1. Choose the right tool for your goal

Different image models have different strengths. Without going deep into benchmarks, you can keep a simple mental model:

  • DALL·E – strong for concept art, illustrations and images that are easy to edit later.
  • Midjourney – strong for stylised, artistic and visually striking results.

2. Build prompts in layers

Instead of one vague sentence, think of your prompt as four parts:

  • Subject: what is in the image (e.g. “a small spaceship above the ocean”).
  • Style: photography, 3D render, anime, oil painting, etc.
  • Composition: close-up, wide shot, top view, centred subject, etc.
  • Mood & lighting: soft light, dramatic, cinematic, warm sunset, neon, etc.

Example prompt

“A small spaceship above a calm ocean at sunset, wide shot, cinematic lighting, soft clouds, detailed 3D render, ultra high resolution.”

3. Iterate instead of starting from zero each time

Most tools let you generate variations based on an image you liked. Use this to move in small steps:

  • First: get the basic idea right (subject & composition).
  • Then: adjust style and mood (more realistic, more stylised, etc.).
  • Finally: refine details like colours, background elements and text.

4. Be explicit about what you do not want

Negative prompts can help avoid common issues. For example: “no text”, “no watermark”, “no extra hands”, “no busy background”.

Tip: Keep a small text file or note with prompts that worked well for you. Reuse and adapt them instead of inventing from scratch every time.