如果你在用 GPT Image 2 做营销素材,真正能把结果拉稳定的,通常不是“更会写形容词”,而是一套能重复使用的结构。
这篇文章把 GPT Image 2 prompt templates 整理成一个可复用的模板库:广告图、产品主图、海报、社媒 quote card、YouTube 缩略图、UI mockup,以及一组用来修边角问题的微模板。你可以把它当成团队内部的提示词 baseline。
- 立即生成:/ai-image-generator
- 纯文生图入口:/text-to-image/gpt-image-2
- 编辑、局部修改和变体:/image-to-image/gpt-image-2
- 返回工作台:/
模板索引
这套 GPT Image 2 提示词模板 主要覆盖:
- 可复用的通用 prompt 结构
- 电商产品主图
- 三张一组的广告创意变体
- 带 headline 和 CTA 的海报 / flyer
- 可批量产出的社媒 quote card
- 大字清晰的 YouTube 缩略图
- 落地页风格的 UI mockup
- before / after 对比图
- 一组快速修复问题的微模板
你真正需要的提示词结构
大多数“提示词列表”之所以不好用,不是模板不够多,而是少了约束。写营销素材时,优先把 prompt 当成规格说明来写:
- 资产类型与平台
- 主体事实与不可变信息
- 版式与布局锚点
- 需要逐字出现的文本
- 风格约束
- 输出约束
只要这六层写清楚,很多 GPT Image 2 prompt templates 的稳定性就会明显提升。
生成前的质量闸门
在 reroll 之前,先快速对一遍:
- 有没有写清楚 asset type 和 platform
- 有没有锁定主体事实
- 有没有给布局、margin、text-safe zone
- 文案是不是逐字引用,且限制了不要额外文字
- 标题、副标题、CTA 的层级是不是明确
- 是否说明了比例、背景和输出要求
- 每轮是否只改一个变量
先记住这几条文字可读性规则
如果你的素材里有文字,先把这几个约束写进去:
- headline 保持短句
- 用 quotes 写 exact strings,并加上 verbatim
- 给文字留出 clean text-safe zone
- 指定 safe margins
- 提高文字与背景的对比度
- 明确写 no extra text
模板 0:通用填空模板
先把这份通用骨架存下来。后面无论做广告图、主图还是海报,都可以从这里起步。
gpt-image-2 prompt templates — reusable fill-in spec
You are generating a {ASSET_TYPE} for {PLATFORM}.
Subject (non-negotiables):
- {PRODUCT/SUBJECT}
- Materials/colors: {FACTS}
- Must include: {LOGO_RULES}
- Must NOT include: {FORBIDDEN_CLAIMS}
Composition + layout:
- Aspect ratio: {AR}
- Camera/angle: {CAMERA}
- Background: {BG}
- Place subject: {SUBJECT_PLACEMENT}
- Leave negative space for text: {TEXT_AREA}
- Safe margins: {MARGINS}
Text (verbatim, no extra text):
Headline: "{HEADLINE}"
Subhead: "{SUBHEAD}"
CTA: "{CTA}"
Style constraints:
- Lighting: {LIGHTING}
- Color palette: {PALETTE}
- Typography style: {TYPE_RULE}
- Overall vibe: {VIBE} (avoid generic adjectives; be specific)
Hard constraints:
- No watermark
- No misspellings
- No additional logos or text
- No trademarks or branded characters unless explicitly requested模板 1:电商产品主图
适合商品详情页、主视觉或广告首图。重点是把主体、构图、文案和 CTA 一次写完整。
gpt-image-2 prompt templates — product hero (e-commerce)
Create a Shopify product hero image.
Subject (non-negotiables):
- Product: matte black insulated water bottle, 750ml
- Must show: stainless steel cap, subtle condensation, no visible brand name
Composition + layout:
- Aspect ratio: 1:1
- Camera: 50mm product photo, eye-level, sharp focus
- Background: clean light gray studio seamless
- Place product: centered, slight 10° rotation
- Leave negative space: top 20% for headline
- Safe margins: 12% padding
Text (verbatim, no extra text):
Headline: "Stay Cold. All Day."
Subhead: "Vacuum insulated • 750ml"
CTA: "Shop Now"
Style constraints:
- Lighting: softbox, gentle shadow under product
- Palette: grayscale with a single cyan accent line
- Typography: bold sans headline, smaller regular subhead, high contrast
Hard constraints:
- No extra text, no watermark, no additional products模板 2:广告创意三连图(一次只改一个变量)
想做可控变体时,最有效的方法还是 variant ladder:每张图只改一个点,其他全部保持一致。
gpt-image-2 prompt templates — ad creative set (variant ladder)
Generate a set of 3 Meta ad creatives for the same product.
Invariant rules (keep identical across all 3 images):
- Product: minimalist desk lamp, brushed aluminum
- Background: clean off-white
- Lighting: soft studio, natural shadow
- Brand palette: black + off-white + one accent color
- Typography: bold sans headline, simple CTA pill button
- Safe margins: 12%
- No extra text beyond provided strings
Layout:
- Aspect ratio: 4:5
- Product placement: right side, 60% height
- Text block: left side, vertically centered
Variation A (variable = angle):
- Camera angle: 3/4 view
Headline: "Better Light. Less Clutter."
Subhead: "Minimal design • Warm glow"
CTA: "Get Yours"
Variation B (variable = accent color):
- Accent color: teal
Headline: "Your Desk, Upgraded."
Subhead: "Aluminum body • Soft shadows"
CTA: "Shop Lamps"
Variation C (variable = headline):
Headline: "Focus Mode Starts Here."
Subhead: "Clean lines • Calm lighting"
CTA: "See Details"模板 3:海报 / flyer(headline + CTA)
海报最容易翻车的地方是文字可读性,所以这个模板先把层级和留白写死。
gpt-image-2 prompt templates — poster/flyer with readable text
Create a modern event poster with readable text.
Composition + layout:
- Aspect ratio: 2:3
- Background: abstract gradient with subtle grain
- Provide a clear text-safe area at the top (30% height)
- Provide a CTA-safe area at the bottom (20% height)
- Keep 10% margins on all sides
Text (verbatim, no extra text):
Headline: "SPRING STUDIO SALE"
Subhead: "Up to 40% off — 3 days only"
CTA: "This Weekend"
Style constraints:
- High contrast typography
- Minimal decorative elements near text
- No random letters, no extra dates, no watermarks模板 4:社媒 quote card(适合系列化)
这类图适合批量生产。只要背景、字号和 accent element 固定住,就很适合连载。
gpt-image-2 prompt templates — social quote card
Create a social quote card for Instagram.
Layout:
- Aspect ratio: 1:1
- Background: solid #0B0F14 with a subtle texture
- Safe margins: 14%
- Center-aligned text block
Text (verbatim, no extra text):
Headline: "Clarity beats intensity."
Subhead: "Build systems, not sprints."
CTA: "@yourbrand"
Style constraints:
- Typography: strong sans headline, lighter subhead
- One accent element: a thin cyan underline under the headline
- No additional text, no watermark模板 5:YouTube 缩略图(大字清晰)
缩略图失败,通常不是画面不够热闹,而是文字太多、太小、太乱。
gpt-image-2 prompt templates — YouTube thumbnail text
Create a YouTube thumbnail with very large readable text.
Layout:
- Aspect ratio: 16:9
- Subject: a creator at desk, expressive but clean background
- Leave left 40% as clean text area
- Safe margins: 8%
Text (verbatim, no extra text):
Headline: "STOP GUESSING"
Subhead: "Use templates"
Style constraints:
- Headline extremely large, high contrast, thick stroke
- Subhead smaller, aligned under headline
- No tiny text, no extra words, no watermark模板 6:营销落地页 UI mockup
如果你要出一张“像产品页截图一样”的营销图,请优先约束布局和排版,而不是只写模糊的风格词。
gpt-image-2 prompt templates — landing page UI mockup
Create a clean SaaS landing page hero UI mockup (single screenshot style).
Layout:
- Aspect ratio: 16:9
- Hero section with a left text column and right product preview
- 12-column grid, generous whitespace, safe margins 10%
Text (verbatim, no extra text):
Headline: "Design faster with templates"
Subhead: "Create consistent visuals in minutes."
CTA: "Try it free"
Style constraints:
- Minimalist UI, modern sans typography
- Soft shadows, subtle gradients
- No real brand logos, no fake app store badges, no watermark模板 7:Before / After 对比图
这种图的关键不是华丽,而是 layout 必须稳定,观众一眼就能读懂变化在哪里。
gpt-image-2 prompt templates — before/after comparison layout
Create a before/after comparison image with strict layout.
Layout:
- Aspect ratio: 4:5
- Split screen vertical: left = BEFORE, right = AFTER
- Large labels at the top with 10% margin
- Keep center divider clean and straight
Text (verbatim, no extra text):
Headline: "BEFORE"
Subhead: "AFTER"
CTA: "See the difference"
Style constraints:
- Consistent lighting on both sides
- No extra text, no watermark微模板库:快速修边角
下面这组短模板,适合在你已经有基线图之后,单独补背景、阴影、边距、文案长度或商标风险控制。
gpt-image-2 prompt templates — background cleanup
Replace the background with a clean {COLOR} studio backdrop. Keep the subject unchanged. No extra text.Prompt micro-template — add a subtle shadow
Add a realistic soft shadow under the subject. Keep lighting consistent. No extra objects.Prompt micro-template — remove extra artifacts
Remove any stray text, watermarks, logos, or artifacts. Keep composition unchanged.gpt-image-2 prompt templates — enforce margins
Recompose with {PERCENT}% safe margins on all sides. Keep subject scale similar.gpt-image-2 prompt templates — text-safe zone
Create a clean text-safe zone at {PLACEMENT} occupying {PERCENT}% of the canvas. No patterns behind the zone.gpt-image-2 prompt templates — high-contrast typography
Increase text contrast: darker text on lighter background or vice versa. Keep the exact strings unchanged.gpt-image-2 prompt templates — one-variable rule
Generate 3 variants changing ONLY {VARIABLE}. Keep everything else identical (layout, palette, typography).gpt-image-2 prompt templates — brand palette lock
Use only these colors: {HEX_1}, {HEX_2}, {HEX_3}. Do not introduce new accent colors.gpt-image-2 prompt templates — consistent camera
Keep camera: {LENS} {ANGLE}. Keep lighting: {LIGHTING}. Do not change perspective.gpt-image-2 prompt templates — simplify copy
Rewrite the headline to 3–5 words while preserving meaning. Keep subhead under 8 words.Prompt micro-template — CTA button style
Render the CTA as a simple rounded pill button with high contrast. Keep text verbatim.Prompt micro-template — avoid trademark risk
Do not include any real brand logos, trademarks, or recognizable characters. Use generic placeholders only.一套更稳的迭代流程
如果你要连续做 10 张广告图、20 张产品图,流程比任何单条 prompt 都重要:
- 先写 invariants:配色、字体规则、布局、留白、no extra text
- 先出一张 baseline 图
- 需要定向修改时,切到 /image-to-image/gpt-image-2
- 每轮只改一个变量,例如 headline、角度或 accent color
常见问题
为什么文字总是拼错或跑偏?
大多数情况是文案太长,或者没有写清楚 exact strings、hierarchy 和 placement。先缩短 headline,再把 text-safe zone 和 safe margins 写进去。
为什么每次 reroll 都像换了一个设计系统?
因为你把“感觉”写多了,把约束写少了。把配色、排版、布局、margin 和按钮样式固定下来,结果会稳很多。
什么时候该从文生图切到编辑?
当你的整体布局已经接近可用,只差 headline、间距、背景清理或局部修正时,就该切到 /image-to-image/gpt-image-2 了。
下一步
- 先做第一张素材:/ai-image-generator
- 做纯文生图版本:/text-to-image/gpt-image-2
- 做编辑和局部迭代:/image-to-image/gpt-image-2
- 返回工作台:/

