提示 ChatGPT 的核心不是找一句万能咒语，而是交代角色、真实任务、缺失信息、硬限制、输出样例和迭代规则。做图片时，它应该先帮你整理生产 brief，再交给 Vogue AI 生成。
TL;DR
- 先说明图片用途、发布渠道，以及哪些元素必须保持可识别。
- 让 ChatGPT 先问缺失问题，再写最终提示词。
- 明确裁切、参考图、文字规则、比例、光线和检查顺序。
- 用 ChatGPT 把弱提示词改写成稳定的 Vogue AI 提示词块。
- 在 Vogue AI 生成后，只针对最大失败点继续迭代。
好的 ChatGPT 提示词是一份回答契约：角色、上下文、任务、限制、示例和迭代规则。
好的 ChatGPT 提示词是一份回答契约：角色、上下文、任务、限制、示例和迭代规则。
| Part | What to give ChatGPT | Image prompt example |
|---|---|---|
| Role | The expert lens it should use. | Act as an image creative director. |
| Context | The product, audience, channel, and goal. | This is for a product page and launch post. |
| Constraints | Ratio, reference, text policy, fixed details. | 4:5, no text, preserve product shape. |
| Example | The structure or style of answer you need. | Return one prompt plus one review checklist. |
| Iteration rule | How it should revise after a result. | Change only the largest visible failure. |
Workflow
粗略视觉想法要变成可用提示词时，按这个顺序走：给 brief、补问题、确认目标、写结构化提示词、生成、只修最大失败点。
| Step | ChatGPT instruction | Result you want |
|---|---|---|
| 1 | Ask clarifying questions before writing. | Missing context appears before the prompt. |
| 2 | Write one structured image prompt. | A paste-ready production brief. |
| 3 | Create channel variations. | Product, campaign, and social versions. |
| 4 | Diagnose the first generation. | One visible failure named clearly. |
| 5 | Revise one control. | A comparable second prompt. |
Copyable prompts
The public prompt blocks stay in English so they remain copyable in ChatGPT and Vogue AI.
- Act as an image creative director. I need a [channel] visual for [product or subject]. Ask me up to three missing questions first. Then write one production-ready prompt with subject, context, composition, lighting, style, reference-image instructions, aspect ratio, and a first-result review checklist.
- Rewrite this weak image prompt into a controlled Vogue AI prompt: "[paste weak prompt]". Keep the goal, remove vague filler, add concrete subject details, camera/framing, material or wardrobe, lighting, output rules, and one sentence explaining what to inspect after generation.
- Create three prompt variations for the same visual brief: one product-page version, one campaign-poster version, and one social-feed version. Keep the core subject stable, change only composition, channel ratio, background, and energy level.
- Use this uploaded image as reference. State exactly what the reference must control: [face identity/product shape/UI hierarchy/palette]. Then write a prompt that changes only [background, lighting, wardrobe, camera angle, or mood] while preserving those fixed details.
Reference-image handoff
- Say what the reference controls.
- Say what may change.
- Name what to inspect first after generation.
- Do not rely on style words to preserve identity.
Prompt variations
- Keep subject and fixed details stable.
- Change one channel variable at a time.
- Label each version by job, not by mood.
- Save the version that solved the visible failure.
Worked example
Weak request
Make a cool ad image for my jacket, cinematic and premium.
Better ChatGPT request
Act as an image creative director. Turn this into a Vogue AI GPT Image 2 prompt for a 4:5 launch visual. Ask only if product material, audience, or fixed logo placement is missing. The prompt must include subject, composition, lighting, reference-image handoff, no-text rule, and a review checklist.
Why it works
粗略视觉想法要变成可用提示词时，按这个顺序走：给 brief、补问题、确认目标、写结构化提示词、生成、只修最大失败点。
Mistakes and fixes
| Mistake | Fix first | Avoid |
|---|---|---|
| Prompt is vague | Add audience, channel, subject, and output format. | More impressive adjectives. |
| Result ignores identity | Add reference handoff and fixed details. | Changing style before identity is stable. |
| ChatGPT answer is too broad | Specify answer format and ask for one prompt. | Asking for a long brainstorm first. |
| Revision gets worse | Change one control only. | Rewriting the whole prompt after each result. |
Model choice
- Use GPT Image 2 for close instruction following and reference-sensitive image prompts.
- Use Nano Banana for fast visual branches after the brief is stable.
- Use Midjourney when mood and stylized exploration matter more than exact structure.
- Keep the ChatGPT brief stable before comparing models.
FAQ
提示 ChatGPT 的最佳方式是什么？
给出角色、上下文、任务、限制、输出格式示例和修改规则。
如何让 ChatGPT 写图片提示词？
要求它写出包含主体、构图、光线、风格、参考图交接、比例和检查项的图片 brief。
应该让 ChatGPT 生成图片还是写提示词？
先让 ChatGPT 整理提示词，再在具备模型和参考图控制的图片工作区生成。
为什么提示词很泛？
通常缺少受众、渠道、固定细节、输出规则或答案格式示例。
提示词需要多长？
足够包含任务和限制即可，不要堆砌空泛形容词。
ChatGPT 能改进坏提示词吗？
可以。粘贴弱提示词，说明失败结果，并要求一次只改一个控制点。
Vogue AI 适合做什么？
用来粘贴提示词、选择 GPT Image 2/Nano Banana/Midjourney、添加参考图并比较结果。