How to write prompts to generate images

Image generators do not chat like a chatbot: they turn a visual description into an image. The more concrete the description of what is seen, the better the result.

The basic structure

Order the prompt from most to least important:

1. Image type: photograph, illustration, 3D render, watercolor, logo…
2. Subject: who or what appears, with details (age, clothing, material, color).
3. Action or pose: what it is doing.
4. Setting: where it is, era, weather.
5. Lighting: golden-hour natural light, soft studio light, neon…
6. Camera and composition: wide shot, close-up, 85mm lens, top-down view, rule of thirds.
7. Style and color: palette, aesthetic, art movement references (not living artists).

Example

Professional food photograph of a bowl of ramen with a soft-boiled egg,
on a dark wooden table, steam rising, soft window light from the left,
close-up from a 45-degree angle, shallow depth of field,
warm and moody color palette, editorial magazine style

Differences between tools

• Midjourney accepts parameters at the end of the prompt, such as --ar 16:9 for aspect ratio. These parameters do not work in other tools.
• DALL-E (in ChatGPT) understands natural-language descriptions well and you can request changes conversationally.
• Stable Diffusion often also uses a negative prompt to state what you do not want (e.g. “blurry, extra fingers, watermark”).
• Flux and Ideogram stand out for following detailed descriptions and, in Ideogram's case, rendering text inside images.
• Adobe Firefly is integrated into Adobe tools and lets you pick styles in the interface.

Features change often: check each tool's official documentation.

Practical tips

• Describe what you want to see, not what you do not (“a clear blue sky” rather than “no clouds”), except in negative prompts.
• If in-image text matters, put it in quotes and keep it short.
• Generate several versions and keep the best; then refine by changing one thing at a time.
• Use reference images where supported to keep the same character or product.

Responsible use

Do not generate images of real people without their consent, or images imitating a living artist's style to compete with them. Check each tool's commercial-use terms.

Explore the image generation prompts and the Midjourney prompts.