Nano Banana Pro Conversational Editing: Background, Passersby, Outfit Swaps

The biggest cultural difference between the Nano Banana family and Midjourney is 'edit, don't regenerate': if an image is already 80% right, keep it and edit via conversation — it saves both quota and luck. The key to success is that 'everything else stays exactly the same' preservation clause; without it, the model often redraws the whole image without being asked. In practice it will occasionally pretend to make a change but return the original image unchanged — rephrasing and resending usually fixes this. Changing the background, removing a passerby, swapping an outfit, or changing the weather all follow the same sentence pattern — just swap out the two variables to reuse it.

填下面的欄位，上方 prompt 會即時替換 [方括號] 內容。填好後按「複製組好的 prompt」直接丟進工具。

請對我上傳的這張照片做局部編修，只改我指定的地方，其他全部保持不變。

這次要改：[要修改的部分]
改成：[想要的結果]

編修規則：
1. 主體人物的臉部、姿勢、表情完全不動
2. 新元素必須符合原照片的光源方向與色溫，邊緣自然融合、不能有剪貼感
3. 如果這個改動會影響到地面陰影或玻璃反光，請一併修正，讓畫面物理上合理
4. 不要重新生成整張圖，也不要改變構圖與裁切
5. 輸出解析度與原圖相同

完成後我會繼續在這個對話裡追加修改，每次同樣只動我講的那一處。