First record the four actual Describe suggestions. Select one and change only composition (for example negative space on the left), preserving blue pot, yellow pear and red cloth. No fabricated Describe return, no guarantee of copying the original image.

1. 准备三件静物输入，先人工记录主体、灯光、机位与留白
2. 拖入Describe区域并保存四条真实返回建议，避免刷新丢失
3. 把每条拆成可见主体、风格和构图词，标出遗漏或臆测
4. 选择一条并只改构图目标，记录实际修改词句
5. 用当前账户可用图像模型生成并保存真实输出
6. 逐项对照主体数、留白和机位差异，交付提示词与图片而不承诺复制原图

Target account execution: not performed. Save actual output separately.
