact use, obtain appropriate professional advice.
Write a prompt that identifies one person
A useful prompt combines subject, position and one or two visible traits. “Remove the person” is ambiguous in a crowd; “remove the person in the red coat at the far right” gives the model a narrower target.
Describe what to remove, not what to redesign. The service already asks the model to reconstruct the hidden area while keeping the rest of the composition unchanged.
- Good: Remove the person in a red coat at the far right.
- Good: Remove the blurred pedestrian behind the bicycle.
- Weak: Clean the image and make it better.
- Risky: Remove everyone and completely rebuild the street.
Review the result at the problem areas
Compare before and after rather than checking only the empty space. Look for partial limbs, duplicated objects, warped faces, broken railings, inconsistent paving and shadows that no longer belong to anything.
If the subject overlaps another person or a foreground object, a narrower crop or a more exact description may help. A second successful edit is a separate task and may use additional credits under the published credit rules.
- Zoom around hair, hands and clothing edges.
- Check mirrors, windows, water and polished surfaces for reflections.
- Check the ground for a remaining cast shadow.
- Confirm text, logos and unrelated people did not change.
Know when not to rely on a generative edit
Large subjects hide more background, so the model must invent more pixels. The output can look plausible without matching what was actually behind the person. That makes the edit unsuitable as factual evidence of the original scene.
Keep the original file and disclose material edits when the context requires it. The NIST AI Risk Management Framework is a useful general reference for documenting and reviewing AI-assisted outputs, although it is not a product-specific approval.
People-removal publishing checklist
Before download or publication, record the source, instruction and review date when the image matters to a customer, campaign or public claim.
- Rights or permission checked.
- Target person removed without changing unrelated people.
- Edges, hands, faces, shadows and reflections reviewed.
- Material edit disclosed where appropriate.
- Original and edited files retained under an appropriate access policy.
Frequently asked questions
Can AI remove a person from a crowded photo?
It can attempt the edit, but overlap makes the task harder because the model must reconstruct hidden people and background. Identify one person precisely, inspect neighboring faces and limbs, and treat a plausible-looking result as a generated reconstruction rather than proof of the original scene.
Will the original dimensions stay the same?
Not necessarily. Output dimensions, format and detail can differ from the source. The public test method defines a dimension-retention metric, but no measured baseline is published yet. Check the downloaded file before using it in a fixed-size workflow.
Is it legal to remove someone from a photo?
That depends on ownership, consent, location, purpose and applicable law. The tool does not determine your rights. Confirm that you have the necessary lawful basis, and obtain professional advice for advertising, sensitive people or other consequential uses.
Sources and evidence boundary
Sources were checked on 28 July 2026. External policies can change; use the linked primary source at decision time. Product-specific behavior is bounded by the current ImgObjectRemover code and policies.
- [1] NIST AI Risk Management FrameworkGeneral AI output review and risk documentation.
- [2] U.S. Copyright Office AI initiativePrimary-source background on evolving U.S. copyright and AI policy.
- [3] ImgObjectRemover TermsUploader rights, AI-result limitations and prohibited uses.