想 GPT Image 2 生成啱用嘅 poster、banner、slide 或產品相,prompt 最好當成 6 部分設計 brief:用途、主體、構圖、風格、文字同限制。 做 marketing asset 時,最容易影響成敗嘅位通常係比例、主體位置、留白、畫面文字要原文出現,以及 no extra text、no watermark、no unrelated logos 呢類 constraints。

Create a landscape editorial hero image for this Studio Global article: Prompt GPT Image 2: công thức brief thiết kế cho poster, banner, slide, ảnh sản phẩm. Article summary: Để GPT Image 2 tạo đúng poster, banner, slide hoặc ảnh sản phẩm, hãy viết prompt như một brief thiết kế 6 phần: mục đích, chủ thể, bố cục, phong cách, text và ràng buộc; nếu dùng API, cần kiểm tra đúng tên model vì tà.... Topic tags: ai, openai, chatgpt, image generation, prompt engineering. Reference image context from search candidates: Reference image 1: visual subject "Nếu bạn đang làm marketing và phải liên tục tạo hình ảnh cho quảng cáo, social post, landing page... thì việc biết cách viết prompt tạo ảnh AI sẽ giúp bạn tiết kiệm hàng giờ – và c" source context "Prompt tạo ảnh AI: Vũ khí mới cho marketer 2026" Reference image 2: visual subject "Nếu bạn đang làm marketing và phải liên tục tạo hình ảnh cho quảng
想 GPT Image 2 幫你生到真正可以放入 campaign、presentation deck 或產品頁嘅 asset,唔好由一句「整張靚 poster」開始。更穩陣嘅做法,是把 prompt 寫成一份細但清楚嘅設計 brief:張圖用喺邊度、主 visual 係乜、版面點排、邊段文字要原文出現,以及有乜嘢唔可以亂加或亂改。OpenAI 關於圖像生成和 image prompt 嘅指引,都強調要講清楚用途、構圖、風格、文字同 constraints;如果要求多,最好用有標籤嘅段落分開寫,等 model 更易跟住做。
**關於名稱:**本文沿用搜尋時常見嘅「GPT Image 2」叫法;但實際操作,尤其係用 API 時,應該以你當刻使用嘅官方文件或介面所列 model 名為準。今次提供嘅 OpenAI 來源在不同情境中分別提到 GPT Image Generation Models、API 入面嘅
gpt-image-1.5、ChatGPT Images 同 ChatGPT Images 2.0。
呢個 6 部分唔係 OpenAI 官方命名嘅公式,而係把官方 prompting 建議整理成一個易用框架。目的係先鎖定最容易走樣嘅幾件事:用途、版面、畫面文字,以及需要保留或避免嘅限制。
| 部分 | 要寫清楚乜 | 短例子 |
|---|---|---|
| 用途 | 係 poster、banner、slide、social ad、產品相,定係 infographic | AI workshop 直度 poster |
| 主體 | 產品、人物、場景、物件或畫面中心動作 | 一支 serum 放喺雲石檯面上 |
| 構圖 | 畫面比例、鏡頭角度、framing、主體位置、留白位置 | 16:9,產品喺右邊,左邊留白放 copy |
| 風格 | 真實程度、材質、燈光、色調、mood | Studio product photo,柔光,高級 beige palette |
| 畫面文字 | 要顯示嘅字句原文、字體感覺、顏色、大小同位置 | Text: ‘SUMMER SALE’,粗身 sans-serif,白色,top-left |
| 限制 | 唔可以加乜,或者乜嘢一定要保持不變 | No extra text, no watermark, no unrelated logos, preserve product geometry |
OpenAI 嘅 image prompting 指引入面,對需要嚴格控制設計或包裝嘅情境,會用到 no trademarksno watermarksno logosinclude only this text verbatimdo not redesign the packaging
當 prompt 開始有幾個條件,唔好全部塞入一句長句。用標籤分段會更易管理,亦符合 OpenAI 對複雜 prompt 嘅建議。
Create a [poster/banner/slide/product image] for [purpose].
Subject: [main subject].
Composition: [aspect ratio, placement, negative space, camera angle].
Style: [visual style, lighting, color palette, realism level].
Text: ‘[exact text]’, [font/style/size/color/placement].
Constraints: [no extra text / no watermark / no unrelated logos / preserve layout / preserve product geometry].以下模板可以直接改;中括號位置換成你自己嘅活動、產品或 copy。重點唔係字數長,而係每一項都夠清楚。
Create a vertical 4:5 poster for [event or campaign].
Purpose: promote [goal].
Subject: [describe the main visual].
Composition: place the subject in the center, with clean empty space at the top for the headline.
Style: [editorial / cinematic / minimal], [describe lighting], color palette [main colors].
Text: only show ‘[HEADLINE]’, bold sans-serif font, white, placed at the top of the poster.
Constraints: no extra text, no watermark, no unrelated logos.Create a horizontal 3:1 website banner about [topic].
Subject: place the main subject on the right; keep a clean empty area on the left for copy.
Style: modern, premium, minimal details, color palette [palette], soft lighting.
Text: only show ‘[TEXT]’, placed on the left, high contrast and easy to read.
Constraints: no extra text, no watermark, no unrelated logos.Create a 16:9 slide background for a presentation about [topic].
Style: clean, professional, minimal details, enough contrast for readable text.
Composition: leave a large empty area on the left for a title and 3 bullet points; place the main visual on the right.
If there is text in the image, use only ‘[TEXT]’, with sharp text rendering.
Constraints: do not add icons, labels, logos or watermarks beyond the request.Create a photorealistic product image of [product name] on a [color or material] background.
Camera angle: 3/4 front view, soft studio lighting, realistic reflections, natural shadow.
Packaging: clear and sharp; the product label must be readable: ‘[TEXT]’.
Constraints: if a reference image is provided, preserve product geometry; do not redesign the packaging; do not add objects, watermarks, logos or unrelated trademarks.如果張圖需要 headline、產品 label、CTA 或 slide 內容,最好將文字變成獨立要求。OpenAI 建議寫明文字原文、位置同格式,並盡量保持文字簡短;如果係文字密集嘅 infographic 或複雜 layout,仍然應該仔細檢查,必要時再用設計工具微調。
Text: ‘LAUNCH WEEK’only include this textno extra text如果你 upload 產品圖、layout reference 或 style reference,prompt 一定要講清楚每張圖嘅角色。OpenAI 關於生成同編輯圖片嘅資料,都強調要說明邊部分要保留、邊部分要改,以及加上 constraints,減少不必要改動。
Image 1 is the original product. Image 2 is the style reference.
Keep the shape, proportions, label and composition of Image 1.
Apply only the lighting, color palette and mood from Image 2.
Do not redesign the packaging, add text, watermark or new logo.如果只想改一個細節,prompt 要窄同直接:
Change only the background color to warm beige.
Keep the product shape, label placement, lighting, shadows, camera angle and composition exactly the same.
No extra text, no watermark.對產品相或包裝圖嚟講,preserve product geometrydo not redesign the packagingkeep the label placement unchanged
如果結果已經有 70–80% 接近你想要,唔一定要成段 prompt 推倒重來。OpenAI 建議逐項改善結果;一次過改太多,反而容易令原本啱嘅構圖或風格走位。
make the lighting brighter and softersimplify the backgroundincrease negative spacekeep the same composition, make the style more modernonly include this textpreserve product geometrydo not redesign the packagingkeep label placement unchanged送出 prompt 之前,可以用呢幾條問題過一過:
總結一句:好用嘅 GPT Image 2 prompt 唔係最長嗰個,而係最似清楚 design brief 嗰個。你越清楚寫出用途、主體、構圖、畫面文字同限制,就越有機會得到接近原意、可落地使用嘅 poster、banner、slide 或產品圖。
Studio Global AI
Use this topic as a starting point for a fresh source-backed answer, then compare citations before you share it.
想 GPT Image 2 生成啱用嘅 poster、banner、slide 或產品相,prompt 最好當成 6 部分設計 brief:用途、主體、構圖、風格、文字同限制。
想 GPT Image 2 生成啱用嘅 poster、banner、slide 或產品相,prompt 最好當成 6 部分設計 brief:用途、主體、構圖、風格、文字同限制。 做 marketing asset 時,最容易影響成敗嘅位通常係比例、主體位置、留白、畫面文字要原文出現,以及 no extra text、no watermark、no unrelated logos 呢類 constraints。
用參考圖或編輯現有圖片時,要講清楚邊部分要保留、邊部分可以改;細修最好用 change only X 這類窄範圍指令,減少 layout、產品形狀或包裝走樣。 [1][3][5]