Local AI image editing: masks, boundaries and preserving untouched areas

Choose one target and state what must stay

To make a local AI edit, select the original image, mark the area to change, describe the change and compare the result with the original. This suits replacing an object, changing a material or fixing a shadow. For a new overall composition, prepare a new reference and generation task.

A mask communicates editing intent; it does not guarantee unchanged pixels outside the boundary. For clothing edits, specify the face, pose and background to preserve. For product backgrounds, specify the outline, logo and packaging text. Compare every protected detail, and use conventional masking and compositing when exact preservation is required.

Check the original and input limits

• Use an original image with recognizable edges. A canvas screenshot or blurred thumbnail is a poor starting point for detailed editing.
• Image imports must fit within 20,250,000 total pixels and an 8192px longest edge. If either limit is exceeded, choose to resize for upload or cancel. Keep the original file at its full size.
• Wait for upload to finish before opening local editing. Having an image on the canvas does not mean every model can edit it this way; use the workflow and parameters offered by the current tool.

Draw the selection and submit one edit

1. Select the image, open local editing and zoom until the target outline is clear.
2. Include the whole target with a little room for contact shadows or material transitions.
3. Check that faces, packaging text and logos were not included by mistake; remove unnecessary selected areas.
4. State one main change, its location and what to preserve. Check the tool parameters, then submit.
5. Open the result and compare it with the original at the same display size before adjusting the selection and retrying.

Author-written prompt example, not generation-tested

Scenario: the chair is correctly positioned and only its material needs changing. This writing example has not been tested in a generation run and does not guarantee a result. Change the chair first, then address any remaining shadow problem separately.

Replace only the selected red chair with a light gray fabric chair. Keep its position, orientation and size. Preserve the floor, window and surrounding furniture. Match the chair-leg shadows to the existing light direction. Add no objects.

Common problems and next steps

• The object looks pasted in: check for missing edges or contact shadows, then allow slightly more transition space.
• Faces, text or the background changed: return to the original, narrow the mask and simplify the instruction. If exact preservation still fails, composite with the original.
• Several unwanted changes appeared: keep one primary edit instead of changing the object, lighting and composition together.
• The result looks soft: open the full preview or download the file before deciding whether it is just an unloaded thumbnail.

Review at three scales and save the files

Review composition when zoomed out, subject and lighting at normal size, and outlines, small text and repeated textures when zoomed in. Check both the selected area and everything you asked to preserve. Retain the original and approved intermediate versions so the next revision can return to them.

Save the final file through the download action. Local media used for generation is uploaded, and generated results are returned as remote media URLs. Browser history is a recent activity record, not a permanent file backup. Save the original, final file and edit instructions separately before delivery.