Describe the scene you want to create

1. Decide where the image will be used

A profile picture needs a subject that reads at a small size. A background may need empty space for text or icons. Write down the intended use and the most important element before generating. Start with a single scene instead of combining many characters, props and settings; it will be easier to decide what to change.

2. Describe subject, framing, setting and light

‘A beautiful picture’ does not say what should stand out. This example separates the subject from its placement and surroundings. Choose an available art style in the service.

A small yellow robot stands on a wooden desk. Its entire body is centered in the image, with space on both sides. A plain pale blue wall fills the background, and soft daylight enters from a window on the left.

An original example prompt, not a demonstrated generation result or a promise of a specific output.

3. Change the part that needs improvement

If the robot is too small, first change the framing to ‘a full-body view filling most of the center’. If the background is busy, simplify its description. Changing subject, style, lighting and composition together makes comparison difficult.

• Compare a full-body view with a close view of the upper body for different uses.
• Keep the subject and framing fixed; change soft daylight to warm evening light.
• Record the prompt, style and issue you want to improve. Identical inputs can still produce different results.

4. Compare styles with the same subject

Keep the subject and framing unchanged when comparing available anime, watercolor or crayon styles. Look at outlines, color transitions and readability at the size you will use. A style reference image is not a promise to reproduce a particular prompt.

5. Review before saving or publishing

• Zoom in on hands, small parts and lettering to find mistakes.
• Avoid personal or private information in prompts. Check relevant image, trademark and copyright permissions before publishing.
• Google members can currently generate up to 10 images a day, resetting at midnight in Asia/Seoul. Check the service for the current allowance.
• Results and history are available in the service. This guide does not collect or automatically publish private creations.