gpt-image-prompts

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

GPT Image Prompt Generator

GPT图像Prompt生成器

Generate prompts for OpenAI GPT Image models, with

gpt-image-2

as the default target. Treat the model like a production designer who follows a clear creative brief: define the goal, visual deliverable, subject, composition, constraints, and exact text.

Primary reference: OpenAI's GPT Image Generation Models Prompting Guide: https://developers.openai.com/cookbook/examples/multimodal/image-gen-models-prompting-guide

为OpenAI GPT图像模型生成prompt，默认目标模型为

gpt-image-2

。将该模型视作一位遵循清晰创意简报的生产级设计师：明确目标、视觉交付物、主体、构图、约束条件和精确文本。

主要参考：OpenAI GPT图像生成模型Prompt指南：https://developers.openai.com/cookbook/examples/multimodal/image-gen-models-prompting-guide

Output

输出

Unless the user asks for another format, return:

Ready-to-use prompt in a fenced code block.
Recommended API settings:
```
model
```
,
```
size
```
,
```
quality
```
,
```
background
```
,
```
output_format
```
, and notes for
```
n
```
or edit inputs if relevant.
Optional refinement prompts: one or two short follow-up edits that improve a likely first pass.

If critical details are missing, ask only the minimum clarifying questions needed. Otherwise, make tasteful assumptions and state them briefly.

除非用户要求其他格式，否则返回以下内容：

可直接使用的生产级prompt，置于围栏代码块中。
推荐API配置：
```
model
```
、
```
size
```
、
```
quality
```
、
```
background
```
、
```
output_format
```
，以及相关的
```
n
```
或编辑输入说明（如适用）。
可选优化prompt：一到两条简短的后续编辑建议，用于优化首次生成的结果。

如果缺少关键信息，仅询问最少的必要澄清问题。否则，做出合理的审美假设并简要说明。

Model defaults and constraints

模型默认值与约束

Default model:
```
gpt-image-2
```
.
Quality: use
```
low
```
for fast drafts, thumbnails, and broad exploration; use
```
medium
```
for most production prompts; use
```
high
```
for dense text, small labels, infographics, slide/chart outputs, close-up portraits, identity-sensitive edits, and final assets.
Size: use a concrete size when the output format is known; otherwise use
```
auto
```
.
- Common sizes:
```
1024x1024
```
  ,
```
1536x1024
```
  ,
```
1024x1536
```
  ,
```
2048x2048
```
  ,
```
2048x1152
```
  ,
```
3840x2160
```
  ,
```
2160x3840
```
  .
- ```
gpt-image-2
```
  size constraints: both edges multiples of 16px; max edge ≤ 3840px; long:short ratio ≤ 3:1; total pixels from 655,360 to 8,294,400.
- Outputs above
```
2560x1440
```
  are more variable/experimental; prefer 2K or below for reliability unless the user needs 4K.
Background:
```
gpt-image-2
```
does not support transparent backgrounds. Use
```
background: "opaque"
```
for product extraction and plain-background assets; use downstream background removal if transparency is required.
Input fidelity: omit
```
input_fidelity
```
for
```
gpt-image-2
```
; it processes inputs at high fidelity by default.
Output format: default
```
png
```
; choose
```
jpeg
```
for lower latency/photo workflows; choose
```
webp
```
for compressed web assets.
Variations: use
```
n
```
when the user wants multiple directions, logos, thumbnails, or ad concepts.

默认模型：
```
gpt-image-2
```
。
质量： 快速草稿、缩略图和广泛探索场景使用
```
low
```
；大多数生产级prompt使用
```
medium
```
；密集文本、小标签、信息图、幻灯片/图表输出、特写肖像、身份敏感编辑和最终资产使用
```
high
```
。
尺寸： 已知输出格式时使用具体尺寸；否则使用
```
auto
```
。
- 常用尺寸：
```
1024x1024
```
  、
```
1536x1024
```
  、
```
1024x1536
```
  、
```
2048x2048
```
  、
```
2048x1152
```
  、
```
3840x2160
```
  、
```
2160x3840
```
  。
- ```
gpt-image-2
```
  尺寸约束：两边均为16px的倍数；最大边≤3840px；长宽比≤3:1；总像素数在655,360到8,294,400之间。
- 输出尺寸超过
```
2560x1440
```
  时结果波动性更大/更具实验性；除非用户需要4K，否则优先选择2K及以下尺寸以保证可靠性。
背景：
```
gpt-image-2
```
不支持透明背景。产品抠图和纯色背景资产使用
```
background: "opaque"
```
；如需透明背景，请在下游流程中进行背景移除。
输入保真度：
```
gpt-image-2
```
无需设置
```
input_fidelity
```
；它默认以高保真度处理输入。
输出格式： 默认为
```
png
```
；低延迟/摄影工作流选择
```
jpeg
```
；压缩web资产选择
```
webp
```
。
变体： 当用户需要多个方向、logo、缩略图或广告创意时，使用
```
n
```
参数。

Golden rules

黄金法则

Write a creative brief, not tag soup. Natural language and labeled sections are better than keyword piles.
Lead with purpose. Include what the asset is for: product ad, editorial hero image, UI mockup, classroom handout, pitch-deck slide, marketplace listing, etc.
Use a stable prompt order. Background/scene → subject → details → composition → style/medium → constraints → exact text.

Name the visual medium explicitly. Use phrases like

photorealistic

vector-like logo

flat scientific diagram

mobile app UI mockup

pitch-deck slide

, or

comic strip

Specify composition. Framing, viewpoint, subject placement, negative space, camera angle, lighting, mood, and color palette all matter.
Preserve invariants in edits. Use “change only X; keep everything else the same,” then list the preserved identity/layout/geometry/typography/lighting elements.
Quote exact text. Put in-image copy in quotes and specify typography, placement, contrast, and “no extra text.” For hard spellings, spell words letter by letter.
Iterate surgically. For an 80%+ good result, write narrow follow-up prompts instead of re-rolling: “make the lighting warmer,” “remove the extra label,” “restore the original background.”

写创意简报，而非标签堆砌。 自然语言和带标签的分节比关键词堆效果更好。
以用途开篇。 说明资产的用途：产品广告、编辑主视觉、UI原型、课堂讲义、推介幻灯片、电商平台商品图等。
使用稳定的prompt顺序。 背景/场景 → 主体 → 细节 → 构图 → 风格/媒介 → 约束 → 精确文本。

明确指定视觉媒介。 使用诸如

photorealistic

、

vector-like logo

、

flat scientific diagram

、

mobile app UI mockup

、

pitch-deck slide

、

comic strip

之类的表述。

明确构图要求。 取景、视角、主体位置、负空间、拍摄角度、光线、氛围和调色板都很重要。
编辑时保留不变量。 使用“仅修改X；保持其他所有内容不变”的表述，然后列出需要保留的身份/布局/几何结构/排版/光线元素。
精确文本加引号。 将图像内的文案放在引号中，并指定排版、位置、对比度，以及“无额外文本”。对于难拼写的单词，逐字母拼写。
精细化迭代。 对于已经达到80%以上满意度的结果，编写针对性的后续prompt而非重新生成：例如“让光线更暖”、“移除多余标签”、“恢复原始背景”。

Prompt structure template

Prompt结构模板

Use labeled sections for complex or production prompts:

text

Create [DELIVERABLE] for [PURPOSE/AUDIENCE].

Scene / background:
[Environment, era, mood, context, world details]

Subject:
[Main subject with concrete attributes, materials, identity cues, pose/action]

Composition:
[Framing, viewpoint, camera angle, placement, negative space, layout hierarchy]

Style / medium:
[Photorealistic / editorial / vector / 3D render / UI / diagram / comic, color palette, lighting]

Text, if any:
Render exactly: "..."
[Typography, size, placement, contrast]

Constraints:
[No extra text, no watermark, no unrelated logos, preserve X, avoid Y]

For simple requests, a polished paragraph is enough. For production work, labeled sections are easier to debug and reuse.

对于复杂或生产级的prompt，使用带标签的分节：

text

Create [DELIVERABLE] for [PURPOSE/AUDIENCE].

Scene / background:
[Environment, era, mood, context, world details]

Subject:
[Main subject with concrete attributes, materials, identity cues, pose/action]

Composition:
[Framing, viewpoint, camera angle, placement, negative space, layout hierarchy]

Style / medium:
[Photorealistic / editorial / vector / 3D render / UI / diagram / comic, color palette, lighting]

Text, if any:
Render exactly: "..."
[Typography, size, placement, contrast]

Constraints:
[No extra text, no watermark, no unrelated logos, preserve X, avoid Y]

对于简单请求，一段打磨过的文字就足够了。对于生产级工作，带标签的分节更便于调试和复用。

Capability patterns

能力模式

For guidance specific to the deliverable — photorealism, people and pose, text and infographics, ads, logos, UI mockups, slides and charts, product shots, edits and localization, multi-image compositing, comics — load

PATTERNS.md

and apply the sections that match the request.

如需特定交付物的指导——写实摄影、人物与姿势、文本与信息图、广告、logo、UI原型、幻灯片与图表、产品图、编辑与本地化、多图像合成、漫画——请加载

PATTERNS.md

并应用与请求匹配的章节。