如果你在用 GPT Image 2 做行銷素材,真正讓結果穩定下來的,通常不是更會寫形容詞,而是一套可重複使用的結構。
這篇文章把 GPT Image 2 prompt templates 整理成一個可重用的模板庫:廣告圖、產品主圖、海報、社群 quote card、YouTube 縮圖、UI mockup,以及一組專門用來修小問題的微範本。你可以把它當成團隊內部的提示詞 baseline。
- 立即生成:/ai-image-generator
- 純文生圖入口:/text-to-image/gpt-image-2
- 編輯、局部修改與變體:/image-to-image/gpt-image-2
- 返回工作台:/
範本索引
這套 GPT Image 2 提示詞範本 主要涵蓋:
- 可重複使用的通用 prompt 結構
- 電商產品主圖
- 三張一組的廣告創意變體
- 帶 headline 與 CTA 的海報 / flyer
- 可批量產出的社群 quote card
- 大字清楚的 YouTube 縮圖
- 落地頁風格的 UI mockup
- before / after 對比圖
- 一組快速修邊角的微範本
你真正需要的提示詞結構
多數「提示詞清單」之所以不好用,不是因為模板太少,而是因為少了約束。做行銷素材時,優先把 prompt 當成規格說明來寫:
- 資產類型與平台
- 主體事實與不可變資訊
- 版式與布局錨點
- 需要逐字出現的文字
- 風格約束
- 輸出約束
只要這六層寫清楚,很多 GPT Image 2 prompt templates 的穩定度就會明顯提高。
生成前的品質閘門
在 reroll 之前,先快速對一遍:
- 有沒有寫清楚 asset type 與 platform
- 有沒有鎖定主體事實
- 有沒有給布局、margin、text-safe zone
- 文案是不是逐字引用,而且限制了不要額外文字
- headline、副標與 CTA 的層級是否明確
- 是否說明了比例、背景與輸出要求
- 每輪是否只改一個變數
先記住這幾條文字可讀性規則
如果素材裡有文字,先把這些約束寫進去:
- headline 保持短句
- 用 quotes 寫 exact strings,並加上 verbatim
- 給文字留出 clean text-safe zone
- 指定 safe margins
- 提高文字與背景的對比度
- 明確寫 no extra text
範本 0:通用填空模板
先把這份通用骨架存下來。之後不管做廣告圖、主圖還是海報,都可以從這裡起步。
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範本 1:電商產品主圖
適合商品詳情頁、主視覺或廣告首圖。重點是把主體、構圖、文案與 CTA 一次寫完整。
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範本 2:廣告創意三連圖(一次只改一個變數)
想做可控變體時,最有效的方法仍然是 variant ladder:每張圖只改一個點,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"範本 3:海報 / flyer(headline + CTA)
海報最容易翻車的地方是文字可讀性,所以這個範本先把層級與留白寫死。
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範本 4:社群 quote card(適合系列化)
這類圖片很適合批量生產。只要背景、字級與 accent element 固定住,就很適合長期復用。
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範本 5:YouTube 縮圖(大字清楚)
縮圖失敗,通常不是畫面不夠熱鬧,而是文字太多、太小、太亂。
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範本 6:行銷落地頁 UI mockup
如果你要做一張像產品頁截圖的行銷圖,請優先約束布局與排版,而不是只寫模糊風格詞。
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範本 7:Before / After 對比圖
這種圖片的關鍵不是華麗,而是 layout 必須穩定,讓使用者一眼就知道變化在哪裡。
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微範本庫:快速修邊角
下面這組短模板,適合在你已經有 baseline 圖之後,單獨補背景、陰影、邊距、文案長度或商標風險控制。
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.一套更穩的迭代流程
如果你要連續做 10 張廣告圖、20 張產品圖,流程比任何單條 prompt 都更重要:
- 先寫 invariants:配色、字體規則、布局、留白、no extra text
- 先出一張 baseline 圖
- 需要定向修改時,切到 /image-to-image/gpt-image-2
- 每輪只改一個變數,例如 headline、角度或 accent color
常見問題
為什麼文字總是拼錯或跑位?
大多數情況是文案太長,或沒有寫清楚 exact strings、hierarchy 與 placement。先縮短 headline,再把 text-safe zone 和 safe margins 寫進去。
為什麼每次 reroll 都像換了一套設計系統?
因為你把「感覺」寫太多,把約束寫太少。把配色、排版、布局、margin 與按鈕樣式固定下來,結果就會穩很多。
什麼時候該從文生圖切到編輯?
當整體 layout 已經接近可用,只差 headline、間距、背景整理或局部修正時,就該切到 /image-to-image/gpt-image-2 了。
下一步
- 先做第一張素材:/ai-image-generator
- 做純文生圖版本:/text-to-image/gpt-image-2
- 做編輯與局部迭代:/image-to-image/gpt-image-2
- 返回工作台:/

