Why GPT-Image-2 Outshines Nano Banana in Every Way

The article reviews the full release of GPT-Image-2, showcases dozens of Chinese prompt examples that generate travel guides, recipe flowcharts, scientific infographics, portrait photography, and Chinese‑style posters, and distills five practical prompt‑engineering rules while linking to a popular GitHub prompt repository.

Su San Talks Tech
Su San Talks Tech
Su San Talks Tech
Why GPT-Image-2 Outshines Nano Banana in Every Way

GPT-Image-2 Full Release

GPT-Image-2 became available to all ChatGPT users, removing the previous Plus/Pro restriction. The author observed a generation time of about three minutes for a single image, with visual fidelity comparable to professional designers.

One‑Sentence Prompt Cases

Travel‑Guide Poster

生成【城市】三天旅游攻略

Replacing 【城市】 with a city name (e.g., "生成成都三天旅游攻略") yields a vertical timeline poster containing itinerary, attractions, and food recommendations. Layout, color scheme, and Chinese typography remain consistent across tested cities (Hangzhou, Chongqing, Xi’an).

Recipe Flowchart

帮我制作辣椒炒肉这道菜的详细制作流程图,真实风格,适用于小红书图文比例

The output is a step‑by‑step cooking diagram sized for Xiaohongshu vertical posts, with each step illustrated and captioned. The author notes the clarity exceeds many manually created graphics.

Universal Template Prompts

Scientific Encyclopedia Card

请根据【主题】生成一张高质量竖版「科普百科图」。
这张图不是普通海报,也不是单纯插画,而是一张兼具"图鉴感、百科感、信息结构感、收藏感"的模块化科普信息图。...

Changing 【主题】 (e.g., "柴犬", "咖啡豆", "维生素 C") produces a modular infographic with a central visual, detailed sub‑features, rounded information blocks, hierarchical titles, and concise encyclopedia content. The negative phrasing ("不是普通海报,而是科普百科图") guides the model away from generic templates.

Photography‑Level Portraits

Japanese‑Film‑Style Portrait

Analog 35mm film photography, soft airy Japanese-style aesthetic, gentle diffused natural window light, slight overexposure, pastel tones, low contrast, soft highlights, ... --ar 9:16

The result exhibits film‑grain texture, dreamy pastel tones, and realistic lighting that closely matches a real 35 mm photograph.

JSON‑Structured Prompt

{
  "prompt": {
    "style_and_tech": "mobile phone photo, old CCD camera aesthetic, harsh flash, grainy, dim messy indoor lighting, candid snapshot feeling, slight motion blur",
    "subject": "young Korean female idol, soft innocent look",
    "pose": "mid-action, slightly turning head toward camera as if just noticed being photographed, shoulders slightly raised",
    "expression": "eyes widened slightly, lips parted in surprise, shy and caught‑off‑guard expression",
    "clothing": "loose soft homewear (thin cardigan + inner top), slightly slipping off one shoulder but not revealing",
    "vibe": "unprepared, intimate, accidental moment, evokes curiosity and protectiveness",
    "aspect ratio": "9:16"
  }
}

Explicitly separating dimensions makes later adjustments (e.g., swapping style_and_tech or expression) straightforward.

Chinese‑Style Design

New‑Chinese Ink Poster

新中式水墨山水海报,竖版9:16构图,东方极简美学风格,...(省略详细艺术描述)

The prompt specifies ink‑wash mountains, mist‑filled lake, a red‑clad fisherman, and a calligraphic title. The generated 8K poster matches professional studio output, demonstrating the model’s ability to interpret specialized art terminology such as "水墨晕染" and "飞白".

Creative “Brain‑Storm” Ideas

Historical Event as Social‑Media Screenshot

玄武门之变的朋友圈

The model produces a mock WeChat Moments screenshot with authentic UI elements, avatars, timestamps, and Chinese text, illustrating automatic visual translation of a historical narrative into a modern platform.

Ancient Programming Performance

在计算机博物馆里,一个程序员在展厅中央,正在演示C语言编程,很多参观者在围观,屏幕上的代码清晰可见。...(省略细节)

The output shows a 2D cartoon scene of a programmer demonstrating C code in a museum, with clear on‑screen code and a caption "古法编程,现场表演".

Prompt‑Writing Rules Derived from the Cases

Chinese prompts are highly effective. Simple Chinese commands generate dense visual information without English translation.

Match prompt length to scene complexity. Short phrases (e.g., seven characters) suffice for simple UI mockups; detailed portraits require several hundred words describing lighting, composition, and attire.

Prefer domain‑specific terminology. Terms such as "35mm film", "diffused light", "水墨晕染" improve fidelity compared to generic descriptions.

Structure prompts. JSON, sectioned descriptions, or per‑grid specifications help the model parse long instructions.

Use negation to steer output. Phrases like "不是普通海报,而是科普百科图" prevent the model from defaulting to common templates.

Technical Resources

GitHub repository containing over 70 verified prompt examples: https://github.com/EvoLinkAI/awesome-gpt-image-2-prompts.

Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

Prompt engineeringGitHubAI image generationcreative AIGPT Image 2Chinese prompts
Su San Talks Tech
Written by

Su San Talks Tech

Su San, former staff at several leading tech companies, is a top creator on Juejin and a premium creator on CSDN, and runs the free coding practice site www.susan.net.cn.

0 followers
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.