How to Craft Effective Gemini NanoBanana Pro Prompts for Stunning AI Images

Model Access

Use the Gemini NanoBanana Pro (Gemini 3 Image Pro) model, reachable at https://gemini.google.com/app or any platform that supports the NanoBanana Pro model.

Prompt Construction

Write a concise prompt that specifies the characters, visual style, and image constraints. Example prompt in Chinese:

让大雄和哆啦A梦为主人公（根据内容决定出现一个人物还是两个同时出现），以彩色漫画风格绘制一张图片来展现用户发送的内容，遵循帕累托法则仅提取核心关键词(3-5字),绝不大段文字，图片比例16:9

Alternatively, send a full description directly, e.g.:

我们之前学习某一个知识点，可能比较枯燥。我们只需要创建一个"通俗讲解专家 Skill"。那么它就会用生活化的例子，让我们快速理解 80%；然后用相对通俗的语言进行专业性的讲解；再给我们一些顺口溜或者对比等比较容易记忆的方法；最后画一张图，帮我们非常直观地理解这个知识。

Prompt Optimizer Skill

The prompt-optimizer skill matches the input against dozens of professional prompt frameworks and rewrites it automatically. The skill’s source code is hosted at https://github.com/chujianyun/skills.

Running the original prompt through the optimizer yields a structured prompt such as:

【图像生成提示词】
以可爱的卡通小女孩为主角（根据内容决定单人或多人场景），采用手绘插图风格绘制图片来展现用户发送的内容

【核心规则】
1. 遵循帕累托法则，绝不大段文字
2. 图片比例：16:9
3. 主色调：蓝色系（含浅蓝、天蓝、深蓝等层次）
4. 风格：手绘插图风格，保留笔触质感
5. 人物设定：可爱卡通小女孩形象

【画面构成】
- 主角：一个或多个卡通小女孩，造型简洁可爱
- 色彩：背景白色
- 线条：手绘线条，自然流畅，保留手工绘制质感
- 氛围：温馨、童真、富有想象力
- 文字：中文

【输出格式】
根据用户输入内容 → 提取3-5字核心关键词 → 生成16:9比例的手绘风格图片

If the generated image does not meet expectations, the optimizer can be invoked again or the user can manually tweak parameters such as color palette or character details.

Watermark Removal

Generated images may contain a watermark in the lower‑right corner. Two remediation methods are provided:

Upload the image to the online watermark removal service at https://banana.ovo.re/.

Install the userscript https://banana.ovo.re/userscript/gemini-watermark-remover.user.js (compatible with Tampermonkey/Violentmonkey) which automatically strips the watermark during generation.

Style Consistency Across Multiple Images

To keep visual style consistent:

Make the prompt as detailed as possible, enumerating color, composition, and line‑style constraints.

Continue the conversation within the same chat session rather than starting a new one, allowing the model to retain the previously established style context.

Optionally provide a previously generated satisfactory image as a reference for subsequent generations.

Full Workflow

Select the Gemini NanoBanana Pro model (https://gemini.google.com/app).

Craft a concise yet detailed prompt describing characters, style, aspect ratio, and any domain‑specific rules.

Optionally run the prompt through the prompt-optimizer skill (https://github.com/chujianyun/skills) to obtain a structured prompt.

Submit the prompt to the model and receive the generated image.

If a watermark appears, apply one of the removal methods described above.

Manually adjust style elements (color, character pose, line weight) if fine‑tuning is needed.

For subsequent images, reuse the detailed prompt and stay within the same chat session or supply a reference image.

Following these steps yields high‑quality, stylistically consistent illustrations without manual post‑processing.