AI Image Prompts · 绘画提示词库

A detailed image generation prompt, labeled 'Banana Prompt,' designed to create a close-up, high-angle, vertical composition of a traditional Chinese garden courtyard. The style is specified as a miniature effect, clay render, monochromatic pure white, with embossed paper texture and strong ambient occlusion, suitable for architectural visualization.

A vertical composition close-up shot, high-angle view looking down into a corner of a traditional Chinese garden courtyard. A pavilion connected by a small bridge, surrounded by dense trees. Three-quarter perspective, miniature effect, clay render style, monochromatic pure white, embossed paper texture, strong ambient occlusion, soft studio lighting, rich architectural details. --ar 2:3

A simple instruction indicating that the user needs to upload an image to proceed with the Gemini App's Nano Banana Pro feature.

Upload your image 💫

A simple, descriptive prompt used to generate an image featuring a large yellow feathered dragon and a woman in high-fashion yellow armor on a foggy forest floor, demonstrating the initial image creation capability of Nano Banana Pro.

"green forest floor, ferns, dew, foggy, day, a large yellow feathered dragon, a woman in high fashion yellow armor"

Generates an Abstract Expressionist painting interpretation of a boxing match, emphasizing dynamic gestures, thick impasto paint, and chaotic energy using violent splatters of red, white, and black.

An Abstract Expressionist interpretation of the boxing match between the two intense figures, rendered in dynamic, sweeping gestures and thick impasto paint. Violent splatters of deep red, stark white, and urgent black dominate the canvas, capturing the raw, chaotic energy of the conflict. The figures are distorted into forms of pure kinetic stress and emotional fury, emphasizing texture and rapid movement over realistic depiction.

A tip for increasing the probability of image output: explicitly instructing the model to 'absolutely output in Nano banana Pro image, not text'.

Do not output in text, absolutely output as a Nano banana Pro image

A simple Japanese prompt used with Nano Banana to convert a 2D illustration into a 3D style, specifically requesting a rendering style similar to Blender.

Please generate a 3D style image, referencing this image, rendered like it was done in Blender.

This tweet discusses the concept of adding imperfections (like flour stains) to AI-generated faces to increase texture and realism. It contrasts the performance of Seedream4_5 (better with Chinese prompts) and Nano Banana (better with English prompts) and includes a prompt fragment focused on adding facial flaws.

Add some flaws, facial blemishes, (flour stains on the nose and cheeks). The foreign body sensation on the face will make the skin look more textured.

This tweet describes two distinct styles generated using Nano Banana Pro: an 'Artist Style' self-portrait in an oil painting atelier and an 'Electric Neon Style' integrated into a night city. The user generated two versions of a 'scissors man' (ハサミおじさん) in each style, highlighting the model's wide range of expression.

The 'Artist Style' self-portrait painted in an oil painting atelier, and the 'Electric Neon Style' melting into the night city. I generated two patterns each of the extremely contrasting 'scissors man'.

A simple, high-level prompt instructing the AI to transform any uploaded person's image into a professional sketch style.

Turn any person into a pro sketch

A simple image editing prompt used across multiple AI tools (Qwen, Grok, Nano Banana Pro) to transform an existing image of a girl. The instruction is to modify the scene so that the girl is taking an exam at a magic school. This demonstrates the use of Nano Banana Pro for image-to-image editing based on a textual instruction.

Modify the picture so that this girl is taking an exam at a magic school.

This is an instruction set for using a custom GPT designed to analyze an uploaded image and generate a detailed JSON prompt optimized for Nano Banana Pro. This workflow is useful for reverse-engineering or transferring styles from existing images into the Nano Banana Pro environment.

Analyze this image in exhaustive JSON detail

This is not a prompt itself, but a meta-prompting strategy designed to prevent image generation failures due to triggering Google Nano Banana Pro's safety filters (Safety Filter/Risk Control). It instructs the model to automatically filter dangerous words, use safe alternatives, and ensures the generated prompt meets platform requirements. It also lists explicitly forbidden categories and mandatory safe replacement words that must be included in every prompt.

The application is now capable of automatically filtering all dangerous vocabulary, using safe alternative words, and ensuring that the generated prompt complies with platform risk control requirements.

Absolutely forbidden vocabulary:

Category Disabled Words
Energy/Special Effects: lightning, electric, energy, spark, voltage, current
Combat/Conflict: attack, battle, fight, punch, kick, strike, combat, clash
Hazardous Elements: fire, flame, explosion, weapon, sword, knife, gun
Negative Emotions: fierce, intense, aggressive, violent, rage, angry, scream, panic
Body Horror: glowing eyes, red eyes, blood, wound, transform, distort
Mandatory Safe Alternative Words
Original Word / Alternative Word
lightning/electric: soft light, warm light, ambient light
fight/attack: move toward, approach, interaction
fierce/intense: warm, calm, gentle, peaceful
explosion/fire: bloom, brighten, illuminate, radiate
angry/rage: concerned, surprised, amazed, excited
Every prompt must contain at least 2 safe words:
gentle, soft, calm, peaceful, warm, bright, smooth, quiet, serene, beautiful, lovely, sweet, slowly, smoothly, gracefully