图像提示词：构图优先，细节其次

图像提示词在先确立框架再装饰细节时效果最佳。模型可以将“四分之三肖像，主体位于右侧三分之一，硬侧光”作为一组视觉约束。这与提示词是否富有诗意关系不大。

COPYABLE PROMPT, GENERATED OUTPUT

PROMPT

Editorial portrait of a ceramicist in a cobalt-blue studio,
three-quarter profile positioned on the right third of a vertical frame,
negative space to the left, 85mm portrait lens, eyes in sharp focus,
hard window light from camera left, deep but readable shadows,
subtle clay dust in the air, tactile matte surfaces, restrained film color.

OUTPUT — GENERATED WITH MANIFOLDGEN

从大到小构建

1. 主体与动作：观众在看什么？
2. 构图：肖像、广角、居中、留白、前景/背景。
3. 光学：镜头感、景深、焦点平面、相机高度。
4. 光线与材质：方向、柔和度、反射、表面质感。
5. 收尾：色彩分级或媒介，作为最后的微调。

可复制的提示词

Editorial portrait of a ceramicist in a cobalt-blue studio,
three-quarter profile positioned on the right third of a vertical frame,
negative space to the left, 85mm portrait lens, eyes in sharp focus,
hard window light from camera left, deep but readable shadows,
subtle clay dust in the air, tactile matte surfaces, restrained film color.

对于批量生成，一次只改变一个变量：取景、光线方向或材质。这样输出结果能让你有所收获。如果每次变化同时改变主体、镜头、情绪和色彩，你就无法判断哪个指令起了作用。