Image edit changes part of a picture you already have. You describe the change in words — swap the background, remove an object, change the colour of one item — and the model applies it while leaving the rest of the frame as it was. It is the alternative to rebuilding a scene from scratch when only one thing about it is wrong.
The instruction should name what changes and what stays. "Replace the grey background with a plain sand-coloured wall" is an edit. Re-describing the whole image works against you: the model already sees it, and a long restatement invites it to alter parts you were happy with.
One change at a time is the reliable approach. If you need three things altered, three passes usually give a better result than one instruction carrying all three, and it is much easier to see which pass introduced a problem.
It suits small, specific corrections: a distracting object, a wrong colour, a background that does not fit the product. It is a poor fit for changing the subject of the picture, for producing a person who does not exist, or for anything that requires the model to know facts it cannot see. Treat every edit as an experiment: keep the original, because a second pass on an already-edited image degrades faster than a fresh pass on the source.
Will it change the whole image? The intent is to change only what you describe, but small shifts elsewhere are common — check the whole frame, not just the area you edited.
How much does it cost? The credit cost appears beside the form before you submit. Credit prices are on pricing.
Where do I start? Open the Wan workspace and choose the image-edit tab. To build a picture from nothing first, use text to image.