工作流指南
10 个 ChatGPT 图片提示词:商品图、营销配图与日常工作

实用的 ChatGPT 图片提示词需要写清任务、主体、构图与关键细节。下面十个模板对应常见工作,并为每个输出提供检查方法。它们是为 Images 2.5 编写的起点,不是经过实测的性能承诺。替换方括号中的内容后,再按实际用途审阅生成结果。
来源: 官方参考 1. 核对日期:2026 年 9 月 11 日。
模板为本次原创,官方提示词文档提供能力参考。代码块保留英文原始指令,中文说明帮助理解与调整;某种措辞并不能保证完美出图。
1. 为文案留白的商品图
需要准确展示在售商品时,使用真实参考照片。
Create a landscape product composition using the uploaded [PRODUCT].
Position it in the left third on a warm stone surface.
Use a quiet studio background with generous empty space on the right.
Keep the product's proportions, packaging, colors, and printed details.
Do not generate a headline; I will add the approved copy separately.**检查:**先对照商品再看风格。包装改变了,就不能用来准确展示该商品。
2. 表达明确观点的博客插图
Create an editorial illustration about [ARTICLE'S CENTRAL IDEA].
Show [CONCRETE SUBJECT] performing [VISIBLE ACTION].
Use a restrained paper-cut style with three flat colors.
Make the main relationship understandable at thumbnail size.
Avoid lettering, decorative charts, and tiny secondary objects.**检查:**请没读过文章的人说说图片表达什么。含义不清时,简化视觉比喻。
3. 活动海报概念
Design a poster concept for a neighborhood book exchange.
Include exactly these three lines:
"Bring a Book"
"Take a Story"
"Saturday at 10 AM"
Use bold readable type, a simple stack-of-books illustration,
and generous margins. Do not invent a venue, sponsor, or extra copy.**检查:**逐字核对。示例三行英文应替换成活动已确认的文案。需要可编辑字体或印刷文件时,把结果作为设计概念。
4. 自然可信的背景替换
Edit the uploaded photograph of [OBJECT].
Replace the surroundings with a tidy home workspace at midday.
Keep the object and its visible identifying details.
Make the surface contact and shadow fit the new scene.
Add no people, devices, or text that were not requested.**检查:**关注悬浮物体、不一致的反射,以及与原图不符的透视。
5. 演示文稿章节配图
Create a wide visual for a presentation section about [TOPIC].
Use one large [OBJECT OR SCENE] toward the lower right.
Keep the upper-left half visually calm for a slide heading.
Use a professional editorial style suitable for [AUDIENCE].
Do not draw a slide border, title, logo, or fake interface.**检查:**把图片放进真实幻灯片,确认标题在留白区域清楚可读。
6. 单一视觉焦点的缩略图
Create a thumbnail concept for a video about repairing a bicycle.
Show a clearly visible bicycle wheel and a hand holding a tire lever.
Use a simple workshop setting and strong separation from the background.
Keep the image focused on the repair; omit text and extra tools.
Make the subject readable when the image is displayed very small.**检查:**看清手、工具和车轮结构,再缩小到实际展示尺寸判断构图。
7. 团队页面的人像插画
Use the uploaded portrait as a reference for a warm editorial drawing.
Keep the person's recognizable facial features and hairstyle.
Use visible pencil texture, a plain background, and natural proportions.
Frame the head and shoulders with consistent space above the head.
Avoid changing age, expression, or adding clothing accessories.**检查:**请本人确认是否可识别、是否适合预定用途。
8. 只改一处颜色
In the uploaded room image, change the lampshade from red to muted olive.
Keep the lamp base, surrounding furniture, camera view, and room layout.
Make no other decorative changes and introduce no additional objects.**检查:**对照整张画面,不只看台灯。其他位置的小变化也可能破坏已确认设计。
9. 根据真实脚本生成分镜
Create one storyboard frame for this shot: [SHOT DESCRIPTION].
Use the attached character reference for [CHARACTER].
Show [ACTION] from [CAMERA POSITION].
Use a simple grayscale drawing with clear staging.
Do not add dialogue, captions, extra characters, or later story events.**检查:**与脚本逐项对照。制作连续画面时,检查服装、道具和位置,不要默认它们一致。
10. 新用户引导邮件配图
Create a friendly illustration for an onboarding email about [FIRST TASK].
Show one person completing [SPECIFIC ACTION] with a clear result.
Use a simple rounded illustration style and a light background.
Keep small details minimal so the image works on a narrow phone screen.
Do not invent product screenshots, logos, or interface text.**检查:**图片应帮助解释邮件中的首次任务。通用的漂亮场景也可能表达错误操作。
保留有助于复查的信息
失败时先判断问题属于构图、主体身份、准确文字,还是意外修改,再调整那一项要求,不要只叠加风格形容词。
保存任务、参考图、完整提示词、输出和采用原因。这些模板也可以改用于其他图片工具,但控制项和效果会不同。选一个模板,替换成本周真正需要完成的图片任务。
继续完成图片任务
ChatGPT Images 2.5 更新了什么?从一个实际任务开始 · ChatGPT 图片编辑教程:用 Images 2.5 修改已有图片.
在 Ottermind 图片工作区使用你的创作说明. 请查看界面当前提供的模型。提供的 Ottermind 界面截图显示,模型菜单已列出 Flare 与 Sunburst,均标为 Premium,当前选中 Sunburst。
阅读 GPT Image 2 案例集. 这些属于上一代模型案例,不能作为 Images 2.5 的效果证据。
