Qikuyueser.com

One Job per Reference Image

Added 2026-10

One Job per Reference Image — Chinese original prompt

我手上有[4]张参考图。请按[构图、光线、材质、色调]各分配一张,写明每张只取什么、不取什么,避免互相污染。再指出哪张是主参考(决定骨架)、哪张只作微调。输出分工表,并提醒:参考图越多,正文里越要减少冲突项,否则模型会平均化。
Site original

Chinese original

Open in ChatGPT

Example

Sample input

手上有四张参考图,分别擅长构图、光线、材质、色调。分配每张只取什么、不取什么,指出主参考。

The output should include

  • 四张各管一项,写明只取与不取
  • 指出哪张决定骨架、哪张只微调
  • 提醒参考图越多越要减少冲突项

Replace these placeholders 2

Swap the bracketed parts for your own details before sending.

  • [4]
  • [构图、光线、材质、色调]
How to use

参考图超过四张后,通常要删文字而不是加文字。

Try tweaking it

  • Swap style words for measurable parameters: light, lens, material, composition.
  • Ask for three variants of one prompt, then compare the differences.

What you should get

  • A parameter sheet you can act on (light, lens, material, framing)
  • Two or three alternatives with trade-offs

Common mistakes

  • Stacking words like "premium, cinematic, 8K" — no two results match.
  • Asking for a real brand or celebrity look and getting something close but unusable.

How to check the output

  • Parameters are executable — ready to use in a tool or hand to a designer.
  • It explains the trade-offs, not just the final look.

Where it works

  • Good for: breaking vague "premium feel" into concrete light, material, lens and composition parameters.
  • Watch out: for real brands, celebrities or third-party work, use equivalent descriptions to avoid infringement.

More in this topic