How to get better images from Gemini: Google's six prompt tips

If you use Google's AI assistant Gemini to create or edit images, the way you phrase your request can make a big difference. That is the message of a guide Google published on August 26, 2025. The author is Naina Raisinghani, a product manager at Google DeepMind.

A prompt is the text you write to tell the AI what to do.

Six elements of a good prompt

According to Google's guide, a good prompt should include six things: subject, composition, action, location, style and editing instructions. In other words: Who or what is in the picture, how is it arranged, what is happening, where does it take place, and what style should it have?

When you want to change an image that already exists, the instructions should be direct and precise. Google's examples include changing the color of a sofa to deep navy blue or changing a man's tie to green.

Five techniques

Google's guide shows five techniques. Chrome Unboxed, which has covered Google's tips and example prompts, also mentions them:

- Consistent characters across multiple images. - Step-by-step editing, using follow-up prompts. - Blending concepts, for example an astronaut and a jungle. - Style transfer, for example making the image look like an architect's drawing. - Logical reasoning, for example showing what happens if a person holding a cake falls.

The model has weaknesses

Google itself acknowledges that the model is not perfect. The weaknesses concern consistent stylization, correct rendering of text and preserving the image aspect ratio. Google says it is working to improve them.

So don't expect text in images to always come out right, or the format to always stay unchanged.

The guide dates from August 2025, and newer image models may have arrived in Gemini since then.