A minimal Apple-like flow: upload one photo, paint the mask, describe the edit, then compare and download the final result.
Pick one image to edit. You’ll paint the exact editable area in the next step.
The app will send the original image, your mask image, and a hidden safety prompt plus your own instruction to Gemini.
The hidden precision prompt below will always be added first. The user only writes what should change.
The app sends your image and mask to Gemini, then re-applies the original photo everywhere you did not paint, so changes stay inside the mask even if the model over-edits.
Sending image, mask, and prompt to Gemini.
0.0s
Same frame for both images — drag the line or use the slider. Original below the line, edited above.